Datasets:

jamescalam

/

llama-2-arxiv-papers-chunked

like 16

1301.4083

69

T.M. Mitchell and S.B. Thrun. Explanation-based neural network learning for robot control. Advances in Neural information processing systems , pages 287{287, 1993. R. Montague. Universal grammar. Theoria , 36(3):373{398, 1970. V. Nair and G. E Hinton. Rectied linear units improve restricted Boltzmann machines. In ICML'10 , 2010. L.B.J.H.F.R.A. Olshen and C.J. Stone. Classication and regression trees. Belmont, Calif.: Wadsworth , 1984. Joseph O'Sullivan. Integrating initialization bias and search bias in neural network learning, 1996. F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, et al. Scikit-learn: Machine learning in python. The Journal of Machine Learning Research , 12:2825{2830, 2011. Gail B. Peterson. A day of great illumination: B. F. Skinner's discovery of shaping. Journal of the Experimental Analysis of Behavior , 82(3):317{328, 2004.

1301.4083

Knowledge Matters: Importance of Prior Information for Optimization

We explore the effect of introducing prior information into the intermediate level of neural networks for a learning task on which all the state-of-the-art machine learning algorithms tested failed to learn. We motivate our work from the hypothesis that humans learn such intermediate concepts from other individuals via a form of supervision or guidance using a curriculum. The experiments we have conducted provide positive evidence in favor of this hypothesis. In our experiments, a two-tiered MLP architecture is trained on a dataset with 64x64 binary inputs images, each image with three sprites. The final task is to decide whether all the sprites are the same or one of them is different. Sprites are pentomino tetris shapes and they are placed in an image with different locations using scaling and rotation transformations. The first part of the two-tiered MLP is pre-trained with intermediate-level targets being the presence of sprites at each location, while the second part takes the output of the first part as input and predicts the final task's target binary event. The two-tiered MLP architecture, with a few tens of thousand examples, was able to learn the task perfectly, whereas all other algorithms (include unsupervised pre-training, but also traditional algorithms like SVMs, decision trees and boosting) all perform no better than chance. We hypothesize that the optimization difficulty involved when the intermediate pre-training is not performed is due to the {\em composition} of two highly non-linear tasks. Our findings are also consistent with hypotheses on cultural learning inspired by the observations of optimization problems with deep learning, presumably because of effective local minima.

http://arxiv.org/pdf/1301.4083

37 Pages, 5 figures, 5 tables JMLR Special Topics on Representation Learning Submission

cs.LG

20130117

20130713

1301.4083

70

the Experimental Analysis of Behavior , 82(3):317{328, 2004. Tapani Raiko, Harri Valpola, and Yann LeCun. Deep learning made easier by linear transformations in perceptrons. In International Conference on Articial Intelligence and Statistics , pages 924{932, 2012. Salah Rifai, Pascal Vincent, Xavier Muller, Xavier Glorot, and Yoshua Bengio. Contractive auto-encoders: Explicit invariance during feature extraction. In ICML'2011 , 2011. 30 Salah Rifai, Yoshua Bengio, Yann Dauphin, and Pascal Vincent. A generative process for sampling contractive auto-encoders. In Proceedings of the Twenty-nine International Conference on Machine Learning (ICML'12) . ACM, 2012. URL http://icml.cc/discuss/2012/590. html . R. Salakhutdinov and G.E. Hinton. Deep Boltzmann machines. In Proceedings of the Twelfth International Conference on Articial Intelligence and Statistics (AISTATS 2009) , volume 8, 2009. Burrhus F. Skinner. Reinforcement today. American Psychologist , 13:94{99, 1958.

1301.4083

Knowledge Matters: Importance of Prior Information for Optimization

We explore the effect of introducing prior information into the intermediate level of neural networks for a learning task on which all the state-of-the-art machine learning algorithms tested failed to learn. We motivate our work from the hypothesis that humans learn such intermediate concepts from other individuals via a form of supervision or guidance using a curriculum. The experiments we have conducted provide positive evidence in favor of this hypothesis. In our experiments, a two-tiered MLP architecture is trained on a dataset with 64x64 binary inputs images, each image with three sprites. The final task is to decide whether all the sprites are the same or one of them is different. Sprites are pentomino tetris shapes and they are placed in an image with different locations using scaling and rotation transformations. The first part of the two-tiered MLP is pre-trained with intermediate-level targets being the presence of sprites at each location, while the second part takes the output of the first part as input and predicts the final task's target binary event. The two-tiered MLP architecture, with a few tens of thousand examples, was able to learn the task perfectly, whereas all other algorithms (include unsupervised pre-training, but also traditional algorithms like SVMs, decision trees and boosting) all perform no better than chance. We hypothesize that the optimization difficulty involved when the intermediate pre-training is not performed is due to the {\em composition} of two highly non-linear tasks. Our findings are also consistent with hypotheses on cultural learning inspired by the observations of optimization problems with deep learning, presumably because of effective local minima.

http://arxiv.org/pdf/1301.4083

37 Pages, 5 figures, 5 tables JMLR Special Topics on Representation Learning Submission

cs.LG

20130117

20130713

1301.4083

71

2009. Burrhus F. Skinner. Reinforcement today. American Psychologist , 13:94{99, 1958. R.J. Solomono. A system for incremental learning based on algorithmic probability. In Proceedings of the Sixth Israeli Conference on Articial Intelligence, Computer Vision and Pattern Recognition , pages 515{527. Citeseer, 1989. G.G. Towell and J.W. Shavlik. Knowledge-based articial neural networks. Articial intelligence , 70(1):119{165, 1994. Tommi Vatanen, Tapani Raiko, Harri Valpola, and Yann LeCun. Pushing stochastic gradient towards second-order methods{backpropagation learning with transformations in nonlinearities. arXiv preprint arXiv:1301.3476 , 2013. Pascal Vincent, Hugo Larochelle, Isabelle Lajoie, Yoshua Bengio, and Pierre-Antoine Manzagol. Stacked denoising autoencoders: Learning useful representations in a deep network with a local denoising criterion. Journal of Machine Learning Research , 11:3371{3408, December 2010. Luis Von Ahn, Manuel Blum, Nicholas J Hopper, and John Langford. Captcha: Using hard

1301.4083

Knowledge Matters: Importance of Prior Information for Optimization

We explore the effect of introducing prior information into the intermediate level of neural networks for a learning task on which all the state-of-the-art machine learning algorithms tested failed to learn. We motivate our work from the hypothesis that humans learn such intermediate concepts from other individuals via a form of supervision or guidance using a curriculum. The experiments we have conducted provide positive evidence in favor of this hypothesis. In our experiments, a two-tiered MLP architecture is trained on a dataset with 64x64 binary inputs images, each image with three sprites. The final task is to decide whether all the sprites are the same or one of them is different. Sprites are pentomino tetris shapes and they are placed in an image with different locations using scaling and rotation transformations. The first part of the two-tiered MLP is pre-trained with intermediate-level targets being the presence of sprites at each location, while the second part takes the output of the first part as input and predicts the final task's target binary event. The two-tiered MLP architecture, with a few tens of thousand examples, was able to learn the task perfectly, whereas all other algorithms (include unsupervised pre-training, but also traditional algorithms like SVMs, decision trees and boosting) all perform no better than chance. We hypothesize that the optimization difficulty involved when the intermediate pre-training is not performed is due to the {\em composition} of two highly non-linear tasks. Our findings are also consistent with hypotheses on cultural learning inspired by the observations of optimization problems with deep learning, presumably because of effective local minima.

http://arxiv.org/pdf/1301.4083

37 Pages, 5 figures, 5 tables JMLR Special Topics on Representation Learning Submission

cs.LG

20130117

20130713

1301.4083

72

2010. Luis Von Ahn, Manuel Blum, Nicholas J Hopper, and John Langford. Captcha: Using hard ai problems for security. In Advances in CryptologyEUROCRYPT 2003 , pages 294{311. Springer, 2003. Jason Weston, Fr ed eric Ratle, and Ronan Collobert. Deep learning via semi-supervised embedding. In William W. Cohen, Andrew McCallum, and Sam T. Roweis, editors, Proceedings of the Twenty-fth International Conference on Machine Learning (ICML'08) , pages 1168{1175, New York, NY, USA, 2008. ACM. ISBN 978-1-60558-205-4. doi: 10.1145/1390156.1390303. Matthew D Zeiler. Adadelta: An adaptive learning rate method. arXiv preprint arXiv:1212.5701 , 2012. 5. Appendix 5.1 Binary-Binary RBMs on Pentomino Dataset We trained binary-binary RBMs (both visible and hidden are binary) on 8 8 patches extracted from the Pentomino Dataset using PCD (stochastic maximum likelihood), a weight decay of .0001 and a sparsity penalty9. We used 256 hidden units and trained by SGD with a batch

1301.4083

Knowledge Matters: Importance of Prior Information for Optimization

We explore the effect of introducing prior information into the intermediate level of neural networks for a learning task on which all the state-of-the-art machine learning algorithms tested failed to learn. We motivate our work from the hypothesis that humans learn such intermediate concepts from other individuals via a form of supervision or guidance using a curriculum. The experiments we have conducted provide positive evidence in favor of this hypothesis. In our experiments, a two-tiered MLP architecture is trained on a dataset with 64x64 binary inputs images, each image with three sprites. The final task is to decide whether all the sprites are the same or one of them is different. Sprites are pentomino tetris shapes and they are placed in an image with different locations using scaling and rotation transformations. The first part of the two-tiered MLP is pre-trained with intermediate-level targets being the presence of sprites at each location, while the second part takes the output of the first part as input and predicts the final task's target binary event. The two-tiered MLP architecture, with a few tens of thousand examples, was able to learn the task perfectly, whereas all other algorithms (include unsupervised pre-training, but also traditional algorithms like SVMs, decision trees and boosting) all perform no better than chance. We hypothesize that the optimization difficulty involved when the intermediate pre-training is not performed is due to the {\em composition} of two highly non-linear tasks. Our findings are also consistent with hypotheses on cultural learning inspired by the observations of optimization problems with deep learning, presumably because of effective local minima.

http://arxiv.org/pdf/1301.4083

37 Pages, 5 figures, 5 tables JMLR Special Topics on Representation Learning Submission

cs.LG

20130117

20130713

1301.4083

73

.0001 and a sparsity penalty9. We used 256 hidden units and trained by SGD with a batch size of 32 and a annealing learning rate (Bengio, 2013b) starting from 1e-3 with annealing rate 9. implemented as TorontoSparsity in pylearn2, see the yaml le in the repository for more details 31 1.000015. The RBM is trained with momentum starting from 0.5. The biases are initialized to -2 in order to get a sparse representation. The RBM is trained for 120 epochs (approximately 50 million updates). After pretraining the RBM, its parameters are used to initialize the rst layer of an SMLPnohints network. As in the usual architecture of the SMLP-nohints on top of P1NN, there is an intermediate layer. Both P1NN and the intermediate layer have a sigmoid nonlinearity, and the intermediate layer has 11 units per location. This SMLP-nohints is trained with Adadelta and standardization at the intermediate layer10. 0 5 10 15 20 25 30 35 40 45 Epoch0.00.10.20.30.40.5Error percentageRBM Test and Training Errors Training Error Test Error

1301.4083

Knowledge Matters: Importance of Prior Information for Optimization

We explore the effect of introducing prior information into the intermediate level of neural networks for a learning task on which all the state-of-the-art machine learning algorithms tested failed to learn. We motivate our work from the hypothesis that humans learn such intermediate concepts from other individuals via a form of supervision or guidance using a curriculum. The experiments we have conducted provide positive evidence in favor of this hypothesis. In our experiments, a two-tiered MLP architecture is trained on a dataset with 64x64 binary inputs images, each image with three sprites. The final task is to decide whether all the sprites are the same or one of them is different. Sprites are pentomino tetris shapes and they are placed in an image with different locations using scaling and rotation transformations. The first part of the two-tiered MLP is pre-trained with intermediate-level targets being the presence of sprites at each location, while the second part takes the output of the first part as input and predicts the final task's target binary event. The two-tiered MLP architecture, with a few tens of thousand examples, was able to learn the task perfectly, whereas all other algorithms (include unsupervised pre-training, but also traditional algorithms like SVMs, decision trees and boosting) all perform no better than chance. We hypothesize that the optimization difficulty involved when the intermediate pre-training is not performed is due to the {\em composition} of two highly non-linear tasks. Our findings are also consistent with hypotheses on cultural learning inspired by the observations of optimization problems with deep learning, presumably because of effective local minima.

http://arxiv.org/pdf/1301.4083

37 Pages, 5 figures, 5 tables JMLR Special Topics on Representation Learning Submission

cs.LG

20130117

20130713

1301.4083

74

0 5 10 15 20 25 30 35 40 45 Epoch0.00.10.20.30.40.5Error percentageRBM Test and Training Errors Training Error Test Error Figure 16: Training and test errors of an SMLP-nohints network whose rst layer is pre-trained as an RBM. Training error reduces to 0% at epoch 42, but test error is still chance. 5.2 Experimental Setup and Hyper-parameters 5.2.1 Decision Trees We used the decision tree implementation in the scikit-learn (Pedregosa et al., 2011) python package which is an implementation of the CART (Regression Trees) algorithm. The CART algorithm constructs the decision tree recursively and partitions the input space such that the samples belonging to the same category are grouped together (Olshen and Stone, 1984). We used The Gini index as the impurity criteria. We evaluated the hyper-parameter congurations with a grid-search. We cross-validated the maximum depth ( maxdepth ) of the tree (for preventing the algorithm to severely overt the training set) and minimum number of samples 10. In our auto-encoder experiments we directly fed features to P2NN without standardization and Adadelta. 32

1301.4083

Knowledge Matters: Importance of Prior Information for Optimization

We explore the effect of introducing prior information into the intermediate level of neural networks for a learning task on which all the state-of-the-art machine learning algorithms tested failed to learn. We motivate our work from the hypothesis that humans learn such intermediate concepts from other individuals via a form of supervision or guidance using a curriculum. The experiments we have conducted provide positive evidence in favor of this hypothesis. In our experiments, a two-tiered MLP architecture is trained on a dataset with 64x64 binary inputs images, each image with three sprites. The final task is to decide whether all the sprites are the same or one of them is different. Sprites are pentomino tetris shapes and they are placed in an image with different locations using scaling and rotation transformations. The first part of the two-tiered MLP is pre-trained with intermediate-level targets being the presence of sprites at each location, while the second part takes the output of the first part as input and predicts the final task's target binary event. The two-tiered MLP architecture, with a few tens of thousand examples, was able to learn the task perfectly, whereas all other algorithms (include unsupervised pre-training, but also traditional algorithms like SVMs, decision trees and boosting) all perform no better than chance. We hypothesize that the optimization difficulty involved when the intermediate pre-training is not performed is due to the {\em composition} of two highly non-linear tasks. Our findings are also consistent with hypotheses on cultural learning inspired by the observations of optimization problems with deep learning, presumably because of effective local minima.

