Datasets:

jamescalam

/

llama-2-arxiv-papers-chunked

like 17

1504.06201

37

detection and hierarchical image segmentation. IEEE Trans. Pattern Anal. Mach. Intell. , 33(5):898–916, May 2011. 2 [2] P. Arbelaez, J. Pont-Tuset, J. Barron, F. Marqués, and J. Malik. Multiscale combinatorial grouping. In Computer Vision and Pattern Recognition (CVPR) , 2014. 2, 4 [3] G. Bertasius, J. Shi, and L. Torresani. Deepedge: A multiscale bifurcated deep network for top-down contour detection. In The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , June 2015. 2, 3, 4, 5 [4] C. J. C. Burges. A tutorial on support vector machines for pattern recognition. Data Mining and Knowledge Discovery , 2:121–167, 1998. 2 [5] L. Chen, G. Papandreou, I. Kokkinos, K. Murphy, and A. L. Yuille. Semantic image segmentation with deep convolutional nets and fully connected crfs. CoRR , abs/1412.7062, 2014. 1, 5, 6, 7

1504.06201

High-for-Low and Low-for-High: Efficient Boundary Detection from Deep Object Features and its Applications to High-Level Vision

Most of the current boundary detection systems rely exclusively on low-level features, such as color and texture. However, perception studies suggest that humans employ object-level reasoning when judging if a particular pixel is a boundary. Inspired by this observation, in this work we show how to predict boundaries by exploiting object-level features from a pretrained object-classification network. Our method can be viewed as a "High-for-Low" approach where high-level object features inform the low-level boundary detection process. Our model achieves state-of-the-art performance on an established boundary detection benchmark and it is efficient to run. Additionally, we show that due to the semantic nature of our boundaries we can use them to aid a number of high-level vision tasks. We demonstrate that using our boundaries we improve the performance of state-of-the-art methods on the problems of semantic boundary labeling, semantic segmentation and object proposal generation. We can view this process as a "Low-for-High" scheme, where low-level boundaries aid high-level vision tasks. Thus, our contributions include a boundary detection system that is accurate, efficient, generalizes well to multiple datasets, and is also shown to improve existing state-of-the-art high-level vision methods on three distinct tasks.

http://arxiv.org/pdf/1504.06201

cs.CV

20150423

20150921

1504.06201

38

2014. 1, 5, 6, 7 [6] P. Dollár and C. L. Zitnick. Fast edge detection using structured forests. PAMI , 2015. 2, 3, 4, 5, 8 [7] J. Donahue, Y . Jia, O. Vinyals, J. Hoffman, N. Zhang, E. Tzeng, and T. Darrell. Decaf: A deep convolutional activation feature for generic visual recognition. CoRR , abs/1310.1531, 2013. 5 [8] M. Everingham, L. Van Gool, C. K. I. Williams, J. Winn, and A. Zisserman. The PASCAL Visual Object Classes Challenge 2007 (VOC2007) Results. http://www.pascalnetwork.org/challenges/VOC/voc2007/workshop/index.html. 7 [9] M. Everingham, L. Van Gool, C. K. I. Williams, J. Winn, and A. Zisserman. The PASCAL Visual Object Classes Challenge 2012 (VOC2012) Results. http://www.pascalnetwork.org/challenges/VOC/voc2012/workshop/index.html. 8

1504.06201

High-for-Low and Low-for-High: Efficient Boundary Detection from Deep Object Features and its Applications to High-Level Vision

Most of the current boundary detection systems rely exclusively on low-level features, such as color and texture. However, perception studies suggest that humans employ object-level reasoning when judging if a particular pixel is a boundary. Inspired by this observation, in this work we show how to predict boundaries by exploiting object-level features from a pretrained object-classification network. Our method can be viewed as a "High-for-Low" approach where high-level object features inform the low-level boundary detection process. Our model achieves state-of-the-art performance on an established boundary detection benchmark and it is efficient to run. Additionally, we show that due to the semantic nature of our boundaries we can use them to aid a number of high-level vision tasks. We demonstrate that using our boundaries we improve the performance of state-of-the-art methods on the problems of semantic boundary labeling, semantic segmentation and object proposal generation. We can view this process as a "Low-for-High" scheme, where low-level boundaries aid high-level vision tasks. Thus, our contributions include a boundary detection system that is accurate, efficient, generalizes well to multiple datasets, and is also shown to improve existing state-of-the-art high-level vision methods on three distinct tasks.

http://arxiv.org/pdf/1504.06201

cs.CV

20150423

20150921

1504.06201

39

Challenge 2012 (VOC2012) Results. http://www.pascalnetwork.org/challenges/VOC/voc2012/workshop/index.html. 8 [10] Y . Ganin and V . S. Lempitsky. N4-fields: Neural network nearest neighbor fields for image transforms. ACCV , 2014. 2, 3, 4, 5 [11] B. Hariharan, P. Arbelaez, L. Bourdev, S. Maji, and J. Malik. Semantic contours from inverse detectors. In International Conference on Computer Vision (ICCV) , 2011. 1, 5, 6, 7 [12] B. Hariharan, P. A. Arbeláez, R. B. Girshick, and J. Malik. Hypercolumns for object segmentation and fine-grained localization. CoRR , abs/1411.5752, 2014. 3 [13] P.-J. Hsieh, E. Vul, and N. Kanwisher. Recognition alters the spatial pattern of fmri activation in early retinotopic cortex. Journal of Neurophysiology , 103(3):1501–1507, 2010. 1, 8

1504.06201

High-for-Low and Low-for-High: Efficient Boundary Detection from Deep Object Features and its Applications to High-Level Vision

Most of the current boundary detection systems rely exclusively on low-level features, such as color and texture. However, perception studies suggest that humans employ object-level reasoning when judging if a particular pixel is a boundary. Inspired by this observation, in this work we show how to predict boundaries by exploiting object-level features from a pretrained object-classification network. Our method can be viewed as a "High-for-Low" approach where high-level object features inform the low-level boundary detection process. Our model achieves state-of-the-art performance on an established boundary detection benchmark and it is efficient to run. Additionally, we show that due to the semantic nature of our boundaries we can use them to aid a number of high-level vision tasks. We demonstrate that using our boundaries we improve the performance of state-of-the-art methods on the problems of semantic boundary labeling, semantic segmentation and object proposal generation. We can view this process as a "Low-for-High" scheme, where low-level boundaries aid high-level vision tasks. Thus, our contributions include a boundary detection system that is accurate, efficient, generalizes well to multiple datasets, and is also shown to improve existing state-of-the-art high-level vision methods on three distinct tasks.

http://arxiv.org/pdf/1504.06201

cs.CV

20150423

20150921

1504.06201

40

spatial pattern of fmri activation in early retinotopic cortex. Journal of Neurophysiology , 103(3):1501–1507, 2010. 1, 8 [14] P. Isola, D. Zoran, D. Krishnan, and E. H. Adelson. Crisp boundary detection using pointwise mutual information. In ECCV , 2014. 2, 4[15] Y . Jia, E. Shelhamer, J. Donahue, S. Karayev, J. Long, R. Girshick, S. Guadarrama, and T. Darrell. Caffe: Convolutional architecture for fast feature embedding. arXiv preprint arXiv:1408.5093 , 2014. 4 [16] J. J. Kivinen, C. K. Williams, N. Heess, and D. Technologies. Visual boundary prediction: A deep neural prediction network and quality dissection. AISTATS , 1(2):9, 2014. 2, 3, 4, 5 [17] Z. Kourtzi and N. Kanwisher. Representation of perceived object shape by the human lateral occipital complex. Science, 293:1506–1509, 2001. 1, 8

1504.06201

High-for-Low and Low-for-High: Efficient Boundary Detection from Deep Object Features and its Applications to High-Level Vision

Most of the current boundary detection systems rely exclusively on low-level features, such as color and texture. However, perception studies suggest that humans employ object-level reasoning when judging if a particular pixel is a boundary. Inspired by this observation, in this work we show how to predict boundaries by exploiting object-level features from a pretrained object-classification network. Our method can be viewed as a "High-for-Low" approach where high-level object features inform the low-level boundary detection process. Our model achieves state-of-the-art performance on an established boundary detection benchmark and it is efficient to run. Additionally, we show that due to the semantic nature of our boundaries we can use them to aid a number of high-level vision tasks. We demonstrate that using our boundaries we improve the performance of state-of-the-art methods on the problems of semantic boundary labeling, semantic segmentation and object proposal generation. We can view this process as a "Low-for-High" scheme, where low-level boundaries aid high-level vision tasks. Thus, our contributions include a boundary detection system that is accurate, efficient, generalizes well to multiple datasets, and is also shown to improve existing state-of-the-art high-level vision methods on three distinct tasks.

http://arxiv.org/pdf/1504.06201

cs.CV

20150423

20150921

1504.06201

41

object shape by the human lateral occipital complex. Science, 293:1506–1509, 2001. 1, 8 [18] J. Lim, C. L. Zitnick, and P. Dollár. Sketch tokens: A learned mid-level representation for contour and object detection. In CVPR , 2013. 2 [19] J. Long, E. Shelhamer, and T. Darrell. Fully convolutional networks for semantic segmentation. CVPR (to appear) , Nov. 2015. 3, 5 [20] M. Maire, S. X. Yu, and P. Perona. Reconstructive sparse code transfer for contour detection and semantic labeling. In Asian Conference on Computer Vision (ACCV) , 2014. 2 [21] T. Malisiewicz, A. Gupta, and A. A. Efros. Ensemble of exemplar-svms for object detection and beyond. In ICCV , 2011. 4 [22] D. Martin, C. Fowlkes, D. Tal, and J. Malik. A database of human segmented natural images and its application to evaluating segmentation algorithms and measuring ecological statistics. In Proc. 8th Int’l Conf. Computer Vision , volume 2, pages 416–423, July 2001. 4

1504.06201

High-for-Low and Low-for-High: Efficient Boundary Detection from Deep Object Features and its Applications to High-Level Vision

Most of the current boundary detection systems rely exclusively on low-level features, such as color and texture. However, perception studies suggest that humans employ object-level reasoning when judging if a particular pixel is a boundary. Inspired by this observation, in this work we show how to predict boundaries by exploiting object-level features from a pretrained object-classification network. Our method can be viewed as a "High-for-Low" approach where high-level object features inform the low-level boundary detection process. Our model achieves state-of-the-art performance on an established boundary detection benchmark and it is efficient to run. Additionally, we show that due to the semantic nature of our boundaries we can use them to aid a number of high-level vision tasks. We demonstrate that using our boundaries we improve the performance of state-of-the-art methods on the problems of semantic boundary labeling, semantic segmentation and object proposal generation. We can view this process as a "Low-for-High" scheme, where low-level boundaries aid high-level vision tasks. Thus, our contributions include a boundary detection system that is accurate, efficient, generalizes well to multiple datasets, and is also shown to improve existing state-of-the-art high-level vision methods on three distinct tasks.

http://arxiv.org/pdf/1504.06201

cs.CV

20150423

20150921

1504.06201

42

evaluating segmentation algorithms and measuring ecological statistics. In Proc. 8th Int’l Conf. Computer Vision , volume 2, pages 416–423, July 2001. 4 [23] X. Ren and L. Bo. Discriminatively Trained Sparse Code Gradients for Contour Detection. In Advances in Neural Information Processing Systems , December 2012. 2, 4 [24] O. Russakovsky, J. Deng, H. Su, J. Krause, S. Satheesh, S. Ma, Z. Huang, A. Karpathy, A. Khosla, M. Bernstein, A. C. Berg, and L. Fei-Fei. ImageNet Large Scale Visual Recognition Challenge, 2014. 3 [25] J. L. Sanguinetti, J. J. Allen, and M. A. Peterson. The ground side of an object perceived as shapeless yet processed for semantics. Psychological science , page 0956797613502814, 2013. 1, 8 [26] P. Sermanet, D. Eigen, X. Zhang, M. Mathieu, R. Fergus, and Y . LeCun. Overfeat: Integrated recognition, localization and detection using convolutional networks. CoRR , abs/1312.6229, 2013. 3

1504.06201

High-for-Low and Low-for-High: Efficient Boundary Detection from Deep Object Features and its Applications to High-Level Vision

Most of the current boundary detection systems rely exclusively on low-level features, such as color and texture. However, perception studies suggest that humans employ object-level reasoning when judging if a particular pixel is a boundary. Inspired by this observation, in this work we show how to predict boundaries by exploiting object-level features from a pretrained object-classification network. Our method can be viewed as a "High-for-Low" approach where high-level object features inform the low-level boundary detection process. Our model achieves state-of-the-art performance on an established boundary detection benchmark and it is efficient to run. Additionally, we show that due to the semantic nature of our boundaries we can use them to aid a number of high-level vision tasks. We demonstrate that using our boundaries we improve the performance of state-of-the-art methods on the problems of semantic boundary labeling, semantic segmentation and object proposal generation. We can view this process as a "Low-for-High" scheme, where low-level boundaries aid high-level vision tasks. Thus, our contributions include a boundary detection system that is accurate, efficient, generalizes well to multiple datasets, and is also shown to improve existing state-of-the-art high-level vision methods on three distinct tasks.

