Datasets:

jamescalam

/

llama-2-arxiv-papers-chunked

like 17

1405.3531

44

of data augmentation similar to the CNN-based methods) we are able to improve the performance of shallow IFV to 68.02% (Table 2, [i]). 4.8 Performance Evolution on VOC-2007 A comparative plot of the evolution in the performance of the methods evaluated in this paper, along with a selection from our earlier review of shallow methods [12] is presented in Fig. 1. Classification accuracy over PASCAL VOC was 54.48% mAP for the BoVW model in 2008, 61.7% for the 10 BOW 32K –FK-BL 327K – [I]FK 327K – [a]FK-IN 84K f s [i]DeCAF 4K t t [II]CNN-F 4K f s [m]CNN-M 2K 2K f s [y]CNN-S 4K (TN) f s [γ] 545658606264666870727476788082mAP 68.02 54.4861.6963.6673.4177.1580.132008 2010 2013 2014 ... 82.42 Method Dim. Aug. Ref.

1405.3531

Return of the Devil in the Details: Delving Deep into Convolutional Nets

The latest generation of Convolutional Neural Networks (CNN) have achieved impressive results in challenging benchmarks on image recognition and object detection, significantly raising the interest of the community in these methods. Nevertheless, it is still unclear how different CNN methods compare with each other and with previous state-of-the-art shallow representations such as the Bag-of-Visual-Words and the Improved Fisher Vector. This paper conducts a rigorous evaluation of these new techniques, exploring different deep architectures and comparing them on a common ground, identifying and disclosing important implementation details. We identify several useful properties of CNN-based representations, including the fact that the dimensionality of the CNN output layer can be reduced significantly without having an adverse effect on performance. We also identify aspects of deep and shallow methods that can be successfully shared. In particular, we show that the data augmentation techniques commonly applied to CNN-based methods can also be applied to shallow methods, and result in an analogous performance boost. Source code and models to reproduce the experiments in the paper is made publicly available.

http://arxiv.org/pdf/1405.3531

Published in proceedings of BMVC 2014

cs.CV

20140514

20141105

1405.3531

45

54.4861.6963.6673.4177.1580.132008 2010 2013 2014 ... 82.42 Method Dim. Aug. Ref. Fig. 1. Evolution of Performance on PASCAL VOC-2007 over the recent years. Please refer to Table 2 for details and references. IFV in 2010 [12], and 73.41% for DeCAF [7] and similar [8], [9] CNN-based methods introduced in late 2013. Our best performing CNN-based method (CNN-S with fine-tuning) achieves 82.42%, comparable to the most recent state-of-the-art. 4.9 Timings and dimensionality One of our best-performing CNN representations CNN-M-2048 [x] is 42more compact than the best performing IFV [i] (84K vs. 2K) and CNN-M features are also 50faster to compute ( 120s vs.2:4sper image with augmentation enabled, over a single CPU core). Non-augmented CNN-M features [o] take around 0:3sper image, compared to0:4sfor CNN-S features and 0:13sfor CNNF features. 5 C ONCLUSION

1405.3531

Return of the Devil in the Details: Delving Deep into Convolutional Nets

The latest generation of Convolutional Neural Networks (CNN) have achieved impressive results in challenging benchmarks on image recognition and object detection, significantly raising the interest of the community in these methods. Nevertheless, it is still unclear how different CNN methods compare with each other and with previous state-of-the-art shallow representations such as the Bag-of-Visual-Words and the Improved Fisher Vector. This paper conducts a rigorous evaluation of these new techniques, exploring different deep architectures and comparing them on a common ground, identifying and disclosing important implementation details. We identify several useful properties of CNN-based representations, including the fact that the dimensionality of the CNN output layer can be reduced significantly without having an adverse effect on performance. We also identify aspects of deep and shallow methods that can be successfully shared. In particular, we show that the data augmentation techniques commonly applied to CNN-based methods can also be applied to shallow methods, and result in an analogous performance boost. Source code and models to reproduce the experiments in the paper is made publicly available.

http://arxiv.org/pdf/1405.3531

Published in proceedings of BMVC 2014

cs.CV

20140514

20141105

1405.3531

46

to0:4sfor CNN-S features and 0:13sfor CNNF features. 5 C ONCLUSION In this paper we presented a rigorous empirical evaluation of CNN-based methods for image classification, along with a comparison with more traditional shallow feature encoding methods. We have demonstrated that the performance of shallow representations can be significantly improved by adopting data augmentation, typically used indeep learning. In spite of this improvement, deep architectures still outperform the shallow methods by a large margin. We have shown that the performance of deep representations on the ILSVRC dataset is a good indicator of their performance on other datasets, and that fine-tuning can further improve on already very strong results achieved using the combination of deep representations and a linear SVM. Source code and CNN models to reproduce the experiments presented in the paper are available on the project website [32] in the hope that it would provide common ground for future comparisons, and good baselines for image representation research. ACKNOWLEDGEMENTS This work was supported by the EPSRC and ERC grant VisRec no. 228180. We gratefully acknowledge the support of NVIDIA Corporation with the donation of the GPUs used for this research. REFERENCES

1405.3531

Return of the Devil in the Details: Delving Deep into Convolutional Nets

The latest generation of Convolutional Neural Networks (CNN) have achieved impressive results in challenging benchmarks on image recognition and object detection, significantly raising the interest of the community in these methods. Nevertheless, it is still unclear how different CNN methods compare with each other and with previous state-of-the-art shallow representations such as the Bag-of-Visual-Words and the Improved Fisher Vector. This paper conducts a rigorous evaluation of these new techniques, exploring different deep architectures and comparing them on a common ground, identifying and disclosing important implementation details. We identify several useful properties of CNN-based representations, including the fact that the dimensionality of the CNN output layer can be reduced significantly without having an adverse effect on performance. We also identify aspects of deep and shallow methods that can be successfully shared. In particular, we show that the data augmentation techniques commonly applied to CNN-based methods can also be applied to shallow methods, and result in an analogous performance boost. Source code and models to reproduce the experiments in the paper is made publicly available.

http://arxiv.org/pdf/1405.3531

Published in proceedings of BMVC 2014

cs.CV

20140514

20141105

1405.3531

47

grant VisRec no. 228180. We gratefully acknowledge the support of NVIDIA Corporation with the donation of the GPUs used for this research. REFERENCES [1] G. Csurka, C. Bray, C. Dance, and L. Fan, “Visual categorization with bags of keypoints,” in Workshop on Statistical Learning in Computer Vision, ECCV , 2004, pp. 1–22. 11 [2] J. Sivic and A. Zisserman, “Video Google: A text retrieval approach to object matching in videos,” in Proc. ICCV , vol. 2, 2003, pp. 1470–1477. [3] F. Perronnin, J. S ´anchez, and T. Mensink, “Improving the Fisher kernel for large-scale image classification,” in Proc. ECCV , 2010. [4] Y. LeCun, B. Boser, J. S. Denker, D. Henderson, R. E. Howard, W. Hubbard, and L. D. Jackel, “Backpropagation applied to handwritten zip code recognition,” Neural Computation , vol. 1, no. 4, pp. 541–551, 1989.

1405.3531

Return of the Devil in the Details: Delving Deep into Convolutional Nets

The latest generation of Convolutional Neural Networks (CNN) have achieved impressive results in challenging benchmarks on image recognition and object detection, significantly raising the interest of the community in these methods. Nevertheless, it is still unclear how different CNN methods compare with each other and with previous state-of-the-art shallow representations such as the Bag-of-Visual-Words and the Improved Fisher Vector. This paper conducts a rigorous evaluation of these new techniques, exploring different deep architectures and comparing them on a common ground, identifying and disclosing important implementation details. We identify several useful properties of CNN-based representations, including the fact that the dimensionality of the CNN output layer can be reduced significantly without having an adverse effect on performance. We also identify aspects of deep and shallow methods that can be successfully shared. In particular, we show that the data augmentation techniques commonly applied to CNN-based methods can also be applied to shallow methods, and result in an analogous performance boost. Source code and models to reproduce the experiments in the paper is made publicly available.

http://arxiv.org/pdf/1405.3531

Published in proceedings of BMVC 2014

cs.CV

20140514

20141105

1405.3531

48

Computation , vol. 1, no. 4, pp. 541–551, 1989. [5] J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei, “Imagenet: A large-scale hierarchical image database,” in Proc. CVPR , 2009. [6] M. Everingham, L. Van Gool, C. K. I. Williams, J. Winn, and A. Zisserman, “The PASCAL Visual Object Classes (VOC) challenge,” IJCV , vol. 88, no. 2, pp. 303–338, 2010. [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 , vol. abs/1310.1531, 2013. [8] M. Oquab, L. Bottou, I. Laptev, and J. Sivic, “Learning and Transferring Mid-Level Image Representations using Convolutional Neural Networks,” in Proc. CVPR , 2014.

1405.3531

Return of the Devil in the Details: Delving Deep into Convolutional Nets

The latest generation of Convolutional Neural Networks (CNN) have achieved impressive results in challenging benchmarks on image recognition and object detection, significantly raising the interest of the community in these methods. Nevertheless, it is still unclear how different CNN methods compare with each other and with previous state-of-the-art shallow representations such as the Bag-of-Visual-Words and the Improved Fisher Vector. This paper conducts a rigorous evaluation of these new techniques, exploring different deep architectures and comparing them on a common ground, identifying and disclosing important implementation details. We identify several useful properties of CNN-based representations, including the fact that the dimensionality of the CNN output layer can be reduced significantly without having an adverse effect on performance. We also identify aspects of deep and shallow methods that can be successfully shared. In particular, we show that the data augmentation techniques commonly applied to CNN-based methods can also be applied to shallow methods, and result in an analogous performance boost. Source code and models to reproduce the experiments in the paper is made publicly available.

http://arxiv.org/pdf/1405.3531

Published in proceedings of BMVC 2014

cs.CV

20140514

20141105

1405.3531

49

and Transferring Mid-Level Image Representations using Convolutional Neural Networks,” in Proc. CVPR , 2014. [9] A. Razavian, H. Azizpour, J. Sullivan, and S. Carlsson, “CNN Features off-the-shelf: an Astounding Baseline for Recognition,” CoRR , vol. abs/1403.6382, 2014. [10] P . Sermanet, D. Eigen, X. Zhang, M. Mathieu, R. Fergus, and Y. LeCun, “OverFeat: Integrated Recognition, Localization and Detection using Convolutional Networks,” in Proc. ICLR , 2014. [11] R. B. Girshick, J. Donahue, T. Darrell, and J. Malik, “Rich feature hierarchies for accurate object detection and semantic segmentation,” in Proc. CVPR , 2014. [12] K. Chatfield, V . Lempitsky, A. Vedaldi, and A. Zisserman, “The devil is in the details: an evaluation of recent feature encoding methods,” in Proc. BMVC. , 2011.

1405.3531

Return of the Devil in the Details: Delving Deep into Convolutional Nets

The latest generation of Convolutional Neural Networks (CNN) have achieved impressive results in challenging benchmarks on image recognition and object detection, significantly raising the interest of the community in these methods. Nevertheless, it is still unclear how different CNN methods compare with each other and with previous state-of-the-art shallow representations such as the Bag-of-Visual-Words and the Improved Fisher Vector. This paper conducts a rigorous evaluation of these new techniques, exploring different deep architectures and comparing them on a common ground, identifying and disclosing important implementation details. We identify several useful properties of CNN-based representations, including the fact that the dimensionality of the CNN output layer can be reduced significantly without having an adverse effect on performance. We also identify aspects of deep and shallow methods that can be successfully shared. In particular, we show that the data augmentation techniques commonly applied to CNN-based methods can also be applied to shallow methods, and result in an analogous performance boost. Source code and models to reproduce the experiments in the paper is made publicly available.

http://arxiv.org/pdf/1405.3531

Published in proceedings of BMVC 2014

cs.CV

20140514

20141105

1405.3531

50

“The devil is in the details: an evaluation of recent feature encoding methods,” in Proc. BMVC. , 2011. [13] A. Krizhevsky, I. Sutskever, and G. E. Hinton, “ImageNet classification with deep convolutional neural networks,” inNIPS , 2012, pp. 1106–1114. [14] M. Paulin, J. Revaud, Z. Harchaoui, F. Perronnin, and C. Schmid, “Transformation Pursuit for Image Classification,” in Proc. CVPR , 2014. [15] F. Perronnin, Z. Akata, Z. Harchaoui, and C. Schmid, “Towards good practice in large-scale learning for image classification,” in Proc. CVPR , 2012, pp. 3482–3489. [16] M. Oquab, L. Bottou, I. Laptev, and J. Sivic, “Weakly supervised object recognition with convolutional neural networks,” INRIA, Tech. Rep. HAL-01015140, 2014.