http://arxiv.org/pdf/1301.4083

37 Pages, 5 figures, 5 tables JMLR Special Topics on Representation Learning Submission

cs.LG

20130117

20130713

1301.4083

75

10. In our auto-encoder experiments we directly fed features to P2NN without standardization and Adadelta. 32 Figure 17: Filters learned by the binary-binary RBM after training on the 40k examples. The RBM did learn the edge structure of Pentomino shapes. Figure 18: 100 samples generated from trained RBM. All the generated samples are valid Pentomino shapes. 33 required to create a split ( minsplit). 20 dierent congurations of hyper-parameter values were evaluated. We obtained the best validation error with maxdepth = 300 andminsplit = 8. 5.2.2 Support Vector Machines We used the \Support Vector Classier (SVC)" implementation from the scikit-learn package which in turn uses the libsvm's Support Vector Machine (SVM) implementation. Kernelbased SVMs are non-parametric models that map the data into a high dimensional space and separate dierent classes with hyperplane(s) such that the support vectors for each category will be separated by a large margin. We cross-validated three hyper-parameters of the model using grid-search: C, and the type of kernel( kerneltype).Cis the penalty term (weight

1301.4083

Knowledge Matters: Importance of Prior Information for Optimization

We explore the effect of introducing prior information into the intermediate level of neural networks for a learning task on which all the state-of-the-art machine learning algorithms tested failed to learn. We motivate our work from the hypothesis that humans learn such intermediate concepts from other individuals via a form of supervision or guidance using a curriculum. The experiments we have conducted provide positive evidence in favor of this hypothesis. In our experiments, a two-tiered MLP architecture is trained on a dataset with 64x64 binary inputs images, each image with three sprites. The final task is to decide whether all the sprites are the same or one of them is different. Sprites are pentomino tetris shapes and they are placed in an image with different locations using scaling and rotation transformations. The first part of the two-tiered MLP is pre-trained with intermediate-level targets being the presence of sprites at each location, while the second part takes the output of the first part as input and predicts the final task's target binary event. The two-tiered MLP architecture, with a few tens of thousand examples, was able to learn the task perfectly, whereas all other algorithms (include unsupervised pre-training, but also traditional algorithms like SVMs, decision trees and boosting) all perform no better than chance. We hypothesize that the optimization difficulty involved when the intermediate pre-training is not performed is due to the {\em composition} of two highly non-linear tasks. Our findings are also consistent with hypotheses on cultural learning inspired by the observations of optimization problems with deep learning, presumably because of effective local minima.

http://arxiv.org/pdf/1301.4083

37 Pages, 5 figures, 5 tables JMLR Special Topics on Representation Learning Submission

cs.LG

20130117

20130713

1301.4083

76

using grid-search: C, and the type of kernel( kerneltype).Cis the penalty term (weight decay) for the SVM and is a hyper-parameter that controls the width of the Gaussian for the RBF kernel. For the polynomial kernel, controls the exibility of the classier (degree of the polynomial) as the number of parameters increases (Hsu et al., 2003; Ben-Hur and Weston, 2010). We evaluated forty-two hyper-parameter congurations. That includes, two kernel types:fRBF; Polynomial g; three gammas:f1e2;1e3;1e4gfor the RBF kernel, f1;2;5gfor the polynomial kernel, and seven Cvalues among:f0:1;1;2;4;8;10;16g. As a result of the grid search and cross-validation, we have obtained the best test error by using the RBF kernel, with C= 2 and = 1. 5.2.3 Multi Layer Perceptron We have our own implementation of Multi Layer Perceptron based on the Theano (Bergstra et al., 2010) machine learning libraries. We have selected 2 hidden layers, the rectier activation

1301.4083

Knowledge Matters: Importance of Prior Information for Optimization

We explore the effect of introducing prior information into the intermediate level of neural networks for a learning task on which all the state-of-the-art machine learning algorithms tested failed to learn. We motivate our work from the hypothesis that humans learn such intermediate concepts from other individuals via a form of supervision or guidance using a curriculum. The experiments we have conducted provide positive evidence in favor of this hypothesis. In our experiments, a two-tiered MLP architecture is trained on a dataset with 64x64 binary inputs images, each image with three sprites. The final task is to decide whether all the sprites are the same or one of them is different. Sprites are pentomino tetris shapes and they are placed in an image with different locations using scaling and rotation transformations. The first part of the two-tiered MLP is pre-trained with intermediate-level targets being the presence of sprites at each location, while the second part takes the output of the first part as input and predicts the final task's target binary event. The two-tiered MLP architecture, with a few tens of thousand examples, was able to learn the task perfectly, whereas all other algorithms (include unsupervised pre-training, but also traditional algorithms like SVMs, decision trees and boosting) all perform no better than chance. We hypothesize that the optimization difficulty involved when the intermediate pre-training is not performed is due to the {\em composition} of two highly non-linear tasks. Our findings are also consistent with hypotheses on cultural learning inspired by the observations of optimization problems with deep learning, presumably because of effective local minima.

http://arxiv.org/pdf/1301.4083

37 Pages, 5 figures, 5 tables JMLR Special Topics on Representation Learning Submission

cs.LG

20130117

20130713

1301.4083

77

et al., 2010) machine learning libraries. We have selected 2 hidden layers, the rectier activation function, and 2048 hidden units per layer. We cross-validated three hyper-parameters of the model using random-search, sampling the learning rates in log-domain, and selecting L1 andL2 regularization penalty coecients in sets of xed values, evaluating 64 hyperparameter values. The range of the hyperparameter values are 2[0:0001;1],L12f0:;1e6;1e5;1e4g andL22f0;1e6;1e5g. As a result, the following were selected: L1 = 1e6,L2 = 1e5 and= 0:05. 5.2.4 Random Forests We used scikit-learn's implementation of \Random Forests" decision tree learning. The Random Forests algorithm creates an ensemble of decision trees by randomly selecting for each tree a subset of features and applying bagging to combine the individual decision trees (Breiman, 2001). We have used grid-search and cross-validated the maxdepth ,minsplit, and number

1301.4083

Knowledge Matters: Importance of Prior Information for Optimization

We explore the effect of introducing prior information into the intermediate level of neural networks for a learning task on which all the state-of-the-art machine learning algorithms tested failed to learn. We motivate our work from the hypothesis that humans learn such intermediate concepts from other individuals via a form of supervision or guidance using a curriculum. The experiments we have conducted provide positive evidence in favor of this hypothesis. In our experiments, a two-tiered MLP architecture is trained on a dataset with 64x64 binary inputs images, each image with three sprites. The final task is to decide whether all the sprites are the same or one of them is different. Sprites are pentomino tetris shapes and they are placed in an image with different locations using scaling and rotation transformations. The first part of the two-tiered MLP is pre-trained with intermediate-level targets being the presence of sprites at each location, while the second part takes the output of the first part as input and predicts the final task's target binary event. The two-tiered MLP architecture, with a few tens of thousand examples, was able to learn the task perfectly, whereas all other algorithms (include unsupervised pre-training, but also traditional algorithms like SVMs, decision trees and boosting) all perform no better than chance. We hypothesize that the optimization difficulty involved when the intermediate pre-training is not performed is due to the {\em composition} of two highly non-linear tasks. Our findings are also consistent with hypotheses on cultural learning inspired by the observations of optimization problems with deep learning, presumably because of effective local minima.

http://arxiv.org/pdf/1301.4083

37 Pages, 5 figures, 5 tables JMLR Special Topics on Representation Learning Submission

cs.LG

20130117

20130713

1301.4083

78

2001). We have used grid-search and cross-validated the maxdepth ,minsplit, and number of trees (nestimators ). We have done the grid-search on the following hyperparameter values,nestimators2 f5;10;15;25;50g,maxdepth2 f100;300;600;900g, andminsplits2 f1;4;16g. We obtained the best validation error with maxdepth = 300,minsplit = 4 and nestimators = 10. 5.2.5 k-Nearest Neighbors We used scikit-learn's implementation of k-Nearest Neighbors (k-NN). k-NN is an instancebased, lazy learning algorithm that selects the training examples closest in Euclidean distance to the input query. It assigns a class label to the test example based on the categories of the kclosest neighbors. The hyper-parameters we have evaluated in the cross-validation are the number of neighbors ( k) andweights . Theweights hyper-parameter can be either \uniform" or 34 \distance". With \uniform", the value assigned to the query point is computed by the majority vote of the nearest neighbors. With \distance", each value assigned to the query point is computed by weighted majority votes where the weights are computed with the inverse distance

1301.4083

Knowledge Matters: Importance of Prior Information for Optimization

We explore the effect of introducing prior information into the intermediate level of neural networks for a learning task on which all the state-of-the-art machine learning algorithms tested failed to learn. We motivate our work from the hypothesis that humans learn such intermediate concepts from other individuals via a form of supervision or guidance using a curriculum. The experiments we have conducted provide positive evidence in favor of this hypothesis. In our experiments, a two-tiered MLP architecture is trained on a dataset with 64x64 binary inputs images, each image with three sprites. The final task is to decide whether all the sprites are the same or one of them is different. Sprites are pentomino tetris shapes and they are placed in an image with different locations using scaling and rotation transformations. The first part of the two-tiered MLP is pre-trained with intermediate-level targets being the presence of sprites at each location, while the second part takes the output of the first part as input and predicts the final task's target binary event. The two-tiered MLP architecture, with a few tens of thousand examples, was able to learn the task perfectly, whereas all other algorithms (include unsupervised pre-training, but also traditional algorithms like SVMs, decision trees and boosting) all perform no better than chance. We hypothesize that the optimization difficulty involved when the intermediate pre-training is not performed is due to the {\em composition} of two highly non-linear tasks. Our findings are also consistent with hypotheses on cultural learning inspired by the observations of optimization problems with deep learning, presumably because of effective local minima.

http://arxiv.org/pdf/1301.4083

37 Pages, 5 figures, 5 tables JMLR Special Topics on Representation Learning Submission

cs.LG

20130117

20130713

1301.4083

79

vote of the nearest neighbors. With \distance", each value assigned to the query point is computed by weighted majority votes where the weights are computed with the inverse distance between the query point and the neighbors. We have used nneighbours2f1;2;4;6;8;12gand weights2f"uniform ";"distance "gfor hyper-parameter search. As a result of cross-validation and grid search, we obtained the best validation error with k= 2 andweights =\uniform". 5.2.6 Convolutional Neural Nets We used a Theano (Bergstra et al., 2010) implementation of Convolutional Neural Networks (CNN) from the deep learning tutorial at deeplearning.net , which is based on a vanilla version of a CNN LeCun et al. (1998). Our CNN has two convolutional layers. Following each convolutional layer, we have a max-pooling layer. On top of the convolution-poolingconvolution-pooling layers there is an MLP with one hidden layer. In the cross-validation we have sampled 36 learning rates in log-domain in the range [0 :0001;1] and the number of lters

1301.4083

Knowledge Matters: Importance of Prior Information for Optimization

We explore the effect of introducing prior information into the intermediate level of neural networks for a learning task on which all the state-of-the-art machine learning algorithms tested failed to learn. We motivate our work from the hypothesis that humans learn such intermediate concepts from other individuals via a form of supervision or guidance using a curriculum. The experiments we have conducted provide positive evidence in favor of this hypothesis. In our experiments, a two-tiered MLP architecture is trained on a dataset with 64x64 binary inputs images, each image with three sprites. The final task is to decide whether all the sprites are the same or one of them is different. Sprites are pentomino tetris shapes and they are placed in an image with different locations using scaling and rotation transformations. The first part of the two-tiered MLP is pre-trained with intermediate-level targets being the presence of sprites at each location, while the second part takes the output of the first part as input and predicts the final task's target binary event. The two-tiered MLP architecture, with a few tens of thousand examples, was able to learn the task perfectly, whereas all other algorithms (include unsupervised pre-training, but also traditional algorithms like SVMs, decision trees and boosting) all perform no better than chance. We hypothesize that the optimization difficulty involved when the intermediate pre-training is not performed is due to the {\em composition} of two highly non-linear tasks. Our findings are also consistent with hypotheses on cultural learning inspired by the observations of optimization problems with deep learning, presumably because of effective local minima.

http://arxiv.org/pdf/1301.4083

37 Pages, 5 figures, 5 tables JMLR Special Topics on Representation Learning Submission

cs.LG

20130117

20130713

1301.4083

80

have sampled 36 learning rates in log-domain in the range [0 :0001;1] and the number of lters from the range [10 ;20;30;40;50;60] uniformly. For the rst convolutional layer we used 9 9 receptive elds in order to guarantee that each object ts inside the receptive eld. As a result of random hyperparameter search and doing manual hyperparameter search on the validation dataset, the following values were selected: The number of features used for the rst layer is 30 and the second layer is 60. For the second convolutional layer, 7 7 receptive elds. The stride for both convolutional layers is 1. Convolved images are downsampled by a factor of 2 2 at each pooling operation. The learning rate for CNN is 0.01 and it was trained for 8 epochs. 5.2.7 Maxout Convolutional Neural Nets We used the pylearn2 ( https://github.com/lisa-lab/pylearn2 ) implementation of maxout convolutional networks (Goodfellow et al., 2013). There are two convolutional layers in the

1301.4083

Knowledge Matters: Importance of Prior Information for Optimization

We explore the effect of introducing prior information into the intermediate level of neural networks for a learning task on which all the state-of-the-art machine learning algorithms tested failed to learn. We motivate our work from the hypothesis that humans learn such intermediate concepts from other individuals via a form of supervision or guidance using a curriculum. The experiments we have conducted provide positive evidence in favor of this hypothesis. In our experiments, a two-tiered MLP architecture is trained on a dataset with 64x64 binary inputs images, each image with three sprites. The final task is to decide whether all the sprites are the same or one of them is different. Sprites are pentomino tetris shapes and they are placed in an image with different locations using scaling and rotation transformations. The first part of the two-tiered MLP is pre-trained with intermediate-level targets being the presence of sprites at each location, while the second part takes the output of the first part as input and predicts the final task's target binary event. The two-tiered MLP architecture, with a few tens of thousand examples, was able to learn the task perfectly, whereas all other algorithms (include unsupervised pre-training, but also traditional algorithms like SVMs, decision trees and boosting) all perform no better than chance. We hypothesize that the optimization difficulty involved when the intermediate pre-training is not performed is due to the {\em composition} of two highly non-linear tasks. Our findings are also consistent with hypotheses on cultural learning inspired by the observations of optimization problems with deep learning, presumably because of effective local minima.