http://arxiv.org/pdf/1504.06201

cs.CV

20150423

20150921

1504.06201

43

and Y . LeCun. Overfeat: Integrated recognition, localization and detection using convolutional networks. CoRR , abs/1312.6229, 2013. 3 [27] W. Shen, X. Wang, Y . Wang, X. Bai, and Z. Zhang. Deepcontour: A deep convolutional feature learned by positivesharing loss for contour detection. June 2015. 1, 3, 4, 5 [28] J. Shi and J. Malik. Normalized cuts and image segmentation. IEEE Transactions on Pattern Analysis and Machine Intelligence , 22:888–905, 1997. 2, 6 [29] K. Simonyan and A. Zisserman. Very deep convolutional networks for large-scale image recognition. CoRR , abs/1409.1556, 2014. 1 [30] A. Sironi, V . Lepetit, and P. Fua. Multiscale centerline detection by learning a scale-space distance transform. June 2014. 4 [31] C. L. Zitnick and P. Dollár. Edge boxes: Locating object proposals from edges. In ECCV , 2014. 1, 8

1504.06201

High-for-Low and Low-for-High: Efficient Boundary Detection from Deep Object Features and its Applications to High-Level Vision

Most of the current boundary detection systems rely exclusively on low-level features, such as color and texture. However, perception studies suggest that humans employ object-level reasoning when judging if a particular pixel is a boundary. Inspired by this observation, in this work we show how to predict boundaries by exploiting object-level features from a pretrained object-classification network. Our method can be viewed as a "High-for-Low" approach where high-level object features inform the low-level boundary detection process. Our model achieves state-of-the-art performance on an established boundary detection benchmark and it is efficient to run. Additionally, we show that due to the semantic nature of our boundaries we can use them to aid a number of high-level vision tasks. We demonstrate that using our boundaries we improve the performance of state-of-the-art methods on the problems of semantic boundary labeling, semantic segmentation and object proposal generation. We can view this process as a "Low-for-High" scheme, where low-level boundaries aid high-level vision tasks. Thus, our contributions include a boundary detection system that is accurate, efficient, generalizes well to multiple datasets, and is also shown to improve existing state-of-the-art high-level vision methods on three distinct tasks.

http://arxiv.org/pdf/1504.06201

cs.CV

20150423

20150921

1511.02674

0

Semantic Segmentation with Boundary Neural Fields Gedas Bertasius University of Pennsylvania gberta@seas.upenn.eduJianbo Shi University of Pennsylvania jshi@seas.upenn.eduLorenzo Torresani Dartmouth College lt@dartmouth.edu Abstract The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use colorbased pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

1

segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively. 1. Introduction The recent introduction of fully convolutional networks (FCNs) [22] has led to significant quantitative improvements on the task of semantic segmentation. However, despite their empirical success, FCNs suffer from some limitations. Large receptive fields in the convolutional layers and the presence of pooling layers lead to blurring and segmentation predictions at a significantly lower resolution than the

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

2

the presence of pooling layers lead to blurring and segmentation predictions at a significantly lower resolution than the Figure 1: Examples illustrating shortcomings of prior semantic segmentation methods: the second column shows results obtained with a FCN [22], while the third column shows the output of a Dense-CRF applied to FCN predictions [19, 7]. Segments produced by FCN are blob-like and are poorly localized around object boundaries. Dense-CRF produces spatially disjoint object segments due to the use of a color-based pixel affinity function that is unable to measure semantic similarity between pixels. original image. As a result, their predicted segments tend to be blobby and lack fine object boundary details. We report in Fig. 1 some examples illustrating typical poor localization of objects in the outputs of FCNs. Recently, Chen at al. [7] addressed this issue by applying a Dense-CRF post-processing step [19] on top of coarse FCN segmentations. However, such an approach introduces several problems of its own. First, the Dense-CRF adds new parameters that are difficult to tune and integrate into the original network architecture. Additionally, most methods

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

3

parameters that are difficult to tune and integrate into the original network architecture. Additionally, most methods based on CRFs or MRFs use low-level pixel affinity functions, such as those based on color. These low-level affinities often fail to capture semantic relationships between objects and lead to poor segmentation results (see last column in Fig. 1). We propose to address these shortcomings by means of a Boundary Neural Field (BNF), an architecture that employs a single semantic segmentation FCN to predict semantic boundaries and then use them to produce semantic segmentation maps via a global optimization. We demonstrate that even though the semantic segmentation FCN has not been optimized to detect boundaries, it provides good 1arXiv:1511.02674v2 [cs.CV] 24 May 2016 ... Convolutional Feature Maps RGB Input Image Segmentation Mask in Softmax Layer Minimize Global Energy Boundaries |{z}... wn()++()w1 Xi✓i(zi)+Xij✓ij(zi,zj) Linear Combination of Feature MapsFigure 2: The architecture of our system (best viewed in color). We employ a semantic segmentation FCN [7] for two purposes: 1) to obtain semantic segmentation unaries

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

4

color). We employ a semantic segmentation FCN [7] for two purposes: 1) to obtain semantic segmentation unaries for our global energy; 2) to compute object boundaries. Specifically, we define semantic boundaries as a linear combination of these feature maps (with a sigmoid function applied on top of the sum) and learn individual weights corresponding to each convolutional feature map. We integrate this boundary information in the form of pairwise potentials (pixel affinities) for our energy model. features for boundary detection. Specifically, the contributions of our work are as follows: We show that semantic boundaries can be expressed as a linear combination of interpolated convolutional feature maps inside an FCN. We introduce a boundary detection method that exploits this intuition to predict object boundaries with accuracy superior to the statethe-of-art. We demonstrate that boundary-based pixel affinities are better suited for semantic segmentation than the commonly used color affinity functions. Finally, we introduce a new global energy that decomposes semantic segmentation into multiple binary problems and relaxes the integrality constraint. We show that minimizing our proposed energy yields better qualitative and quantitative results relative to traditional globalization models such as MRFs or CRFs.

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

5

problems and relaxes the integrality constraint. We show that minimizing our proposed energy yields better qualitative and quantitative results relative to traditional globalization models such as MRFs or CRFs. 2. Related Work Boundary Detection. Spectral methods comprise one of the most prominent categories for boundary detection. In a typical spectral framework, one formulates a generalized eigenvalue system to solve a low-level pixel groupingproblem. The resulting eigenvectors are then used to predict the boundaries. Some of the most notable approaches in this genre are MCG [2], gPb [1], PMI [17], and Normalized Cuts [29]. A weakness of spectral approaches is that they tend to be slow as they perform a global inference over the entire image. To address this issue, recent approaches cast boundary detection as a classification problem and predict the boundaries in a local manner with high efficiency. The most notable examples in this genre include sketch tokens (ST) [20] and structured edges (SE) [9], which employ fast random forests. However, many of these methods are based on hand-constructed features, which are difficult to tune.

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

6

forests. However, many of these methods are based on hand-constructed features, which are difficult to tune. The issue of hand-constructed features have been recently addressed by several approaches based on deep learning, such as N4fields [11], DeepNet [18], DeepContour [27], DeepEdge [3], H FL [4] and HED [33]. All of these methods use CNNs in some way to predict the boundaries. Whereas DeepNet and DeepContour optimize ordinary CNNs to a boundary based optimization criterion from scratch, DeepEdge and H FL employ pretrained models to compute boundaries. The most recent of these methods is HED [33], which shows the benefit of deeply supervised learning for boundary detection. In comparison to prior deep learning approaches, our method offers several contributions. First, we exploit the inherent relationship between boundary detection and semantic segmentation to predict semantic boundaries. Specifically, we show that even though the semantic FCN has not been explicitly trained to predict boundaries, the convolutional filters inside the FCN provide good features for boundary detection. Additionally, unlike DeepEdge [3] and HFL [4], our method does not require a pre-processing step to select candidate contour points, as we predict boundaries

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

7

HFL [4], our method does not require a pre-processing step to select candidate contour points, as we predict boundaries on all pixels in the image. We demonstrate that our approach allows us to achieve state-of-the-art boundary detection results according to both F-score and Average Precision metrics. Additionally, due to the semantic nature of our boundaries, we can successfully use them as pairwise potentials for semantic segmentation in order to improve object localization and recover fine structural details, typically lost by pure FCN-based approaches. Semantic Segmentation. We can group most semantic segmentation methods into three broad categories. The first category can be described as “two-stage” approaches, where an image is first segmented and then each segment is classified as belonging to a certain object class. Some of the most notable methods that belong to this genre include [24, 6, 12, 14]. The primary weakness of the above methods is that they are unable to recover from errors made by the segmentation algorithm. Several recent papers [15, 10] address this issue by proposing to use deep per-pixel CNN features and then classify each pixel as belonging to a certain class. While these approaches partially address the incorrect segmentation problem, they perform predictions independently on

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

8

classify each pixel as belonging to a certain class. While these approaches partially address the incorrect segmentation problem, they perform predictions independently on each pixel. This leads to extremely local predictions, where the relationships between pixels are not exploited in any way, and thus the resulting segmentations may be spatially disjoint. The third and final group of semantic segmentation methods can be viewed as front-to-end schemes where segmentation maps are predicted directly from raw pixels without any intermediate steps. One of the earliest examples of such methods is the FCN introduced in [22]. This approach gave rise to a number of subsequent related approaches which have improved various aspects of the original semantic segmentation [7, 34, 8, 16, 21]. There have also been attempts at integrating the CRF mechanism into the network architecture [7, 34]. Finally, it has been shown that semantic segmentation can also be improved using additional training data in the form of bounding boxes [8]. Our BNF offers several contributions over prior work. To the best of our knowledge, we are the first to present a model that exploits the relationship between boundary detection and semantic segmentation within a FCN framework . We introduce pairwise pixel affinities computed from semantic boundaries inside an FCN, and use these boundaries to

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

9

and semantic segmentation within a FCN framework . We introduce pairwise pixel affinities computed from semantic boundaries inside an FCN, and use these boundaries to predict the segmentations in a global fashion. Unlike [21], which requires a large number of additional parameters to learn for the pairwise potentials, our global model only needs5Kextra parameters, which is about 3orders of magnitudes less than the number of parameters in a typical deep convolutional network (e.g. VGG [30]). We empirically show that our proposed boundary-based affinities are better suited for semantic segmentation than color-based affinities. Additionally, unlike in [7, 34, 21], the solution to our proposed global energy can be obtained in closed-form, which makes global inference easier. Finally we demonstrate that our method produces better results than traditional globalization models such as CRFs or MRFs. 3. Boundary Neural Fields In this section, we describe Boundary Neural Fields. Similarly to traditional globalization methods, Boundary Neural Fields are defined by an energy including unary and pairwise potentials. Minimization of the global energy yields the semantic segmentation. BNFs build both unary and pairwise potentials from the input RGB image

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

10

and pairwise potentials. Minimization of the global energy yields the semantic segmentation. BNFs build both unary and pairwise potentials from the input RGB image and then combine them in a global manner. More precisely, the coarse segmentations predicted by a semantic FCN are used to define the unary potentials of our BNF. Next, we show that the convolutional feature maps of the FCN can be used to accurately predict semantic boundaries. These boundaries are then used to build pairwise pixel affinities, which are used as pairwise potentials by the BNF. Finally,we introduce a global energy function, which minimizes the energy corresponding to the unary and pairwise terms and improves the initial FCN segmentation. The detailed illustration of our architecture is presented in Figure 2. We now explain each of these steps in more detail. 3.1. FCN Unary Potentials To predict semantic unary potentials we employ the DeepLab model [7], which is a fully convolutional adaptation of the VGG network [30]. The FCN consists of 16convolutional layers and 3fully convolutional layers. There are more recent FCN-based methods that have demonstrated even better semantic segmentation results [8, 34, 16, 21].