1405.3531

Return of the Devil in the Details: Delving Deep into Convolutional Nets

The latest generation of Convolutional Neural Networks (CNN) have achieved impressive results in challenging benchmarks on image recognition and object detection, significantly raising the interest of the community in these methods. Nevertheless, it is still unclear how different CNN methods compare with each other and with previous state-of-the-art shallow representations such as the Bag-of-Visual-Words and the Improved Fisher Vector. This paper conducts a rigorous evaluation of these new techniques, exploring different deep architectures and comparing them on a common ground, identifying and disclosing important implementation details. We identify several useful properties of CNN-based representations, including the fact that the dimensionality of the CNN output layer can be reduced significantly without having an adverse effect on performance. We also identify aspects of deep and shallow methods that can be successfully shared. In particular, we show that the data augmentation techniques commonly applied to CNN-based methods can also be applied to shallow methods, and result in an analogous performance boost. Source code and models to reproduce the experiments in the paper is made publicly available.

http://arxiv.org/pdf/1405.3531

Published in proceedings of BMVC 2014

cs.CV

20140514

20141105

1405.3531

51

supervised object recognition with convolutional neural networks,” INRIA, Tech. Rep. HAL-01015140, 2014. [17] Y. Wei, W. Xia, J. Huang, B. Ni, J. Dong, Y. Zhao, and S. Yan, “CNN: Single-label to multi-label,” CoRR , vol. abs/1406.5726, 2014. [18] H. J ´egou, F. Perronnin, M. Douze, J. S ´anchez, P . P ´erez, and C. Schmid, “Aggregating local images descriptors into compact codes,” IEEE P AMI , 2012. [19] M. D. Zeiler and R. Fergus, “Visualizing and understanding convolutional networks,” CoRR , vol. abs/1311.2901, 2013. [20] Y. Jia, “Caffe: An open source convolutional architecture for fast feature embedding,” http://caffe.berkeleyvision. org/, 2013. [21] L. Fei-Fei, R. Fergus, and P . Perona, “Learning generative

1405.3531

Return of the Devil in the Details: Delving Deep into Convolutional Nets

The latest generation of Convolutional Neural Networks (CNN) have achieved impressive results in challenging benchmarks on image recognition and object detection, significantly raising the interest of the community in these methods. Nevertheless, it is still unclear how different CNN methods compare with each other and with previous state-of-the-art shallow representations such as the Bag-of-Visual-Words and the Improved Fisher Vector. This paper conducts a rigorous evaluation of these new techniques, exploring different deep architectures and comparing them on a common ground, identifying and disclosing important implementation details. We identify several useful properties of CNN-based representations, including the fact that the dimensionality of the CNN output layer can be reduced significantly without having an adverse effect on performance. We also identify aspects of deep and shallow methods that can be successfully shared. In particular, we show that the data augmentation techniques commonly applied to CNN-based methods can also be applied to shallow methods, and result in an analogous performance boost. Source code and models to reproduce the experiments in the paper is made publicly available.

http://arxiv.org/pdf/1405.3531

Published in proceedings of BMVC 2014

cs.CV

20140514

20141105

1405.3531

52

org/, 2013. [21] L. Fei-Fei, R. Fergus, and P . Perona, “Learning generative visual models from few training examples: An incremental bayesian approach tested on 101 object categories,” inIEEE CVPR Workshop of Generative Model Based Vision , 2004.[22] G. Griffin, A. Holub, and P . Perona, “Caltech-256 object category dataset,” California Institute of Technology, Tech. Rep. 7694, 2007. [Online]. Available: http:// authors.library.caltech.edu/7694 [23] J. S ´anchez, F. Perronnin, and T. Em ´ıdio de Campos, “Modeling the spatial layout of images beyond spatial pyramids,” Pattern Recognition Letters , vol. 33, no. 16, pp. 2216–2223, 2012. [24] R. Arandjelovi ´c and A. Zisserman, “Three things everyone should know to improve object retrieval,” in Proc. CVPR , 2012. [25] A. Vedaldi and A. Zisserman, “Efficient additive kernels via explicit feature maps,” IEEE P AMI , 2011.

1405.3531

Return of the Devil in the Details: Delving Deep into Convolutional Nets

The latest generation of Convolutional Neural Networks (CNN) have achieved impressive results in challenging benchmarks on image recognition and object detection, significantly raising the interest of the community in these methods. Nevertheless, it is still unclear how different CNN methods compare with each other and with previous state-of-the-art shallow representations such as the Bag-of-Visual-Words and the Improved Fisher Vector. This paper conducts a rigorous evaluation of these new techniques, exploring different deep architectures and comparing them on a common ground, identifying and disclosing important implementation details. We identify several useful properties of CNN-based representations, including the fact that the dimensionality of the CNN output layer can be reduced significantly without having an adverse effect on performance. We also identify aspects of deep and shallow methods that can be successfully shared. In particular, we show that the data augmentation techniques commonly applied to CNN-based methods can also be applied to shallow methods, and result in an analogous performance boost. Source code and models to reproduce the experiments in the paper is made publicly available.

http://arxiv.org/pdf/1405.3531

Published in proceedings of BMVC 2014

cs.CV

20140514

20141105

1405.3531

53

[25] A. Vedaldi and A. Zisserman, “Efficient additive kernels via explicit feature maps,” IEEE P AMI , 2011. [26] S. Lazebnik, C. Schmid, and J. Ponce, “Beyond Bags of Features: Spatial Pyramid Matching for Recognizing Natural Scene Categories,” in Proc. CVPR , 2006. [27] R. Arandjelovi ´c and A. Zisserman, “All about VLAD,” in Proc. CVPR , 2013. [28] H. J ´egou, M. Douze, and C. Schmid, “On the burstiness of visual elements,” in Proc. CVPR , Jun 2009. [29] K. He, A. Zhang, S. Ren, and J. Sun, “Spatial pyramid pooling in deep convolutional networks for visual recognition,” in Proc. ECCV , 2014. [30] M.-M. Cheng, Z. Zhang, W.-Y. Lin, and P . H. S. Torr, “BING: Binarized normed gradients for objectness estimation at 300fps,” in Proc. CVPR , 2014.

1405.3531

Return of the Devil in the Details: Delving Deep into Convolutional Nets

The latest generation of Convolutional Neural Networks (CNN) have achieved impressive results in challenging benchmarks on image recognition and object detection, significantly raising the interest of the community in these methods. Nevertheless, it is still unclear how different CNN methods compare with each other and with previous state-of-the-art shallow representations such as the Bag-of-Visual-Words and the Improved Fisher Vector. This paper conducts a rigorous evaluation of these new techniques, exploring different deep architectures and comparing them on a common ground, identifying and disclosing important implementation details. We identify several useful properties of CNN-based representations, including the fact that the dimensionality of the CNN output layer can be reduced significantly without having an adverse effect on performance. We also identify aspects of deep and shallow methods that can be successfully shared. In particular, we show that the data augmentation techniques commonly applied to CNN-based methods can also be applied to shallow methods, and result in an analogous performance boost. Source code and models to reproduce the experiments in the paper is made publicly available.

http://arxiv.org/pdf/1405.3531

Published in proceedings of BMVC 2014

cs.CV

20140514

20141105

1405.3531

54

“BING: Binarized normed gradients for objectness estimation at 300fps,” in Proc. CVPR , 2014. [31] S. Yan, J. Dong, Q. Chen, Z. Song, Y. Pan, W. Xia, H. Zhongyang, Y. Hua, and S. Shen, “Generalized hierarchical matching for subcategory aware object classification,” in The P ASCAL Visual Object Classes Challenge Workshop , 2012. [32] K. Chatfield, K. Simonyan, A. Vedaldi, and A. Zisserman, “Return of the devil in the details: delving deep into convolutional nets webpage,” 2014. [Online]. Available: http://www.robots.ox.ac.uk/vgg/research/deep eval

1405.3531

Return of the Devil in the Details: Delving Deep into Convolutional Nets

The latest generation of Convolutional Neural Networks (CNN) have achieved impressive results in challenging benchmarks on image recognition and object detection, significantly raising the interest of the community in these methods. Nevertheless, it is still unclear how different CNN methods compare with each other and with previous state-of-the-art shallow representations such as the Bag-of-Visual-Words and the Improved Fisher Vector. This paper conducts a rigorous evaluation of these new techniques, exploring different deep architectures and comparing them on a common ground, identifying and disclosing important implementation details. We identify several useful properties of CNN-based representations, including the fact that the dimensionality of the CNN output layer can be reduced significantly without having an adverse effect on performance. We also identify aspects of deep and shallow methods that can be successfully shared. In particular, we show that the data augmentation techniques commonly applied to CNN-based methods can also be applied to shallow methods, and result in an analogous performance boost. Source code and models to reproduce the experiments in the paper is made publicly available.

http://arxiv.org/pdf/1405.3531

Published in proceedings of BMVC 2014

cs.CV

20140514

20141105

1406.1078

0

Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation Kyunghyun Cho Bart van Merri ¨enboer Caglar Gulcehre Universit ´e de Montr ´eal firstname.lastname@umontreal.caDzmitry Bahdanau Jacobs University, Germany d.bahdanau@jacobs-university.de Fethi Bougares Holger Schwenk Universit ´e du Maine, France firstname.lastname@lium.univ-lemans.frYoshua Bengio Universit ´e de Montr ´eal, CIFAR Senior Fellow find.me@on.the.web Abstract In this paper, we propose a novel neural network model called RNN Encoder– Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixedlength vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

1

decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder–Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases. 1 Introduction Deep neural networks have shown great success in various applications such as objection recognition (see, e.g., (Krizhevsky et al., 2012)) and speech recognition (see, e.g., (Dahl et al., 2012)). Furthermore, many recent works showed that neural networks can be successfully used in a number of tasks in natural language processing (NLP). These include, but are not limited to, language modeling (Bengio et al., 2003), paraphrase detection (Socher et al., 2011) and word embedding extraction (Mikolov et al., 2013). In the field of statistical machine translation (SMT), deep neural networks have begun to show promising results. (Schwenk, 2012) summarizes a successful usage of feedforward neural networks in the framework

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

2

networks have begun to show promising results. (Schwenk, 2012) summarizes a successful usage of feedforward neural networks in the framework of phrase-based SMT system.Along this line of research on using neural networks for SMT, this paper focuses on a novel neural network architecture that can be used as a part of the conventional phrase-based SMT system. The proposed neural network architecture, which we will refer to as an RNN Encoder–Decoder , consists of two recurrent neural networks (RNN) that act as an encoder and a decoder pair. The encoder maps a variable-length source sequence to a fixed-length vector, and the decoder maps the vector representation back to a variable-length target sequence. The two networks are trained jointly to maximize the conditional probability of the target sequence given a source sequence. Additionally, we propose to use a rather sophisticated hidden unit in order to improve both the memory capacity and the ease of training. The proposed RNN Encoder–Decoder with a novel hidden unit is empirically evaluated on the task of translating from English to French. We train the model to learn the translation probability of an English phrase to a corresponding French phrase. The model is then used as a part of a standard phrase-based SMT system by scoring each

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

3

train the model to learn the translation probability of an English phrase to a corresponding French phrase. The model is then used as a part of a standard phrase-based SMT system by scoring each phrase pair in the phrase table. The empirical evaluation reveals that this approach of scoring phrase pairs with an RNN Encoder–Decoder improves the translation performance. We qualitatively analyze the trained RNN Encoder–Decoder by comparing its phrase scores with those given by the existing translation model. The qualitative analysis shows that the RNN Encoder–Decoder is better at capturing the linguistic regularities in the phrase table, indirectly explaining the quantitative improvements in the overall translation performance. The further analysis of the model reveals that the RNN Encoder– Decoder learns a continuous space representation of a phrase that preserves both the semantic and syntactic structure of the phrase.arXiv:1406.1078v3 [cs.CL] 3 Sep 2014 2 RNN Encoder–Decoder 2.1 Preliminary: Recurrent Neural Networks A recurrent neural network (RNN) is a neural network that consists of a hidden state hand an optional output ywhich operates on a variablelength sequence x= (x1;:::;x T). At each time stept, the hidden state hhtiof the RNN is updated by

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

4

stept, the hidden state hhtiof the RNN is updated by hhti=f hht1i;xt
; (1) wherefis a non-linear activation function.fmay be as simple as an elementwise logistic sigmoid function and as complex as a long short-term memory (LSTM) unit (Hochreiter and Schmidhuber, 1997). An RNN can learn a probability distribution over a sequence by being trained to predict the next symbol in a sequence. In that case, the output at each timestep tis the conditional distribution p(xtjxt1;:::;x 1). For example, a multinomial distribution ( 1-of-Kcoding) can be output using a softmax activation function p(xt;j= 1jxt1;:::;x 1) =exp wjhhti
PK j0=1exp wj0hhti
; (2) for all possible symbols j= 1;:::;K , where wj are the rows of a weight matrix W. By combining these probabilities, we can compute the probability of the sequence xusing p(x) =TY

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

5

are the rows of a weight matrix W. By combining these probabilities, we can compute the probability of the sequence xusing p(x) =TY t=1p(xtjxt1;:::;x 1): (3) From this learned distribution, it is straightforward to sample a new sequence by iteratively sampling a symbol at each time step. 2.2 RNN Encoder–Decoder In this paper, we propose a novel neural network architecture that learns to encode a variable-length sequence into a fixed-length vector representation and to decode a given fixed-length vector representation back into a variable-length sequence. From a probabilistic perspective, this new model is a general method to learn the conditional distribution over a variable-length sequence conditioned on yet another variable-length sequence, e.g.p(y1;:::;y T0jx1;:::;x T), where one x1x2 xTyT' y2 y1 cDecoder EncoderFigure 1: An illustration of the proposed RNN Encoder–Decoder. should note that the input and output sequence lengthsTandT0may differ. The encoder is an RNN that reads each symbol of an input sequence xsequentially. As it reads

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

6

lengthsTandT0may differ. The encoder is an RNN that reads each symbol of an input sequence xsequentially. As it reads each symbol, the hidden state of the RNN changes according to Eq. (1). After reading the end of the sequence (marked by an end-of-sequence symbol), the hidden state of the RNN is a summary c of the whole input sequence. The decoder of the proposed model is another RNN which is trained to generate the output sequence by predicting the next symbol ytgiven the hidden state hhti. However, unlike the RNN described in Sec. 2.1, both ytandhhtiare also conditioned onyt1and on the summary cof the input sequence. Hence, the hidden state of the decoder at timetis computed by, hhti=f hht1i;yt1;c
; and similarly, the conditional distribution of the next symbol is P(ytjyt1;yt2;:::;y 1;c) =g hhti;yt1;c
: for given activation functions fandg(the latter must produce valid probabilities, e.g. with a softmax). See Fig. 1 for a graphical depiction of the proposed model architecture.

