Sample Distillation for Object Detection and Image Classification

Olivier Canevet, Leonidas Lefakis, Francois Fleuret
Proceedings of the Sixth Asian Conference on Machine Learning, PMLR 39:64-79, 2015.

Abstract

We propose a novel approach to efficiently select informative samples for large-scale learning. Instead of directly feeding a learning algorithm with a very large amount of samples, as it is usually done to reach state-of-the-art performance, we have developed a "distillation" procedure to recursively reduce the size of an initial training set using a criterion that ensures the maximization of the information content of the selected sub-set. We demonstrate the performance of this procedure for two different computer vision problems. First, we show that distillation can be used to improve the traditional bootstrapping approach to object detection. Second, we apply distillation to a classification problem with artificial distortions. We show that in both cases, using the result of a distillation process instead of a random sub-set taken uniformly in the original sample set improves performance significantly.

Cite this Paper


BibTeX
@InProceedings{pmlr-v39-canevet14b, title = {Sample Distillation for Object Detection and Image Classification}, author = {Canevet, Olivier and Lefakis, Leonidas and Fleuret, Francois}, booktitle = {Proceedings of the Sixth Asian Conference on Machine Learning}, pages = {64--79}, year = {2015}, editor = {Phung, Dinh and Li, Hang}, volume = {39}, series = {Proceedings of Machine Learning Research}, address = {Nha Trang City, Vietnam}, month = {26--28 Nov}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v39/canevet14b.pdf}, url = {https://proceedings.mlr.press/v39/canevet14b.html}, abstract = {We propose a novel approach to efficiently select informative samples for large-scale learning. Instead of directly feeding a learning algorithm with a very large amount of samples, as it is usually done to reach state-of-the-art performance, we have developed a "distillation" procedure to recursively reduce the size of an initial training set using a criterion that ensures the maximization of the information content of the selected sub-set. We demonstrate the performance of this procedure for two different computer vision problems. First, we show that distillation can be used to improve the traditional bootstrapping approach to object detection. Second, we apply distillation to a classification problem with artificial distortions. We show that in both cases, using the result of a distillation process instead of a random sub-set taken uniformly in the original sample set improves performance significantly.} }
Endnote
%0 Conference Paper %T Sample Distillation for Object Detection and Image Classification %A Olivier Canevet %A Leonidas Lefakis %A Francois Fleuret %B Proceedings of the Sixth Asian Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2015 %E Dinh Phung %E Hang Li %F pmlr-v39-canevet14b %I PMLR %P 64--79 %U https://proceedings.mlr.press/v39/canevet14b.html %V 39 %X We propose a novel approach to efficiently select informative samples for large-scale learning. Instead of directly feeding a learning algorithm with a very large amount of samples, as it is usually done to reach state-of-the-art performance, we have developed a "distillation" procedure to recursively reduce the size of an initial training set using a criterion that ensures the maximization of the information content of the selected sub-set. We demonstrate the performance of this procedure for two different computer vision problems. First, we show that distillation can be used to improve the traditional bootstrapping approach to object detection. Second, we apply distillation to a classification problem with artificial distortions. We show that in both cases, using the result of a distillation process instead of a random sub-set taken uniformly in the original sample set improves performance significantly.
RIS
TY - CPAPER TI - Sample Distillation for Object Detection and Image Classification AU - Olivier Canevet AU - Leonidas Lefakis AU - Francois Fleuret BT - Proceedings of the Sixth Asian Conference on Machine Learning DA - 2015/02/16 ED - Dinh Phung ED - Hang Li ID - pmlr-v39-canevet14b PB - PMLR DP - Proceedings of Machine Learning Research VL - 39 SP - 64 EP - 79 L1 - http://proceedings.mlr.press/v39/canevet14b.pdf UR - https://proceedings.mlr.press/v39/canevet14b.html AB - We propose a novel approach to efficiently select informative samples for large-scale learning. Instead of directly feeding a learning algorithm with a very large amount of samples, as it is usually done to reach state-of-the-art performance, we have developed a "distillation" procedure to recursively reduce the size of an initial training set using a criterion that ensures the maximization of the information content of the selected sub-set. We demonstrate the performance of this procedure for two different computer vision problems. First, we show that distillation can be used to improve the traditional bootstrapping approach to object detection. Second, we apply distillation to a classification problem with artificial distortions. We show that in both cases, using the result of a distillation process instead of a random sub-set taken uniformly in the original sample set improves performance significantly. ER -
APA
Canevet, O., Lefakis, L. & Fleuret, F.. (2015). Sample Distillation for Object Detection and Image Classification. Proceedings of the Sixth Asian Conference on Machine Learning, in Proceedings of Machine Learning Research 39:64-79 Available from https://proceedings.mlr.press/v39/canevet14b.html.

Related Material