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

11

more recent FCN-based methods that have demonstrated even better semantic segmentation results [8, 34, 16, 21]. Although these more advanced architectures could be integrated into our framework to improve our unary potentials, in this work we focus on two aspects orthogonal to this prior work: 1) demonstrating that our boundary-based affinity function is better suited for semantic segmentation than the common color-based affinities and 2) showing that our proposed global energy achieves better qualitative and quantitative semantic segmentation results in comparison to prior globalization models. 3.2. Boundary Pairwise Potentials In this section, we describe our approach for building pairwise pixel affinities using semantic boundaries. The basic idea behind our boundary detection approach is to express semantic boundaries as a function of convolutional feature maps inside the FCN. Due to the close relationship between the tasks of semantic segmentation and boundary detection, we hypothesize that convolutional feature maps from the semantic segmentation FCN can be employed as features for boundary detection. 3.2.1 Learning to Predict Semantic Boundaries. We propose to express semantic boundaries as a linear combination of interpolated FCN feature maps with a non-linear function applied on top of this sum. We note that interpolation of feature maps has been successfully used in prior

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

12

function applied on top of this sum. We note that interpolation of feature maps has been successfully used in prior work (see e.g. [15]) in order to obtain dense pixel-level features from the low-resolution outputs of deep convolutional layers. Here we adopt interpolation to produce pixel-level boundary predictions. There are several advantages to our proposed formulation. First, because we express boundaries as a linear combination of feature maps, we only need to learn a small number of parameters, corresponding to the individual weight values of each feature map in the FCN. This amounts to5Klearning parameters, which is much smaller than the number of parameters in the entire network (15M). In comparison, DeepEdge [3] and HFL [4] need 17M and 6M additional parameters to predict boundaries. Furthermore, expressing semantic boundaries as a linear combination of FCN feature maps allows us to efficiently predict boundary probabilities for all pixels in the image (we resize the FCN feature maps to the original image dimensions). This eliminates the need to select candidate boundary points in a pre-processing stage, which was instead required in prior boundary detection work [3, 4]. Our boundary prediction pipeline can be described as

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

13

boundary points in a pre-processing stage, which was instead required in prior boundary detection work [3, 4]. Our boundary prediction pipeline can be described as follows. First we use use SBD segmentations [13] to optimize our FCN for semantic segmentation task. We then treat FCN convolutional maps as features for the boundary detection task and use the boundary annotations from BSDS 500 dataset [23] to learn the weights for each feature map. BSDS 500 dataset contains 200training, 100validation, 200testing images, and ground truth annotations by 5 human labelers for each of these images. To learn the weights corresponding to each convolutional feature map we first sample 80Kpoints from the dataset. We define the target labels for each point as the fraction of human annotators agreeing on that point being a boundary. To fix the issue of label imbalance (there are many more non-boundaries than boundaries), we divide the label space into four quartiles, and select an equal number of samples for each quartile to balance the training dataset. Given these sampled points, we then define our features as the values in the interpolated convolutional feature maps corresponding to these points. To predict semantic boundaries we weigh each convolutional feature map by its weight, sum them up

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

14

the interpolated convolutional feature maps corresponding to these points. To predict semantic boundaries we weigh each convolutional feature map by its weight, sum them up and apply a sigmoid function on top of it. We obtain the weights corresponding to each convolutional feature map by minimizing the cross-entropy loss using a stochastic batch gradient descent for 50epochs. To obtain crisper boundaries at test-time we post-process the boundary probabilities using non-maximum suppression. To give some intuition on how FCN feature maps contribute to boundary detection, in Fig. 3 we visualize the feature maps corresponding to the highest weight magnitudes. It is clear that many of these maps contain highly localized boundary information. Boundary Detection Results Before discussing how boundary information is integrated in our energy for semantic segmentation, here we present experimental results assessing the accuracy of our boundary detection scheme. We tested our boundary detector on the BSDS500 dataset [23], which is the standard benchmark for boundary detection. The quality of the predicted boundaries is evaluated using three standard measures: fixed contour threshold (ODS), per-image best threshold (OIS), and average precision (AP). In Table 1 we show that our algorithm outperforms all prior methods according to both F-score measures and the

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

15

per-image best threshold (OIS), and average precision (AP). In Table 1 we show that our algorithm outperforms all prior methods according to both F-score measures and the Average Precision metric. In Fig. 4, we also visualize our predicted boundaries. The second column shows the pixellevel softmax output computed from the linear combination of feature maps, while the third column depicts our fiFigure 3: An input image and convolutional feature maps corresponding to the largest weight magnitude values. Intuitively these are the feature maps that contribute most heavily to the task of boundary detection. Method ODS OIS AP SCG [25] 0.739 0.758 0.773 SE [9] 0.746 0.767 0.803 MCG [2] 0.747 0.779 0.759 N4-fields [11] 0.753 0.769 0.784 DeepEdge [3] 0.753 0.772 0.807 DeepContour [27] 0.756 0.773 0.797 HFL [4] 0.767 0.788 0.795 HED [33] 0.782 0.804 0.833 BNF 0.788 0.807 0.851

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

16

HED [33] 0.782 0.804 0.833 BNF 0.788 0.807 0.851 Table 1: Boundary detection results on BSDS500 benchmark. Our proposed method outperforms all prior algorithms according to all three evaluation metrics. nal boundaries after applying a non-maximum suppression post-processing step. We note that our predicted boundaries achieve highconfidence predictions around objects. This is important as we employ these boundaries to improve semantic segmentation results, as discussed in the next subsection. 3.2.2 Constructing Pairwise Pixel Affinities. We can use the predicted boundaries to build pairwise pixel affinities. Intuitively, we declare two pixels as similar (i.e., likely to belong to the same segment) if there is no boundary crossing the straight path between these two pixels. Conversely, two pixels are dissimilar if there is a boundary crossing their connecting path. The larger the boundary magnitude of the crossed path, the more dissimilar the two pixels should be, since a strong boundary is likely to mark the separation of two distinct segments. Similarly to [1], we encode this intuition with a following formulation: wsb ij= exp (Mij sb) (1)

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

17

encode this intuition with a following formulation: wsb ij= exp (Mij sb) (1) Figure 4: A figure illustrating our boundary detection results. In the second column, we visualize the raw probability output of our boundary detector. In the third column, we present the final boundary maps after non-maximum suppression. While most prior methods predict the boundaries where the sharpest change in color occurs, our method captures semantic object-level boundaries, which we subsequently use to aid semantic segmentation. whereMijdenotes the maximum boundary value that crosses the straight line path between pixels iandj,sbdepicts the smoothing parameter and wsb ijdenotes the semantic boundary-based affinity between pixels iandj. Similarly, we want to exploit high-level object information in the network to define another type of pixel similarity. Specifically, we use object class probabilities from the softmax (SM) layer to achieve this goal. Intuitively, if pixels i andjhave different hard segmentation labels from the softmax layer, we set their similarity ( wsm ij) to0. Otherwise, we compute their similarity using the following equation: wsm ij= exp (Dij sm) (2)

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

18

ij) to0. Otherwise, we compute their similarity using the following equation: wsm ij= exp (Dij sm) (2) whereDijdenotes the difference in softmax output values corresponding to the most likely object class for pixels iandj, andsmis a smoothing parameter. Then we can write the final affinity measure as: wij= exp (wsm ij)wsb ij (3) We exponentiate the term corresponding to the objectlevel affinity because our boundary-based affinity may be too aggressive in declaring two pixels as dissimilar. To address this issue, we increase the importance of the objectlevel affinity in (3) using the exponential function. However, in the experimental results section, we demonstrate that most of the benefit from modeling pairwise potentials comes from wsb ijrather thanwsm ij.We then use this pairwise pixel affinity measure to build a global affinity matrix Wthat encodes relationships between pixels in the entire image. For a given pixel, we sample10% of points in the neighborhood of radius 20 around that pixel, and store the resulting affinities into W.

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

19

sample10% of points in the neighborhood of radius 20 around that pixel, and store the resulting affinities into W. 3.3. Global Inference The last step in our proposed method is to combine semantic boundary information with the coarse segmentation from the FCN softmax layer to produce an improved segmentation. We do this by introducing a global energy function that utilizes the affinity matrix constructed in the previous section along with the segmentation from the FCN softmax layer. Using this energy, we perform a global inference to get segmentations that are well localized around the object boundaries and that are also spatially smooth. Typical globalization models such as MRFs [31], CRFs [19] or Graph Cuts [5] produce a discrete label assignment for the segmentation problem by jointly modeling a multi-label distribution and solving a non-convex optimization. The common problem in doing so is that the optimization procedure may get stuck in local optima. We introduce a new global energy function, which overcomes this issue and achieves better segmentation in comparison to prior globalization models. Similarly to prior globalization approaches, our goal is to minimize the energy corresponding to the sum of unary and pairwise potentials. However, the key difference in our approach comes from the

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

20

globalization approaches, our goal is to minimize the energy corresponding to the sum of unary and pairwise potentials. However, the key difference in our approach comes from the relaxation of some of the constraints. Specifically, instead of modeling our problem as a joint multi-label distribution, we propose to decompose it into multiple binary problems, which can be solved concurrently. This decomposition can be viewed as assigning pixels to foreground and background labels for each of the different object classes. Additionally, we relax the integrality constraint. Both of these relaxations make our problem more manageable and allow us to formulate a global energy function that is differentiable, and has a closed form solution. In [35], the authors introduce the idea of learning with global and local consistency in the context of semisupervised problems. Inspired by this work, we incorporate some of these ideas in the context of semantic segmentation. Before defining our proposed global energy function, we introduce some relevant notation. For the purpose of illustration, suppose that we only have two classes: foreground and background. Then we can denote an optimal continuous solution to such a segmentation problem with variable z. To denote similarity between pixelsiandjwe usewij. Then,diindicates the degree of a

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

21

problem with variable z. To denote similarity between pixelsiandjwe usewij. Then,diindicates the degree of a pixeli. In graph theory, the degree of a node denotes the number of edges incident to that node. Thus, we set the degree of a pixel to di=Pn j=1wijfor alljexcepti6=j. Finally, with fiwe denote an initial segmentation probaInput Softmax Dense-CRF BNF Boundaries BNF Segmentation Figure 5: A figure illustrating semantic segmentation results. Images in columns two and three represent FCN softmax and Dense-CRF predictions, respectively. Note that all methods use the same FCN unary potentials. Additionally, observe that unlike FCN and Dense-CRF, our methods predicts segmentation that are both well localized around object boundaries and that are also spatially smooth. bility, which in our case is obtained from the FCN softmax layer. Using this notation, we can then formulate our global inference objective as: z= argmin z 2X idi(zifi di)2+1 2X ijwij(zizj)2(4) This energy consists of two different terms. Similar to

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

22

idi(zifi di)2+1 2X ijwij(zizj)2(4) This energy consists of two different terms. Similar to the general globalization framework, our first term encodes the unary energy while the second term includes the pairwise energy. We now explain the intuition behind each of these terms. The unary term attempts to find a segmentation assignment ( zi) that deviates little from the initial candidate segmentation computed from the softmax layer (denoted by fi). Theziin the unary term is weighted by the degree di of the pixel in order to produce larger unary costs for pixels that have many similar pixels within the neighborhood. Instead, the pairwise term ensures that pixels that are similar should be assigned similar zvalues. To balance the energies of the two terms we introduce a parameter and set it to0:025throughout all our experiments. We can also express the same global energy function in matrix notation: z= argmin z 2D(zD1f)T(zD1f)+1 2zT(DW)z (5) where zis an1vector containing an optimal continuous assignment for all npixels, Dis a diagonal degree

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

23

2zT(DW)z (5) where zis an1vector containing an optimal continuous assignment for all npixels, Dis a diagonal degree matrix, and Wis thennpixel affinity matrix. Finally, f denotes an1vector containing the probabilities from the softmax layer corresponding to a particular object class. An advantage of our energy is that it is differentiable. Ifwe denote the above energy as E(z)then the derivative of this energy can be written as follows: @E(z) @z=D(zD1f) + (DW)z=0 (6) With simple algebraic manipulations we can then obtain a closed form solution to this optimization: z= (DW)1f (7) where=1 1+and= 1+. In the general case where we havekobject classes we can write the solution as: Z= (DW)1F (8) where Znow depicts a nkmatrix containing assignments for all kobject classes, while Fdenotesnkmatrix with object class probabilities from softmax layer. Due to

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

24

where Znow depicts a nkmatrix containing assignments for all kobject classes, while Fdenotesnkmatrix with object class probabilities from softmax layer. Due to the large size of DWit is impractical to invert it. However, if we consider an image as a graph where each pixel denotes a vertex in the graph, we can observe that the term DWin our optimization is equivalent to a Laplacian matrix of such graph. Since we know that a Laplacian matrix is positive semi-definite, we can use the preconditioned conjugate gradient method [28] to solve the system in Eq. (9). Alternatively, because our defined global energy in Eq. (5) is differentiable, we can efficiently solve this optimization problem using stochastic gradient descent. We choose the former option and solve the following system: (DW)z=f (9) To obtain the final discrete segmentation, for each pixel we assign the object class that corresponds to the largest column value in the row of Z(note that each row in Zrepresents a single pixel in the image, and each column in Z

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

25

we assign the object class that corresponds to the largest column value in the row of Z(note that each row in Zrepresents a single pixel in the image, and each column in Z Metric Inference Method RGB Affinity BNF Affinity PP-IOUBelief Propagation [31] 75.4 75.6 ICM 74.2 75.8 TRWS [32] 75.9 76.7 QPBO [26] 76.9 77.2 BNF 74.6 77.6 PI-IOUBelief Propagation [31] 45.9 46.2 ICM 45.7 48.8 TRWS [32] 51.5 52.0 QPBO [26] 55.3 57.2 BNF 53.0 58.5 Table 2: We compare semantic segmentation results when using a color-based pixel affinity and our proposed boundary-based affinity. We note that our proposed affinity improves the performance of all globalization techniques. Note that all of the inference methods use the same FCN unary potentials . This suggests that for every method our

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

26

improves the performance of all globalization techniques. Note that all of the inference methods use the same FCN unary potentials . This suggests that for every method our boundary-based affinity is more beneficial for semantic segmentation than the color-based affinity. represents one of the object classes). In the experimental section, we show that this solution produces better quantitative and qualitative results in comparison to commonly used globalization techniques. 4. Experimental Results In this section we present quantitative and qualitative results for semantic segmentation on the SBD [13] dataset, which contains objects and their per-pixel annotations for 20Pascal VOC classes. We evaluate semantic segmentation results using two evaluation metrics. The first metric measures accuracy based on pixel intersection-over-union averaged per pixels (PP-IOU) across the 20 classes. According to this metric, the accuracy is computed on a perpixel basis. As a result, the images that contain large object regions are given more importance. However, for certain applications we may need to accurately segment small objects. Therefore, similar to [4] we also consider the PIIOU metric (pixel intersection-over-union averaged per image across the 20 classes), which gives equal weight to each of the images.