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

7

for given activation functions fandg(the latter must produce valid probabilities, e.g. with a softmax). See Fig. 1 for a graphical depiction of the proposed model architecture. The two components of the proposed RNN Encoder–Decoder are jointly trained to maximize the conditional log-likelihood max 1 NNX n=1logp(ynjxn); (4) where is the set of the model parameters and each (xn;yn)is an (input sequence, output sequence) pair from the training set. In our case, as the output of the decoder, starting from the input, is differentiable, we can use a gradient-based algorithm to estimate the model parameters. Once the RNN Encoder–Decoder is trained, the model can be used in two ways. One way is to use the model to generate a target sequence given an input sequence. On the other hand, the model can be used to score a given pair of input and output sequences, where the score is simply a probability p(yjx)from Eqs. (3) and (4). 2.3 Hidden Unit that Adaptively Remembers and Forgets In addition to a novel model architecture, we also propose a new type of hidden unit ( fin Eq. (1))

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

8

and Forgets In addition to a novel model architecture, we also propose a new type of hidden unit ( fin Eq. (1)) that has been motivated by the LSTM unit but is much simpler to compute and implement.1Fig. 2 shows the graphical depiction of the proposed hidden unit. Let us describe how the activation of the j-th hidden unit is computed. First, the reset gaterjis computed by rj= [Wrx]j+ Urhht1i j ; (5) whereis the logistic sigmoid function, and [:]j denotes the j-th element of a vector. xandht1 are the input and the previous hidden state, respectively. WrandUrare weight matrices which are learned. Similarly, the update gatezjis computed by zj= [Wzx]j+ Uzhht1i j : (6) The actual activation of the proposed unit hjis then computed by hhti j=zjhht1i j + (1zj)~hhti j; (7) where ~hhti j= [Wx]j+ U

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

9

j; (7) where ~hhti j= [Wx]j+ U rhht1i
 j :(8) In this formulation, when the reset gate is close to 0, the hidden state is forced to ignore the previous hidden state and reset with the current input 1The LSTM unit, which has shown impressive results in several applications such as speech recognition, has a memory cell and four gating units that adaptively control the information flow inside the unit, compared to only two gating units in the proposed hidden unit. For details on LSTM networks, see, e.g., (Graves, 2012). z rh h~xFigure 2: An illustration of the proposed hidden activation function. The update gate zselects whether the hidden state is to be updated with a new hidden state ~h. The reset gate rdecides whether the previous hidden state is ignored. See Eqs. (5)–(8) for the detailed equations of r,z,h and~h. only. This effectively allows the hidden state to drop any information that is found to be irrelevant later in the future, thus, allowing a more compact representation. On the other hand, the update gate controls how much information from the previous hidden state

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

10

later in the future, thus, allowing a more compact representation. On the other hand, the update gate controls how much information from the previous hidden state will carry over to the current hidden state. This acts similarly to the memory cell in the LSTM network and helps the RNN to remember longterm information. Furthermore, this may be considered an adaptive variant of a leaky-integration unit (Bengio et al., 2013). As each hidden unit has separate reset and update gates, each hidden unit will learn to capture dependencies over different time scales. Those units that learn to capture short-term dependencies will tend to have reset gates that are frequently active, but those that capture longer-term dependencies will have update gates that are mostly active. In our preliminary experiments, we found that it is crucial to use this new unit with gating units. We were not able to get meaningful result with an oft-used tanh unit without any gating. 3 Statistical Machine Translation In a commonly used statistical machine translation system (SMT), the goal of the system (decoder, specifically) is to find a translation fgiven a source sentence e, which maximizes p(fje)/p(ejf)p(f); where the first term at the right hand side is called

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

11

sentence e, which maximizes p(fje)/p(ejf)p(f); where the first term at the right hand side is called translation model and the latter language model (see, e.g., (Koehn, 2005)). In practice, however, most SMT systems model logp(fje)as a loglinear model with additional features and corresponding weights: logp(fje) =NX n=1wnfn(f;e) + logZ(e);(9) wherefnandwnare then-th feature and weight, respectively. Z(e)is a normalization constant that does not depend on the weights. The weights are often optimized to maximize the BLEU score on a development set. In the phrase-based SMT framework introduced in (Koehn et al., 2003) and (Marcu and Wong, 2002), the translation model logp(ejf)is factorized into the translation probabilities of matching phrases in the source and target sentences.2These probabilities are once again considered additional features in the log-linear model (see Eq. (9)) and are weighted accordingly to maximize the BLEU score. Since the neural net language model was proposed in (Bengio et al., 2003), neural networks

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

12

accordingly to maximize the BLEU score. Since the neural net language model was proposed in (Bengio et al., 2003), neural networks have been used widely in SMT systems. In many cases, neural networks have been used to rescore translation hypotheses ( n-best lists) (see, e.g., (Schwenk et al., 2006)). Recently, however, there has been interest in training neural networks to score the translated sentence (or phrase pairs) using a representation of the source sentence as an additional input. See, e.g., (Schwenk, 2012), (Son et al., 2012) and (Zou et al., 2013). 3.1 Scoring Phrase Pairs with RNN Encoder–Decoder Here we propose to train the RNN Encoder– Decoder (see Sec. 2.2) on a table of phrase pairs and use its scores as additional features in the loglinear model in Eq. (9) when tuning the SMT decoder. When we train the RNN Encoder–Decoder, we ignore the (normalized) frequencies of each phrase pair in the original corpora. This measure was taken in order (1) to reduce the computational expense of randomly selecting phrase pairs from a

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

13

pair in the original corpora. This measure was taken in order (1) to reduce the computational expense of randomly selecting phrase pairs from a large phrase table according to the normalized frequencies and (2) to ensure that the RNN Encoder– Decoder does not simply learn to rank the phrase pairs according to their numbers of occurrences. One underlying reason for this choice was that the existing translation probability in the phrase table already reflects the frequencies of the phrase 2Without loss of generality, from here on, we refer to p(ejf)for each phrase pair as a translation model as wellpairs in the original corpus. With a fixed capacity of the RNN Encoder–Decoder, we try to ensure that most of the capacity of the model is focused toward learning linguistic regularities, i.e., distinguishing between plausible and implausible translations, or learning the “manifold” (region of probability concentration) of plausible translations. Once the RNN Encoder–Decoder is trained, we add a new score for each phrase pair to the existing phrase table. This allows the new scores to enter into the existing tuning algorithm with minimal additional overhead in computation. As Schwenk pointed out in (Schwenk, 2012), it is possible to completely replace the existing

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

14

additional overhead in computation. As Schwenk pointed out in (Schwenk, 2012), it is possible to completely replace the existing phrase table with the proposed RNN Encoder– Decoder. In that case, for a given source phrase, the RNN Encoder–Decoder will need to generate a list of (good) target phrases. This requires, however, an expensive sampling procedure to be performed repeatedly. In this paper, thus, we only consider rescoring the phrase pairs in the phrase table. 3.2 Related Approaches: Neural Networks in Machine Translation Before presenting the empirical results, we discuss a number of recent works that have proposed to use neural networks in the context of SMT. Schwenk in (Schwenk, 2012) proposed a similar approach of scoring phrase pairs. Instead of the RNN-based neural network, he used a feedforward neural network that has fixed-size inputs (7 words in his case, with zero-padding for shorter phrases) and fixed-size outputs (7 words in the target language). When it is used specifically for scoring phrases for the SMT system, the maximum phrase length is often chosen to be small. However, as the length of phrases increases or as we apply neural

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

15

phrases for the SMT system, the maximum phrase length is often chosen to be small. However, as the length of phrases increases or as we apply neural networks to other variable-length sequence data, it is important that the neural network can handle variable-length input and output. The proposed RNN Encoder–Decoder is well-suited for these applications. Similar to (Schwenk, 2012), Devlin et al. (Devlin et al., 2014) proposed to use a feedforward neural network to model a translation model, however, by predicting one word in a target phrase at a time. They reported an impressive improvement, but their approach still requires the maximum length of the input phrase (or context words) to be fixed a priori. Although it is not exactly a neural network they train, the authors of (Zou et al., 2013) proposed to learn a bilingual embedding of words/phrases. They use the learned embedding to compute the distance between a pair of phrases which is used as an additional score of the phrase pair in an SMT system. In (Chandar et al., 2014), a feedforward neural network was trained to learn a mapping from a bag-of-words representation of an input phrase to an output phrase. This is closely related to both the

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

16

network was trained to learn a mapping from a bag-of-words representation of an input phrase to an output phrase. This is closely related to both the proposed RNN Encoder–Decoder and the model proposed in (Schwenk, 2012), except that their input representation of a phrase is a bag-of-words. A similar approach of using bag-of-words representations was proposed in (Gao et al., 2013) as well. Earlier, a similar encoder–decoder model using two recursive neural networks was proposed in (Socher et al., 2011), but their model was restricted to a monolingual setting, i.e. the model reconstructs an input sentence. More recently, another encoder–decoder model using an RNN was proposed in (Auli et al., 2013), where the decoder is conditioned on a representation of either a source sentence or a source context. One important difference between the proposed RNN Encoder–Decoder and the approaches in (Zou et al., 2013) and (Chandar et al., 2014) is that the order of the words in source and target phrases is taken into account. The RNN Encoder–Decoder naturally distinguishes between sequences that have the same words but in a different order, whereas the aforementioned approaches effectively ignore order information. The closest approach related to the proposed

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

17

sequences that have the same words but in a different order, whereas the aforementioned approaches effectively ignore order information. The closest approach related to the proposed RNN Encoder–Decoder is the Recurrent Continuous Translation Model (Model 2) proposed in (Kalchbrenner and Blunsom, 2013). In their paper, they proposed a similar model that consists of an encoder and decoder. The difference with our model is that they used a convolutional n-gram model (CGM) for the encoder and the hybrid of an inverse CGM and a recurrent neural network for the decoder. They, however, evaluated their model on rescoring the n-best list proposed by the conventional SMT system and computing the perplexity of the gold standard translations. 4 Experiments We evaluate our approach on the English/French translation task of the WMT’14 workshop.4.1 Data and Baseline System Large amounts of resources are available to build an English/French SMT system in the framework of the WMT’14 translation task. The bilingual corpora include Europarl (61M words), news commentary (5.5M), UN (421M), and two crawled corpora of 90M and 780M words respectively. The last two corpora are quite noisy. To train

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

18

corpora of 90M and 780M words respectively. The last two corpora are quite noisy. To train the French language model, about 712M words of crawled newspaper material is available in addition to the target side of the bitexts. All the word counts refer to French words after tokenization. It is commonly acknowledged that training statistical models on the concatenation of all this data does not necessarily lead to optimal performance, and results in extremely large models which are difficult to handle. Instead, one should focus on the most relevant subset of the data for a given task. We have done so by applying the data selection method proposed in (Moore and Lewis, 2010), and its extension to bitexts (Axelrod et al., 2011). By these means we selected a subset of 418M words out of more than 2G words for language modeling and a subset of 348M out of 850M words for training the RNN Encoder–Decoder. We used the test set newstest2012 and 2013 for data selection and weight tuning with MERT, and newstest2014 as our test set. Each set has more than 70 thousand words and a single reference translation. For training the neural networks, including the proposed RNN Encoder–Decoder, we limited the source and target vocabulary to the most frequent