http://arxiv.org/pdf/1301.4083

37 Pages, 5 figures, 5 tables JMLR Special Topics on Representation Learning Submission

cs.LG

20130117

20130713

1301.4083

81

convolutional networks (Goodfellow et al., 2013). There are two convolutional layers in the selected architecture, without any pooling. In the last convolutional layer, there is a maxout non-linearity. The following were selected by cross-validation: learning rate, number of channels for the both convolution layers, number of kernels for the second layer and number of units and pieces per maxout unit in the last layer, a linearly decaying learning rate, momentum starting from 0.5 and saturating to 0.8 at the 200'th epoch. Random search for the hyperparameters was used to evaluate 48 dierent hyperparameter congurations on the validation dataset. For the rst convolutional layer, 8 8 kernels were selected to make sure that each Pentomino shape ts into the kernel. Early stopping was used and test error on the model that has the best validation error is reported. Using norm constraint on the fan-in of the nal softmax units yields slightly better result on the validation dataset. As a result of cross-validation and manually tuning the hyperparameters we used the following hyperparameters: 16 channels per convolutional layer. 600 hidden units for the maxout layer. 6x6 kernels for the second convolutional layer. 35

1301.4083

Knowledge Matters: Importance of Prior Information for Optimization

We explore the effect of introducing prior information into the intermediate level of neural networks for a learning task on which all the state-of-the-art machine learning algorithms tested failed to learn. We motivate our work from the hypothesis that humans learn such intermediate concepts from other individuals via a form of supervision or guidance using a curriculum. The experiments we have conducted provide positive evidence in favor of this hypothesis. In our experiments, a two-tiered MLP architecture is trained on a dataset with 64x64 binary inputs images, each image with three sprites. The final task is to decide whether all the sprites are the same or one of them is different. Sprites are pentomino tetris shapes and they are placed in an image with different locations using scaling and rotation transformations. The first part of the two-tiered MLP is pre-trained with intermediate-level targets being the presence of sprites at each location, while the second part takes the output of the first part as input and predicts the final task's target binary event. The two-tiered MLP architecture, with a few tens of thousand examples, was able to learn the task perfectly, whereas all other algorithms (include unsupervised pre-training, but also traditional algorithms like SVMs, decision trees and boosting) all perform no better than chance. We hypothesize that the optimization difficulty involved when the intermediate pre-training is not performed is due to the {\em composition} of two highly non-linear tasks. Our findings are also consistent with hypotheses on cultural learning inspired by the observations of optimization problems with deep learning, presumably because of effective local minima.

http://arxiv.org/pdf/1301.4083

37 Pages, 5 figures, 5 tables JMLR Special Topics on Representation Learning Submission

cs.LG

20130117

20130713

1301.4083

82

16 channels per convolutional layer. 600 hidden units for the maxout layer. 6x6 kernels for the second convolutional layer. 35 5 pieces for the convolution layers and 4 pieces for the maxout layer per maxout units. We decayed the learning rate by the factor of 0.001 and the initial learning rate is 0.026367. But we scaled the learning rate of the second convolutional layer by a constant factor of 0.6. The norm constraint (on the incoming weights of each unit) is 1.9365. Figure 19 shows the rst layer lters of the maxout convolutional net, after being trained on the 80k training set for 85 epochs. Figure 19: Maxout convolutional net rst layer lters. Most of the lters were able to learn the basic edge structure of the Pentomino shapes. 5.2.8 Stacked Denoising Auto-Encoders Denoising Auto-Encoders (DAE) are a form of regularized auto-encoder (Bengio et al., 2013). The DAE forces the hidden layer to discover more robust features and prevents it from simply learning the identity by reconstructing the input from a corrupted version of it (Vincent et al.,

1301.4083

Knowledge Matters: Importance of Prior Information for Optimization

We explore the effect of introducing prior information into the intermediate level of neural networks for a learning task on which all the state-of-the-art machine learning algorithms tested failed to learn. We motivate our work from the hypothesis that humans learn such intermediate concepts from other individuals via a form of supervision or guidance using a curriculum. The experiments we have conducted provide positive evidence in favor of this hypothesis. In our experiments, a two-tiered MLP architecture is trained on a dataset with 64x64 binary inputs images, each image with three sprites. The final task is to decide whether all the sprites are the same or one of them is different. Sprites are pentomino tetris shapes and they are placed in an image with different locations using scaling and rotation transformations. The first part of the two-tiered MLP is pre-trained with intermediate-level targets being the presence of sprites at each location, while the second part takes the output of the first part as input and predicts the final task's target binary event. The two-tiered MLP architecture, with a few tens of thousand examples, was able to learn the task perfectly, whereas all other algorithms (include unsupervised pre-training, but also traditional algorithms like SVMs, decision trees and boosting) all perform no better than chance. We hypothesize that the optimization difficulty involved when the intermediate pre-training is not performed is due to the {\em composition} of two highly non-linear tasks. Our findings are also consistent with hypotheses on cultural learning inspired by the observations of optimization problems with deep learning, presumably because of effective local minima.

http://arxiv.org/pdf/1301.4083

37 Pages, 5 figures, 5 tables JMLR Special Topics on Representation Learning Submission

cs.LG

20130117

20130713

1301.4083

83

The DAE forces the hidden layer to discover more robust features and prevents it from simply learning the identity by reconstructing the input from a corrupted version of it (Vincent et al., 2010). Two DAEs were stacked, resulting in an unsupervised transformation with two hidden layers of 1024 units each. Parameters of all layers are then ne-tuned with supervised netuning using logistic regression as the classier and SGD as the gradient-based optimization algorithm. The stochastic corruption process is binomial (0 or 1 replacing each input value, with probability 0.2). The selected learning rate is 0= 0:01 for the DAe and 1= 0:1 for supervised ne-tuning. Both L1 and L2 penalty for the DAEs and for the logistic regression layer are set to 1e-6. CAE+MLP with Supervised Finetuning: A regularized auto-encoder which sometimes outperforms the DAE is the Contractive Auto-Encoder (CAE), (Rifai et al., 2012), which penalizes the Frobenius norm of the Jacobian matrix of derivatives of the hidden units with respect to the CAE's inputs. The CAE serves as pre-training for an MLP, and in the supervised

1301.4083

Knowledge Matters: Importance of Prior Information for Optimization

We explore the effect of introducing prior information into the intermediate level of neural networks for a learning task on which all the state-of-the-art machine learning algorithms tested failed to learn. We motivate our work from the hypothesis that humans learn such intermediate concepts from other individuals via a form of supervision or guidance using a curriculum. The experiments we have conducted provide positive evidence in favor of this hypothesis. In our experiments, a two-tiered MLP architecture is trained on a dataset with 64x64 binary inputs images, each image with three sprites. The final task is to decide whether all the sprites are the same or one of them is different. Sprites are pentomino tetris shapes and they are placed in an image with different locations using scaling and rotation transformations. The first part of the two-tiered MLP is pre-trained with intermediate-level targets being the presence of sprites at each location, while the second part takes the output of the first part as input and predicts the final task's target binary event. The two-tiered MLP architecture, with a few tens of thousand examples, was able to learn the task perfectly, whereas all other algorithms (include unsupervised pre-training, but also traditional algorithms like SVMs, decision trees and boosting) all perform no better than chance. We hypothesize that the optimization difficulty involved when the intermediate pre-training is not performed is due to the {\em composition} of two highly non-linear tasks. Our findings are also consistent with hypotheses on cultural learning inspired by the observations of optimization problems with deep learning, presumably because of effective local minima.

http://arxiv.org/pdf/1301.4083

37 Pages, 5 figures, 5 tables JMLR Special Topics on Representation Learning Submission

cs.LG

20130117

20130713

1301.4083

84

respect to the CAE's inputs. The CAE serves as pre-training for an MLP, and in the supervised ne-tuning state, the Adagrad method was used to automatically tune the learning rate (Duchi et al., 2010). After training a CAE with 100 sigmoidal units patch-wise, the features extracted on each patch are concatenated and fed as input to an MLP. The selected Jacobian penalty coecient is 2, the learning rate for pre-training is 0.082 with batch size of 200 and 200 epochs of unsupervised learning are performed on the training set. For supervised netuning, the learning rate is 0.12 over 100 epochs, L1 and L2 regularization penalty terms respectively are 1e-4 and 1e-6, and the top-level MLP has 6400 hidden units. 36 Greedy Layerwise CAE+DAE Supervised Finetuning: For this experiment we stack a CAE with sigmoid non-linearities and then a DAE with rectier non-linearities during the pretraining phase. As recommended by Glorot et al. (2011b) we have used a softplus nonlinearity for reconstruction, softplus (x) =log(1 +ex). We used an L1 penalty on the rectier outputs

1301.4083

Knowledge Matters: Importance of Prior Information for Optimization

We explore the effect of introducing prior information into the intermediate level of neural networks for a learning task on which all the state-of-the-art machine learning algorithms tested failed to learn. We motivate our work from the hypothesis that humans learn such intermediate concepts from other individuals via a form of supervision or guidance using a curriculum. The experiments we have conducted provide positive evidence in favor of this hypothesis. In our experiments, a two-tiered MLP architecture is trained on a dataset with 64x64 binary inputs images, each image with three sprites. The final task is to decide whether all the sprites are the same or one of them is different. Sprites are pentomino tetris shapes and they are placed in an image with different locations using scaling and rotation transformations. The first part of the two-tiered MLP is pre-trained with intermediate-level targets being the presence of sprites at each location, while the second part takes the output of the first part as input and predicts the final task's target binary event. The two-tiered MLP architecture, with a few tens of thousand examples, was able to learn the task perfectly, whereas all other algorithms (include unsupervised pre-training, but also traditional algorithms like SVMs, decision trees and boosting) all perform no better than chance. We hypothesize that the optimization difficulty involved when the intermediate pre-training is not performed is due to the {\em composition} of two highly non-linear tasks. Our findings are also consistent with hypotheses on cultural learning inspired by the observations of optimization problems with deep learning, presumably because of effective local minima.

http://arxiv.org/pdf/1301.4083

37 Pages, 5 figures, 5 tables JMLR Special Topics on Representation Learning Submission

cs.LG

20130117

20130713

1301.4083

85

for reconstruction, softplus (x) =log(1 +ex). We used an L1 penalty on the rectier outputs to obtain a sparser representation with rectier non-linearity and L2 regularization to keep the non-zero weights small. The main dierence between the DAE and CAE is that the DAE yields more robust reconstruction whereas the CAE obtains more robust features (Rifai et al., 2011). As seen on Figure 7 the weights U and V are shared on each patch and we concatenate the outputs of the last auto-encoder on each patch to feed it as an input to an MLP with a large hidden layer. We used 400 hidden units for the CAE and 100 hidden units for DAE. The learning rate used for the CAE is 0.82 and for DAE it is 9*1e-3. The corruption level for the DAE (binomial noise) is 0.25 and the contraction level for the CAE is 2.0. The L1 regularization penalty for the DAE is 2.25*1e-4 and the L2 penalty is 9.5*1e-5. For the supervised netuning phase the

1301.4083

Knowledge Matters: Importance of Prior Information for Optimization

We explore the effect of introducing prior information into the intermediate level of neural networks for a learning task on which all the state-of-the-art machine learning algorithms tested failed to learn. We motivate our work from the hypothesis that humans learn such intermediate concepts from other individuals via a form of supervision or guidance using a curriculum. The experiments we have conducted provide positive evidence in favor of this hypothesis. In our experiments, a two-tiered MLP architecture is trained on a dataset with 64x64 binary inputs images, each image with three sprites. The final task is to decide whether all the sprites are the same or one of them is different. Sprites are pentomino tetris shapes and they are placed in an image with different locations using scaling and rotation transformations. The first part of the two-tiered MLP is pre-trained with intermediate-level targets being the presence of sprites at each location, while the second part takes the output of the first part as input and predicts the final task's target binary event. The two-tiered MLP architecture, with a few tens of thousand examples, was able to learn the task perfectly, whereas all other algorithms (include unsupervised pre-training, but also traditional algorithms like SVMs, decision trees and boosting) all perform no better than chance. We hypothesize that the optimization difficulty involved when the intermediate pre-training is not performed is due to the {\em composition} of two highly non-linear tasks. Our findings are also consistent with hypotheses on cultural learning inspired by the observations of optimization problems with deep learning, presumably because of effective local minima.

http://arxiv.org/pdf/1301.4083

37 Pages, 5 figures, 5 tables JMLR Special Topics on Representation Learning Submission

cs.LG

20130117

20130713

1301.4083

86

the DAE is 2.25*1e-4 and the L2 penalty is 9.5*1e-5. For the supervised netuning phase the learning rate used is 4*1e-4 with L1 and L2 penalties respectively 1e-5 and 1e-6. The top-level MLP has 6400 hidden units. The auto-encoders are each trained for 150 epochs while the whole MLP is ne-tuned for 50 epochs. Greedy Layerwise DAE+DAE Supervised Finetuning: For this architecture, we have trained two layers of denoising auto-encoders greedily and performed supervised netuning after unsupervised pre-training. The motivation for using two denoising auto-encoders is the fact that rectier nonlinearities work well with the deep networks but it is dicult to train CAEs with the rectier non-linearity. We have used the same type of denoising auto-encoder that is used for the greedy layerwise CAE+DAE supervised netuning experiment. In this experiment we have used 400 hidden units for the rst layer DAE and 100 hidden units for the second layer DAE. The other hyperparameters for DAE and supervised netuning

1301.4083

Knowledge Matters: Importance of Prior Information for Optimization

We explore the effect of introducing prior information into the intermediate level of neural networks for a learning task on which all the state-of-the-art machine learning algorithms tested failed to learn. We motivate our work from the hypothesis that humans learn such intermediate concepts from other individuals via a form of supervision or guidance using a curriculum. The experiments we have conducted provide positive evidence in favor of this hypothesis. In our experiments, a two-tiered MLP architecture is trained on a dataset with 64x64 binary inputs images, each image with three sprites. The final task is to decide whether all the sprites are the same or one of them is different. Sprites are pentomino tetris shapes and they are placed in an image with different locations using scaling and rotation transformations. The first part of the two-tiered MLP is pre-trained with intermediate-level targets being the presence of sprites at each location, while the second part takes the output of the first part as input and predicts the final task's target binary event. The two-tiered MLP architecture, with a few tens of thousand examples, was able to learn the task perfectly, whereas all other algorithms (include unsupervised pre-training, but also traditional algorithms like SVMs, decision trees and boosting) all perform no better than chance. We hypothesize that the optimization difficulty involved when the intermediate pre-training is not performed is due to the {\em composition} of two highly non-linear tasks. Our findings are also consistent with hypotheses on cultural learning inspired by the observations of optimization problems with deep learning, presumably because of effective local minima.

http://arxiv.org/pdf/1301.4083

37 Pages, 5 figures, 5 tables JMLR Special Topics on Representation Learning Submission

cs.LG

20130117

20130713

1301.4083

87

units for the second layer DAE. The other hyperparameters for DAE and supervised netuning are the same as with the CAE+DAE MLP Supervised Finetuning experiment. 37