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

27

of the images. We compare Boundary Neural Fields with other commonly used global inference methods. These methods include Belief Propagation [31], Iterated Conditional Mode (ICM), Graph Cuts [5], and Dense-CRF [19]. Note that in all of our evaluations we use the same FCN unary potentials for every model. Our evaluations provide evidence for three conclusions: In Subsection 4.1, we show that our boundary-based pixel affinities are better suited for semantic segmentation than the traditional color-based affinities. In Subsection 4.2, we demonstrate that our global minimization leads to better results than those achieved byother inference schemes. In Fig. 5, we qualitatively compare the outputs of FCN and Dense-CRF to our predicted segmentations. This comparison shows that the BNF segments are better localized around the object boundaries and that they are also spatially smooth. 4.1. Comparing Affinity Functions for Semantic Segmentation In Table 2, we consider two global models. Both models use the same unary potentials obtained from the FCN softmax layer. However, the first model uses the popular color-based pairwise affinities, while the second employs

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

28

softmax layer. However, the first model uses the popular color-based pairwise affinities, while the second employs our boundary-based affinities. Each of these two models is optimized using several inference strategies. The table shows that using our boundary based-affinity function improves the results of all global inference methods according to both evaluation metrics. Note that we cannot include Dense-CRF [19] in this comparison because it employs an efficient message-passing technique and integrating our affinities into this technique is a non-trivial task. However, we compare our method with Dense-CRF in Subsection 4.2. The results in Table 2 suggest that our semantic boundary based pixel affinity function yields better semantic segmentation results compared to the commonly-used color based affinities. We note that we also compared the results of our inference technique using other edge detectors, notably UCM [1] and H FL [4]. In comparison to UCM edges, we observed that our boundaries provide 1:0%and6:0% according to both evaluation metrics respectively. When comparing our boundaries with H FL method, we observed similar segmentation performance, which suggests that our method works best with the high quality semantic boundaries.

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

29

comparing our boundaries with H FL method, we observed similar segmentation performance, which suggests that our method works best with the high quality semantic boundaries. 4.2. Comparing Inference Methods for Semantic Segmentation Additionally, we also present semantic segmentation results for both of the metrics (PP-IOU and PI-IOU) in Table 3. In this comparison, all the techniques use the same FCN unary potentials. Additionally, all inference methods except Dense-CRF use our affinity measure (since the previous analysis suggested that our affinities yield better performance). We use BNF-SB to denote the variant of our method that uses only semantic boundary based affinities. Additionally, we use BNF-SB-SM to indicate the version of our method that uses both boundary and softmax -based affinities (see Eq. (3)). Based on these results, we observe that our proposed technique outperforms all the other globalization methods according to both metrics, by 0:3%and1:3%respectively. Metric Method aero bike bird boat bottle bus car cat chair cow table dog horse mbike person plant sheep sofa train tv mean

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

30

Metric Method aero bike bird boat bottle bus car cat chair cow table dog horse mbike person plant sheep sofa train tv mean PP-IOUFCN-Softmax 80.7 71.6 80.7 71.3 72.9 88.1 81.8 86.6 47.4 82.9 57.9 83.9 79.6 80.4 81.0 64.7 78.2 54.5 80.9 69.9 74.8 Belief Propagation [31] 81.4 72.2 82.4 72.2 74.3 88.8 82.4 87.2 48.4 83.8 58.4 84.6 80.5 80.9 81.5 65.1 79.5 55.5 81.5 71.2 75.6 ICM 81.7 72.2 82.8 72.1 75.3 89.6 83.4 87.7 46.3 83.3 58.4 84.6 80.6 81.4 81.5 65.8 79.5 56.0 80.7 74.1 75.8

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

31

TRWS [32] 81.6 70.9 83.8 72.0 75.1 89.5 82.5 88.0 51.7 86.6 61.9 85.8 83.3 80.8 81.1 65.3 81.5 58.8 77.6 75.9 76.7 Graph Cuts [5] 82.5 72.4 84.6 73.3 77.2 89.7 83.3 88.8 49.3 84.0 60.3 85.4 82.2 81.2 81.9 66.7 79.8 58.0 82.3 74.9 76.9 QPBO [26] 82.6 72.3 84.7 73.1 76.7 89.9 83.6 89.3 49.7 85.0 61.1 86.2 82.9 81.3 82.3 67.1 80.5 58.8 82.2 75.1 77.2 Dense-CRF [19] 83.4 71.5 84.9 72.6 76.2 89.5 83.3 89.1 50.4 86.7 61.0 86.8 83.5 81.8 82.3 66.9 82.2 58.2 81.9 75.1 77.3

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

32

BNF-SB 81.9 72.5 84.9 73.3 76.0 90.3 83.1 89.2 51.2 86.7 61.5 86.6 83.2 81.3 81.9 66.2 81.7 58.6 81.6 75.8 77.4 BNF-SB-SM 82.2 73.1 85.1 73.8 76.7 90.6 83.4 89.5 51.3 86.7 61.4 86.8 83.3 81.7 82.3 67.7 81.9 58.4 82.4 75.4 77.6 PI-IOUFCN-Softmax 56.9 35.1 47.8 41.1 27.4 51.1 43.4 52.7 22.2 43.1 29.2 54.2 40.5 45.6 59.1 24.2 43.6 24.8 55.9 37.2 41.8 Belief Propagation [31] 68.0 38.6 52.9 45.8 31.9 55.9 47.2 58.2 24.6 49.9 31.7 60.2 44.9 50.1 62.4 25.2 49.9 27.6 62.3 42.2 46.2

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

33

ICM 65.3 40.9 56.4 45.3 33.7 58.9 49.5 61.9 25.8 53.5 33.2 62.1 48.0 53.2 63.4 24.1 54.8 34.0 63.7 47.7 48.8 TRWS [32] 67.5 40.7 60.3 46.3 35.6 63.4 49.6 69.3 29.7 58.9 37.8 67.4 57.3 53.8 64.1 26.3 62.0 36.9 63.1 49.9 52.0 Graph Cuts [5] 72.1 47.8 64.5 50.8 36.0 70.8 51.4 71.6 31.7 65.8 34.4 71.8 62.0 59.4 64.8 29.0 60.9 38.7 70.3 51.6 55.3 QPBO [26] 71.6 46.8 65.6 49.6 38.0 72.6 52.7 76.7 32.5 69.6 38.9 74.4 61.4 61.0 66.2 30.3 68.7 41.4 72.2 52.8 57.2

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

34

Dense-CRF [19] 68.0 39.5 58.0 45.0 33.4 62.8 47.7 66.0 29.4 60.9 36.0 68.5 54.6 51.4 63.7 28.3 57.6 37.1 65.9 48.2 51.1 BNF-SB 71.6 48.1 67.2 52.3 37.8 79.5 52.9 80.8 33.3 71.5 39.5 75.1 65.7 63.4 65.1 31.1 67.5 39.6 73.2 54.7 58.5 BNF-SB-SM 72.0 48.9 66.5 52.9 39.1 79.0 53.4 78.6 32.9 72.2 39.4 74.6 65.9 64.2 65.8 31.7 66.9 39.0 73.1 53.9 58.5 Table 3: Semantic segmentation results on the SBD dataset according to PP-IOU (per pixel) and PI-IOU (per image) evaluation metrics. We use BNF-SB to denote the variant of our method that uses only semantic boundary based affinities.

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

35

Additionally, we use BNF-SB-SM to indicate our method that uses boundary and softmax based affinities (See Eq. (3)). We observe that our proposed globalization method outperforms other globalization techniques according to both metrics by at least 0:3%and1:3%respectively. Note that in this experiment, all of the inference methods use the same FCN unary potentials . Additionally, for each method except Dense-CRF (it is challenging to incorporate boundary based affinities into the Dense-CRF framework) we use our boundary based affinities, since those lead to better results. Additionally, these results indicate that most benefit comes from the semantic boundary affinity term rather than the softmax affinity term. In Fig. 5, we also present qualitative semantic segmentation results. Note that, compared to the segmentation output from the softmax layer, our segmentation is much better localized around the object boundaries. Additionally, in comparison to Dense-CRF predictions, our method produces segmentations that are much spatially smoother. 4.3. Semantic Boundary Classification We can also label our boundaries with a specific object

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

36

4.3. Semantic Boundary Classification We can also label our boundaries with a specific object class, using the same classification strategy as in the H FL system [4]. Since the SBD dataset provides annotations for semantic boundary classification, we can test our results against the state-of-the-art H FL [4] method for this task. Due to the space limitation, we do not include full results for each category. However, we observe that our produced results achieve mean Max F-Score of 54:5%(averaged across all20classes) whereas H FL method obtains 51:7%. 5. Conclusions In this work we introduced a Boundary Neural Field (BNF), an architecture that employs a semantic segmentation FCN to predict semantic boundaries and then uses the predicted boundaries and the FCN output to produce an improved semantic segmentation maps a global optimization. We showed that our predicted boundaries are better suited for semantic segmentation than the commonly used low-level color based affinities. Additionally, we introduced a global energy function that decomposes semantic segmentation into multiple binary problems and relaxes an integrality constraint. We demonstrated that the minimization of this global energy allows us to predict segmentations that are better localized around the object boundaries and

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

37

of this global energy allows us to predict segmentations that are better localized around the object boundaries and that are spatially smoother compared to the segmentations achieved by prior methods. We made the code of our globalization technique available at http://www.seas.upenn. edu/~gberta/publications.html . The main goal of this work was to show the effectiveness of boundary-based affinities for semantic segmentation. However, due to differentiability of our global energy, it may be possible to add more parameters inside the BNFs and learn them in a front-to-end fashion. We believe that optimizing the entire architecture jointly could capture the inherent relationship between semantic segmentation and boundary detection even better and further improve the performance of BNFs. We will investigate this possibility in our future work. 6. Acknowledgements This research was funded in part by NSF award CNS1205521. References [1] Pablo Arbelaez, Michael Maire, Charless Fowlkes, and Jitendra Malik. Contour detection and hierarchical image segmentation. IEEE Trans. Pattern Anal. Mach. Intell. , 33(5):898–916, May 2011. 2, 4, 7

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

38

Trans. Pattern Anal. Mach. Intell. , 33(5):898–916, May 2011. 2, 4, 7 [2] Pablo Arbelaez, J. Pont-Tuset, Jon Barron, F. Marqués, and Jitendra Malik. Multiscale combinatorial grouping. In Computer Vision and Pattern Recognition (CVPR) , 2014. 2, 4 [3] Gedas Bertasius, Jianbo Shi, and Lorenzo Torresani. Deepedge: A multi-scale bifurcated deep network for top-down contour detection. InThe IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , June 2015. 2, 3, 4 [4] Gedas Bertasius, Jianbo Shi, and Lorenzo Torresani. High-for-low and low-for-high: Efficient boundary detection from deep object features and its applications to high-level vision. In The IEEE International Conference on Computer Vision (ICCV) , December 2015. 2, 3, 4, 7, 8 [5] Yuri Boykov, Olga Veksler, and Ramin Zabih. Fast approximate energy minimization via graph cuts. IEEE Trans. Pattern Anal. Mach. Intell. , 23(11):1222–1239, November 2001. 5, 7, 8

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

39

Intell. , 23(11):1222–1239, November 2001. 5, 7, 8 [6] João Carreira, Rui Caseiro, Jorge Batista, and Cristian Sminchisescu. Semantic segmentation with second-order pooling. In Proceedings of the 12th European Conference on Computer Vision - Volume Part VII, ECCV’12, pages 430–443, Berlin, Heidelberg, 2012. SpringerVerlag. 2 [7] Liang-Chieh Chen, George Papandreou, Iasonas Kokkinos, Kevin Murphy, and Alan L. Yuille. Semantic image segmentation with deep convolutional nets and fully. In ICLR , 2015. 1, 2, 3 [8] Jifeng Dai, Kaiming He, and Jian Sun. Boxsup: Exploiting bounding boxes to supervise convolutional networks for semantic segmentation. In The IEEE International Conference on Computer Vision (ICCV) , December 2015. 3 [9] Piotr Dollár and C. Lawrence Zitnick. Fast edge detection using structured forests. PAMI , 2015. 2, 4