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

19

For training the neural networks, including the proposed RNN Encoder–Decoder, we limited the source and target vocabulary to the most frequent 15,000 words for both English and French. This covers approximately 93% of the dataset. All the out-of-vocabulary words were mapped to a special token ( [UNK ]). The baseline phrase-based SMT system was built using Moses with default settings. This system achieves a BLEU score of 30.64 and 33.3 on the development and test sets, respectively (see Table 1). 4.1.1 RNN Encoder–Decoder The RNN Encoder–Decoder used in the experiment had 1000 hidden units with the proposed gates at the encoder and at the decoder. The input matrix between each input symbol xhtiand the hidden unit is approximated with two lower-rank matrices, and the output matrix is approximated ModelsBLEU dev test Baseline 30.64 33.30 RNN 31.20 33.87 CSLM + RNN 31.48 34.64 CSLM + RNN + WP 31.50 34.54 Table 1: BLEU scores computed on the development and test sets using different combinations of approaches. WP denotes a word penalty , where we penalizes the number of unknown words to

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

20

Table 1: BLEU scores computed on the development and test sets using different combinations of approaches. WP denotes a word penalty , where we penalizes the number of unknown words to neural networks. similarly. We used rank-100 matrices, equivalent to learning an embedding of dimension 100 for each word. The activation function used for ~hin Eq. (8) is a hyperbolic tangent function. The computation from the hidden state in the decoder to the output is implemented as a deep neural network (Pascanu et al., 2014) with a single intermediate layer having 500 maxout units each pooling 2 inputs (Goodfellow et al., 2013). All the weight parameters in the RNN Encoder– Decoder were initialized by sampling from an isotropic zero-mean (white) Gaussian distribution with its standard deviation fixed to 0:01, except for the recurrent weight parameters. For the recurrent weight matrices, we first sampled from a white Gaussian distribution and used its left singular vectors matrix, following (Saxe et al., 2014). We used Adadelta and stochastic gradient descent to train the RNN Encoder–Decoder with hyperparameters = 106and= 0:95(Zeiler, 2012). At each update, we used 64

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

21

with hyperparameters = 106and= 0:95(Zeiler, 2012). At each update, we used 64 randomly selected phrase pairs from a phrase table (which was created from 348M words). The model was trained for approximately three days. Details of the architecture used in the experiments are explained in more depth in the supplementary material. 4.1.2 Neural Language Model In order to assess the effectiveness of scoring phrase pairs with the proposed RNN Encoder– Decoder, we also tried a more traditional approach of using a neural network for learning a target language model (CSLM) (Schwenk, 2007). Especially, the comparison between the SMT system using CSLM and that using the proposed approach of phrase scoring by RNN Encoder–Decoder will clarify whether the contributions from multiple neural networks in different parts of the SMT sys-tem add up or are redundant. We trained the CSLM model on 7-grams from the target corpus. Each input word was projected into the embedding space R512, and they were concatenated to form a 3072dimensional vector. The concatenated vector was fed through two rectified layers (of size 1536 and 1024) (Glorot et al., 2011). The output layer was

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

22

fed through two rectified layers (of size 1536 and 1024) (Glorot et al., 2011). The output layer was a simple softmax layer (see Eq. (2)). All the weight parameters were initialized uniformly between0:01and0:01, and the model was trained until the validation perplexity did not improve for 10 epochs. After training, the language model achieved a perplexity of 45.80. The validation set was a random selection of 0.1% of the corpus. The model was used to score partial translations during the decoding process, which generally leads to higher gains in BLEU score than n-best list rescoring (Vaswani et al., 2013). To address the computational complexity of using a CSLM in the decoder a buffer was used to aggregate n-grams during the stacksearch performed by the decoder. Only when the buffer is full, or a stack is about to be pruned, the n-grams are scored by the CSLM. This allows us to perform fast matrixmatrix multiplication on GPU using Theano (Bergstra et al., 2010; Bastien et al., 2012). −60 −50 −40 −30 −20 −10 0−14−12−10−8−6−4−20

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

23

−60 −50 −40 −30 −20 −10 0−14−12−10−8−6−4−20 RNN Scores (log)TM Scores (log) Figure 3: The visualization of phrase pairs according to their scores (log-probabilities) by the RNN Encoder–Decoder and the translation model. 4.2 Quantitative Analysis We tried the following combinations: 1. Baseline configuration 2. Baseline + RNN 3. Baseline + CSLM + RNN 4. Baseline + CSLM + RNN + Word penalty Source Translation Model RNN Encoder–Decoder at the end of the [a la fin de la] [ ´r la fin des ann ´ees] [ ˆetre supprim ´es`a la fin de la][`a la fin du] [ `a la fin des] [ `a la fin de la] for the first time [rc pour la premi r¨ere fois] [ ´et´e donn ´es pour la premi `ere fois] [ ´et´e comm ´emor ´ee pour la

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

24

la premi `ere fois] [ ´et´e comm ´emor ´ee pour la premi `ere fois][pour la premi `ere fois] [pour la premi `ere fois ,] [pour la premi `ere fois que] in the United States and[?aux?tats-Unis et] [ ´et´e ouvertes aux ´EtatsUnis et] [ ´et´e constat ´ees aux ´Etats-Unis et][aux Etats-Unis et] [des Etats-Unis et] [des ´Etats-Unis et] , as well as [?s , qu’] [ ?s , ainsi que] [ ?re aussi bien que] [, ainsi qu’] [, ainsi que] [, ainsi que les] one of the most [?t?l’ un des plus] [ ?l’ un des plus] [ ˆetre retenue comme un de ses plus][l’ un des] [le] [un des] (a) Long, frequent source phrases Source Translation Model RNN Encoder–Decoder

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

25

(a) Long, frequent source phrases Source Translation Model RNN Encoder–Decoder , Minister of Communications and Transport[Secr ´etaire aux communications et aux transports :] [Secr ´etaire aux communications et aux transports][Secr ´etaire aux communications et aux transports] [Secr ´etaire aux communications et aux transports :] did not comply with the[vestimentaire , ne correspondaient pas `a des] [susmentionn ´ee n’ ´etait pas conforme aux] [pr´esent ´ees n’ ´etaient pas conformes `a la][n’ ont pas respect ´e les] [n’ ´etait pas conforme aux] [n’ ont pas respect ´e la] parts of the world . [c gions du monde .] [r ´egions du monde consid´er´ees .] [r ´egion du monde consid ´er´ee .][parties du monde .] [les parties du monde .] [des parties du monde .] the past few days . [le petit texte .] [cours des tout derniers jours .]

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

26

[des parties du monde .] the past few days . [le petit texte .] [cours des tout derniers jours .] [les tout derniers jours .][ces derniers jours .] [les derniers jours .] [cours des derniers jours .] on Friday and Saturday[vendredi et samedi `a la] [vendredi et samedi `a] [se d ´eroulera vendredi et samedi ,][le vendredi et le samedi] [le vendredi et samedi] [vendredi et samedi] (b) Long, rare source phrases Table 2: The top scoring target phrases for a small set of source phrases according to the translation model (direct translation probability) and by the RNN Encoder–Decoder. Source phrases were randomly selected from phrases with 4 or more words. ?denotes an incomplete (partial) character. ris a Cyrillic letter ghe. The results are presented in Table 1. As expected, adding features computed by neural networks consistently improves the performance over the baseline performance. The best performance was achieved when we used both CSLM and the phrase scores from the RNN Encoder–Decoder. This suggests that the

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

27

the baseline performance. The best performance was achieved when we used both CSLM and the phrase scores from the RNN Encoder–Decoder. This suggests that the contributions of the CSLM and the RNN Encoder– Decoder are not too correlated and that one can expect better results by improving each method independently. Furthermore, we tried penalizing the number of words that are unknown to the neural networks (i.e. words which are not in the shortlist). We do so by simply adding the number of unknown words as an additional feature the loglinear model in Eq. (9).3However, in this case we 3To understand the effect of the penalty, consider the set of all words in the 15,000 large shortlist, SL. All words xi=2 SL are replaced by a special token [UNK]before being scored by the neural networks. Hence, the conditional probability of anyxi t=2SL is actually given by the model as p(xt= [UNK]jx<t) =p(xt=2SLjx<t) =X xj t=2SLp xj tjx<t p xi tjx<t ;

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

28

=X xj t=2SLp xj tjx<t p xi tjx<t ; where x<tis a shorthand notation for xt1; : : : ; x 1.were not able to achieve better performance on the test set, but only on the development set. 4.3 Qualitative Analysis In order to understand where the performance improvement comes from, we analyze the phrase pair scores computed by the RNN Encoder–Decoder against the corresponding p(fje)from the translation model. Since the existing translation model relies solely on the statistics of the phrase pairs in the corpus, we expect its scores to be better estimated for the frequent phrases but badly estimated for rare phrases. Also, as we mentioned earlier in Sec. 3.1, we further expect the RNN Encoder– Decoder which was trained without any frequency information to score the phrase pairs based rather on the linguistic regularities than on the statistics of their occurrences in the corpus. We focus on those pairs whose source phrase is long (more than 3 words per source phrase) and As a result, the probability of words not in the shortlist is always overestimated. It is possible to address this issue by backing off to an existing model that contain non-shortlisted

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

29

always overestimated. It is possible to address this issue by backing off to an existing model that contain non-shortlisted words (see (Schwenk, 2007)) In this paper, however, we opt for introducing a word penalty instead, which counteracts the word probability overestimation. Source Samples from RNN Encoder–Decoder at the end of the [`a la fin de la] (11) for the first time [pour la premi `ere fois] (24) [pour la premi `ere fois que] (2) in the United States and [aux ´Etats-Unis et] (6) [dans les ´Etats-Unis et] (4) , as well as [, ainsi que] [,] [ainsi que] [, ainsi qu’] [et UNK] one of the most [l’ un des plus] (9) [l’ un des] (5) [l’ une des plus] (2) (a) Long, frequent source phrases Source Samples from RNN Encoder–Decoder , Minister of Communications and Transport[ , ministre des communications et le transport] ( 13)

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

30

Source Samples from RNN Encoder–Decoder , Minister of Communications and Transport[ , ministre des communications et le transport] ( 13) did not comply with the [n’ tait pas conforme aux] [n’ a pas respect l’] ( 2) [n’ a pas respect la] ( 3) parts of the world . [arts du monde .] (11) [des arts du monde .] ( 7) the past few days . [quelques jours .] (5) [les derniers jours .] ( 5) [ces derniers jours .] ( 2) on Friday and Saturday [vendredi et samedi] ( 5) [le vendredi et samedi] ( 7) [le vendredi et le samedi] ( 4) (b) Long, rare source phrases Table 3: Samples generated from the RNN Encoder–Decoder for each source phrase used in Table 2. We show the top-5 target phrases out of 50 samples. They are sorted by the RNN Encoder–Decoder scores. Figure 4: 2–D embedding of the learned word representation. The left one shows the full embedding

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

31

Figure 4: 2–D embedding of the learned word representation. The left one shows the full embedding space, while the right one shows a zoomed-in view of one region (color–coded). For more plots, see the supplementary material. frequent . For each such source phrase, we look at the target phrases that have been scored high either by the translation probability p(fje)or by the RNN Encoder–Decoder. Similarly, we perform the same procedure with those pairs whose source phrase is long butrare in the corpus. Table 2 lists the top- 3target phrases per source phrase favored either by the translation model or by the RNN Encoder–Decoder. The source phrases were randomly chosen among long ones having more than 4 or 5 words. In most cases, the choices of the target phrases by the RNN Encoder–Decoder are closer to actual or literal translations. We can observe that the RNN Encoder–Decoder prefers shorter phrases in general. Interestingly, many phrase pairs were scored similarly by both the translation model and the RNN Encoder–Decoder, but there were as manyother phrase pairs that were scored radically different (see Fig. 3). This could arise from the proposed approach of training the RNN Encoder–

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

32

proposed approach of training the RNN Encoder– Decoder on a set of unique phrase pairs, discouraging the RNN Encoder–Decoder from learning simply the frequencies of the phrase pairs from the corpus, as explained earlier. Furthermore, in Table 3, we show for each of the source phrases in Table 2, the generated samples from the RNN Encoder–Decoder. For each source phrase, we generated 50 samples and show the top-five phrases accordingly to their scores. We can see that the RNN Encoder–Decoder is able to propose well-formed target phrases without looking at the actual phrase table. Importantly, the generated phrases do not overlap completely with the target phrases from the phrase table. This encourages us to further investigate the possibility of replacing the whole or a part of the phrase table Figure 5: 2–D embedding of the learned phrase representation. The top left one shows the full representation space (5000 randomly selected points), while the other three figures show the zoomed-in view of specific regions (color–coded). with the proposed RNN Encoder–Decoder in the future. 4.4 Word and Phrase Representations Since the proposed RNN Encoder–Decoder is not