1301.4083

Knowledge Matters: Importance of Prior Information for Optimization

We explore the effect of introducing prior information into the intermediate level of neural networks for a learning task on which all the state-of-the-art machine learning algorithms tested failed to learn. We motivate our work from the hypothesis that humans learn such intermediate concepts from other individuals via a form of supervision or guidance using a curriculum. The experiments we have conducted provide positive evidence in favor of this hypothesis. In our experiments, a two-tiered MLP architecture is trained on a dataset with 64x64 binary inputs images, each image with three sprites. The final task is to decide whether all the sprites are the same or one of them is different. Sprites are pentomino tetris shapes and they are placed in an image with different locations using scaling and rotation transformations. The first part of the two-tiered MLP is pre-trained with intermediate-level targets being the presence of sprites at each location, while the second part takes the output of the first part as input and predicts the final task's target binary event. The two-tiered MLP architecture, with a few tens of thousand examples, was able to learn the task perfectly, whereas all other algorithms (include unsupervised pre-training, but also traditional algorithms like SVMs, decision trees and boosting) all perform no better than chance. We hypothesize that the optimization difficulty involved when the intermediate pre-training is not performed is due to the {\em composition} of two highly non-linear tasks. Our findings are also consistent with hypotheses on cultural learning inspired by the observations of optimization problems with deep learning, presumably because of effective local minima.

http://arxiv.org/pdf/1301.4083

37 Pages, 5 figures, 5 tables JMLR Special Topics on Representation Learning Submission

cs.LG

20130117

20130713

1312.4659

0

DeepPose: Human Pose Estimation via Deep Neural Networks Alexander Toshev Christian Szegedy Google 1600 Amphitheatre Pkwy Mountain View, CA 94043 toshev,szegedy@google.com Figure 1. Besides extreme variability in articulations, many of the joints are barely visible. We can guess the location of the right arm in the left image only because we see the rest of the pose and anticipate the motion or activity of the person. Similarly, the left body half of the person on the right is not visible at all. These are examples of the need for holistic reasoning . We believe that DNNs can naturally provide such type of reasoning. Abstract We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-ofart or better performance on four academic benchmarks of diverse real-world images. 1. Introduction The problem of human pose estimation, defined as the

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

1

diverse real-world images. 1. Introduction The problem of human pose estimation, defined as the problem of localization of human joints, has enjoyed substantial attention in the computer vision community. In Fig. 1, one can see some of the challenges of this problem – strong articulations, small and barely visible joints, occlusions and the need to capture the context. The main stream of work in this field has been motivatedmainly by the first challenge, the need to search in the large space of all possible articulated poses. Part-based models lend themselves naturally to model articulations ([16, 8]) and in the recent years a variety of models with efficient inference have been proposed ([6, 19]). The above efficiency, however, is achieved at the cost of limited expressiveness – the use of local detectors, which reason in many cases about a single part, and most importantly by modeling only a small subset of all interactions between body parts. These limitations, as exemplified in Fig. 1, have been recognized and methods reasoning about pose in a holistic manner have been proposed [15, 21] but with limited success in real-world problems. In this work we ascribe to this holistic view of human pose estimation. We capitalize on recent developments of

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

2

with limited success in real-world problems. In this work we ascribe to this holistic view of human pose estimation. We capitalize on recent developments of deep learning and propose a novel algorithm based on a Deep Neural Network (DNN). DNNs have shown outstanding performance on visual classification tasks [14] and more recently on object localization [23, 9]. However, the question of applying DNNs for precise localization of articulated objects has largely remained unanswered. In this paper we attempt to cast a light on this question and present a simple and yet powerful formulation of holistic human pose estimation as a DNN. We formulate the pose estimation as a joint regression problem and show how to successfully cast it in DNN settings. The location of each body joint is regressed to using as an input the full image and a 7-layered generic convolutional DNN. There are two advantages of this formulation. First, the DNN is capable of capturing the full context of each body joint – each joint regressor uses the full image as a signal. Second, the approach is substantially simpler to formulate than methods based on graphical models – no need to explicitly design feature representations and detectors for parts; no need to explicitly design a model topology and interactions between joints. Instead, we show that a generic convolutional DNN can be learned for this problem.

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

3

and interactions between joints. Instead, we show that a generic convolutional DNN can be learned for this problem. Further, we propose a cascade of DNN-based pose predictors. Such a cascade allows for increased precision of 1arXiv:1312.4659v3 [cs.CV] 20 Aug 2014 joint localization. Starting with an initial pose estimation, based on the full image, we learn DNN-based regressors which refines the joint predictions by using higher resolution sub-images. We show state-of-art results or better than state-of-art on four widely used benchmarks against all reported results. We show that our approach performs well on images of people which exhibit strong variation in appearance as well as articulations. Finally, we show generalization performance by cross-dataset evaluation. 2. Related Work The idea of representing articulated objects in general, and human pose in particular, as a graph of parts has been advocated from the early days of computer vision [16]. The so called Pictorial Strictures (PSs), introduced by Fishler and Elschlager [8], were made tractable and practical by Felzenszwalb and Huttenlocher [6] using the distance transform trick. As a result, a wide variety of PS-based models

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

4

Felzenszwalb and Huttenlocher [6] using the distance transform trick. As a result, a wide variety of PS-based models with practical significance were subsequently developed. The above tractability, however, comes with the limitation of having a tree-based pose models with simple binary potential not depending on image data. As a result, research has focused on enriching the representational power of the models while maintaining tractability. Earlier attempts to achieve this were based on richer part detectors [19, 1, 4]. More recently, a wide variety of models expressing complex joint relationships were proposed. Yang and Ramanan [27] use a mixture model of parts. Mixture models on the full model scale, by having mixture of PSs, have been studied by Johnson and Everingham [13]. Richer higher-order spatial relationships were captured in a hierarchical model by Tian et al. [25]. A different approach to capture higherorder relationship is through image-dependent PS models, which can be estimated via a global classifier [26, 20, 18]. Approaches which ascribe to our philosophy of reasoning about pose in a holistic manner have shown limited practicality. Mori and Malik [15] try to find for each test image the closest exemplar from a set of labeled images

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

5

practicality. Mori and Malik [15] try to find for each test image the closest exemplar from a set of labeled images and transfer the joint locations. A similar nearest neighbor setup is employed by Shakhnarovich et al. [21], who however use locality sensitive hashing. More recently, Gkioxari et al. [10] propose a semi-global classifier for part configuration. This formulation has shown very good results on real-world data, however, it is based on linear classifiers with less expressive representation than ours and is tested on arms only. Finally, the idea of pose regression has been employed by Ionescu et al. [11], however they reason about 3D pose. The closest work to ours uses convolution NNs together with Neighborhood Component Analysis to regress toward a point in an embedding representing pose [24]. However, this work does not employ a cascade of networks. Cascadesof DNN regressors have been used for localization, however of facial points [22]. On the related problem of face pose estimation, Osadchy et al. [17] employ a NN-based pose embedding trained with a contrastive loss. 3. Deep Learning Model for Pose Estimation

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

6

estimation, Osadchy et al. [17] employ a NN-based pose embedding trained with a contrastive loss. 3. Deep Learning Model for Pose Estimation We use the following notation. To express a pose, we encode the locations of all kbody joints in pose vector defined asy= (:::;yT i;:::)T;i2f1;:::;kg, where yicontains thexandycoordinates of the ithjoint. A labeled image is denoted by (x;y)wherexstands for the image data and y is the ground truth pose vector. Further, since the joint coordinates are in absolute image coordinates, it proves beneficial to normalize them w. r. t. a boxbbounding the human body or parts of it. In a trivial case, the box can denote the full image. Such a box is defined by its center bc2R2as well as width bwand height bh:b= (bc;bw;bh). Then the joint yican be translated by the box center and scaled by the box size which we refer to as normalization by b: N(yi;b) =1=bw 0 0 1=bh (yibc) (1)

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

7

as normalization by b: N(yi;b) =1=bw 0 0 1=bh (yibc) (1) Further, we can apply the same normalization to the elements of pose vector N(y;b) = (:::;N (yi;b)T;:::)Tresulting in a normalized pose vector . Finally, with a slight abuse of notation, we use N(x;b)to denote a crop of the imagexby the bounding box b, which de facto normalizes the image by the box. For brevity we denote by N(
)normalization with bbeing the full image box. 3.1. Pose Estimation as DNN-based Regression In this work, we treat the problem of pose estimation as regression, where the we train and use a function (x;)2 R2kwhich for an image xregresses to a normalized pose vector, where denotes the parameters of the model. Thus, using the normalization transformation from Eq. (1) the pose prediction yin absolute image coordinates reads y=N1( (N(x);)) (2) Despite its simple formulation, the power and complexity of the method is in , which is based on a convolutional

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

8

Despite its simple formulation, the power and complexity of the method is in , which is based on a convolutional Deep Neural Network (DNN). Such a convolutional network consists of several layers – each being a linear transformation followed by a non-linear one. The first layer takes as input an image of predefined size and has a size equal to the number of pixels times three color channels. The last layer outputs the target values of the regression, in our case 2kjoint coordinates. We base the architecture of the on the work by Krizhevsky et al. [14] for image classification since it has shown outstanding results on object localization as well (xi, yi) (x(s-1)i, y (s-1) i)xsi - x(s-1)iysi - y(s-1)iInitial stageStage s send refined values to next stage220 x 220 DNN-based regressor 27 x 27 x 128 13 x 13 x 192 13 x 13 x192 13 x 13 x192 4096 4096 55 x 55 x 48xiyi...DNN-based refiner 27 x 27 x 128 13 x 13 x 192 13 x 13 x192

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

9

4096 55 x 55 x 48xiyi...DNN-based refiner 27 x 27 x 128 13 x 13 x 192 13 x 13 x192 13 x 13 x192 4096 4096 55 x 55 x 48Figure 2. Left: schematic view of the DNN-based pose regression. We visualize the network layers with their corresponding dimensions, where convolutional layers are in blue, while fully connected ones are in green. We do not show the parameter free layers. Right: at stage s, a refining regressor is applied on a sub image to refine a prediction from the previous stage. [23]. In a nutshell, the network consists of 7layers (see Fig. 2 left). Denote by Ca convolutional layer, by LRN a local response normalization layer, Pa pooling layer and byFa fully connected layer. Only CandFlayers contain learnable parameters, while the rest are parameter free. Both CandFlayers consist of a linear transformation followed by a nonlinear one, which in our case is a rectified linear unit. For Clayers, the size is defined as widthheightdepth, where the first two dimensions have a spatial meaning while the depth defines

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

10

the number of filters. If we write the size of each layer in parentheses, then the network can be described concisely asC(555596)LRNPC(2727256) LRNPC(1313384)C(1313384) C(1313256)PF(4096)F(4096) . The filter size for the first two Clayers is 1111and55and for the remaining three is 33. Pooling is applied after three layers and contributes to increased performance despite the reduction of resolution. The input to the net is an image of220220which via stride of 4is fed into the network. The total number of parameters in the above model is about 40M. For further details, we refer the reader to [14]. The use of a generic DNN architecture is motivated by its outstanding results on both classification and localization problems. In the experimental section we show that such a generic architecture can be used to learn a model resulting in state-of-art or better performance on pose estimation as well. Further, such a model is a truly holistic one — the

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

11

generic architecture can be used to learn a model resulting in state-of-art or better performance on pose estimation as well. Further, such a model is a truly holistic one — the final joint location estimate is based on a complex nonlinear transformation of the full image. Additionally, the use of a DNN obviates the need to design a domain specific pose model. Instead such a model and the features are learned from the data. Although the regression loss does not model explicit interactions between joints, such are implicitly captured by all of the 7hidden layers – all the internal features are shared by all joint regressors. Training The difference to [14] is the loss. Instead of a classification loss, we train a linear regression on top of thelast network layer to predict a pose vector by minimizing L2distance between the prediction and the true pose vector. Since the ground truth pose vector is defined in absolute image coordinates and poses vary in size from image to image, we normalize our training set Dusing the normalization from Eq. (1): DN=f(N(x);N(y))j(x;y)2Dg (3) Then theL2loss for obtaining optimal network parameters reads: arg min X

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

12

Then theL2loss for obtaining optimal network parameters reads: arg min X (x;y)2DNkX i=1jjyi i(x;)jj2 2 (4) For clarity we write out the optimization over individual joints. It should be noted, that the above objective can be used even if for some images not all joints are labeled. In this case, the corresponding terms in the sum would be omitted. The above parameters are optimized for using Backpropagation in a distributed online implementation. For each mini-batch of size 128, adaptive gradient updates are computed [3]. The learning rate, as the most important parameter, is set to 0:0005 . Since the model has large number of parameters and the used datasets are of relatively small size, we augment the data using large number of randomly translated image crops (see Sec. 3.2), left/right flips as well as DropOut regularization for the Flayers set to 0:6. 3.2. Cascade of Pose Regressors The pose formulation from the previous section has the advantage that the joint estimation is based on the full image and thus relies on context. However, due to its fixed input size of 220220, the network has limited capacity

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

13

input size of 220220, the network has limited capacity to look at detail – it learns filters capturing pose properties at coarse scale. These are necessary to estimate rough pose but insufficient to always precisely localize the body joints. Note that we cannot easily increase the input size since this will increase the already large number of parameters. In order to achieve better precision, we propose to train a cascade of pose regressors. At the first stage, the cascade starts off by estimating an initial pose as outlined in the previous section. At subsequent stages, additional DNN regressors are trained to predict a displacement of the joint locations from previous stage to the true location. Thus, each subsequent stage can be thought of as a refinement of the currently predicted pose, as shown in Fig. 2. Further, each subsequent stage uses the predicted joint locations to focus on the relevant parts of the image – subimages are cropped around the predicted joint location from previous stage and the pose displacement regressor for this joint is applied on this sub-image. In this way, subsequent pose regressors see higher resolution images and thus learn features for finer scales which ultimately leads to higher precision. We use the same network architecture for all stages of the cascade but learn different network parameters. For

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

14

features for finer scales which ultimately leads to higher precision. We use the same network architecture for all stages of the cascade but learn different network parameters. For stages2 f1;:::;Sgof totalScascade stages, we denote bysthe learned network parameters. Thus, the pose displacement regressor reads (x;s). To refine a given joint location yiwe will consider a joint bounding boxbicapturing the sub-image around yi:bi(y;) = (yi;diam(y);diam(y))having as center the i-th joint and as dimension the pose diameter scaled by . The diameter diam (y)of the pose is defined as the distance between opposing joints on the human torso, such as left shoulder and right hip, and depends on the concrete pose definition and dataset. Using the above notation, at the stage s= 1we start with a bounding box b0which either encloses the full image or is obtained by a person detector. We obtain an initial pose: Stage 1 : y1 N1( (N(x;b0);1);b0)(5)

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

15

is obtained by a person detector. We obtain an initial pose: Stage 1 : y1 N1( (N(x;b0);1);b0)(5) At each subsequent stage s2, for all joints i2f1;:::;kg we regress first towards a refinement displacement ys i y(s1) i by applying a regressor on the sub image defined byb(s1) i from previous stage (s1). Then, we estimate new joint boxes bs i: Stages:ys i y(s1) i +N1( i(N(x;b);s);b)(6) forb=b(s1) i bs i (ys i;diam(ys);diam(ys)) (7) We apply the cascade for a fixed number of stages S, which is determined as explained in Sec. 4.1. Training The network parameters 1are trained as outlined in Sec. 3.1, Eq. (4). At subsequent stages