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

40

[9] Piotr Dollár and C. Lawrence Zitnick. Fast edge detection using structured forests. PAMI , 2015. 2, 4 [10] Clement Farabet, Camille Couprie, Laurent Najman, and Yann LeCun. Learning hierarchical features for scene labeling. IEEE Transactions on Pattern Analysis and Machine Intelligence , August 2013. 2 [11] Yaroslav Ganin and Victor S. Lempitsky. N4-fields: Neural network nearest neighbor fields for image transforms. ACCV , 2014. 2, 4 [12] Saurabh Gupta, Ross Girshick, Pablo Arbeláez, and Jitendra Malik. Learning rich features from RGB-D images for object detection and segmentation. In Proceedings of the European Conference on Computer Vision (ECCV) , 2014. 2 [13] Bharath Hariharan, Pablo Arbelaez, Lubomir Bourdev, Subhransu Maji, and Jitendra Malik. Semantic contours from inverse detectors. InInternational Conference on Computer Vision (ICCV) , 2011. 4, 7

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

41

Maji, and Jitendra Malik. Semantic contours from inverse detectors. InInternational Conference on Computer Vision (ICCV) , 2011. 4, 7 [14] Bharath Hariharan, Pablo Arbeláez, Ross Girshick, and Jitendra Malik. Simultaneous detection and segmentation. In European Conference on Computer Vision (ECCV) , 2014. 2 [15] Bharath Hariharan, Pablo Andrés Arbeláez, Ross B. Girshick, and Jitendra Malik. Hypercolumns for object segmentation and finegrained localization. In The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , June 2015. 2, 3 [16] Seunghoon Hong, Hyeonwoo Noh, and Bohyung Han. Decoupled deep neural network for semi-supervised semantic segmentation. In NIPS) , December 2015. 3 [17] Phillip Isola, Daniel Zoran, Dilip Krishnan, and Edward H. Adelson. Crisp boundary detection using pointwise mutual information. In ECCV , 2014. 2 [18] Jyri J Kivinen, Christopher KI Williams, and Nicolas Heess. Visual boundary prediction: A deep neural prediction network and quality

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

42

[18] Jyri J Kivinen, Christopher KI Williams, and Nicolas Heess. Visual boundary prediction: A deep neural prediction network and quality dissection. AISTATS , 1(2):9, 2014. 2 [19] Philipp Krähenbühl and Vladlen Koltun. Efficient inference in fully connected crfs with gaussian edge potentials. In J. Shawe-Taylor, R.S. Zemel, P.L. Bartlett, F. Pereira, and K.Q. Weinberger, editors, Advances in Neural Information Processing Systems 24 , pages 109– 117. Curran Associates, Inc., 2011. 1, 5, 7, 8[20] Joseph Lim, C. Lawrence Zitnick, and Piotr Dollár. Sketch tokens: A learned mid-level representation for contour and object detection. InCVPR , 2013. 2 [21] Guosheng Lin, Chunhua Shen, Ian D. Reid, and Anton van den Hengel. Efficient piecewise training of deep structured models for semantic segmentation. CoRR , abs/1504.01013, 2015. 3 [22] Jonathan Long, Evan Shelhamer, and Trevor Darrell. Fully convolutional networks for semantic segmentation. In The IEEE Conference

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

43

[22] Jonathan Long, Evan Shelhamer, and Trevor Darrell. Fully convolutional networks for semantic segmentation. In The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , June 2015. 1, 3 [23] D. Martin, C. Fowlkes, D. Tal, and J. Malik. A database of human segmented natural images and its application to evaluating segmentation algorithms and measuring ecological statistics. In Proc. 8th Int’l Conf. Computer Vision , volume 2, pages 416–423, July 2001. 4 [24] Mohammadreza Mostajabi, Payman Yadollahpour, and Gregory Shakhnarovich. Feedforward semantic segmentation with zoom-out features. CoRR , abs/1412.0774, 2014. 2 [25] X. Ren and L. Bo. Discriminatively Trained Sparse Code Gradients for Contour Detection. In Advances in Neural Information Processing Systems , December 2012. 4 [26] Carsten Rother, Vladimir Kolmogorov, Victor Lempitsky, and Martin Szummer. Optimizing binary mrfs via extended roof duality. In Proc Comp. Vision Pattern Recogn. (CVPR) , June 2007. 7, 8 [27] Wei Shen, Xinggang Wang, Yan Wang, Xiang Bai, and Zhijiang

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

44

Comp. Vision Pattern Recogn. (CVPR) , June 2007. 7, 8 [27] Wei Shen, Xinggang Wang, Yan Wang, Xiang Bai, and Zhijiang Zhang. Deepcontour: A deep convolutional feature learned by positive-sharing loss for contour detection. June 2015. 2, 4 [28] Jonathan R Shewchuk. An introduction to the conjugate gradient method without the agonizing pain. Technical report, Pittsburgh, PA, USA, 1994. 6 [29] Jianbo Shi and Jitendra Malik. Normalized cuts and image segmentation. IEEE Transactions on Pattern Analysis and Machine Intelligence , 22:888–905, 1997. 2 [30] K. Simonyan and A. Zisserman. Very deep convolutional networks for large-scale image recognition. CoRR , abs/1409.1556, 2014. 3 [31] Marshall F. Tappen and William T. Freeman. Comparison of graph cuts with belief propagation for stereo, using identical mrf parameters. In Proceedings of the Ninth IEEE International Conference on Computer Vision - Volume 2 , ICCV ’03, pages 900–, Washington, DC, USA, 2003. IEEE Computer Society. 5, 7, 8

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

45

Computer Vision - Volume 2 , ICCV ’03, pages 900–, Washington, DC, USA, 2003. IEEE Computer Society. 5, 7, 8 [32] Martin Wainwright, Tommi Jaakkola, and Alan Willsky. Map estimation via agreement on (hyper)trees: Message-passing and linear programming approaches. IEEE Transactions on Information Theory, 51:3697–3717, 2002. 7, 8 [33] Saining Xie and Zhuowen Tu. Holistically-nested edge detection. InThe IEEE International Conference on Computer Vision (ICCV) , December 2015. 2, 4 [34] Shuai Zheng, Sadeep Jayasumana, Bernardino Romera-Paredes, Vibhav Vineet, Zhizhong Su, Dalong Du, Chang Huang, and Philip Torr. Conditional random fields as recurrent neural networks. In International Conference on Computer Vision (ICCV) , 2015. 3 [35] Dengyong Zhou, Olivier Bousquet, Thomas N. Lal, Jason Weston, and Bernhard Schölkopf. Learning with local and global consistency. In S. Thrun, L.K. Saul, and B. Schölkopf, editors, Advances in Neural Information Processing Systems 16 , pages 321–328. MIT Press,

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.02674

46

2004. 5

1511.02674

Semantic Segmentation with Boundary Neural Fields

The state-of-the-art in semantic segmentation is currently represented by fully convolutional networks (FCNs). However, FCNs use large receptive fields and many pooling layers, both of which cause blurring and low spatial resolution in the deep layers. As a result FCNs tend to produce segmentations that are poorly localized around object boundaries. Prior work has attempted to address this issue in post-processing steps, for example using a color-based CRF on top of the FCN predictions. However, these approaches require additional parameters and low-level features that are difficult to tune and integrate into the original network architecture. Additionally, most CRFs use color-based pixel affinities, which are not well suited for semantic segmentation and lead to spatially disjoint predictions. To overcome these problems, we introduce a Boundary Neural Field (BNF), which is a global energy model integrating FCN predictions with boundary cues. The boundary information is used to enhance semantic segment coherence and to improve object localization. Specifically, we first show that the convolutional filters of semantic FCNs provide good features for boundary detection. We then employ the predicted boundaries to define pairwise potentials in our energy. Finally, we show that our energy decomposes semantic segmentation into multiple binary problems, which can be relaxed for efficient global optimization. We report extensive experiments demonstrating that minimization of our global boundary-based energy yields results superior to prior globalization methods, both quantitatively as well as qualitatively.

http://arxiv.org/pdf/1511.02674

cs.CV

20151109

20160524

1511.06464

0

Unitary Evolution Recurrent Neural Networks Martin ArjovskyMARJOVSKY @DC.UBA.AR Amar ShahAS793@ CAM .AC.UK Yoshua Bengio Universidad de Buenos Aires, University of Cambridge, Universit ´e de Montr ´eal. Yoshua Bengio is a CIFAR Senior Fellow. Indicates first authors. Ordering determined by coin flip. Abstract Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishingandexploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

1

expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very longterm dependencies. 1. Introduction Deep Neural Networks have shown remarkably good performance on a wide range of complex data problems including speech recognition (Hinton et al., 2012), image recognition (Krizhevsky et al., 2012) and natural language processing (Collobert et al., 2011). However, training very deep models remains a difficult task. The main issue surrounding the training of deep networks is the vanishing andexploding gradients problems introduced by HochreProceedings of the 33rdInternational Conference on Machine Learning , New York, NY , USA, 2016. JMLR: W&CP volume 48. Copyright 2016 by the author(s).iter (1991) and shown by Bengio et al. (1994) to be necessarily arising when trying to learn to reliably store bits of information in any parametrized dynamical system. If gradients propagated back through a network vanish, the credit assignment role of backpropagation is lost, as information about small changes in states in the far past has no

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

2

gradients propagated back through a network vanish, the credit assignment role of backpropagation is lost, as information about small changes in states in the far past has no influence on future states. If gradients explode, gradientbased optimization algorithms struggle to traverse down a cost surface, because gradient-based optimization assumes small changes in parameters yield small changes in the objective function. As the number of time steps considered in the sequence of states grows, the shrinking or expanding effects associated with the state-to-state transformation at individual time steps can grow exponentially, yielding respectively vanishing or exploding gradients. See Pascanu et al. (2010) for a review. Although the long-term dependencies problem appears intractable in the absolute (Bengio et al., 1994) for parametrized dynamical systems, several heuristics have recently been found to help reduce its effect, such as the use of self-loops and gating units in the LSTM (Hochreiter & Schmidhuber, 1997) and GRU (Cho et al., 2014) recurrent architectures. Recent work also supports the idea of using orthogonal weight matrices to assist optimization (Saxe et al., 2014; Le et al., 2015). In this paper, we explore the use of orthogonal and unitary

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

3

(Saxe et al., 2014; Le et al., 2015). In this paper, we explore the use of orthogonal and unitary matrices in recurrent neural networks. We start in Section 2 by showing a novel bound on the propagated gradients in recurrent nets when the recurrent matrix is orthogonal. Section 3 discusses the difficulties of parameterizing real valued orthogonal matrices and how they can be alleviated by moving to the complex domain. We discuss a novel approach to constructing expressive unitary matrices as the composition of simple unitary matrices which require at most O(nlogn)computation and O(n)memory, when the state vector has dimension n. These are unlike general matrices, which require O(n2) computation and memory. Complex valued representationsarXiv:1511.06464v4 [cs.LG] 25 May 2016 Unitary Evolution Recurrent Neural Networks have been considered for neural networks in the past, but with limited success and adoption (Hirose, 2003; Zimmermann et al., 2011). We hope our findings will change this. Whilst our model uses complex valued matrices and parameters, all implementation and optimization is possible with real numbers and has been done in Theano (Bergstra

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

4

Whilst our model uses complex valued matrices and parameters, all implementation and optimization is possible with real numbers and has been done in Theano (Bergstra et al., 2010). This along with other implementation details are discussed in Section 4, and the code used for the experiments is available online. The potential of the developed model for learning long term dependencies with relatively few parameters is explored in Section 5. We find that the proposed architecture generally outperforms LSTMs and previous approaches based on orthogonal initialization. 2. Orthogonal Weights and Bounding the Long-Term Gradient A matrix, W, is orthogonal if W>W=WW>=I. Orthogonal matrices have the property that they preserve norm (i.e.kWhk2=khk2) and hence repeated iterative multiplication of a vector by an orthogonal matrix leaves the norm of the vector unchanged. LethTandhtbe the hidden unit vectors for hidden layers Tandtof a neural network with Thidden layers and T t. IfCis the objective we are trying to minimize, then thevanishing andexploding gradient problems refer to the decay or growth of@C @htas the number of layers, T, grows. Letbe a pointwise nonlinearity function, and

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

5

decay or growth of@C @htas the number of layers, T, grows. Letbe a pointwise nonlinearity function, and zt+1=Wtht+Vtxt+1 ht+1=(zt+1) (1) then by the chain rule @C @ht=@C @hT@hT @ht =@C @hTT1Y k=t@hk+1 @hk=@C @hTT1Y k=tDk+1WT k (2) where Dk+1=diag(0(zk+1))is the Jacobian matrix of the pointwise nonlinearity. In the following we define the norm of a matrix to refer to the spectral radius norm (or operator 2-norm) and the norm of a vector to mean L2-norm. By definition of the operator norms, for any matrices A;Band vectorvwe have kAvkk AkkvkandkABkk AkkBk. If the weight matrices Wkare norm preserving (i.e. orthogonal), thenweprove @C