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

33

with the proposed RNN Encoder–Decoder in the future. 4.4 Word and Phrase Representations Since the proposed RNN Encoder–Decoder is not specifically designed only for the task of machine translation, here we briefly look at the properties of the trained model. It has been known for some time that continuous space language models using neural networks are able to learn semantically meaningful embeddings (See, e.g., (Bengio et al., 2003; Mikolov et al., 2013)). Since the proposed RNN Encoder–Decoder also projects to and maps back from a sequence of words into a continuous space vector, we expect to see a similar property with the proposed model as well. The left plot in Fig. 4 shows the 2–D embedding of the words using the word embedding matrix learned by the RNN Encoder–Decoder. The projection was done by the recently proposed BarnesHut-SNE (van der Maaten, 2013). We can clearly see that semantically similar words are clusteredwith each other (see the zoomed-in plots in Fig. 4). The proposed RNN Encoder–Decoder naturally generates a continuous-space representation of a phrase. The representation ( cin Fig. 1) in this

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

34

The proposed RNN Encoder–Decoder naturally generates a continuous-space representation of a phrase. The representation ( cin Fig. 1) in this case is a 1000-dimensional vector. Similarly to the word representations, we visualize the representations of the phrases that consists of four or more words using the Barnes-Hut-SNE in Fig. 5. From the visualization, it is clear that the RNN Encoder–Decoder captures both semantic and syntactic structures of the phrases. For instance, in the bottom-left plot, most of the phrases are about the duration of time, while those phrases that are syntactically similar are clustered together. The bottom-right plot shows the cluster of phrases that are semantically similar (countries or regions). On the other hand, the top-right plot shows the phrases that are syntactically similar. 5 Conclusion In this paper, we proposed a new neural network architecture, called an RNN Encoder–Decoder that is able to learn the mapping from a sequence of an arbitrary length to another sequence, possibly from a different set, of an arbitrary length. The proposed RNN Encoder–Decoder is able to either score a pair of sequences (in terms of a conditional probability) or generate a target sequence given a source sequence. Along with the new architecture,

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

35

score a pair of sequences (in terms of a conditional probability) or generate a target sequence given a source sequence. Along with the new architecture, we proposed a novel hidden unit that includes a reset gate and an update gate that adaptively control how much each hidden unit remembers or forgets while reading/generating a sequence. We evaluated the proposed model with the task of statistical machine translation, where we used the RNN Encoder–Decoder to score each phrase pair in the phrase table. Qualitatively, we were able to show that the new model is able to capture linguistic regularities in the phrase pairs well and also that the RNN Encoder–Decoder is able to propose well-formed target phrases. The scores by the RNN Encoder–Decoder were found to improve the overall translation performance in terms of BLEU scores. Also, we found that the contribution by the RNN Encoder– Decoder is rather orthogonal to the existing approach of using neural networks in the SMT system, so that we can improve further the performance by using, for instance, the RNN Encoder– Decoder and the neural net language model together. Our qualitative analysis of the trained model shows that it indeed captures the linguistic regularities in multiple levels i.e. at the word level as well as phrase level. This suggests that there may

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

36

Our qualitative analysis of the trained model shows that it indeed captures the linguistic regularities in multiple levels i.e. at the word level as well as phrase level. This suggests that there may be more natural language related applications that may benefit from the proposed RNN Encoder– Decoder. The proposed architecture has large potential for further improvement and analysis. One approach that was not investigated here is to replace the whole, or a part of the phrase table by letting the RNN Encoder–Decoder propose target phrases. Also, noting that the proposed model is not limited to being used with written language, it will be an important future research to apply the proposed architecture to other applications such as speech transcription. Acknowledgments KC, BM, CG, DB and YB would like to thank NSERC, Calcul Qu ´ebec, Compute Canada, the Canada Research Chairs and CIFAR. FB and HS were partially funded by the European Commis-sion under the project MateCat, and by DARPA under the BOLT project. References [Auli et al.2013] Michael Auli, Michel Galley, Chris Quirk, and Geoffrey Zweig. 2013. Joint language and translation modeling with recurrent neural networks. In Proceedings of the ACL Conference on Empirical Methods in Natural Language Processing

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

37

Quirk, and Geoffrey Zweig. 2013. Joint language and translation modeling with recurrent neural networks. In Proceedings of the ACL Conference on Empirical Methods in Natural Language Processing (EMNLP) , pages 1044–1054. [Axelrod et al.2011] Amittai Axelrod, Xiaodong He, and Jianfeng Gao. 2011. Domain adaptation via pseudo in-domain data selection. In Proceedings of the ACL Conference on Empirical Methods in Natural Language Processing (EMNLP) , pages 355–362. [Bastien et al.2012] Fr ´ed´eric Bastien, Pascal Lamblin, Razvan Pascanu, James Bergstra, Ian J. Goodfellow, Arnaud Bergeron, Nicolas Bouchard, and Yoshua Bengio. 2012. Theano: new features and speed improvements. Deep Learning and Unsupervised Feature Learning NIPS 2012 Workshop. [Bengio et al.2003] Yoshua Bengio, R ´ejean Ducharme, Pascal Vincent, and Christian Janvin. 2003. A neural probabilistic language model. J. Mach. Learn. Res., 3:1137–1155, March.

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

38

Pascal Vincent, and Christian Janvin. 2003. A neural probabilistic language model. J. Mach. Learn. Res., 3:1137–1155, March. [Bengio et al.2013] Y . Bengio, N. BoulangerLewandowski, and R. Pascanu. 2013. Advances in optimizing recurrent networks. In Proceedings of the 38th International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2013) , May. [Bergstra et al.2010] James Bergstra, Olivier Breuleux, Fr´ed´eric Bastien, Pascal Lamblin, Razvan Pascanu, Guillaume Desjardins, Joseph Turian, David WardeFarley, and Yoshua Bengio. 2010. Theano: a CPU and GPU math expression compiler. In Proceedings of the Python for Scientific Computing Conference (SciPy) , June. Oral Presentation. [Chandar et al.2014] Sarath Chandar, Stanislas Lauly, Hugo Larochelle, Mitesh Khapra, Balaraman Ravindran, Vikas Raykar, and Amrita Saha. 2014. An autoencoder approach to learning bilingual word representations. arXiv: 1402.1454 [cs.CL] , February.

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

39

[Dahl et al.2012] George E. Dahl, Dong Yu, Li Deng, and Alex Acero. 2012. Context-dependent pretrained deep neural networks for large vocabulary speech recognition. IEEE Transactions on Audio, Speech, and Language Processing , 20(1):33–42. [Devlin et al.2014] Jacob Devlin, Rabih Zbib, Zhongqiang Huang, Thomas Lamar, Richard Schwartz, , and John Makhoul. 2014. Fast and robust neural network joint models for statistical machine translation. In Proceedings of the ACL 2014 Conference , ACL ’14, pages 1370–1380. [Gao et al.2013] Jianfeng Gao, Xiaodong He, Wen tau Yih, and Li Deng. 2013. Learning semantic representations for the phrase translation model. Technical report, Microsoft Research. [Glorot et al.2011] X. Glorot, A. Bordes, and Y . Bengio. 2011. Deep sparse rectifier neural networks. In AISTATS’2011 . [Goodfellow et al.2013] Ian J. Goodfellow, David Warde-Farley, Mehdi Mirza, Aaron Courville, and Yoshua Bengio. 2013. Maxout networks. In ICML’2013 .

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

40

Warde-Farley, Mehdi Mirza, Aaron Courville, and Yoshua Bengio. 2013. Maxout networks. In ICML’2013 . [Graves2012] Alex Graves. 2012. Supervised Sequence Labelling with Recurrent Neural Networks . Studies in Computational Intelligence. Springer. [Hochreiter and Schmidhuber1997] S. Hochreiter and J. Schmidhuber. 1997. Long short-term memory. Neural Computation , 9(8):1735–1780. [Kalchbrenner and Blunsom2013] Nal Kalchbrenner and Phil Blunsom. 2013. Two recurrent continuous translation models. In Proceedings of the ACL Conference on Empirical Methods in Natural Language Processing (EMNLP) , pages 1700–1709. [Koehn et al.2003] Philipp Koehn, Franz Josef Och, and Daniel Marcu. 2003. Statistical phrase-based translation. In Proceedings of the 2003 Conference of the North American Chapter of the Association for Computational Linguistics on Human Language Technology - Volume 1 , NAACL ’03, pages 48–54. [Koehn2005] P. Koehn. 2005. Europarl: A parallel corpus for statistical machine translation. In Machine

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

41

[Koehn2005] P. Koehn. 2005. Europarl: A parallel corpus for statistical machine translation. In Machine Translation Summit X , pages 79–86, Phuket, Thailand. [Krizhevsky et al.2012] Alex Krizhevsky, Ilya Sutskever, and Geoffrey Hinton. 2012. ImageNet classification with deep convolutional neural networks. In Advances in Neural Information Processing Systems 25 (NIPS’2012) . [Marcu and Wong2002] Daniel Marcu and William Wong. 2002. A phrase-based, joint probability model for statistical machine translation. In Proceedings of the ACL-02 Conference on Empirical Methods in Natural Language Processing - Volume 10, EMNLP ’02, pages 133–139. [Mikolov et al.2013] Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg Corrado, and Jeff Dean. 2013. Distributed representations of words and phrases and their compositionality. In Advances in Neural Information Processing Systems 26 , pages 3111–3119. [Moore and Lewis2010] Robert C. Moore and William Lewis. 2010. Intelligent selection of language model training data. In Proceedings of the ACL 2010 Conference Short Papers , ACLShort ’10,

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

42

Lewis. 2010. Intelligent selection of language model training data. In Proceedings of the ACL 2010 Conference Short Papers , ACLShort ’10, pages 220–224, Stroudsburg, PA, USA.[Pascanu et al.2014] R. Pascanu, C. Gulcehre, K. Cho, and Y . Bengio. 2014. How to construct deep recurrent neural networks. In Proceedings of the Second International Conference on Learning Representations (ICLR 2014) , April. [Saxe et al.2014] Andrew M. Saxe, James L. McClelland, and Surya Ganguli. 2014. Exact solutions to the nonlinear dynamics of learning in deep linear neural networks. In Proceedings of the Second International Conference on Learning Representations (ICLR 2014) , April. [Schwenk et al.2006] Holger Schwenk, Marta R. CostaJuss`a, and Jos ´e A. R. Fonollosa. 2006. Continuous space language models for the iwslt 2006 task. In IWSLT , pages 166–173. [Schwenk2007] Holger Schwenk. 2007. Continuous space language models. Comput. Speech Lang. , 21(3):492–518, July. [Schwenk2012] Holger Schwenk. 2012. Continuous

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

43

space language models. Comput. Speech Lang. , 21(3):492–518, July. [Schwenk2012] Holger Schwenk. 2012. Continuous space translation models for phrase-based statistical machine translation. In Martin Kay and Christian Boitet, editors, Proceedings of the 24th International Conference on Computational Linguistics (COLIN) , pages 1071–1080. [Socher et al.2011] Richard Socher, Eric H. Huang, Jeffrey Pennington, Andrew Y . Ng, and Christopher D. Manning. 2011. Dynamic pooling and unfolding recursive autoencoders for paraphrase detection. In Advances in Neural Information Processing Systems 24. [Son et al.2012] Le Hai Son, Alexandre Allauzen, and Franc ¸ois Yvon. 2012. Continuous space translation models with neural networks. In Proceedings of the 2012 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies , NAACL HLT ’12, pages 39–48, Stroudsburg, PA, USA. [van der Maaten2013] Laurens van der Maaten. 2013. Barnes-hut-sne. In Proceedings of the First International Conference on Learning Representations (ICLR 2013) , May.