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

16

Training The network parameters 1are trained as outlined in Sec. 3.1, Eq. (4). At subsequent stages s2, the training is done identically with one important difference. Each joint ifrom a training example(x;y)is normalized using a different bounding box(y(s1) i;diam(y(s1));diam(y(s1)))– the one centered at the prediction for the same joint obtained from previous stage – so that we condition the training of the stage based on the model from previous stage. Since deep learning methods have large capacity, we augment the training data by using multiple normalizations for each image and joint. Instead of using the prediction from previous stage only, we generate simulated predictions. This is done by randomly displacing the ground truth location for joint iby a vector sampled at random from a 2-dimensional Normal distribution N(s1) i with mean and variance equal to the mean and variance of the observed displacements (y(s1) iyi)across all examples in the training data. The full augmented training data can be defined by first sampling an example and a joint from the original data at uniform and then generating a simulated prediction

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

17

by first sampling an example and a joint from the original data at uniform and then generating a simulated prediction based on a sampled displacement fromN(s1) i : Ds A=f(N(x;b);N(yi;b))j (x;yi)D;N(s1) i; b= (yi+;diam(y))g The training objective for cascade stage sis done as in Eq. (4) by taking extra care to use the correct normalization for each joint: s= arg min X (x;yi)2Ds Ajjyi i(x;)jj2 2 (8) 4. Empirical Evaluation 4.1. Setup Datasets There is a wide variety of benchmarks for human pose estimation. In this work we use datasets, which have large number of training examples sufficient to train a large model such as the proposed DNN, as well as are realistic and challenging. The first dataset we use is Frames Labeled In Cinema (FLIC), introduced by [20], which consists of 4000 training and 1000 test images obtained from popular Hollywood movies. The images contain people in diverse poses and especially diverse clothing. For each labeled human, 10upper

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

18

movies. The images contain people in diverse poses and especially diverse clothing. For each labeled human, 10upper body joints are labeled. The second dataset we use is Leeds Sports Dataset [12] and its extension [13], which we will jointly denote by LSP. Combined they contain 11000 training and 1000 testing images. These are images from sports activities and as such are quite challenging in terms of appearance and especially articulations. In addition, the majority of people have 150 pixel height which makes the pose estimation even more challenging. In this dataset, for each person the full body is labeled with total 14joints. For all of the above datasets, we define the diameter of a poseyto be the distance between a shoulder and hip from opposing sides and denote it by diam (y). It should be noted, that the joints in all datasets are arranged in a tree kinematically mimicking the human body. This allows for a definition of a limb being a pair of neighboring joints in the pose tree. Metrics In order to be able to compare with published results we will use two widely accepted evaluation metrics. Percentage of Correct Parts (PCP) measures detection rate of limbs, where a limb is considered detected if the distance between the two predicted joint locations and the true limb

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

19

Percentage of Correct Parts (PCP) measures detection rate of limbs, where a limb is considered detected if the distance between the two predicted joint locations and the true limb joint locations is at most half of the limb length [5]. PCP was the initially preferred metric for evaluation, however it has the drawback of penalizing shorter limbs, such as lower arms, which are usually harder to detect. To address this drawback, recently detection rates of joints are being reported using a different detection criterion – a joint is considered detected if the distance between the predicted and the true joint is within a certain fraction of the torso diameter. By varying this fraction, detection rates are obtained for varying degrees of localization precision. This metric alleviates the drawback of PCP since the detection criteria for all joints are based on the same distance threshold. We refer to this metric as Percent of Detected Joints (PDJ). Experimental Details For all the experiments we use the same network architecture. Inspired by [7], we use a body detector on FLIC to obtain initially a rough estimate of the human body bounding box. It is based on a face detector – the detected face rectangle is enlarged by a fixed scaler. This scaler is determined on the training data such that it contains all labeled joints. This face-based body detector results in

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

20

scaler is determined on the training data such that it contains all labeled joints. This face-based body detector results in a rough estimate, which however presents a good starting point for our approach. For LSP we use the full image as initial bounding box since the humans are relatively tightly cropped by design. Using a small held-out set of 50 images for both datasets to determine the algorithm hyperparameters. To measure optimality of the parameters we used average over PDJ at 0:2across all joints. The scaler , which defines the size of the refinement joint bounding box as a fraction of the pose size, is determined as follows: for FLIC we chose = 1:0 after exploring values f0:8;1:0;1:2g, for LSP we use = 2:0after tryingf1:5;1:7;2:0;2:3g. The number of cascade stagesSis determined by training stages until the algorithm stopped improving on the held-out set. For both FLIC and LSP we arrived at S= 3. To improve generalization, for each cascade stage starting ats= 2 we augment the training data by sampling 40 randomly translated crop boxes for each joint as explained

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

21

To improve generalization, for each cascade stage starting ats= 2 we augment the training data by sampling 40 randomly translated crop boxes for each joint as explained in Sec. 3.2. Thus, for LSP with 14joints and after mirroring the images and sampling the number training examples is1100040214 = 12M, which is essential for training a large network as ours.The presented algorithm allows for an efficient implementation. The running time is approx. 0:1sper image, as measured on a 12core CPU. This compares favorably to other approaches, as some of the current state-of-art approaches have higher complexity: [20] runs in approx. 4s, while [27] runs in 1:5s. The training complexity, however, is higher. The initial stage was trained within 3days on approx. 100 workers, most of the final performance was achieved after 12hours though. Each refinement stage was trained for 7days since the amount of data was 40larger than the one for the initial stage due to the data augmentation in Sec. 3.2. Note that using more data led to increased performance. 4.2. Results and Discussion

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

22

than the one for the initial stage due to the data augmentation in Sec. 3.2. Note that using more data led to increased performance. 4.2. Results and Discussion Comparisons We present comparative results to other approaches. We compare on LSP using PCP metric in Fig. 1. We show results for the four most challenging limbs – lower and upper arms and legs – as well as the average value across these limbs for all compared algorithms. We clearly outperform all other approaches, especially achieving better estimation for legs. For example, for upper legs we obtain0:78up from 0:74for the next best performing method. It is worth noting that while the other approaches exhibit strengths for particular limbs, none of the other dataset consistently dominates across all limbs. In contrary, DeepPose shows strong results for all challenging limbs. Using the PDJ metric allows us to vary the threshold for the distance between prediction and ground truth, which defines a detection. This threshold can be thought of as a localization precision at which detection rates are plotted. Thus one could compare approaches across different desired precisions. We present results on FLIC in Fig. 3 comparing against additional four methods as well is on LSP in Fig. 4. For each dataset we train and test according the protocol for each dataset. Similarly to previous experiment we

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

23

Fig. 4. For each dataset we train and test according the protocol for each dataset. Similarly to previous experiment we outperform all five algorithms. Our gains are bigger in the low precision domain, in the cases where we detect rough pose without precisely localizing the joints. On FLIC, at normalized distance 0:2we obtain a an increase of detection rates by 0:15and0:2for elbow and wrists against the next best performing method. On LSP, at normalized distance 0:5we get an absolute increase of 0:1. At low precision regime of normalized distance of 0:2for LSP we show comparable performance for legs and slightly worse arms. This can be attributed to the fact that the DNN-based approach computes joint coordinates using 7 layers of transformation, some of which contain max pooling. Another observation is that our approach works well for both appearance heavy movie data as well as string articulation such as the sports images in LSP. Effects of cascade-based refinement A single DNNbased joint regressor gives rough joint location. However, 0 0.05 0.1 0.15 0.20.10.30.50.70.9Elbows Normalized distance to true jointDetection rate

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

24

0 0.05 0.1 0.15 0.20.10.30.50.70.9Elbows Normalized distance to true jointDetection rate DeepPoseMODECEichner et al.Yang et al.Sapp et al. 0 0.05 0.1 0.15 0.20.10.30.50.70.9Wrists Normalized distance to true jointDetection rateFigure 3. Percentage of detected joints (PDJ) on FLIC for two joints: elbow and wrist. We compare DeepPose, after two cascade stages, with four other approaches. MethodArm LegAve.Upper Lower Upper Lower DeepPose-st1 0.5 0.27 0.74 0.65 0.54 DeepPose-st2 0.56 0.36 0.78 0.70 0.60 DeepPose-st3 0.56 0.38 0.77 0.71 0.61 Dantone et al. [2] 0.45 0.25 0.65 0.61 0.49 Tian et al. [25]0.52 0.33 0.70 0.60 0.56 Johnson et al. [13] 0.54 0.38 0.75 0.66 0.58

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

25

Johnson et al. [13] 0.54 0.38 0.75 0.66 0.58 Wang et al. [26]0.565 0.37 0.76 0.68 0.59 Pishchulin [18]0.49 0.32 0.74 0.70 0.56 Table 1. Percentage of Correct Parts (PCP) at 0.5 on LSP for DeepPose as well as five state-of-art approaches.The authors use a slightly looser version of PCP in which for each limb the average distance from predicted limb joints belonging to the corresponding true joints is used to determine whether the limb is correctly detected.The authors use person-centric joint annotations. 0 0.1 0.2 0.3 0.4 0.50.10.30.50.70.9Arms Normalized distance to true jointDetection rate DeepPose − wristsDeepPose − elbowsDantone et al. − wristsDantone et al. − elbows 0 0.1 0.2 0.3 0.4 0.50.10.30.50.70.9Legs Normalized distance to true jointDetection rate DeepPose − ankleDeepPose − kneeDantone et al. − ankleDantone et al. − knee

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

26

Normalized distance to true jointDetection rate DeepPose − ankleDeepPose − kneeDantone et al. − ankleDantone et al. − knee Figure 4. Percentage of detected joints (PDJ) on LSP for four limbs for DeepPose and Dantone et al. [2] over an extended range of distances to true joint: [0;0:5]of the torso diameter. Results of DeepPose are plotted with solid lines while all the results by [2] are plotted in dashed lines. Results for the same joint from both algorithms are colored with same color. to obtain higher precision the subsequent stages of the cascade, which serve as a refinement of the initial prediction, are of paramount importance. To see this, in Fig. 5 we present the joint detections at different precisions for the initial prediction as well as two subsequent cascade stages. As expected, we can see that the major gains of the refinement 00.05 0.10.15 0.20.25 0.30.35 0.40.10.30.50.70.9Wrists Normalized distance to true jointDetection rate DeepPose − initial stage 1DeepPose − stage 2DeepPose − stage 3

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

27

Normalized distance to true jointDetection rate DeepPose − initial stage 1DeepPose − stage 2DeepPose − stage 3 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.40.10.30.50.70.9Elbows Normalized distance to true jointDetection rateFigure 5. Percent of detected joints (PDJ) on FLIC or the first three stages of the DNN cascade. We present results over larger spectrum of normalized distances between prediction and ground truth. procedure are at high-precision regime of at normalized distances of [0:15;0:2]. Further, the major gains are achieved after one stage of refinement. The reason being that subsequent stages end up using smaller sub-images around each joint. And although the subsequent stages look at higher resolution inputs, they have more limited context. Examples of cases, where refinement helps, are visualized in Fig. 6. The initial stage is usually successful at estimating a roughly correct pose, however, this pose is not ”snapped” to the correct one. For example, in row three the pose has the right shape but incorrect scale. In the second row, the predicted pose is translated north from the ideal

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

28

pose has the right shape but incorrect scale. In the second row, the predicted pose is translated north from the ideal one. In most cases, the second stage of the cascade resolves this snapping problem and better aligns the joints. In more rare cases, such as in first row, further facade stages improve on individual joints. Cross-dataset Generalization To evaluate the generalization properties of our algorithm, we used the trained models on LSP and FLIC on two related datasets. The fullbody model trained on LSP is tested on the test portion of the Image Parse dataset [19] with results presented in Table 2. The ImageParse dataset is similar to LSP as it contains people doing sports, however it contains a lot of people from personal photo collections involved in other activities. Further, the upper-body model trained on FLIC was applied on the whole Buffy dataset [7]. We can see that our Initial stage 1stage 2stage 3Figure 6. Predicted poses in red and ground truth poses in green for the first three stages of a cascade for three examples. 0 0.05 0.1 0.15 0.20.10.30.50.70.9Elbows Normalized distance to true jointDetection rate Eichner et al.Yang et al.Sapp et al.MODECDeepPose

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

29

Normalized distance to true jointDetection rate Eichner et al.Yang et al.Sapp et al.MODECDeepPose 0 0.05 0.1 0.15 0.20.10.30.50.70.9Wrists Normalized distance to true jointDetection rate Figure 7. Percentage of detected joints (PDJ) on Buffy dataset for two joints: elbow and wrist. The models have been trained on FLIC. We compare DeepPose, after two cascade stages, with four other approaches. MethodArm LegAve.Upper Lower Upper Lower DeepPose 0.8 0.75 0.71 0.5 0.69 Pishchulin [18] 0.80 0.70 0.59 037 0.62 Johnson et al. [13] 0.75 0.67 0.67 0.46 0.64 Yang et al. [27] 0.69 0.64 0.55 0.35 0.56 Table 2. Percentage of Correct Parts (PCP) at 0.5 on Image Parse dataset for DeepPose as well as two state-of-art approaches on Image Parse dataset. Results obtained from [18]. approach can retain state-of-art performance compared to other approaches. This shows good generalization abilities.

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

30

approach can retain state-of-art performance compared to other approaches. This shows good generalization abilities. Example poses To get a better idea of the performance of our algorithm, we visualize a sample of estimated poses on images from LSP in Fig. 8. We can see that our algorithm is able to get correct pose for most of the joints under variety of conditions: upside-down people (row 1, column 1), severe foreshortening (row1, column 3), unusual poses (row 3, column 5), occluded limbs as the occluded arms in row 3, columns 2 and 6, unusual illumination conditions (row 3, column 3). In most of the cases, when the estimated pose isnot precise, it still has a correct shape. For example, in the last row some of the predicted limbs are not aligned with the true locations, however the overall shape of the pose is correct. A common failure mode is confusing left with right side when the person was photographed from the back (row 6, column 6). Results on FLIC (see Fig. 9) are usually better with occasional visible mistakes on lower arms. 5. Conclusion We present, to our knowledge, the first application of Deep Neural Networks (DNNs) to human pose estimation. Our formulation of the problem as DNN-based regression to

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

31

Deep Neural Networks (DNNs) to human pose estimation. Our formulation of the problem as DNN-based regression to joint coordinates and the presented cascade of such regressors has the advantage of capturing context and reasoning about pose in a holistic manner. As a result, we are able to achieve state-of-art or better results on several challenging academic datasets. Further, we show that using a generic convolutional neural network, which was originally designed for classification tasks, can be applied to the different task of localization. In future, we plan to investigate novel architectures which could be potentially better tailored towards localization problems in general, and in pose estimation in particular. Acknowledgements I would like to thank Luca Bertelli, Ben Sapp and Tianli Yu for assistance with data and fruitful discussions. References [1] M. Andriluka, S. Roth, and B. Schiele. Pictorial structures revisited: People detection and articulated pose estimation. InCVPR , 2009. [2] M. Dantone, J. Gall, C. Leistner, and L. Van Gool. Human pose estimation using body parts dependent joint regressors. InCVPR , 2013. [3] J. Duchi, E. Hazan, and Y . Singer. Adaptive subgradient

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

32

pose estimation using body parts dependent joint regressors. InCVPR , 2013. [3] J. Duchi, E. Hazan, and Y . Singer. Adaptive subgradient methods for online learning and stochastic optimization. In COLT . ACL, 2010. [4] M. Eichner and V . Ferrari. Better appearance models for pictorial structures. 2009. [5] M. Eichner, M. Marin-Jimenez, A. Zisserman, and V . Ferrari. Articulated human pose estimation and search in (almost) unconstrained still images. ETH Zurich, D-ITET, BIWI, Technical Report No , 272, 2010. [6] P. F. Felzenszwalb and D. P. Huttenlocher. Pictorial structures for object recognition. International Journal of Computer Vision , 61(1):55–79, 2005. [7] V . Ferrari, M. Marin-Jimenez, and A. Zisserman. Progressive search space reduction for human pose estimation. In CVPR , 2008. [8] M. A. Fischler and R. A. Elschlager. The representation and matching of pictorial structures. Computers, IEEE Transactions on , 100(1):67–92, 1973.