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

6

matrices Wkare norm preserving (i.e. orthogonal), thenweprove @C @ht = @C @hTT1Y k=tDk+1WT k  @C @hT T1Y k=t Dk+1WT k = @C @hT T1Y k=tkDk+1k: (3) Since Dkis diagonal,kDkk= max j=1;:::;nj0(z(j) k)j, withz(j) kthej-th pre-activation of the k-th hidden layer. If the absolute value of the derivative 0can take some value > 1, then this bound is useless, since k@C @htk @C @hT Ttwhich grows exponentially in T. We therefore cannot effectively bound@C @htfor deep networks, resulting potentially in exploding gradients. In the casej0j<  < 1, equation 3 proves that that@C

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

7

@htfor deep networks, resulting potentially in exploding gradients. In the casej0j<  < 1, equation 3 proves that that@C @httends to 0 exponentially fast as Tgrows, resulting in guaranteed vanishing gradients. This argument makes the rectified linear unit (ReLU) nonlinearity an attractive choice (Glorot et al., 2011; Nair & Hinton, 2010). Unless all the activations are killed at one layer, the maximum entry of Dkis 1, resulting inkDkk= 1 for all layers k. With ReLU nonlinearities, we thus have @C @ht  @C @hT T1Y k=tkDk+1k= @C @hT : (4) Most notably, this result holds for a network of arbitrary depth and renders engineering tricks like gradient clipping unnecessary (Pascanu et al., 2010). To the best of our knowledge, this analysis is a novel contribution and the first time a neural network architecture has been mathematically proven to avoid exploding gradients. 3. Unitary Evolution RNNs Unitary matrices generalize orthogonal matrices to the

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

8

been mathematically proven to avoid exploding gradients. 3. Unitary Evolution RNNs Unitary matrices generalize orthogonal matrices to the complex domain. A complex valued, norm preserving matrix,U, is called a unitary matrix and is such that UU= UU=I, where Uis the conjugate transpose of U. Directly parametrizing the set of unitary matrices in such a way that gradient-based optimization can be applied is not straightforward because a gradient step will typically yield a matrix that is not unitary, and projecting on the set of unitary matrices (e.g., by performing an eigendecomposition) generally costsO(n3)computation when Uisnn. The most important feature of unitary and orthogonal matrices for our purpose is that they have eigenvalues jwith Unitary Evolution Recurrent Neural Networks absolute value 1. The following lemma, proved in (Hoffman & Kunze, 1971), may shed light on a method which can be used to efficiently span a large set of unitary matrices. Lemma 1. A complex square matrix Wis unitary if and only if it has an eigendecomposition of the form W=

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

9

Lemma 1. A complex square matrix Wis unitary if and only if it has an eigendecomposition of the form W= VDV, wheredenotes the conjugate transpose. Here, V;D2Cnnare complex matrices, where Vis unitary, andDis a diagonal such that jDj;jj= 1. Furthermore, W is a real orthogonal matrix if and only if for every eigenvalue Dj;j=jwith eigenvector vj, there is also a complex conjugate eigenvalue k=jwith corresponding eigenvectorvk=vj. Writingj=eiwjwithwj2R, a naive method to learn a unitary matrix would be to fix a basis of eigenvectors V2 Cnnand set W=VDV; (5) where Dis a diagonal such that Dj;j=j. Lemma 1 informs us how to construct a real orthogonal matrix, W. We must (i) ensure the columns of Vcome in complex conjugate pairs, vk=vj, and (ii) tie weights

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

10

matrix, W. We must (i) ensure the columns of Vcome in complex conjugate pairs, vk=vj, and (ii) tie weights wk=wjin order to achieve eiwj=eiwk. Most neural network objective functions are differentiable with respect to the weight matrices, and consequently wjmay be learned by gradient descent. Unfortunately the above approach has undesirable properties. Fixing Vand learning wrequiresO n2
memory, which is unacceptable given that the number of learned parameters isO(n). Further note that calculating Vufor an arbitrary vector urequiresO(n2)computation. Setting V to the identity would satisfy the conditions of the lemma, whilst reducing memory and computation requirements to O(n), however, Wwould remain diagonal, and have poor representation capacity. We propose an alternative strategy to parameterize unitary matrices. Since the product of unitary matrices is itself a unitary matrix, we compose several simple, parameteric, unitary matrices to construct a single, expressive unitary matrix. The four unitary building blocks considered are D, a diagonal matrix with Dj;j=eiwj, with parameterswj2R, R=I2vv

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

11

D, a diagonal matrix with Dj;j=eiwj, with parameterswj2R, R=I2vv kvk2, a reflection matrix in the complex vectorv2Cn, , a fixed random index permutation matrix, and  F andF1, the Fourier and inverse Fourier transforms.Appealingly, D,Randall permitO(n)storage and O(n)computation for matrix vector products. FandF1 require no storage and O(nlogn)matrix vector multiplication using the Fast Fourier Transform algorithm. A major advantage of composing unitary matrices of the form listed above, is that the number of parameters, memory and computational cost increase almost linearly in the size of the hidden layer. With such a weight matrix, immensely large hidden layers are feasible to train, whilst being impossible in traditional neural networks. With this in mind, in this work we choose to consider recurrent neural networks with unitary hidden to hidden weight matrices. Our claim is that the ability to have large hidden layers where hidden states norms are preserved provides a powerful tool for modeling long term dependencies in sequence data. (Bengio et al., 1994) suggest that having a large memory may be crucial for solving difficult tasks

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

12

sequence data. (Bengio et al., 1994) suggest that having a large memory may be crucial for solving difficult tasks with long ranging dependencies: the smaller the state dimension, the more information necessarily has to be eliminated when mapping a long sequence to a fixed-dimension state. We call any RNN architecture which uses a unitary hidden to hidden matrix a unitary evolution RNN (uRNN). After experimenting with several structures, we settled on the following composition W=D3R2F1D2R 1FD1: (6) Whilst each but the permutation matrix is complex, we parameterize and represent them with real numbers for implementation purposes. When the final cost is real and differentiable, we may perform gradient descent optimization to learn the parameters. (Yang et al., 2015) construct a real valued, non-orthogonal matrix using a similar parameterization with the motivation of parameter reduction by an order of magnitude on an industrial sized network. This combined with earlier work (Le et al., 2010) suggests that it is possible to create highly expressive matrices by composing simple matrices with few parameters. In the following section, we explain details on how to implement our model and illustrate how we bypass the potential difficulties of working in the complex domain.

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

13

section, we explain details on how to implement our model and illustrate how we bypass the potential difficulties of working in the complex domain. 4. Architecture details In this section, we describe the nonlinearity we used, how we incorporate real valued inputs with complex valued hidden units and map from complex hidden states to real outputs. 4.1. Complex hidden units Our implementation represents all complex numbers using real values in terms of their real and imaginary parts. Unitary Evolution Recurrent Neural Networks Under this framework, we sidestep the lack of support for complex numbers by most deep learning frameworks. Consider multiplying the complex weight matrix W= A+iBby the complex hidden vector h=x+iy, where A;B;x;y are real. It is trivially true that Wh= (AxBy) +i(Ay+Bx). When we represent v2Cnas Re(v)>;Im(v)>
>2R2n, we compute complex matrix vector products with real numbers as follows Re(Wh) Im(Wh) =AB B ARe(h) Im(h) : (7) More generally, let f:Cn!Cnbe any complex function andz=x+iyany complex vector. We may write

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

14

Im(h) : (7) More generally, let f:Cn!Cnbe any complex function andz=x+iyany complex vector. We may write f(z) =(x;y) +i(x;y)where; :Rn!Rn. This allows us to implement everything using real valued operations, compatible with any any deep learning framework with automatic differentiation such as Theano. 4.2. Input to Hidden, Nonlinearity, Hidden to Output As is the case with most recurrent networks, our uRNN follows the same hidden to hidden mapping as equation 1 with Vt=VandWt=W. Denote the size of the complex valued hidden states as nh. The input to hidden matrix is complex valued, V2Cnhnin. We learn the initial hidden stateh02Cnhas a parameter of the model. Choosing an appropriate nonlinearity is not trivial in the complex domain. As discussed in the introduction, using a ReLU is a natural choice in combination with a norm preserving weight matrix. We first experimented with placing separate ReLU activations on the real and imaginary parts

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

15

ReLU is a natural choice in combination with a norm preserving weight matrix. We first experimented with placing separate ReLU activations on the real and imaginary parts of the hidden states. However, we found that such a nonlinearity usually performed poorly. Our intuition is that applying separate ReLU nonlinearities to the real and imaginary parts brutally impacts the phase of a complex number, making it difficult to learn structure. We speculate that maintaining the phase of hidden states may be important for storing information across a large number of time steps, and our experiments supported this claim. A variation of the ReLU that we name modReLU , is what we finally chose. It is a pointwise nonlinearity, modReLU (z) :C!C, which affects only the absolute value of a complex number, defined as modReLU (z) =(jzj+b)z jzjifjzj+b0 0 ifjzj+b<0(8) whereb2Ris a bias parameter of the nonlinearity. For a nhdimensional hidden space we learn nhnonlinearity bias parameters, one per dimension. Note that the modReLU is similar to the ReLU in spirit, in fact more concretely

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

16

parameters, one per dimension. Note that the modReLU is similar to the ReLU in spirit, in fact more concretely modReLU (z) =ReLU (jzj+b)z jzj.To map hidden states to output, we define a matrix U2 Rno2nh, wherenois the output dimension. We calculate a linear output as ot=URe(ht) Im(ht) +bo; (9) wherebo2Rnois the output bias. The linear output is real valued (ot2Rno) and can be used for prediction and loss function calculation akin to typical neural networks (e.g. it may be passed through a softmax which is used for cross entropy calculation for classification tasks). 4.3. Initialization Due to the stability of the norm preserving operations of our network, we found that performance was not very sensitive to initialization of parameters. For full disclosure and reproducibility, we explain our initialization strategy for each parameter below. We initialize VandU(the input and output matrices) as in (Glorot & Bengio, 2010), with weights sampled independently from uniforms, Uh p 6pnin+nout;p 6pnin+nouti .

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

17

with weights sampled independently from uniforms, Uh p 6pnin+nout;p 6pnin+nouti . The biases,bandboare initialized to 0. This implies that at initialization, the network is linear with unitary weights, which seems to help early optimization (Saxe et al., 2014). The reflection vectors for R1andR2are initialized coordinate-wise from a uniform U[1;1]. Note that the reflection matrices are invariant to scalar multiplication of the parameter vector, hence the width of the uniform initialization is unimportant. The diagonal weights for D1;D2andD3are sampled from a uniform, U[;]. This ensures that the diagonal entries Dj;jare sampled uniformly over the complex unit circle. We initialize h0with a uniform,Uh q 3 2nh;q 3 2nhi , which results in E kh0k2 = 1. Since the norm of the hidden units are roughly preserved through unitary evolution and inputs are typically whitened to have norm 1, we have hidden states, inputs and linear outputs of the same order of magnitude, which seems to help optimization. 5. Experiments

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

18

norm 1, we have hidden states, inputs and linear outputs of the same order of magnitude, which seems to help optimization. 5. Experiments In this section we explore the performance of our uRNN in relation to (a) RNN with tanh activations, (b) IRNN (Le et al., 2015), that is an RNN with ReLU activations and with the recurrent weight matrix initialized to the identity, Unitary Evolution Recurrent Neural Networks Figure 1. Results of the copying memory problem for time lags of 100;200;300;500. The LSTM is able to beat the baseline only for 100times steps. Conversely the uRNN is able to completely solve each time length in very few training iterations, without getting stuck at the baseline. and (c) LSTM (Hochreiter & Schmidhuber, 1997) models. We show that the uRNN shines quantitatively when it comes to modeling long term dependencies and exhibits qualitatively different learning properties to the other models. We chose a handful of tasks to evaluate the performance of the various models. The tasks were especially created to be be pathologically hard, and have been used as benchmarks for testing the ability of a model to capture long-term memory (Hochreiter & Schmidhuber, 1997; Le et al., 2015;

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

19

memory (Hochreiter & Schmidhuber, 1997; Le et al., 2015; Graves et al., 2014; Martens & Sutskever, 2011) Of the handful of optimization algorithms we tried on the various models, RMSProp (Tieleman & Hinton, 2012) lead to fastest convergence and is what we stuck to for all experiments here on in. However, we found the IRNN to be particularly unstable; it only ran without blowing up with incredibly low learning rates and gradient clipping. Since the performance was so poor relative to other models we compare against, we do not show IRNN curves in the figures. In each experiment we use a learning rate of 103 and a decay rate of 0:9. For the LSTM and RNN models, we had to clip gradients at 1 to avoid exploding gradients.Gradient clipping was unnecessary for the uRNN. 5.1. Copying memory problem Recurrent networks have been known to have trouble remembering information about inputs seen many time steps previously (Bengio et al., 1994; Pascanu et al., 2010). We therefore want to test the uRNN’s ability to recall exactly data seen a long time ago. Following a similar setup to (Hochreiter & Schmidhuber,

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

20

data seen a long time ago. Following a similar setup to (Hochreiter & Schmidhuber, 1997), we outline the copy memory task. Consider 10 categories,faig9 i=0. The input takes the form of a T+20 length vector of categories, where we test over a range of values ofT. The first 10entries are sampled uniformly, independently and with replacement from faig7 i=0, and represent the sequence which will need to be remembered. The next T1entries are set to a8, which can be thought of as the ’blank’ category. The next single entry is a9, which represents a delimiter, which should indicate to the algorithm that it is now required to reproduce the initial 10categories in the output. The remaining 10entries are set to a8. The required output sequence consists of T+ 10 repeated entries ofa8, followed by the first 10categories of the input Unitary Evolution Recurrent Neural Networks Figure 2. Results of the adding problem for T= 100 ;200;400;750. The RNN with tanh is not able to beat the baseline for any time length. The LSTM and the uRNN show similar performance across time lengths, consistently beating the baseline.