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

44

Barnes-hut-sne. In Proceedings of the First International Conference on Learning Representations (ICLR 2013) , May. [Vaswani et al.2013] Ashish Vaswani, Yinggong Zhao, Victoria Fossum, and David Chiang. 2013. Decoding with large-scale neural language models improves translation. Proceedings of the Conference on Empirical Methods in Natural Language Processing , pages 1387–1392. [Zeiler2012] Matthew D. Zeiler. 2012. ADADELTA: an adaptive learning rate method. Technical report, arXiv 1212.5701. [Zou et al.2013] Will Y . Zou, Richard Socher, Daniel M. Cer, and Christopher D. Manning. 2013. Bilingual word embeddings for phrase-based machine translation. In Proceedings of the ACL Conference on Empirical Methods in Natural Language Processing (EMNLP) , pages 1393–1398. A RNN Encoder–Decoder In this document, we describe in detail the architecture of the RNN Encoder–Decoder used in the experiments. Let us denote an source phrase by X= (x1;x2;:::;xN)and a target phrase by Y=

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

45

Let us denote an source phrase by X= (x1;x2;:::;xN)and a target phrase by Y= (y1;y2;:::;yM). Each phrase is a sequence of K-dimensional one-hot vectors, such that only one element of the vector is 1and all the others are 0. The index of the active ( 1) element indicates the word represented by the vector. A.1 Encoder Each word of the source phrase is embedded in a 500-dimensional vector space: e(xi)2R500.e(x)is used in Sec. 4.4 to visualize the words. The hidden state of an encoder consists of 1000 hidden units, and each one of them at time tis computed by hhti j=zjhht1i j + (1zj)~hhti j; where ~hhti j= tanh [We(xt)]j+ U rhht1i
 j ; zj= [Wze(xt)]j+ Uzhht1i j ; rj= [Wre(xt)]j+ Urhht1i j :

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

46

j ; rj= [Wre(xt)]j+ Urhht1i j : andare a logistic sigmoid function and an element-wise multiplication, respectively. To make the equations uncluttered, we omit biases. The initial hidden state hh0i jis fixed to 0. Once the hidden state at the Nstep (the end of the source phrase) is computed, the representation of the source phrase cis c= tanh VhhNi : A.1.1 Decoder The decoder starts by initializing the hidden state with h0h0i= tanh V0c
; where we will use
0to distinguish parameters of the decoder from those of the encoder. The hidden state at time tof the decoder is computed by h0hti j=z0jh0ht1i j + (1z0j)~h0hti j; where ~h0hti j= tanh W0e(yt1) j+r0j U0h0 ht1i+Cc ; z0j=

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

47

j+r0j U0h0 ht1i+Cc ; z0j= W0ze(yt1) j+ U0zh0 ht1i j+ [Czc]j ; r0j= W0re(yt1) j+ U0rh0 ht1i j+ [Crc]j ; ande(y0)is an all-zero vector. Similarly to the case of the encoder, e(y)is an embedding of a target word. Unlike the encoder which simply encodes the source phrase, the decoder is learned to generate a target phrase. At each time t, the decoder computes the probability of generating j-th word by p(yt;j= 1jyt1;:::;y1;X) =exp gjshti
PK j0=1exp gj0shti
; where thei-element of shtiis shti i= maxn s0hti 2i1;s0hti 2io and s0hti=Ohh0hti+Oyyt1+Occ:

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.1078

48

s0hti 2i1;s0hti 2io and s0hti=Ohh0hti+Oyyt1+Occ: In short, the shti iis a so-called maxout unit. For the computational efficiency, instead of a single-matrix output weight G, we use a product of two matrices such that G=GlGr; where Gl2RK500andGr2R5001000. B Word and Phrase Representations Here, we show enlarged plots of the word and phrase representations in Figs. 4–5. Figure 6: 2–D embedding of the learned word representation. The top left one shows the full embedding space, while the other three figures show the zoomed-in view of specific regions (color–coded). Figure 7: 2–D embedding of the learned phrase representation. The top left one shows the full representation space (1000 randomly selected points), while the other three figures show the zoomed-in view of specific regions (color–coded).

1406.1078

Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of the proposed model are jointly trained to maximize the conditional probability of a target sequence given a source sequence. The performance of a statistical machine translation system is empirically found to improve by using the conditional probabilities of phrase pairs computed by the RNN Encoder-Decoder as an additional feature in the existing log-linear model. Qualitatively, we show that the proposed model learns a semantically and syntactically meaningful representation of linguistic phrases.

http://arxiv.org/pdf/1406.1078

EMNLP 2014

cs.CL

20140603

20140903

1406.4729

0

1 Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun Abstract —Existing deep convolutional neural networks (CNNs) require a fixed-size ( e.g., 224224) input image. This requirement is “artificial” and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, “spatial pyramid pooling”, to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-theart classification results using a single full-image representation and no fine-tuning.

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

1

The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition. Index Terms —Convolutional Neural Networks, Spatial Pyramid Pooling, Image Classification, Object Detection F 1 I NTRODUCTION We are witnessing a rapid, revolutionary change in our vision community, mainly caused by deep convolutional neural networks (CNNs) [1] and the availability of large scale training data [2]. Deep-networksbased approaches have recently been substantially improving upon the state of the art in image classification [3], [4], [5], [6], object detection [7], [8], [5],

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

2

improving upon the state of the art in image classification [3], [4], [5], [6], object detection [7], [8], [5], many other recognition tasks [9], [10], [11], [12], and even non-recognition tasks. However, there is a technical issue in the training and testing of the CNNs: the prevalent CNNs require afixed input image size ( e.g., 224224), which limits both the aspect ratio and the scale of the input image. When applied to images of arbitrary sizes, current methods mostly fit the input image to the fixed size, either via cropping [3], [4] or via warping [13], [7], as shown in Figure 1 (top). But the cropped region may not contain the entire object, while the warped content may result in unwanted geometric distortion. Recognition accuracy can be compromised due to the content loss or distortion. Besides, a pre-defined scale K. He and J. Sun are with Microsoft Research, Beijing, China. E-mail: fkahe,jiansung@microsoft.com X. Zhang is with Xi’an Jiaotong University, Xi’an, China. Email:

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

3

fkahe,jiansung@microsoft.com X. Zhang is with Xi’an Jiaotong University, Xi’an, China. Email: xyz.clx@stu.xjtu.edu.cn S. Ren is with University of Science and Technology of China, Hefei, China. Email: sqren@mail.ustc.edu.cn This work was done when X. Zhang and S. Ren were interns at Microsoft Research. crop warp spatial pyramid poolingcrop / warp conv layers image fc layers outputimage conv layers fc layers outputFigure 1: Top: cropping or warping to fit a fixed size. Middle: a conventional CNN. Bottom: our spatial pyramid pooling network structure. may not be suitable when object scales vary. Fixing input sizes overlooks the issues involving scales. So why do CNNs require a fixed input size? A CNN mainly consists of two parts: convolutional layers, and fully-connected layers that follow. The convolutional layers operate in a sliding-window manner and output feature maps which represent the spatial arrangement of the activations (Figure 2). In fact, convolutional layers do not require a fixed image size and

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

4

and output feature maps which represent the spatial arrangement of the activations (Figure 2). In fact, convolutional layers do not require a fixed image size and can generate feature maps of any sizes. On the other hand, the fully-connected layers need to have fixedsize/length input by their definition. Hence, the fixedsize constraint comes only from the fully-connected layers, which exist at a deeper stage of the network. In this paper, we introduce a spatial pyramid pooling (SPP) [14], [15] layer to remove the fixed-size constraint of the network. Specifically, we add anarXiv:1406.4729v4 [cs.CV] 23 Apr 2015 2 SPP layer on top of the last convolutional layer. The SPP layer pools the features and generates fixedlength outputs, which are then fed into the fullyconnected layers (or other classifiers). In other words, we perform some information “aggregation” at a deeper stage of the network hierarchy (between convolutional layers and fully-connected layers) to avoid the need for cropping or warping at the beginning.

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

5

deeper stage of the network hierarchy (between convolutional layers and fully-connected layers) to avoid the need for cropping or warping at the beginning. Figure 1 (bottom) shows the change of the network architecture by introducing the SPP layer. We call the new network structure SPP-net . Spatial pyramid pooling [14], [15] (popularly known as spatial pyramid matching or SPM [15]), as an extension of the Bag-of-Words (BoW) model [16], is one of the most successful methods in computer vision. It partitions the image into divisions from finer to coarser levels, and aggregates local features in them. SPP has long been a key component in the leading and competition-winning systems for classification ( e.g., [17], [18], [19]) and detection ( e.g., [20]) before the recent prevalence of CNNs. Nevertheless, SPP has not been considered in the context of CNNs. We note that SPP has several remarkable properties for deep CNNs: 1) SPP is able to generate a fixedlength output regardless of the input size, while the sliding window pooling used in the previous deep networks [3] cannot; 2) SPP uses multi-level spatial

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

6

sliding window pooling used in the previous deep networks [3] cannot; 2) SPP uses multi-level spatial bins, while the sliding window pooling uses only a single window size. Multi-level pooling has been shown to be robust to object deformations [15]; 3) SPP can pool features extracted at variable scales thanks to the flexibility of input scales. Through experiments we show that all these factors elevate the recognition accuracy of deep networks. SPP-net not only makes it possible to generate representations from arbitrarily sized images/windows for testing, but also allows us to feed images with varying sizes or scales during training. Training with variable-size images increases scale-invariance and reduces over-fitting. We develop a simple multi-size training method. For a single network to accept variable input sizes, we approximate it by multiple networks that share all parameters, while each of these networks is trained using a fixed input size. In each epoch we train the network with a given input size, and switch to another input size for the next epoch. Experiments show that this multi-size training converges just as the traditional single-size training, and leads to better testing accuracy.

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

7

epoch. Experiments show that this multi-size training converges just as the traditional single-size training, and leads to better testing accuracy. The advantages of SPP are orthogonal to the specific CNN designs. In a series of controlled experiments on the ImageNet 2012 dataset, we demonstrate that SPP improves four different CNN architectures in existing publications [3], [4], [5] (or their modifications), over the no-SPP counterparts. These architectures have various filter numbers/sizes, strides, depths, or other designs. It is thus reasonable for us to conjecture that SPP should improve more sophisticated (deeperand larger) convolutional architectures. SPP-net also shows state-of-the-art classification results on Caltech101 [21] and Pascal VOC 2007 [22] using only a single full-image representation and no fine-tuning. SPP-net also shows great strength in object detection. In the leading object detection method R-CNN [7], the features from candidate windows are extracted via deep convolutional networks. This method shows remarkable detection accuracy on both the VOC and ImageNet datasets. But the feature computation in RCNN is time-consuming, because it repeatedly applies

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

8

remarkable detection accuracy on both the VOC and ImageNet datasets. But the feature computation in RCNN is time-consuming, because it repeatedly applies the deep convolutional networks to the raw pixels of thousands of warped regions per image. In this paper, we show that we can run the convolutional layers only once on the entire image (regardless of the number of windows), and then extract features by SPP-net on the feature maps. This method yields a speedup of over one hundred times over R-CNN. Note that training/running a detector on the feature maps (rather than image regions) is actually a more popular idea [23], [24], [20], [5]. But SPP-net inherits the power of the deep CNN feature maps and also the flexibility of SPP on arbitrary window sizes, which leads to outstanding accuracy and efficiency. In our experiment, the SPP-net-based system (built upon the R-CNN pipeline) computes features 24-102 faster than R-CNN, while has better or comparable accuracy. With the recent fast proposal method of EdgeBoxes [25], our system takes 0.5 seconds processing an image (including all steps ). This makes our method practical for real-world applications.

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

9

[25], our system takes 0.5 seconds processing an image (including all steps ). This makes our method practical for real-world applications. A preliminary version of this manuscript has been published in ECCV 2014. Based on this work, we attended the competition of ILSVRC 2014 [26], and ranked #2 in object detection and #3 in image classification (both are provided-data-only tracks) among all 38 teams. There are a few modifications made for ILSVRC 2014. We show that the SPP-nets can boost various networks that are deeper and larger (Sec. 3.1.2-3.1.4) over the no-SPP counterparts. Further, driven by our detection framework, we find that multi-view testing on feature maps with flexibly located/sized windows (Sec. 3.1.5) can increase the classification accuracy. This manuscript also provides the details of these modifications. We have released the code to facilitate future research ( http://research.microsoft.com/en-us/um/people/kahe/ ). 2 D EEP NETWORKS WITH SPATIAL PYRAMIDPOOLING 2.1 Convolutional Layers and Feature Maps

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

10

2 D EEP NETWORKS WITH SPATIAL PYRAMIDPOOLING 2.1 Convolutional Layers and Feature Maps Consider the popular seven-layer architectures [3], [4]. The first five layers are convolutional, some of which are followed by pooling layers. These pooling layers can also be considered as “convolutional”, in the sense that they are using sliding windows. The last two 3 filter #175 filter #55 (a) image (b) feature maps (c) strongest activations filter #66 filter #118 (a) image (b) feature maps (c) strongest activations Figure 2: Visualization of the feature maps. (a) Two images in Pascal VOC 2007. (b) The feature maps of some conv 5filters. The arrows indicate the strongest responses and their corresponding positions in the images. (c) The ImageNet images that have the strongest responses of the corresponding filters. The green rectangles mark the receptive fields of the strongest responses. layers are fully connected, with an N-way softmax as the output, where N is the number of categories. The deep network described above needs a fixed

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

11

layers are fully connected, with an N-way softmax as the output, where N is the number of categories. The deep network described above needs a fixed image size. However, we notice that the requirement of fixed sizes is only due to the fully-connected layers that demand fixed-length vectors as inputs. On the other hand, the convolutional layers accept inputs of arbitrary sizes. The convolutional layers use sliding filters, and their outputs have roughly the same aspect ratio as the inputs. These outputs are known as feature maps [1] - they involve not only the strength of the responses, but also their spatial positions. In Figure 2, we visualize some feature maps. They are generated by some filters of the conv 5layer. Figure 2(c) shows the strongest activated images of these filters in the ImageNet dataset. We see a filter can be activated by some semantic content. For example, the 55-th filter (Figure 2, bottom left) is most activated by a circle shape; the 66-th filter (Figure 2, top right) is