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

33

matching of pictorial structures. Computers, IEEE Transactions on , 100(1):67–92, 1973. Figure 8. Visualization of pose results on images from LSP. Each pose is represented as a stick figure, inferred from predicted joints. Different limbs in the same image are colored differently, same limb across different images has the same color. Figure 9. Visualization of pose results on images from FLIC. Meaning of stick figures is the same as in Fig. 8 above. [9] R. Girshick, J. Donahue, T. Darrell, and J. Malik. Rich feature hierarchies for accurate object detection and semantic segmentation. In CVPR , 2014. [10] G. Gkioxari, P. Arbel ´aez, L. Bourdev, and J. Malik. Articulated pose estimation using discriminative armlet classifiers. InCVPR , 2013. [11] C. Ionescu, F. Li, and C. Sminchisescu. Latent structured models for human pose estimation. In ICCV , 2011. [12] S. Johnson and M. Everingham. Clustered pose and nonlinear appearance models for human pose estimation. In BMVC , 2010. [13] S. Johnson and M. Everingham. Learning effective human

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

34

2010. [13] S. Johnson and M. Everingham. Learning effective human pose estimation from inaccurate annotation. In CVPR , 2011. [14] A. Krizhevsky, I. Sutskever, and G. Hinton. Imagenet classification with deep convolutional neural networks. In NIPS , 2012. [15] G. Mori and J. Malik. Estimating human body configurations using shape context matching. In ECCV , 2002.[16] R. Nevatia and T. O. Binford. Description and recognition of curved objects. Artificial Intelligence , 8(1):77–98, 1977. [17] M. Osadchy, Y . LeCun, and M. L. Miller. Synergistic face detection and pose estimation with energy-based models. The Journal of Machine Learning Research , 8:1197–1215, 2007. [18] L. Pishchulin, M. Andriluka, P. Gehler, and B. Schiele. Poselet conditioned pictorial structures. In CVPR , 2013. [19] D. Ramanan. Learning to parse images of articulated bodies. InNIPS , 2006.

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

35

[19] D. Ramanan. Learning to parse images of articulated bodies. InNIPS , 2006. [20] B. Sapp and B. Taskar. Modec: Multimodal decomposable models for human pose estimation. In CVPR , 2013. [21] G. Shakhnarovich, P. Viola, and T. Darrell. Fast pose estimation with parameter-sensitive hashing. In CVPR , 2003. [22] Y . Sun, X. Wang, and X. Tang. Deep convolutional network cascade for facial point detection. In Computer Vision and Pattern Recognition (CVPR), 2013 IEEE Conference on , pages 3476–3483. IEEE, 2013. [23] C. Szegedy, A. Toshev, and D. Erhan. Object detection via deep neural networks. In NIPS 26 , 2013. [24] G. W. Taylor, R. Fergus, G. Williams, I. Spiro, and C. Bregler. Pose-sensitive embedding by nonlinear nca regression. InNIPS , 2010. [25] Y . Tian, C. L. Zitnick, and S. G. Narasimhan. Exploring the spatial hierarchy of mixture models for human pose estimation. In ECCV , 2012.

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1312.4659

36

spatial hierarchy of mixture models for human pose estimation. In ECCV , 2012. [26] F. Wang and Y . Li. Beyond physical connections: Tree models in human pose estimation. In CVPR , 2013. [27] Y . Yang and D. Ramanan. Articulated pose estimation with flexible mixtures-of-parts. In CVPR , 2011.

1312.4659

DeepPose: Human Pose Estimation via Deep Neural Networks

We propose a method for human pose estimation based on Deep Neural Networks (DNNs). The pose estimation is formulated as a DNN-based regression problem towards body joints. We present a cascade of such DNN regressors which results in high precision pose estimates. The approach has the advantage of reasoning about pose in a holistic fashion and has a simple but yet powerful formulation which capitalizes on recent advances in Deep Learning. We present a detailed empirical analysis with state-of-art or better performance on four academic benchmarks of diverse real-world images.

http://arxiv.org/pdf/1312.4659

IEEE Conference on Computer Vision and Pattern Recognition, 2014

cs.CV

20131217

20140820

1405.3531

0

1 Return of the Devil in the Details: Delving Deep into Convolutional Nets Ken Chatfield, Karen Simonyan, Andrea Vedaldi, and Andrew Zisserman Visual Geometry Group, Department of Engineering Science, University of Oxford fken,karen,vedaldi,az g@robots.ox.ac.uk Abstract —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

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

1

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. F 1 I NTRODUCTION PERHAPS the single most important design choice in current state-of-the-art image classification and object recognition systems is the choice of visual features, or image representation. In fact, most of the quantitative improvements to image understanding obtained in the past dozen years can be ascribed to the introduction of improved representations, from the Bag-of-Visual-Words (BoVW) [1], [2] to the (Improved) Fisher Vector (IFV) [3]. A common characteristic of these methods is that they are largely handcrafted . They are also relatively simple, comprising dense sampling of local image patches, describing them by means of visual descriptors such as SIFT, encoding them into a highdimensional representation, and then pooling over the image. Recently, these handcrafted approaches have been substantially outperformed by the introduction of the latest generation of Convolutional Neural Networks (CNNs) [4] to the computer vision field. These networks have a substantially more sophisticated structure than standard representations, comprising several layers of non-linear feature extractors, and are therefore said to be deep

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

2

field. These networks have a substantially more sophisticated structure than standard representations, comprising several layers of non-linear feature extractors, and are therefore said to be deep (in contrast, classical representation will be referred to as shallow ). Furthermore, while their structureis handcrafted, they contain a very large number of parameters learnt from data. When applied to standard image classification and object detection benchmark datasets such as ImageNet ILSVRC [5] and PASCAL VOC [6] such networks have demonstrated excellent performance [7], [8], [9], [10], [11], significantly better than standard image encodings [12]. Despite these impressive results, it remains unclear how different deep architectures compare to each other and to shallow computer vision methods such as IFV . Most papers did not test these representations extensively on a common ground, so a systematic evaluation of the effect of different design and implementation choices remains largely missing. As noted in our previous work [12], which compared the performance of various shallow visual encodings, the performance of computer vision systems depends significantly on implementation details . For example, state-of-the-art methods such as [13] not only involve the use of a CNN, but also include other improvements such as the use

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

3

details . For example, state-of-the-art methods such as [13] not only involve the use of a CNN, but also include other improvements such as the use of very large scale datasets, GPU computation, and data augmentation (also known as data jittering or virtual sampling). These improvements could also transfer to shallow representations such as the IFV , potentially explaining a part of the performancearXiv:1405.3531v4 [cs.CV] 5 Nov 2014 2 gap [14]. In this study we analyse and empirically clarify these issues, conducting a large set of rigorous experiments (Sect. 4), in many ways picking up the story where it last ended in [12] with the comparison of shallow encoders. We focus on methods to construct image representations ,i.e. encoding functions mapping an image Ito a vector (I)2Rdsuitable for analysis with a linear classifier, such as an SVM. We consider three scenarios (Sect. 2, Sect. 3): shallow image representations, deep representations pre-trained on outside data, and deep representation pre-trained and then finetuned on the target dataset. As part of our tests, we explore generally-applicable best practices that

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

4

and deep representation pre-trained and then finetuned on the target dataset. As part of our tests, we explore generally-applicable best practices that are nevertheless more often found in combination with CNNs [13] or, alternatively, with shallow encoders [12], porting them with mutual benefit. These are (Sect. 2): the use of colour information , feature normalisation , and, most importantly, the use of substantial data augmentation . We also determine scenario-specific best-practices , improving the ones in [12], [15] and others, including dimensionality reduction for deep features. Finally, we achieve performance competitive with the state of the art [16], [17] on PASCAL VOC classification using less additional training data and significantly simpler techniques. As in [12], the source code and models to reproduce all experiments in this paper is available on the project website1. 2 S CENARIOS This section introduces the three types of image representation (I)considered in this paper, describing them within the context of three different scenarios. Having outlined details specific to each, general methodologies which apply to all three scenarios are reviewed, such as data augmentation

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

5

scenarios. Having outlined details specific to each, general methodologies which apply to all three scenarios are reviewed, such as data augmentation and feature normalisation, together with the linear classifier (trained with a standard hinge loss). We also specify here the benchmark datasets used in the evaluation. 2.1 Scenario 1: Shallow representation (IFV) Our reference shallow image representation is the IFV [3]. Our choice is motivated by the fact that IFV usually outperforms related encoding methods such as BoVW, LLC [12], and VLAD [18]. Given an imageI, the IFVFV(I)is obtained by extracting 1. http://www.robots.ox.ac.uk/vgg/research/deep eval/a dense collection of patches and corresponding local descriptors xi2RD(e.g. SIFT) from the image at multiple scales. Each descriptor xiis then softquantized using a Gaussian Mixture Model with Kcomponents. First and second order differences between each descriptor xiand its Gaussian cluster meankare accumulated in corresponding blocks uk,vkin the vector FV(I)2R2KD, appropriately weighed by the Gaussian soft-assignments and

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

6

uk,vkin the vector FV(I)2R2KD, appropriately weighed by the Gaussian soft-assignments and covariance, leading to a 2KD-dimensional image representation FV(I) = [u> 1;v> 1;:::u> K;v> K]>:The improved version of the Fisher vector involves postprocessing FVby computing the signed squareroot of its scalar components and normalising the result to a unit `2norm. The details of this construction can be found in [3]; here we follow the notation of [12]. 2.2 Scenario 2: Deep representation (CNN) with pre-training Our deep representations are inspired by the success of the CNN of Krizhevsky et al. [13]. As shown in [7], [19], the vector of activities CNN(I)of the penultimate layer of a deep CNN, learnt on a large dataset such as ImageNet [5], can be used as a powerful image descriptor applicable to other datasets. Numerous CNN architectures that improve the previous state of the art obtained using shallow representations have been proposed, but choosing

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

7

Numerous CNN architectures that improve the previous state of the art obtained using shallow representations have been proposed, but choosing the best one remains an open question. Many are inspired by [13]: DeCAF [7], [11], Caffe [20], Oquab et al. [8]. Others use larger networks with a smaller stride of the first convolutional layer: Zeiler and Fergus [19] and OverFeat [10], [9]. Other differences include the CNN pre-training protocols. Here we adopt a single learning framework and experiment with architectures of different complexity exploring their performance-speed trade-off. 2.3 Scenario 3: Deep representation (CNN) with pre-training and fine-tuning In Scenario 2 features are trained on one (large) dataset and applied to another (usually smaller). However, it was demonstrated [11] that fine-tuning a pre-trained CNN on the target data can significantly improve the performance. We consider this scenario separately from that of Scenario 2, as the image features become dataset-specific after the fine-tuning. 3 2.4 Commonalities We now turn to what is in common across the scenarios.

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

8

fine-tuning. 3 2.4 Commonalities We now turn to what is in common across the scenarios. 2.4.1 Data augmentation Data augmentation is a method applicable to shallow and deep representations, but that has been so far mostly applied to the latter [13], [19]. By augmentation we mean perturbing an image Iby transformations that leave the underlying class unchanged ( e.g. cropping and flipping) in order to generate additional examples of the class. Augmentation can be applied at training time, at test time, or both. The augmented samples can either be taken as-is or combined to form a single feature, e.g. using sum/max-pooling or stacking. 2.4.2 Linear predictors All the representations (I)in the three scenarios are used to construct linear predictorshw;(I)ifor each class to be recognized. These predictors are learnt using Support Vector Machines (SVM) by fitting wto the available training data by minimizing an objective function balancing a quadratic regularizer and the hinge-loss. The parameter C in the SVM, trading-off regularizer and loss, is determined using an held-off validation subset of

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

9

regularizer and the hinge-loss. The parameter C in the SVM, trading-off regularizer and loss, is determined using an held-off validation subset of the data. Here we use the same learning framework with all representations. It is common experience that linear classifiers are particularly sensitive to thenormalisation of the data and that, in particular, SVMs tend to benefit from `2normalisation [3] (an interpretation is that after normalisation the inner product corresponds to the cosine similarly). 2.5 Benchmark data As reference benchmark we use the PASCAL VOC [6] data as already done in [12]. The VOC2007 edition contains about 10,000 images split into train, validation, and test sets, and labelled with twenty object classes. A one-vs-rest SVM classifier for each class is learnt and evaluated independently and the performance is measured as mean Average Precision (mAP) across all classes. The VOC-2012 edition contains roughly twice as many images and does not include test labels; instead, evaluation uses the official PASCAL Evaluation Server. To train deep representations we use the ILSVRC-2012 challenge dataset. This contains 1,000 object categories

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

10

uses the official PASCAL Evaluation Server. To train deep representations we use the ILSVRC-2012 challenge dataset. This contains 1,000 object categories from ImageNet [5] with roughly 1.2M training images, 50,000 validation images, and 100,000 testimages. Performance is evaluated using the top-5 classification error. Finally, we also evaluate over theCaltech-101 and Caltech-256 image classification benchmarks [21], [22]. For Caltech-101, we followed the protocol of [12], and considered three random splits into training and testing data, each of which comprises 30 training and up to 30 testing images per class. For Caltech-256, two random splits were generated, each of which contains 60 training images per class, and the rest are used for testing. On both Caltech datasets, performance is measured using mean class accuracy. 3 D ETAILS This section gives the implementation details of the methods introduced in Sect. 2. 3.1 Improved Fisher Vector details Our IFV representation uses a slightly improved setting compared to the best result of [12]. Computation starts by upscaling the image Iby a factor of 2 [23], followed by SIFT features extraction

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

11

setting compared to the best result of [12]. Computation starts by upscaling the image Iby a factor of 2 [23], followed by SIFT features extraction with a stride of 3 pixels at 7 different scales withp 2 scale increments. These features are square-rooted as suggested by [24], and decorrelated and reduced in dimension from 128Dto80Dusing PCA. A GMM with K= 256 components is learnt from features sampled from the training images. Hence the Fisher Vector FV(I)has dimension 2KD= 40;960. Before use in classification, the vector is signedsquare-rooted and l2-normalised (square rooting correspond to the Hellinger’s kernel map [25]). As in [12], square-rooting is applied twice, once to the raw encodings, and once again after sum pooling and normalisation. In order to capture weak geometrical information, the IFV representation is used in a spatial pyramid [26]. As in [12], the image is divided into 11,31, and 22spatial subdivisions and corresponding IFVs are computed and stacked with an overall dimension of 82KD= 327;680 elements.