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

21

length. The LSTM and the uRNN show similar performance across time lengths, consistently beating the baseline. sequence in exactly the same order. The goal is to minimize the average cross entropy of category predictions at each time step of the sequence. The task amounts to having to remember a categorical sequence of length 10, for T time steps. A simple baseline can be established by considering an optimal strategy when no memory is available, which we deem the memoryless strategy. The memoryless strategy would be to predict a8forT+ 10 entries and then predict each of the final 10categories from the set faig7 i=0independently and uniformly at random. The categorical cross entropy of this strategy is10 log(8) T+20. We ran experiments where the RNN with tanh activations, IRNN, LSTM and uRNN had hidden layers of size 80, 80, 40 and 128 respectively. This equates to roughly 6500 parameters per model. In Figure 1, we see that aside from the simplest case, both the RNN with tanh and more surprisingly the LSTMs get almost exactly the same cost as the memoryless strategy. This behaviour is consistent with the results of (Graves et al., 2014), in which poor performance

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

22

memoryless strategy. This behaviour is consistent with the results of (Graves et al., 2014), in which poor performance is reported for the LSTM for a very similar long term memory problem. The uRNN consistently achieves perfect performance inrelatively few iterations, even when having to recall sequences after 500 time steps. What is remarkable is that the uRNN does not get stuck at the baseline at all, whilst the LSTM and RNN do. This behaviour suggests that the representations learned by the uRNN have qualitatively different properties from both the LSTM and classical RNNs. 5.2. Adding Problem We closely follow the adding problem defined in (Hochreiter & Schmidhuber, 1997) to explain the task at hand. Each input consists of two sequences of length T. The first sequence, which we denote x, consists of numbers sampled uniformly at random U[0;1]. The second sequence is an indicator sequence consisting of exactly two entries of 1 and remaining entries 0. The first 1 entry is located uniformly at random in the first half of the sequence, whilst the second 1 entry is located uniformly at random in the second half. The output is the sum of the two entries of the first sequence, corresponding to where the 1 entries are located in

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

23

half. The output is the sum of the two entries of the first sequence, corresponding to where the 1 entries are located in the second sequence. A naive strategy of predicting 1 as the output regardless of the input sequence gives an expected mean squared error of 0:167, the variance of the sum of two independent uniform distributions. This is our baseline to beat. Unitary Evolution Recurrent Neural Networks Figure 3. Results on pixel by pixel MNIST classification tasks. The uRNN is able to converge in a fraction of the iterations that the LSTM requires. The LSTM performs better on MNIST classification, but the uRNN outperforms on the more complicated task of permuted pixels. We chose to use 128 hidden units for the RNN with tanh, IRNN and LSTM and 512 for the uRNN. This equates to roughly 16K parameters for the RNN with tanh and IRNN, 60K for the LSTM and almost 9K for the uRNN. All models were trained using batch sizes of 20 and 50 with the best results being reported. Our results are shown in Figure 2. The LSTM and uRNN models are able to convincingly beat the baseline up to T= 400 time steps. Both models do well

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

24

The LSTM and uRNN models are able to convincingly beat the baseline up to T= 400 time steps. Both models do well whenT= 750 , but the mean squared error does not reach close to 0. The uRNN achieves lower test error, but it’s curve is more noisy. Despite having vastly more parameters, we monitored the LSTM performance to ensure no overfitting. The RNN with tanh and IRNN were not able to beat the baseline for any number of time steps. (Le et al., 2015) report that their RNN solve the problem for T= 150 and the IRNN forT= 300 , but they require over a million iterations before they start learning. Neither of the two models came close to either the uRNN or the LSTM in performance. The stark difference in our findings are best explained by our use of RMSprop with significantly higher learning rates ( 103as opposed to 108) than (Le et al., 2015) use for SGD with momentum. 5.3. Pixel-by-pixel MNIST In this task, suggested by (Le et al., 2015), algorithms are fed pixels of MNIST (LeCun et al., 1998) sequentially and

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

25

In this task, suggested by (Le et al., 2015), algorithms are fed pixels of MNIST (LeCun et al., 1998) sequentially and required to output a class label at the end. We consider two tasks: one where pixels are read in order (from left to right, bottom to top) and one where the pixels are all randomly permuted using the same randomly generated permutation matrix. The same model architectures as for the addingproblem were used for this task, except we now use a softmax for category classification. We ran the optimization algorithms until convergence of the mean categorical cross entropy on test data, and plot test accuracy in Figure 3. Both the uRNN and LSTM perform applaudably well here. On the correct unpermuted MNIST pixels, the LSTM performs better, achieving 98.2 % test accurracy versus 95.1% for the uRNN. However, when we permute the ordering of the pixels, the uRNN dominates with 91.4% of accuracy in contrast to the 88% of the LSTM, despite having less than a quarter of the parameters. This result is state of the art on this task, beating the IRNN (Le et al., 2015), which reaches close to 82% after 1 million training iterations. Notice that

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

26

this task, beating the IRNN (Le et al., 2015), which reaches close to 82% after 1 million training iterations. Notice that uRNN reaches convergence in less than 20 thousand iterations, while it takes the LSTM from 5 to 10 times as many to finish learning. Permuting the pixels of MNIST images creates many longer term dependencies across pixels than in the original pixel ordering, where a lot of structure is local. This makes it necessary for a network to learn and remember more complicated dependencies across varying time scales. The results suggest that the uRNN is better able to deal with such structure over the data, where the LSTM is better suited to more local sequence structure tasks. 5.4. Exploratory experiments Norms of hidden state gradients. As discussed in Section 2, key to being able to learn long term dependencies is in controlling@C @ht. With this in mind, we explored how each model propagated gradients, by examining @C @ht as a function of t. Gradient norms were computed at the beUnitary Evolution Recurrent Neural Networks Figure 4. From left to right. Norms of the gradients with respect to hidden states i.e. @C

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

27

Figure 4. From left to right. Norms of the gradients with respect to hidden states i.e. @C @ht at (i) beginning of training, (ii) after 100 iterations. (iii) Norms of the hidden states and (iv) L2distance between hidden states and final hidden state. The gradient norms of uRNNs do not decay as fast as for other models as training progresses. uRNN hidden state norms stay much more consistent over time than the LSTM. LSTM hidden states stay almost the same after a number of time steps, suggesting that it is not able to use new input information. ginning of training and again after 100 iterations of training on the adding problem. The curves are plotted in Figure 4. It is clear that at first, the uRNN propagates gradients perfectly, while each other model has exponentially vanishing gradients. After 100 iterations of training, each model experiences vanishing gradients, but the uRNN is best able to propagate information, having much less decay. Hidden state saturation. We claim that typical recurrent architectures saturate, in the sense that after they acquire some information, it becomes much more difficult to acquire further information pertaining to longer dependencies. We took the uRNN and LSTM models trained on

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

28

some information, it becomes much more difficult to acquire further information pertaining to longer dependencies. We took the uRNN and LSTM models trained on the adding problem with T= 200 , and computed a forward pass with newly generated data for the adding problem withT= 1000 . In order to show saturation effects, we plot the norms of the hidden states and the L2distance between each state and the last in Figure 4. In our experiments, it is clear that the uRNN does not suffer as much as other models do. Notice that whilst the norms of hidden states in the uRNN grow very steadily over time, in the LSTM they grow very fast, and then stay constant after about 500time steps. This behaviour may suggest that the LSTM hidden states saturate in their ability to incorporate new information, which is vital for modeling long complicated sequences. It is interesting to see that the LSTM hidden state at t= 500 , is close to that of t= 1000 , whilst this is far from the case in the uRNN. Again, this suggests that the LSTM’s capacity to use new information to alter its hidden state severly degrades with sequence length. The uRNN does not suffer from this difficulty nearly as badly. A clear example of this phenomenon was observed in the

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

29

uRNN does not suffer from this difficulty nearly as badly. A clear example of this phenomenon was observed in the adding problem with T= 750 . We found that the Pearson correlation between the LSTM output prediction and the first of the two uniform samples (whose sum is the target output) was= 0:991. This suggests that the LSTM learnt to simply find and store the first sample, as it was unable to incorporate any more information by the time it reachedthe second, due to saturation of the hidden states. 6. Discussion There are a plethora of further ideas that may be explored from our findings, both with regards to learning representation and efficient implementation. For example, one hurdle of modeling long sequences with recurrent networks is the requirement of storing all hidden state values for the purpose of gradient backpropagation. This can be prohibitive, since GPU memory is typically a limiting factor of neural network optimization. However, since our weight matrix is unitary, its inverse is its conjugate transpose, which is just as easy to operate with. If further we were to use an invertible nonlinearity function, we would no longer need to store

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

30

as easy to operate with. If further we were to use an invertible nonlinearity function, we would no longer need to store hidden states, since they can be recomputed in the backward pass. This could have potentially huge implications, as we would be able to reduce memory usage by an order ofT, the number of time steps. This would make having immensely large hidden layers possible, perhaps enabling vast memory representations. In this paper we demonstrate state of the art performance on hard problems requiring long term reasoning and memory. These results are based on a novel parameterization of unitary matrices which permit efficient matrix computations and parameter optimization. Whilst complex domain modeling has been widely succesful in the signal processing community (e.g. Fourier transforms, wavelets), we have yet to exploit the power of complex valued representation in the deep learning community. Our hope is that this work will be a step forward in this direction. We motivate the idea of unitary evolution as a novel way to mitigate the problems of vanishing and exploding gradients. Empirical evidence suggests that our uRNN is better able to pass gradient information through long sequences and does not suffer from saturating hidden states as much as LSTMs, typical RNNs, or RNNs initialized with the identity weight matrix (IRNNs).

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

31

suffer from saturating hidden states as much as LSTMs, typical RNNs, or RNNs initialized with the identity weight matrix (IRNNs). Unitary Evolution Recurrent Neural Networks Acknowledgments : We thank the developers of Theano (Bergstra et al., 2010) for their great work. We thank NSERC, Compute Canada, Canada Research Chairs and CIFAR for their support. We would also like to thank C ¸ aglar Gulc ¸ehre, David Krueger, Soroush Mehri, Marcin Moczulski, Mohammad Pezeshki and Saizheng Zhang for helpful discussions, comments and code sharing. References Bengio, Yoshua, Simard, Patrice, and Frasconi, Paolo. Learning long-term dependencies with gradient descent is difficult. IEE Transactions on Neural Networks , 5, 1994. Bergstra, James, Breuleux, Olivier, Bastien, Fr ´ed´eric, Lamblin, Pascal, Pascanu, Razvan, Desjardins, Guillaume, Turian, Joseph, Warde-Farley, David, and Bengio, Yoshua. Theano: a CPU and GPU math expression

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

32

compiler. Proceedings of the Python for Scientific Computing Conference (SciPy) , 2010. Cho, Kyunghyun, van Merri ¨enboer, Bart, Bahdanau, Dzmitry, and Bengio, Yoshua. On the properties of neural machine translation: Encoder–Decoder approaches. InEighth Workshop on Syntax, Semantics and Structure in Statistical Translation , October 2014. Collobert, Ronan, Weston, Jason, Bottou, L ´eon, Karlen, Michael, Kavukcuoglu, Koray, and Kuksa, Pavel. Natural language processing (almost) from scratch. Journal of Machine Learning Research , 12:2493–2537, 2011. Glorot, Xavier and Bengio, Yoshua. Understanding the difficulty of training deep feedforward neural networks. International Conference on Artificial Intelligence and Statistics (AISTATS) , 2010. Glorot, Xavier, Bordes, Antoine, and Bengio, Yoshua. Deep sparse rectifier neural networks. International Conference on Artificial Intelligence and Statistics (AISTATS) , 2011. Graves, Alex, Wayne, Greg, and Danihelka, Ivo. Neural

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

33

Conference on Artificial Intelligence and Statistics (AISTATS) , 2011. Graves, Alex, Wayne, Greg, and Danihelka, Ivo. Neural turing machines. arXiv preprint arXiv:1410.5401 , 2014. Hinton, Geoffrey, Deng, Li, Yu, Dong, Dahl, George, Mohamed, Abdel-rahman, Jaitly, Navdeep, Senior, Andrew, Vanhoucke, Vincent, Nguyen, Patrick, Sainath, Tara, and Kingsbury, Brian. Deep neural networks for acoustic modeling in speech recognition. Signal Processing Magazine , 2012. Hirose, Akira. Complex-valued neural networks: theories and applications , volume 5. World Scientific Publishing Company Incorporated, 2003.Hochreiter, S. Untersuchungen zu dynamischen neuronalen Netzen. Diploma thesis, T.U. M ¨unich, 1991. Hochreiter, Sepp and Schmidhuber, J ¨urgen. Long shortterm memory. Neural Computation , 8(9):1735–1780, 1997. Hoffman, Kenneth and Kunze, Ray. Linear Algebra . Pearson, second edition, 1971.