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

12

a circle shape; the 66-th filter (Figure 2, top right) is most activated by a ^-shape; and the 118-th filter (Figure 2, bottom right) is most activated by a _-shape. These shapes in the input images (Figure 2(a)) activate the feature maps at the corresponding positions (the arrows in Figure 2). It is worth noticing that we generate the feature maps in Figure 2 without fixing the input size. These feature maps generated by deep convolutional layers are analogous to the feature maps in traditional methods [27], [28]. In those methods, SIFT vectors [29] or image patches [28] are densely extracted and then encoded, e.g., by vector quantization [16], [15], [30], sparse coding [17], [18], or Fisher kernels [19]. These encoded features consist of the feature maps, and are then pooled by Bag-of-Words (BoW) [16] or spatial pyramids [14], [15]. Analogously, the deep convolutional features can be pooled in a similar way. 2.2 The Spatial Pyramid Pooling Layer The convolutional layers accept arbitrary input sizes,

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

13

convolutional features can be pooled in a similar way. 2.2 The Spatial Pyramid Pooling Layer The convolutional layers accept arbitrary input sizes, but they produce outputs of variable sizes. The classifiers (SVM/softmax) or fully-connected layers require convolutional layersfeature maps of conv 5 (arbitrary size) fixed-length representation input image16×256-d 4×256-d 256-d…... …... spatial pyramid pooling layerfully-connected layers (fc 6, fc 7)Figure 3: A network structure with a spatial pyramid pooling layer . Here 256 is the filter number of the conv 5layer, and conv 5is the last convolutional layer. fixed-length vectors. Such vectors can be generated by the Bag-of-Words (BoW) approach [16] that pools the features together. Spatial pyramid pooling [14], [15] improves BoW in that it can maintain spatial information by pooling in local spatial bins. These spatial bins have sizes proportional to the image size, so the number of bins is fixed regardless of the image size. This is in contrast to the sliding window pooling of the previous deep networks [3], where the number

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

14

so the number of bins is fixed regardless of the image size. This is in contrast to the sliding window pooling of the previous deep networks [3], where the number of sliding windows depends on the input size. To adopt the deep network for images of arbitrary sizes, we replace the last pooling layer ( e.g., pool 5, after the last convolutional layer) with a spatial pyramid pooling layer . Figure 3 illustrates our method. In each spatial bin, we pool the responses of each filter (throughout this paper we use max pooling). The outputs of the spatial pyramid pooling are kMdimensional vectors with the number of bins denoted asM(kis the number of filters in the last convolutional layer). The fixed-dimensional vectors are the input to the fully-connected layer. With spatial pyramid pooling, the input image can 4 be of any sizes. This not only allows arbitrary aspect ratios, but also allows arbitrary scales. We can resize the input image to any scale ( e.g.,min(w;h)=180, 224, ...) and apply the same deep network. When the input image is at different scales, the network (with the same filter sizes) will extract features at different

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

15

...) and apply the same deep network. When the input image is at different scales, the network (with the same filter sizes) will extract features at different scales. The scales play important roles in traditional methods, e.g., the SIFT vectors are often extracted at multiple scales [29], [27] (determined by the sizes of the patches and Gaussian filters). We will show that the scales are also important for the accuracy of deep networks. Interestingly, the coarsest pyramid level has a single bin that covers the entire image. This is in fact a “global pooling” operation, which is also investigated in several concurrent works. In [31], [32] a global average pooling is used to reduce the model size and also reduce overfitting; in [33], a global average pooling is used on the testing stage after all fc layers to improve accuracy; in [34], a global max pooling is used for weakly supervised object recognition. The global pooling operation corresponds to the traditional Bag-of-Words method. 2.3 Training the Network Theoretically, the above network structure can be trained with standard back-propagation [1], regardless of the input image size. But in practice the GPU

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

16

Theoretically, the above network structure can be trained with standard back-propagation [1], regardless of the input image size. But in practice the GPU implementations (such as cuda-convnet [3] and Caffe [35]) are preferably run on fixed input images. Next we describe our training solution that takes advantage of these GPU implementations while still preserving the spatial pyramid pooling behaviors. Single-size training As in previous works, we first consider a network taking a fixed-size input (224 224) cropped from images. The cropping is for the purpose of data augmentation. For an image with a given size, we can pre-compute the bin sizes needed for spatial pyramid pooling. Consider the feature maps after conv 5that have a size ofaa(e.g., 1313). With a pyramid level of nn bins, we implement this pooling level as a sliding window pooling, where the window size win=da=ne and stride str=ba=ncwithd
eandb
cdenoting ceiling and floor operations. With an l-level pyramid, we implement lsuch layers. The next fully-connected layer (fc 6) will concatenate the loutputs. Figure 4

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

17

we implement lsuch layers. The next fully-connected layer (fc 6) will concatenate the loutputs. Figure 4 shows an example configuration of 3-level pyramid pooling (33, 22, 11) in the cuda-convnet style [3]. The main purpose of our single-size training is to enable the multi-level pooling behavior. Experiments show that this is one reason for the gain of accuracy. Multi-size training Our network with SPP is expected to be applied on images of any sizes. To address the issue of varying [fc6] type=fcoutputs=4096inputs=pool3x3,pool2x2,pool1x1[pool1x1] type=poolpool=maxinputs=conv5sizeX=13stride=13[pool3x3] type=poolpool=maxinputs=conv5sizeX=5stride=4[pool2x2] type=poolpool=maxinputs=conv5sizeX=7stride=6Figure 4: An example 3-level pyramid pooling in the cuda-convnet style [3]. Here sizeX is the size of the pooling window. This configuration is for a network

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

18

cuda-convnet style [3]. Here sizeX is the size of the pooling window. This configuration is for a network whose feature map size of conv 5is 1313, so the pool 33, pool 22, and pool 11layers will have 3 3, 22, and 11 bins respectively. image sizes in training, we consider a set of predefined sizes. We consider two sizes: 180 180 in addition to 224224. Rather than crop a smaller 180 180 region, we resize the aforementioned 224 224 region to 180180. So the regions at both scales differ only in resolution but not in content/layout. For the network to accept 180 180 inputs, we implement another fixed-size-input (180 180) network. The feature map size after conv 5isaa= 1010 in this case. Then we still usewin=da=neandstr=ba=ncto implement each pyramid pooling level. The output of the spatial pyramid pooling layer of this 180-network has the same fixed length as the 224-network. As such, this 180-network has exactly the same parameters as the

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

19

same fixed length as the 224-network. As such, this 180-network has exactly the same parameters as the 224-network in each layer. In other words, during training we implement the varying-input-size SPP-net by two fixed-size networks that share parameters. To reduce the overhead to switch from one network (e.g., 224) to the other ( e.g., 180), we train each full epoch on one network, and then switch to the other one (keeping all weights) for the next full epoch. This is iterated. In experiments, we find the convergence rate of this multi-size training to be similar to the above single-size training. The main purpose of our multi-size training is to simulate the varying input sizes while still leveraging the existing well-optimized fixed-size implementations. Besides the above two-scale implementation, we have also tested a variant using ssas input where sis randomly and uniformly sampled from [180;224] at each epoch. We report the results of both variants in the experiment section. Note that the above single/multi-size solutions are for training only. At the testing stage, it is straightforward to apply SPP-net on images of any sizes. 5

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

20

Note that the above single/multi-size solutions are for training only. At the testing stage, it is straightforward to apply SPP-net on images of any sizes. 5 model conv 1 conv 2 conv 3 conv 4 conv 5 conv 6 conv 7 ZF-5 9672, str 2 25652, str 2 384323843225632 LRN, pool 32, str 2 LRN, pool 32, str 2 - map size 5555 2727 1313 1313 1313 Convnet*-5 96112, str 4 25652384323843225632 LRN, LRN, pool 32, str 2 pool 32, 2 - map size 5555 2727 1313 1313 1313 Overfeat-5/7 9672, str 2 256525123251232512325123251232 pool 32, str 3, LRN pool 22, str 2 map size 3636 1818 1818 1818 1818 1818 1818 Table 1: Network architectures: filter number filter size ( e.g.,9672), filter stride ( e.g., str 2), pooling window

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

21

size ( e.g., pool 32), and the output feature map size ( e.g., map size 5555). LRN represents Local Response Normalization. The padding is adjusted to produce the expected output feature map size. 3 SPP- NET FOR IMAGE CLASSIFICATION 3.1 Experiments on ImageNet 2012 Classification We train the networks on the 1000-category training set of ImageNet 2012. Our training algorithm follows the practices of previous work [3], [4], [36]. The images are resized so that the smaller dimension is 256, and a 224224 crop is picked from the center or the four corners from the entire image1. The data are augmented by horizontal flipping and color altering [3]. Dropout [3] is used on the two fully-connected layers. The learning rate starts from 0.01, and is divided by 10 (twice) when the error plateaus. Our implementation is based on the publicly available code of cuda-convnet [3] and Caffe [35]. All networks in this paper can be trained on a single GeForce GTX Titan GPU (6 GB memory) within two to four weeks. 3.1.1 Baseline Network Architectures The advantages of SPP are independent of the convolutional network architectures used. We investigate

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

22

memory) within two to four weeks. 3.1.1 Baseline Network Architectures The advantages of SPP are independent of the convolutional network architectures used. We investigate four different network architectures in existing publications [3], [4], [5] (or their modifications), and we show SPP improves the accuracy of all these architectures. These baseline architectures are in Table 1 and briefly introduced below: ZF-5 : this architecture is based on Zeiler and Fergus’s (ZF) “fast” (smaller) model [4]. The number indicates five convolutional layers. Convnet*-5 : this is a modification on Krizhevsky et al.’s network [3]. We put the two pooling layers after conv 2and conv 3(instead of after conv 1and conv 2). As a result, the feature maps after each layer have the same size as ZF-5. Overfeat-5/7 : this architecture is based on the Overfeat paper [5], with some modifications as in [6]. In contrast to ZF-5/Convnet*-5, this architecture produces a larger feature map ( 1818instead

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

23

[6]. In contrast to ZF-5/Convnet*-5, this architecture produces a larger feature map ( 1818instead of1313) before the last pooling layer. A larger filter number (512) is used in conv 3and the following convolutional layers. We also investigate 1. In [3], the four corners are picked from the corners of the central 256256 crop.a deeper architecture with 7 convolutional layers, where conv 3to conv 7have the same structures. In the baseline models, the pooling layer after the last convolutional layer generates 66feature maps, with two 4096-d fc layers and a 1000-way softmax layer following. Our replications of these baseline networks are in Table 2 (a). We train 70 epochs for ZF-5 and 90 epochs for the others. Our replication of ZF-5 is better than the one reported in [4]. This gain is because the corner crops are from the entire image, as is also reported in [36]. 3.1.2 Multi-level Pooling Improves Accuracy In Table 2 (b) we show the results using singlesize training. The training and testing sizes are both 224224. In these networks, the convolutional layers

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

24

In Table 2 (b) we show the results using singlesize training. The training and testing sizes are both 224224. In these networks, the convolutional layers have the same structures as the corresponding baseline models, whereas the pooling layer after the final convolutional layer is replaced with the SPP layer. For the results in Table 2, we use a 4-level pyramid. The pyramid isf66, 33, 22, 11g(totally 50 bins). For fair comparison, we still use the standard 10view prediction with each view a 224 224 crop. Our results in Table 2 (b) show considerable improvement over the no-SPP baselines in Table 2 (a). Interestingly, the largest gain of top-1 error (1.65%) is given by the most accurate architecture. Since we are still using the same 10 cropped views as in (a), these gains are solely because of multi-level pooling. It is worth noticing that the gain of multi-level pooling is notsimply due to more parameters; rather, it is because the multi-level pooling is robust to the variance in object deformations and spatial layout [15]. To show this, we train another ZF-5 network with

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

25

it is because the multi-level pooling is robust to the variance in object deformations and spatial layout [15]. To show this, we train another ZF-5 network with a different 4-level pyramid: f44, 33, 22, 11g (totally 30 bins). This network has fewer parameters than its no-SPP counterpart, because its fc 6layer has 30256-d inputs instead of 36 256-d. The top-1/top5 errors of this network are 35.06/14.04. This result is similar to the 50-bin pyramid above (34.98/14.14), but considerably better than the no-SPP counterpart (35.99/14.76). 6 top-1 error (%) ZF-5 Convnet*-5 Overfeat-5 Overfeat-7 (a) no SPP 35.99 34.93 34.13 32.01 (b) SPP single-size trained 34.98 (1.01) 34.38 (0.55) 32.87 (1.26) 30.36 (1.65) (c) SPP multi-size trained 34.60 (1.39) 33.94 (0.99) 32.26 (1.87) 29.68 (2.33)

