Entropy-driven Unsupervised Keypoint Representation Learning in Videos

Ali Younes, Simone Schaub-Meyer, Georgia Chalvatzaki
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:40225-40253, 2023.

Abstract

Extracting informative representations from videos is fundamental for effectively learning various downstream tasks. We present a novel approach for unsupervised learning of meaningful representations from videos, leveraging the concept of image spatial entropy (ISE) that quantifies the per-pixel information in an image. We argue that local entropy of pixel neighborhoods and their temporal evolution create valuable intrinsic supervisory signals for learning prominent features. Building on this idea, we abstract visual features into a concise representation of keypoints that act as dynamic information transmitters, and design a deep learning model that learns, purely unsupervised, spatially and temporally consistent representations directly from video frames. Two original information-theoretic losses, computed from local entropy, guide our model to discover consistent keypoint representations; a loss that maximizes the spatial information covered by the keypoints and a loss that optimizes the keypoints’ information transportation over time. We compare our keypoint representation to strong baselines for various downstream tasks, e.g., learning object dynamics. Our empirical results show superior performance for our information-driven keypoints that resolve challenges like attendance to static and dynamic objects or objects abruptly entering and leaving the scene.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-younes23a, title = {Entropy-driven Unsupervised Keypoint Representation Learning in Videos}, author = {Younes, Ali and Schaub-Meyer, Simone and Chalvatzaki, Georgia}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {40225--40253}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/younes23a/younes23a.pdf}, url = {https://proceedings.mlr.press/v202/younes23a.html}, abstract = {Extracting informative representations from videos is fundamental for effectively learning various downstream tasks. We present a novel approach for unsupervised learning of meaningful representations from videos, leveraging the concept of image spatial entropy (ISE) that quantifies the per-pixel information in an image. We argue that local entropy of pixel neighborhoods and their temporal evolution create valuable intrinsic supervisory signals for learning prominent features. Building on this idea, we abstract visual features into a concise representation of keypoints that act as dynamic information transmitters, and design a deep learning model that learns, purely unsupervised, spatially and temporally consistent representations directly from video frames. Two original information-theoretic losses, computed from local entropy, guide our model to discover consistent keypoint representations; a loss that maximizes the spatial information covered by the keypoints and a loss that optimizes the keypoints’ information transportation over time. We compare our keypoint representation to strong baselines for various downstream tasks, e.g., learning object dynamics. Our empirical results show superior performance for our information-driven keypoints that resolve challenges like attendance to static and dynamic objects or objects abruptly entering and leaving the scene.} }
Endnote
%0 Conference Paper %T Entropy-driven Unsupervised Keypoint Representation Learning in Videos %A Ali Younes %A Simone Schaub-Meyer %A Georgia Chalvatzaki %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-younes23a %I PMLR %P 40225--40253 %U https://proceedings.mlr.press/v202/younes23a.html %V 202 %X Extracting informative representations from videos is fundamental for effectively learning various downstream tasks. We present a novel approach for unsupervised learning of meaningful representations from videos, leveraging the concept of image spatial entropy (ISE) that quantifies the per-pixel information in an image. We argue that local entropy of pixel neighborhoods and their temporal evolution create valuable intrinsic supervisory signals for learning prominent features. Building on this idea, we abstract visual features into a concise representation of keypoints that act as dynamic information transmitters, and design a deep learning model that learns, purely unsupervised, spatially and temporally consistent representations directly from video frames. Two original information-theoretic losses, computed from local entropy, guide our model to discover consistent keypoint representations; a loss that maximizes the spatial information covered by the keypoints and a loss that optimizes the keypoints’ information transportation over time. We compare our keypoint representation to strong baselines for various downstream tasks, e.g., learning object dynamics. Our empirical results show superior performance for our information-driven keypoints that resolve challenges like attendance to static and dynamic objects or objects abruptly entering and leaving the scene.
APA
Younes, A., Schaub-Meyer, S. & Chalvatzaki, G.. (2023). Entropy-driven Unsupervised Keypoint Representation Learning in Videos. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:40225-40253 Available from https://proceedings.mlr.press/v202/younes23a.html.

Related Material