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

12

and corresponding IFVs are computed and stacked with an overall dimension of 82KD= 327;680 elements. In addition to this standard formulation, we experiment with a few modifications. The first one is the use of intra-normalisation of the descriptor blocks, an idea recently proposed for the VLAD descriptor [27]. In this case, the `2normalisation is applied to the individual sub-blocks (uk;vk)of the vector FV(I), which helps to alleviate the local feature burstiness [28]. In the case of the improved intra-normalised features, it was found 4 that applying the square-rooting only once to the final encoding produced the best results. The second modification is the use of spatiallyextended local descriptors [23] instead of a spatial pyramid. Here descriptors xiare appended with their image location (xi;yi)before quantization with the GMM. Formally, xiis extended, after PCA projection, with its normalised spatial coordinates: [x> i;xi=W0:5;yi=H0:5]>, whereWHare the dimensions of the image. Since the GMM quantizes

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

13

i;xi=W0:5;yi=H0:5]>, whereWHare the dimensions of the image. Since the GMM quantizes both appearance and location, this allows for spatial information to be captured directly by the softquantization process. This method is significantly more memory-efficient than using a spatial pyramid. Specifically, the PCA-reduced SIFT features are spatially augmented by appending (x;y)yieldingD= 82 dimensional descriptors pooled in a 2KD= 41;984dimensional IFV . The third modification is the use of colour features in addition to SIFT descriptors. While colour information is used in CNNs [13] and by the original FV paper [3], it was not explored in our previous comparison [12]. We do so here by adopting the same Local Colour Statistics (LCS) features as used by [3]. LCS is computed by dividing an input patch into a 44spatial grid (akin to SIFT), and computing the mean and variance of each of the Lab colour channels for each cell of the grid. The LCS dimensionality is thus 4423 = 96 . This is then encoded in a similar manner to SIFT.

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

14

Lab colour channels for each cell of the grid. The LCS dimensionality is thus 4423 = 96 . This is then encoded in a similar manner to SIFT. 3.2 Convolutional neural networks details The CNN-based features are based on three CNN architectures representative of the state of the art (shown in Table 1) each exploring a different accuracy/speed trade-off. To ensure a fair comparison between them, these networks are trained using the same training protocol and the same implementation, which we developed based on the opensource Caffe framework [20]. `2-normalising the CNN features CNN(I)before use in the SVM was found to be important for performance. Our Fast (CNN-F) architecture is similar to the one used by Krizhevsky et al . [13]. It comprises 8 learnable layers, 5 of which are convolutional, and the last 3 are fully-connected. The input image size is 224224. Fast processing is ensured by the 4 pixel stride in the first convolutional layer. The main differences between our architecture and that of [13] are the reduced number of convolutional layers and the dense connectivity between convolu-tional layers ([13] used sparse connections to enable training on two GPUs).

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

15

of [13] are the reduced number of convolutional layers and the dense connectivity between convolu-tional layers ([13] used sparse connections to enable training on two GPUs). Our Medium (CNN-M) architecture is similar to the one used by Zeiler and Fergus [19]. It is characterised by the decreased stride and smaller receptive field of the first convolutional layer, which was shown to be beneficial on the ILSVRC dataset. At the same time, conv2 uses larger stride (2 instead of 1) to keep the computation time reasonable. The main difference between our net and that of [19] is we use less filters in the conv4 layer (512 vs. 1024). Our Slow (CNN-S) architecture is related to the ‘accurate’ network from the OverFeat package [10]. It also uses 77filters with stride 2in conv1. Unlike CNN-M and [19], the stride in conv2 is smaller (1 pixel), but the max-pooling window in conv1 and conv5 is larger ( 33) to compensate for the increased spatial resolution. Compared to [10], we use 5 convolutional layers as in the previous

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

16

conv1 and conv5 is larger ( 33) to compensate for the increased spatial resolution. Compared to [10], we use 5 convolutional layers as in the previous architectures ([10] used 6), and less filters in conv5 (512 instead of 1024); we also incorporate an LRN layer after conv1 ([10] did not use contrast normalisation). 3.2.1 CNN training In general, our CNN training procedure follows that of [13], learning on ILSVRC-2012 using gradient descent with momentum. The hyperparameters are the same as used by [13]: momentum0:9; weight decay 5
104; initial learning rate 102, which is decreased by a factor of 10, when the validation error stop decreasing. The layers are initialised from a Gaussian distribution with a zero mean and variance equal to 102. We also employ similar data augmentation in the form of random crops, horizontal flips, and RGB colour jittering. Test time crop sampling is discussed in Sect. 3.3; at training time, 224224 crops are sampled randomly, rather than deterministically. Thus, the only notable difference to [13] is that the crops are taken from the whole training image P256;P256,

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

17

notable difference to [13] is that the crops are taken from the whole training image P256;P256, rather than its 256256 centre. Training was performed on a single NVIDIA GTX Titan GPU and the training time varied from 5 days for CNN-F to 3 weeks for CNN-S. 3.2.2 CNN fine-tuning on the target dataset In our experiments, we fine-tuned CNN-S using VOC-2007, VOC-2012, or Caltech-101 as the target data. Fine-tuning was carried out using the same framework (and the same data augmentation), as we used for CNN training on ILSVRC. The last 5 Arch. conv1 conv2 conv3 conv4 conv5 full6 full7 full8 CNN-F64x11x11 256x5x5 256x3x3 256x3x3 256x3x3 4096 4096 1000 st. 4, pad 0 st. 1, pad 2 st. 1, pad 1 st. 1, pad 1 st. 1, pad 1 drop- drop- softLRN, x2 pool LRN, x2 pool - - x2 pool out out max

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

18

CNN-M96x7x7 256x5x5 512x3x3 512x3x3 512x3x3 4096 4096 1000 st. 2, pad 0 st. 2, pad 1 st. 1, pad 1 st. 1, pad 1 st. 1, pad 1 drop- drop- softLRN, x2 pool LRN, x2 pool - - x2 pool out out max CNN-S96x7x7 256x5x5 512x3x3 512x3x3 512x3x3 4096 4096 1000 st. 2, pad 0 st. 1, pad 1 st. 1, pad 1 st. 1, pad 1 st. 1, pad 1 drop- drop- softLRN, x3 pool x2 pool - - x3 pool out out max TABLE 1 CNN architectures. Each architecture contains 5 convolutional layers (conv 1–5) and three fully-connected layers (full 1–3). The details of each of the convolutional layers are given in three sub-rows: the first specifies the number of convolution filters and their receptive field size as “num x size x size”; the second indicates the convolution stride (“st.”) and spatial padding (“pad”); the third indicates if Local

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

19

second indicates the convolution stride (“st.”) and spatial padding (“pad”); the third indicates if Local Response Normalisation (LRN) [13] is applied, and the max-pooling downsampling factor. For full 1–3, we specify their dimensionality, which is the same for all three architectures. Full6 and full7 are regularised using dropout [13], while the last layer acts as a multi-way soft-max classifier. The activation function for all weight layers (except for full8) is the REctification Linear Unit (RELU) [13]. fully-connected layer (conv8) has output dimensionality equal to the number of classes, which differs between datasets, so we initialised it from a Gaussian distribution (as used for CNN training above). Now we turn to dataset-specific fine-tuning details. VOC-2007 and VOC-2012. Considering that PASCAL VOC is a multi-label dataset ( i.e. a single image might have multiple labels), we replaced the softmax regression loss with a more appropriate loss function, for which we considered two options: one-vs-rest classification hinge loss (the same loss

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

20

softmax regression loss with a more appropriate loss function, for which we considered two options: one-vs-rest classification hinge loss (the same loss as used in the SVM experiments) and ranking hinge loss. Both losses define constraints on the scores of positive (Ipos) and negative ( Ineg) images for each class:wc(Ipos)>1;wc(Ineg)<1 +for the classification loss, wc(Ipos)> w c(Ineg) + 1 for the ranking loss ( wcis thec-th row of the last fully-connected layer, which can be seen as a linear classifier on deep features (I);is a slack variable). Our fine-tuned networks are denoted as “CNN S TUNE-CLS” (for the classification loss) and “CNN S TUNE-RNK” (for the ranking loss). In the case of both VOC datasets, the training and validation subsets were combined to form a single training set. Given the smaller size of the training data when compared to ILSVRC-2012, we

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

21

and validation subsets were combined to form a single training set. Given the smaller size of the training data when compared to ILSVRC-2012, we controlled for over-fitting by using lower initial learning rates for the fine-tuned hidden layers.The learning rate schedule for the last layer / hidden layers was: 102=104!103=104! 104=104!105=105. Caltech-101 dataset contains a single class label per image, so fine-tuning was performed using the softmax regression loss. Other settings (including the learning rate schedule) were the same as used for the VOC fine-tuning experiments. 3.2.3 Low-dimensional CNN feature training Our baseline networks (Table 1) have the same dimensionality of the last hidden layer (full7): 4096. This design choice is in accordance with the stateof-the-art architectures [13], [19], [10], and leads to a 4096-D dimensional image representation, which is already rather compact compared to IFV . We further trained three modifications of the CNN-M network, with lower dimensional full7 layers of: 2048, 1024, and 128 dimensions respectively. The

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

22

further trained three modifications of the CNN-M network, with lower dimensional full7 layers of: 2048, 1024, and 128 dimensions respectively. The networks were learnt on ILSVRC-2012. To speedup training, all layers aside from full7/full8 were set to those of the CNN-M net and a lower initial learning rate of 103was used. The initial learning rate of full7/full8 was set to 102. 3.3 Data augmentation details We explore three data augmentation strategies. The first strategy is to use no augmentation . In contrast to IFV , however, CNNs require images to 6 be transformed to a fixed size ( 224224) even when no augmentation is used. Hence the image is downsized so that the smallest dimension is equal to224 pixels and a 224224 crop is extracted from the centre.2The second strategy is to use flip augmentation , mirroring images about the yaxis producing two samples from each image. The third strategy, termed C+F augmentation , combines cropping and flipping. For CNN-based representations, the image is downsized so that the smallest dimension is equal to 256 pixels. Then 224224

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

23

cropping and flipping. For CNN-based representations, the image is downsized so that the smallest dimension is equal to 256 pixels. Then 224224 crops are extracted from the four corners and the centre of the image. Note that the crops are sampled from the whole image, rather than its 256256 centre, as done by [13]. These crops are then flipped about they-axis, producing 10perturbed samples per input image. In the case of the IFV encoding, the same crops are extracted, but at the original image resolution. 4 A NALYSIS This section describes the experimental results, comparing different features and data augmentation schemes. The results are given in Table 2 for VOC-2007 and analysed next, starting from generally applicable methods such as augmentation and then discussing the specifics of each scenario. We then move onto other datasets and the state of the art in Sect. 4.7. 4.1 Data augmentation We experiment with no data augmentation (denoted Image Aug=– in Tab. 2), flip augmentation ( Image Aug=F ), and C+F augmentation ( Image Aug=C ). Augmented images are used as standalone samples ( f), or by fusing the corresponding

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

24

Aug=C ). Augmented images are used as standalone samples ( f), or by fusing the corresponding descriptors using sum ( s) or max ( m) pooling or stacking ( t). So for example Image Aug=(C) f s in row [f] of Tab. 2 means that C+F augmentation is used to generate additional samples in training ( f), and is combined with sum-pooling in testing ( s). Augmentation consistently improves performance by3%for both IFV ( e.g. [d] vs. [f]) and CNN ( e.g. [o] vs. [p]). Using additional samples for training and sum-pooling for testing works best ([p]) followed by sum-pooling [r], max pooling [q], and stacking [s]. In terms of the choice of transformations, flipping improves only marginally ([o] vs. 2. Extracting a 224224 centre crop from a 256256 image [13] resulted in worse performance.[u]), but using the more expensive C+F sampling improves, as seen, by about 23%([o]vs. [p]). We experimented with sampling more transformations, taking a higher density of crops from the centre of the image, but observed no benefit. 4.2 Colour

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

25

experimented with sampling more transformations, taking a higher density of crops from the centre of the image, but observed no benefit. 4.2 Colour Colour information can be added and subtracted in CNN and IFV . In IFV replacing SIFT with the colour descriptors of [3] (denoted COL inMethod ) yields significantly worse performance ([j] vs. [h]). However, when SIFT and colour descriptors are combined by stacking the corresponding IFVs ( COL+ ) there is a small but significant improvement of around1%in the non-augmented case ( e.g. [h] vs. [k]) but little impact in the augmented case (e.g. [i] vs. [l]). For CNNs, retraining the network after converting all the input images to grayscale (denoted GSinMethods ) has a more significant impact, resulting in a performance drop of 3% ([w] vs. [p], [v] vs. [o]). 4.3 Scenario 1: Shallow representation (IFV) The baseline IFV encoding using a spatial pyramid [a] performs slightly better than the results [I] taken

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

26

4.3 Scenario 1: Shallow representation (IFV) The baseline IFV encoding using a spatial pyramid [a] performs slightly better than the results [I] taken from Chatfield et al. [12], primarily due to a larger number of spatial scales being used during SIFT feature extraction, and the resultant SIFT features being square-rooted. Intra-normalisation , denoted as INin the Method column of the table, improves the performance by 1%(e.g. [c] vs. [d]). More interestingly, switching from spatial pooling (denoted spm in the SPool column) to feature spatial augmentation ( SPool=(x,y) ) has either little effect on the performance or results in a marginal increase ([a] vs. [c], [b] vs. [d]), whilst resulting in a representation which is over 10 smaller. We also experimented with augmenting with scale in addition to position as in [23] but observed no improvement. Finally, we investigate pushing the parameters of the representation setting K= 512 (rows [h]-[l]). Increasing the number of GMM centres in the model from K= 256 to512 results in a further performance increase (e.g. [h] vs. [d]), but at the expense of