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

34

1997. Hoffman, Kenneth and Kunze, Ray. Linear Algebra . Pearson, second edition, 1971. Krizhevsky, Alex, Sutskever, Ilya, and Hinton, Geoffrey E. Imagenet classification with deep convolutional neural networks. Neural Information Processing Systems , 2012. Le, Quoc, Sarl ´os, Tam ´as, and Smola, Alex. Fastfood - approximating kernel expansions in loglinear time. International Conference on Machine Learning , 2010. Le, Quoc V ., Navdeep, Jaitly, and Hinton, Geoffrey E. A simple way to initialize recurrent networks of rectified linear units. arXiv preprint arXiv:1504.00941 , 2015. LeCun, Yann, Bottou, L ´eon, Bengio, Yoshua, and Haffner, Patrick. Gradient-based learning applied to document recognition. Proceedings of the IEEE , 1998. Martens, James and Sutskever, Ilya. Learning recurrent neural networks with hessian-free optimization. International Conference on Machine Learning , 2011. Nair, Vinod and Hinton, Geoffrey E. Rectified linear units improve restricted boltzmann machines. International Conference on Machine Learning , 2010.

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.06464

35

Nair, Vinod and Hinton, Geoffrey E. Rectified linear units improve restricted boltzmann machines. International Conference on Machine Learning , 2010. Pascanu, Razvan, Mikolov, Tomas, and Bengio, Yoshua. On the difficulty of training recurrent neural networks. International Conference on Machine Learning , 2010. Saxe, Andrew M., McLelland, James L., and Ganguli, Surya. Exact solutions to the nonlinear dynamics of learning in deep linear neural networks. International Conference in Learning Representations , 2014. Tieleman, Tijmen and Hinton, Geoffrey. Lecture 6.5rmsprop: Divide the gradient by a running average of its recent magnitude. Coursera: Neural Networks for Machine Learning , 2012. Yang, Zichao, Moczulski, Marcin, Denil, Misha, de Freitas, Nando, Smola, Alex, Song, Le, and Wang, Ziyu. Deep fried convnets. International Conference on Computer Vision (ICCV) , 2015. Zimmermann, Hans-Georg, Minin, Alexey, and Kusherbaeva, Victoria. Comparison of the complex valued and real valued neural networks trained with gradient descent and random search algorithms. In ESANN , 2011.

1511.06464

Unitary Evolution Recurrent Neural Networks

Recurrent neural networks (RNNs) are notoriously difficult to train. When the eigenvalues of the hidden to hidden weight matrix deviate from absolute value 1, optimization becomes difficult due to the well studied issue of vanishing and exploding gradients, especially when trying to learn long-term dependencies. To circumvent this problem, we propose a new architecture that learns a unitary weight matrix, with eigenvalues of absolute value exactly 1. The challenge we address is that of parametrizing unitary matrices in a way that does not require expensive computations (such as eigendecomposition) after each weight update. We construct an expressive unitary weight matrix by composing several structured matrices that act as building blocks with parameters to be learned. Optimization with this parameterization becomes feasible only when considering hidden states in the complex domain. We demonstrate the potential of this architecture by achieving state of the art results in several hard tasks involving very long-term dependencies.

http://arxiv.org/pdf/1511.06464

cs.LG

20151120

20160525

1511.08119

0

Higher Order Conditional Random Fields in Deep Neural Networks Anurag Arnab, Sadeep Jayasumana, Shuai Zheng, Philip H.S. Torr University of Oxford {firstname.lastname }@eng.ox.ac.uk Abstract. We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image’s visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

1

We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials. Keywords: Semantic Segmentation, Conditional Random Fields, Deep Learning, Convolutional Neural Networks 1 Introduction Semantic segmentation involves assigning a visual object class label to every pixel in an image, resulting in a segmentation with a semantic meaning for each segment. While a strong pixel-level classifier is critical for obtaining high accuracy in this task, it is also important to enforce the consistency of the semantic segmentation output with visual features of the image. For example, segmentation boundaries should usually coincide with strong edges in the image, and regions in the image with similar appearance should have the same label. Recent advances in deep learning have enabled researchers to create stronger classifiers, with automatically learned features, within a Convolutional Neural Network (CNN) [1 –3]. This has resulted in large improvements in semantic segmentation accuracy on widely used benchmarks such as PASCAL VOC [4]. CNN classifiers are now considered the standard choice for pixel-level classifiers used in semantic segmentation. On the other hand, probabilistic graphical models have long been popular for structured prediction of labels, with constraints enforcing label consistency. Conditional

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

2

On the other hand, probabilistic graphical models have long been popular for structured prediction of labels, with constraints enforcing label consistency. Conditional Random Fields (CRFs) have been the most common framework, and various rich and arXiv:1511.08119v4 [cs.CV] 29 Jul 2016

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

3

Higher Order Conditional Random Fields in Deep Neural Networks 3 We evaluate our higher order potentials on the PASCAL VOC 2012 semantic segmentation benchmark as well as the PASCAL Context dataset, to show significant improvements over our baseline and achieve state-of-the art results. 2 Related Work Before deep learning became prominent, semantic segmentation was performed with dense hand-crafted features which were fed into a per-pixel or region classifier [14]. The individual predictions made by these classifiers were often noisy as they lacked global context, and were thus post-processed with a CRF, making use of prior knowledge such as the fact that nearby pixels, as well as pixels of similar appearance, are likely to share the same class label [14, 15]. The CRF model of [14] initially contained only unary and pairwise terms in an 8-neighbourhood, which [16] showed can result in shrinkage bias. Numerous improvements to this model were subsequently proposed including: densely connected pairwise potentials facilitating interactions between all pairs of image pixels [17], formulating higher order potentials defined over cliques larger than two nodes [5, 16] in order to capture more context, modelling co-occurrence of object classes [18 –20], and utilising

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

4

capture more context, modelling co-occurrence of object classes [18 –20], and utilising the results of object detectors [6, 21, 22]. Recent advances in deep learning have allowed us to replace hand-crafted features with features learned specifically for semantic segmentation. The strength of these representations was illustrated by [3] who achieved significant improvements over previous hand-crafted methods without using any CRF post-processing. Chen et al. [12] showed further improvements by post-processing the results of a CNN with a CRF. Subsequent works [9 –11, 23] have taken this idea further by incorporating a CRF as layers within a deep network and then learning parameters of both the CRF and CNN together via backpropagation. In terms of enhancements to conventional CRF models, Ladicky et al. [6] proposed using an off-the-shelf object detector to provide additional cues for semantic segmentation. Unlike other approaches that refine a bounding-box detection to produce a segmentation [8, 24], this method used detector outputs as a soft constraint and can thus recover from object detection errors. Their formulation, however, used graph-cut inference, which was only tractable due to the absence of dense pairwise potentials.

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

5

thus recover from object detection errors. Their formulation, however, used graph-cut inference, which was only tractable due to the absence of dense pairwise potentials. Object detectors have also been used by [21, 25], who also modelled variables that describe the degree to which an object hypothesis is accepted. We formulate the detection potential in a different manner to [6, 21, 25] so that it is amenable to mean field inference. Mean field permits inference with dense pairwise connections, which results in substantial accuracy improvements [10,12, 17]. Furthermore, mean field updates related to our potentials are differentiable and its parameters can thus be learned in our end-to-end trainable architecture. We also note that while the semantic segmentation problem has mostly been formulated in terms of pixels [3,10,14], some have expressed it in terms of superpixels [26 –28]. Superpixels can capture more context than a single pixel and computational costs can also be reduced if one considers pairwise interactions between superpixels rather than individual pixels [21]. However, such superpixel representations assume that the segments 4 Arnab et al. share boundaries with objects in an image, which is not always true. As a result, several

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

6

4 Arnab et al. share boundaries with objects in an image, which is not always true. As a result, several authors [5, 7] have employed higher order potentials defined over superpixels that encourage label consistency over regions, but do not strictly enforce it. This approach also allows multiple, non-hierarchical layers of superpixels to be integrated. Our formulation uses this kind of higher order potential, but in an end-to-end trainable CNN. Graphical models have been used with CNNs in other areas besides semantic segmentation, such as in pose-estimation [29] and group activity recognition [30]. Alternatively, Ionescu et al. [31] incorporated structure into a deep network with structured matrix layers and matrix backpropagation. However, the nature of models used in these works is substantially different to ours. Some early works that advocated gradient backpropagation through graphical model inference for parameter optimisation include [32, 33] and [34]. Our work differentiates from the above works since, to our knowledge, we are the first to propose and conduct a thorough experimental investigation of higher order potentials that are based on detection outputs and superpixel segmentation in a CRF which is learned end-to-end in a deep network. Note that although [7] formulated mean field

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

7

learned end-to-end in a deep network. Note that although [7] formulated mean field inference with higher order potentials, they did not consider object detection potentials at all, nor were the parameters learned. 3 Conditional Random Fields We now review conditional random fields used in semantic segmentation and introduce the notation used in the paper. Take an image IwithNpixels, indexed 1,2,...,N . In semantic segmentation, we attempt to assign every pixel a label from a predefined set of labels L={l1,l2,...,l L}. Define a set of random variables X1,X2,...,X N, one for each pixel, where each Xi∈ L. LetX= [X1X2... X N]T. Any particular assignment xtoXis thus a solution to the semantic segmentation problem. We use notations {V}, andV(i)to represent the set of elements of a vector V, and theithelement of V, respectively. Given a graph Gwhere the vertices are from {X} and the edges define connections among these variables, the pair (I,X)is modelled as a

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

8

and the edges define connections among these variables, the pair (I,X)is modelled as a CRF characterised by Pr(X=x|I) = (1/Z(I)) exp( −E(x|I)), whereE(x|I)is the energy of the assignment xandZ(I)is the normalisation factor known as the partition function. We drop the conditioning on Ihereafter to keep the notation uncluttered. The energyE(x)of an assignment is defined using the set of cliques Cin the graph G. More specifically, E(x) =/summationtext c∈Cψc(xc),where xcis a vector formed by selecting elements ofxthat correspond to random variables belonging to the clique c, andψc(.)is the cost function for the clique c. The function, ψc(.), usually uses prior knowledge about a good segmentation, as well as information from the image, the observation the CRF is conditioned on. Minimising the energy yields the maximum a posteriori (MAP) labelling of the image i.e.the most probable label assignment given the observation (image). When dense pairwise potentials are used in the CRF to obtain higher accuracy, exact inference

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729

1511.08119

9

image i.e.the most probable label assignment given the observation (image). When dense pairwise potentials are used in the CRF to obtain higher accuracy, exact inference is impracticable, and one has to resort to an approximate inference method such as mean field inference [17]. Mean field inference is particularly appealing in a deep learning setting since it is possible to formulate it as a Recurrent Neural Network [10]. Higher Order Conditional Random Fields in Deep Neural Networks 5 4 CRF with Higher Order Potentials Many CRF models that have been incorporated into deep learning frameworks [10, 12] have so far used only unary and pairwise potentials. However, potentials defined on higher order cliques have been shown to be useful in previous works such as [7, 16]. The key contribution of this paper is to show that a number of explicit higher order potentials can be added to CRFs to improve image segmentation, while staying compatible with deep learning. We formulate these higher order potentials in a manner that mean field inference can still be used to solve the CRF. Advantages of mean field inference are twofold: First, it enables efficient inference when using densely-connected pairwise

1511.08119

Higher Order Conditional Random Fields in Deep Neural Networks

We address the problem of semantic segmentation using deep learning. Most segmentation systems include a Conditional Random Field (CRF) to produce a structured output that is consistent with the image's visual features. Recent deep learning approaches have incorporated CRFs into Convolutional Neural Networks (CNNs), with some even training the CRF end-to-end with the rest of the network. However, these approaches have not employed higher order potentials, which have previously been shown to significantly improve segmentation performance. In this paper, we demonstrate that two types of higher order potential, based on object detections and superpixels, can be included in a CRF embedded within a deep network. We design these higher order potentials to allow inference with the differentiable mean field algorithm. As a result, all the parameters of our richer CRF model can be learned end-to-end with our pixelwise CNN classifier. We achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark with these trainable higher order potentials.

http://arxiv.org/pdf/1511.08119

ECCV 2016

cs.CV

20151125

20160729