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

26

top-5 error (%) ZF-5 Convnet*-5 Overfeat-5 Overfeat-7 (a) no SPP 14.76 13.92 13.52 11.97 (b) SPP single-size trained 14.14 (0.62) 13.54 (0.38) 12.80 (0.72) 11.12 (0.85) (c) SPP multi-size trained 13.64 (1.12) 13.33 (0.59) 12.33 (1.19) 10.95 (1.02) Table 2: Error rates in the validation set of ImageNet 2012. All the results are obtained using standard 10-view testing. In the brackets are the gains over the “no SPP” baselines. SPP on test view top-1 val ZF-5, single-size trained 1 crop 38.01 ZF-5, single-size trained 1 full 37.55 ZF-5, multi-size trained 1 crop 37.57 ZF-5, multi-size trained 1 full 37.07 Overfeat-7, single-size trained 1 crop 33.18 Overfeat-7, single-size trained 1 full 32.72 Overfeat-7, multi-size trained 1 crop 32.57

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

27

Overfeat-7, single-size trained 1 full 32.72 Overfeat-7, multi-size trained 1 crop 32.57 Overfeat-7, multi-size trained 1 full 31.25 Table 3: Error rates in the validation set of ImageNet 2012 using a single view. The images are resized so min(w;h) = 256 . The crop view is the central 224 224 of the image. 3.1.3 Multi-size Training Improves Accuracy Table 2 (c) shows our results using multi-size training. The training sizes are 224 and 180, while the testing size is still 224. We still use the standard 10-view prediction. The top-1/top-5 errors of all architectures further drop. The top-1 error of SPP-net (Overfeat-7) drops to 29.68%, which is 2.33% better than its noSPP counterpart and 0.68% better than its single-size trained counterpart. Besides using the two discrete sizes of 180 and 224, we have also evaluated using a random size uniformly sampled from [180;224]. The top-1/5 error of SPP-net (Overfeat-7) is 30.06%/10.96%. The top1 error is slightly worse than the two-size version,

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

28

of SPP-net (Overfeat-7) is 30.06%/10.96%. The top1 error is slightly worse than the two-size version, possibly because the size of 224 (which is used for testing) is visited less. But the results are still better the single-size version. There are previous CNN solutions [5], [36] that deal with various scales/sizes, but they are mostly based on testing. In Overfeat [5] and Howard’s method [36], the single network is applied at multiple scales in the testing stage, and the scores are averaged. Howard further trains two different networks on low/highresolution image regions and averages the scores. To our knowledge, our method is the first one that trains a single network with input images of multiple sizes.3.1.4 Full-image Representations Improve Accuracy Next we investigate the accuracy of the full-image views. We resize the image so that min(w;h)=256 while maintaining its aspect ratio. The SPP-net is applied on this full image to compute the scores of the full view. For fair comparison, we also evaluate the accuracy of the single view in the center 224 224 crop (which is used in the above evaluations). The comparisons of single-view testing accuracy are in

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

29

the accuracy of the single view in the center 224 224 crop (which is used in the above evaluations). The comparisons of single-view testing accuracy are in Table 3. Here we evaluate ZF-5/Overfeat-7. The top-1 error rates are all reduced by the full-view representation. This shows the importance of maintaining the complete content. Even though our network is trained using square images only, it generalizes well to other aspect ratios. Comparing Table 2 and Table 3, we find that the combination of multiple views is substantially better than the single full-image view. However, the fullimage representations are still of good merits. First, we empirically find that (discussed in the next subsection) even for the combination of dozens of views, the additional two full-image views (with flipping) can still boost the accuracy by about 0.2%. Second, the full-image view is methodologically consistent with the traditional methods [15], [17], [19] where the encoded SIFT vectors of the entire image are pooled together. Third, in other applications such as image retrieval [37], an image representation, rather than a classification score, is required for similarity ranking. A full-image representation can be preferred.

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

30

retrieval [37], an image representation, rather than a classification score, is required for similarity ranking. A full-image representation can be preferred. 3.1.5 Multi-view Testing on Feature Maps Inspired by our detection algorithm (described in the next section), we further propose a multi-view testing method on the feature maps. Thanks to the flexibility of SPP , we can easily extract the features from windows (views) of arbitrary sizes from the convolutional feature maps. On the testing stage, we resize an image so min(w;h) =swheresrepresents a predefined scale (like 256). Then we compute the convolutional feature maps from the entire image. For the usage of 7 method test scales test views top-1 val top-5 val top-5 test Krizhevsky et al. [3] 1 10 40.7 18.2 Overfeat (fast) [5] 1 - 39.01 16.97 Overfeat (fast) [5] 6 - 38.12 16.27 Overfeat (big) [5] 4 - 35.74 14.18 Howard (base) [36] 3 162 37.0 15.8 Howard (high-res) [36] 3 162 36.8 16.2

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

31

Howard (base) [36] 3 162 37.0 15.8 Howard (high-res) [36] 3 162 36.8 16.2 Zeiler & Fergus (ZF) (fast) [4] 1 10 38.4 16.5 Zeiler & Fergus (ZF) (big) [4] 1 10 37.5 16.0 Chatfield et al. [6] 1 10 - 13.1 ours (SPP O-7) 1 10 29.68 10.95 ours (SPP O-7) 6 96+2full 27.86 9.14 9.08 Table 4: Error rates in ImageNet 2012. All the results are based on a single network . The number of views in Overfeat depends on the scales and strides, for which there are several hundreds at the finest scale. flipped views, we also compute the feature maps of the flipped image. Given any view (window) in the image, we map this window to the feature maps (the way of mapping is in Appendix), and then use SPP to pool the features from this window (see Figure 5). The pooled features are then fed into the fc layers to compute the softmax score of this window. These scores are averaged for the final prediction. For the

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

32

The pooled features are then fed into the fc layers to compute the softmax score of this window. These scores are averaged for the final prediction. For the standard 10-view, we use s= 256 and the views are 224224 windows on the corners or center. Experiments show that the top-5 error of the 10-view prediction on feature maps is within 0.1% around the original 10-view prediction on image crops. We further apply this method to extract multiple views from multiple scales. We resize the image to six scaless2 f224;256;300;360;448;560gand compute the feature maps on the entire image for each scale. We use 224224 as the view size for any scale, so these views have different relative sizes on the original image for different scales. We use 18 views for each scale: one at the center, four at the corners, and four on the middle of each side, with/without flipping (when s= 224 there are 6 different views). The combination of these 96 views reduces the top-5 error from 10.95% to 9.36%. Combining the two fullimage views (with flipping) further reduces the top-5 error to 9.14%.

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

33

error from 10.95% to 9.36%. Combining the two fullimage views (with flipping) further reduces the top-5 error to 9.14%. In the Overfeat paper [5], the views are also extracted from the convolutional feature maps instead of image crops. However, their views cannot have arbitrary sizes; rather, the windows are those where the pooled features match the desired dimensionality. We empirically find that these restricted windows are less beneficial than our flexibly located/sized windows. 3.1.6 Summary and Results for ILSVRC 2014 In Table 4 we compare with previous state-of-theart methods. Krizhevsky et al .’s [3] is the winning method in ILSVRC 2012; Overfeat [5], Howard’s [36], and Zeiler and Fergus’s [4] are the leading methodsrank team top-5 test 1 GoogLeNet [32] 6.66 2 VGG [33] 7.32 3 ours 8.06 4 Howard 8.11 5 DeeperVision 9.50 6 NUS-BST 9.79 7 TTIC ECP 10.22

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

34

3 ours 8.06 4 Howard 8.11 5 DeeperVision 9.50 6 NUS-BST 9.79 7 TTIC ECP 10.22 Table 5: The competition results of ILSVRC 2014 classification [26]. The best entry of each team is listed. in ILSVRC 2013. We only consider single-network performance for manageable comparisons. Our best single network achieves 9.14% top-5 error on the validation set. This is exactly the single-model entry we submitted to ILSVRC 2014 [26]. The top-5 error is 9.08% on the testing set (ILSVRC 2014 has the same training/validation/testing data as ILSVRC 2012). After combining eleven models, our team’s result ( 8.06%) is ranked #3 among all 38 teams attending ILSVRC 2014 (Table 5). Since the advantages of SPPnet should be in general independent of architectures, we expect that it will further improve the deeper and larger convolutional architectures [33], [32]. 3.2 Experiments on VOC 2007 Classification Our method can generate a full-view image representation. With the above networks pre-trained on ImageNet, we extract these representations from the

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

35

Our method can generate a full-view image representation. With the above networks pre-trained on ImageNet, we extract these representations from the images in the target datasets and re-train SVM classifiers [38]. In the SVM training, we intentionally do not use any data augmentation (flip/multi-view). We l2-normalize the features for SVM training. The classification task in Pascal VOC 2007 [22] involves 9,963 images in 20 categories. 5,011 images are for training, and the rest are for testing. The performance is evaluated by mean Average Precision (mAP). Table 6 summarizes the results. 8 (a) (b) (c) (d) (e) model no SPP (ZF-5) SPP (ZF-5) SPP (ZF-5) SPP (ZF-5) SPP (Overfeat-7) crop crop full full full size 224224 224224 224- 392- 364conv 4 59.96 57.28 - - conv 5 66.34 65.43 - - pool 5=7(66) 69.14 68.76 70.82 71.67 76.09

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

36

fc6=8 74.86 75.55 77.32 78.78 81.58 fc7=9 75.90 76.45 78.39 80.10 82.44 Table 6: Classification mAP in Pascal VOC 2007. For SPP-net, the pool 5=7layer uses the 66 pyramid level. (a) (b) (c) (d) model no SPP (ZF-5) SPP (ZF-5) SPP (ZF-5) SPP (Overfeat-7) crop crop full full size 224224 224224 224- 224conv 4 80.12 81.03 - conv 5 84.40 83.76 - pool 5=7(66) 87.98 87.60 89.46 91.46 SPP pool 5=7 - 89.47 91.44 93.42 fc6=8 87.86 88.54 89.50 91.83 fc7=9 85.30 86.10 87.08 90.00 Table 7: Classification accuracy in Caltech101. For SPP-net, the pool 5=7layer uses the 66 pyramid level. We start from a baseline in Table 6 (a). The model is

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

37

We start from a baseline in Table 6 (a). The model is ZF-5 without SPP . To apply this model, we resize the image so that its smaller dimension is 224, and crop the center 224224 region. The SVM is trained via the features of a layer. On this dataset, the deeper the layer is, the better the result is. In Table 6 (b), we replace the no-SPP net with our SPP-net. As a first-step comparison, we still apply the SPP-net on the center 224224 crop. The results of the fc layers improve. This gain is mainly due to multi-level pooling. Table 6 (c) shows our results on full images, where the images are resized so that the shorter side is 224. We find that the results are considerably improved (78.39% vs. 76.45%). This is due to the full-image representation that maintains the complete content. Because the usage of our network does not depend on scale, we resize the images so that the smaller dimension is sand use the same network to extract features. We find that s= 392 gives the best results (Table 6 (d)) based on the validation set. This is mainly because the objects occupy smaller regions in VOC

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

38

(Table 6 (d)) based on the validation set. This is mainly because the objects occupy smaller regions in VOC 2007 but larger regions in ImageNet, so the relative object scales are different between the two sets. These results indicate scale matters in the classification tasks, and SPP-net can partially address this “scale mismatch” issue. In Table 6 (e) the network architecture is replaced with our best model (Overfeat-7, multi-size trained), and the mAP increases to 82.44% . Table 8 summarizes our results and the comparisons with the state-of-theart methods. Among these methods, VQ [15], LCC [18], and FK [19] are all based on spatial pyramids matching, and [13], [4], [34], [6] are based on deepnetworks. In these results, Oquab et al.’s (77.7%) and Chatfield et al .’s (82.42%) are obtained by network fine-tuning and multi-view testing. Our result is comparable with the state of the art, using only a single full-image representation and without fine-tuning. 3.3 Experiments on Caltech101

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423

1406.4729

39

full-image representation and without fine-tuning. 3.3 Experiments on Caltech101 The Caltech101 dataset [21] contains 9,144 images in 102 categories (one background). We randomly sample 30 images per category for training and up to 50 images per category for testing. We repeat 10 random splits and average the accuracy. Table 7 summarizes our results. There are some common observations in the Pascal VOC 2007 and Caltech101 results: SPP-net is better than the no-SPP net (Table 7 (b) vs. (a)), and the fullview representation is better than the crop ((c) vs. (b)). But the results in Caltech101 have some differences with Pascal VOC. The fully-connected layers are less accurate, and the SPP layers are better. This is possibly because the object categories in Caltech101 are less related to those in ImageNet, and the deeper layers are more category-specialized. Further, we find that the scale 224 has the best performance among the scales we tested on this dataset. This is mainly because the objects in Caltech101 also occupy large regions of the images, as is the case of ImageNet. Besides cropping, we also evaluate warping the image to fit the 224 224 size. This solution maintains

1406.4729

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224x224) input image. This requirement is "artificial" and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, "spatial pyramid pooling", to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102x faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

http://arxiv.org/pdf/1406.4729

This manuscript is the accepted version for IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 2015. See Changelog

cs.CV

20140618

20150423