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

27

K= 256 to512 results in a further performance increase (e.g. [h] vs. [d]), but at the expense of higher-dimensional codes (125K dimensional). 4.4 Scenario 2: Deep representation (CNN) with pre-training CNN-based methods consistently outperform the shallow encodings, even after the improvements 7 Method SPool Image Aug. Dim mAP (I) FK BL spm – 327K 61.69 79.0 67.4 51.9 70.9 30.8 72.2 (II) DECAF – (C) t t 327K 73.41 87.4 79.3 84.1 78.4 42.3 73.7 (a) FK spm – 327K 63.66 83.4 68.8 59.6 74.1 35.7 71.2 (b) FK IN spm – 327K 64.18 82.1 69.7 59.7 75.2 35.7 71.3 (c) FK (x,y) – 42K 63.51 83.2 69.4 60.6 73.9 36.3 68.6 (d) FK IN (x,y) – 42K 64.36 83.1 70.4 62.4 75.2 37.1 69.1

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

28

(d) FK IN (x,y) – 42K 64.36 83.1 70.4 62.4 75.2 37.1 69.1 (e) FK IN (x,y) (F) f - 42K 64.35 83.1 70.5 62.3 75.4 37.1 69.1 (f) FK IN (x,y) (C) f s 42K 67.17 85.5 71.6 64.6 77.2 39.0 70.8 (g) FK IN (x,y) (C) s s 42K 66.68 84.9 70.1 64.7 76.3 39.2 69.8 (h) FK IN 512 (x,y) – 84K 65.36 84.1 70.4 65.0 76.7 37.2 71.3 (i) FK IN 512 (x,y) (C) f s 84K 68.02 85.9 71.8 67.1 77.1 38.8 72.3 (j) FK IN COL 512 – – 82K 52.18 69.5 52.1 47.5 64.0 24.6 49.8

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

29

(j) FK IN COL 512 – – 82K 52.18 69.5 52.1 47.5 64.0 24.6 49.8 (k) FK IN 512 COL+ (x,y) – 166K 66.37 82.9 70.1 67.0 77.0 36.1 70.0 (l) FK IN 512 COL+ (x,y) (C) f s 166K 67.93 85.1 70.5 67.5 77.4 35.7 71.2 (m) CNN F – (C) f s 4K 77.38 88.7 83.9 87.0 84.7 46.9 77.5 (n) CNN S – (C) f s 4K 79.74 90.7 85.7 88.9 86.6 50.5 80.1 (o) CNN M – – 4K 76.97 89.5 84.3 88.8 83.2 48.4 77.0 (p) CNN M – (C) f s 4K 79.89 91.7 85.4 89.5 86.6 51.6 79.3 (q) CNN M – (C) f m 4K 79.50 90.9 84.6 89.4 85.8 50.3 78.4

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

30

(q) CNN M – (C) f m 4K 79.50 90.9 84.6 89.4 85.8 50.3 78.4 (r) CNN M – (C) s s 4K 79.44 91.4 85.2 89.1 86.1 52.1 78.0 (s) CNN M – (C) t t 41K 78.77 90.7 85.0 89.2 85.8 51.0 77.8 (t) CNN M – (C) f - 4K 77.78 90.5 84.3 88.8 84.5 47.9 78.0 (u) CNN M – (F) f - 4K 76.99 90.1 84.2 89.0 83.5 48.1 77.2 (v) CNN M GS – – 4K 73.59 87.4 80.8 82.4 82.1 44.5 73.5 (w) CNN M GS – (C) f s 4K 77.00 89.4 83.8 85.1 84.4 49.4 77.6 (x) CNN M 2048 – (C) f s 2K 80.10 91.3 85.8 89.9 86.7 52.4 79.7

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

31

(x) CNN M 2048 – (C) f s 2K 80.10 91.3 85.8 89.9 86.7 52.4 79.7 (y) CNN M 1024 – (C) f s 1K 79.91 91.4 86.9 89.3 85.8 53.3 79.8 (z) CNN M 128 – (C) f s 128 78.60 91.3 83.9 89.2 86.9 52.1 81.0 (a) FK+CNN F (x,y) (C) f s 88K 77.95 89.6 83.1 87.1 84.5 48.0 79.4 (b) FK+CNN M 2048 (x,y) (C) f s 86K 80.14 90.9 85.9 88.8 85.5 52.3 81.4 (g) CNN S TUNE-RNK – (C) f s 4K 82.42 95.3 90.4 92.5 89.6 54.4 81.9 TABLE 2 VOC 2007 results (continued overleaf) . See Sect. 4 for details. discussed above, by a large 10% mAP margin ([i]vs. [p]). Our small architecture CNN-F, which is

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

32

discussed above, by a large 10% mAP margin ([i]vs. [p]). Our small architecture CNN-F, which is similar to DeCAF [7], performs significantly better than the latter ([II] vs. [s]), validating our implementation. Both medium CNN-M [m] and slow CNN-S [p] outperform the fast CNN-F [m] by a significant 23% margin. Since the accuracy of CNN-S and CNN-M is nearly the same, we focus on the latter as it is simpler and marginally (25%) faster. Remarkably, these good networks work very well even with no augmentation [o]. Another advantage of CNNs compared to IFV is the small dimensionality of the output features,although IFV can be compressed to an extent. We explored retraining the CNNs such that the final layer was of a lower dimensionality, and reducing from 4096 to 2048 actually resulted in a marginal performance boost ([x] vs. [p]). What is surprising is that we can reduce the output dimensionality further to 1024D [y] and even 128D [z] with only a drop of2%for codes that are 32smaller

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

33

further to 1024D [y] and even 128D [z] with only a drop of2%for codes that are 32smaller (650smaller than our best performing IFV [i]). Note,`2-normalising the features accounted for up to5%of their performance over VOC 2007; it should be applied before input to the SVM and after pooling the augmented descriptors (where 8 (I) 79.9 61.4 56.0 49.6 58.4 44.8 78.8 70.8 85.0 31.7 51.0 56.4 80.2 57.5 (II) 83.7 83.7 54.3 61.9 70.2 79.5 85.3 77.2 90.9 51.1 73.8 57.0 86.4 68.0 (a) 80.7 64.4 53.8 53.8 60.2 47.8 79.9 68.9 86.1 37.3 51.1 55.8 83.7 56.9 (b) 80.6 64.8 53.9 54.9 60.7 50.5 80.4 69.5 86.2 38.3 54.4 56.3 82.7 56.7

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

34

(c) 81.1 64.2 51.1 53.4 61.9 50.0 80.0 67.5 85.3 35.7 51.9 53.8 83.5 58.9 (d) 80.5 66.9 50.9 53.9 62.1 51.5 80.5 68.5 85.9 37.2 55.2 54.3 83.3 59.2 (e) 80.5 66.8 51.0 54.1 62.2 51.5 80.4 68.2 86.0 37.3 55.1 54.2 83.3 59.2 (f) 82.4 71.6 52.8 62.4 63.4 57.1 81.6 70.9 86.9 41.2 61.2 56.9 85.2 61.5 (g) 81.9 71.0 52.8 61.6 62.2 56.8 81.8 70.0 86.5 41.5 61.0 56.5 84.3 60.9 (h) 81.1 67.9 52.6 55.4 61.4 51.2 80.5 69.1 86.4 41.2 56.0 56.2 83.7 59.9

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

35

(i) 82.5 73.2 54.7 62.7 64.5 56.6 82.2 71.3 87.5 43.0 62.0 59.3 85.7 62.4 (j) 66.1 46.6 42.5 35.8 41.1 45.5 75.4 58.3 83.9 39.8 47.3 35.6 69.2 49.0 (k) 80.0 65.9 52.8 56.1 61.0 56.9 81.4 69.6 88.4 49.0 59.2 56.4 84.7 62.8 (l) 81.6 70.8 52.9 59.6 63.1 59.9 82.1 70.5 88.9 50.6 63.7 57.5 86.1 64.1 (m) 86.3 85.4 58.6 71.0 72.6 82.0 87.9 80.7 91.8 58.5 77.4 66.3 89.1 71.3 (n) 87.8 88.3 61.3 74.8 74.7 87.2 89.0 83.7 92.3 58.8 80.5 69.4 90.5 74.0

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

36

(o) 85.1 87.4 58.1 70.4 73.1 83.5 85.5 80.9 90.8 54.1 78.9 61.1 89.0 70.4 (p) 87.7 88.6 60.3 80.1 74.4 85.9 88.2 84.6 92.1 60.3 80.5 66.2 91.3 73.5 (q) 87.6 88.6 60.7 78.2 73.6 86.0 87.4 83.8 92.3 59.3 81.0 66.8 91.3 74.0 (r) 87.5 88.1 60.4 76.9 74.8 85.8 88.1 84.3 92.2 59.5 79.3 65.8 90.8 73.5 (s) 87.3 87.6 60.1 72.3 75.3 85.2 86.9 82.6 91.9 58.5 77.9 66.5 90.5 73.4 (t) 85.7 87.9 58.3 74.2 73.9 84.7 86.6 82.0 91.0 55.8 79.2 62.1 89.3 71.0

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

37

(u) 85.3 87.3 58.1 70.0 73.4 83.5 86.0 80.8 90.9 53.9 78.1 61.2 88.8 70.6 (v) 85.0 84.9 57.8 65.9 69.8 79.5 82.9 77.4 89.2 42.8 71.7 60.2 86.3 67.8 (w) 87.2 86.5 59.5 72.4 74.1 81.7 86.0 82.3 90.8 48.9 73.7 66.8 89.6 71.0 (x) 87.6 88.4 60.2 76.9 75.4 85.5 88.0 83.4 92.1 61.1 83.1 68.5 91.9 74.2 (y) 87.8 88.6 59.0 77.2 73.1 85.9 88.3 83.5 91.8 59.9 81.4 68.3 93.0 74.1 (z) 86.6 87.5 59.1 70.0 72.9 84.6 86.7 83.6 89.4 57.0 81.5 64.8 90.4 73.4

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

38

(a) 86.8 85.6 59.9 72.0 73.4 81.4 88.6 80.5 92.1 60.6 77.3 66.4 89.3 73.3 (b) 87.7 88.4 61.2 76.9 76.6 84.9 89.1 82.9 92.4 61.9 80.9 68.7 91.5 75.1 (g) 91.5 91.9 64.1 76.3 74.9 89.7 92.2 86.9 95.2 60.7 82.9 68.0 95.5 74.4 TABLE 2 VOC 2007 results (continued from previous page) applicable). 4.5 Scenario 3: Deep representation (CNN) with pre-training and fine-tuning We fine-tuned our CNN-S architecture on VOC2007 using the ranking hinge loss, and achieved a significant improvement: 2:7%([g]vs. [n]). This demonstrates that in spite of the small amount of VOC training data (5,011 images), fine-tuning is able to adjust the learnt deep representation to better suit the dataset in question.4.6 Combinations For the CNN-M 2048 representation [x], stacking

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

39

is able to adjust the learnt deep representation to better suit the dataset in question.4.6 Combinations For the CNN-M 2048 representation [x], stacking deep and shallow representations to form a higherdimensional descriptor makes little difference ([x] vs. [b]). For the weaker CNN-F it results in a small boost of0:8%([m] vs. [a]). 4.7 Comparison with the state of the art In Table 3 we report our results on ILSVRC-2012, VOC-2007, VOC-2012, Caltech-101, and Caltech-256 datasets, and compare them to the state of the art. First, we note that the ILSVRC error rates of our CNN-F, CNN-M, and CNN-S networks are better 9 ILSVRC-2012 VOC-2007 VOC-2012 Caltech-101 Caltech-256 (top-5 error) (mAP) (mAP) (accuracy) (accuracy) (a) FK IN 512 - 68.0 – – – (b) CNN F 16.7 77.4 79.9 – – (c) CNN M 13.7 79.9 82.5 87.15 0.80 77.030.46

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

40

(c) CNN M 13.7 79.9 82.5 87.15 0.80 77.030.46 (d) CNN M 2048 13.5 80.1 82.4 86.64 0.53 76.880.35 (e) CNN S 13.1 79.7 82.9 87.76 0.66 77.610.12 (f) CNN S TUNE-CLS 13.1 - 83.0 88.350.56 77.330.56 (g) CNN S TUNE-RNK 13.1 82.4 83.2 – – (h) Zeiler & Fergus [19] 16.1 - 79.0 86.5 0.5 74.20.3 (i) Razavian et al. [9], [10] 14.7 77.2 – – – (j) Oquab et al. [8] 18 77.7 78.7 (82.8*) – – (k) Oquab et al. [16] - - 86.3*– – (l) Wei et al. [17] - 81.5 ( 85.2*) 81.7 ( 90.3*) – –

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

41

(l) Wei et al. [17] - 81.5 ( 85.2*) 81.7 ( 90.3*) – – (m) He et al. [29] 13.6 80.1 - 91.40.7 – TABLE 3 Comparison with the state of the art on ILSVRC2012, VOC2007, VOC2012, Caltech-101, and Caltech-256. Results marked with * were achieved using models pre-trained on the extended ILSVRC datasets (1512 classes in [8], [16], 2000 classes in [17]). All other results were achieved using CNNs pre-trained on ILSVRC-2012 (1000 classes). than those reported by [13], [19], and [10] for the related configurations. This validates our implementation, and the difference is likely to be due to the sampling of image crops from the uncropped image plane (instead of the centre). When using our CNN features on other datasets, the relative performance generally follows the same pattern as on ILSVRC, where the nets are trained – the CNN-F architecture exhibits the worst performance, with CNN-M and CNN-S performing considerably better. Further fine-tuning of CNN-S on the VOC

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

42

CNN-F architecture exhibits the worst performance, with CNN-M and CNN-S performing considerably better. Further fine-tuning of CNN-S on the VOC datasets turns out to be beneficial; on VOC-2012, using the ranking loss is marginally better than the classification loss ([g] vs. [f]), which can be explained by the ranking-based VOC evaluation criterion. Fine-tuning on Caltech-101 also yields a small improvement, but no gain is observed over Caltech-256. Our CNN-S net is competitive with recent CNNbased approaches [19], [9], [8], [16], [17], [29] and on a number of datasets (VOC-2007, VOC-2012, Caltech-101, Caltech-256) and sets the state of the art on VOC-2007 and VOC-2012 across methods pre-trained solely on ILSVRC-2012 dataset. While the CNN-based methods of [16], [17] achieve better

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

43

pre-trained solely on ILSVRC-2012 dataset. While the CNN-based methods of [16], [17] achieve better performance on VOC (86.3% and 90.3% respectively), they were trained using extended ILSVRCdatasets, enriched with additional categories semantically close to the ones in VOC. Additionally, [17] used a significantly more complex classification pipeline, driven by bounding box proposals [30], pre-trained on ILSVRC-2013 detection dataset. Their best reported result on VOC-2012 (90.3%) was achieved by the late fusion with a complex hand-crafted method of [31]; without fusion, they get 84.2%. On Caltech-101, [29] achieves the state of the art using spatial pyramid pooling of conv5 layer features, while we used full7 layer features consistently across all datasets (for full7 features, they report 87:08%). In addition to achieving performance comparable to the state of the art with a very simple approach (but powerful CNN-based features), with the modifications outlined in the paper (primarily the use of data augmentation similar to the CNN-based methods) we are able to improve the performance

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