Simple Disentanglement of Style and Content in Visual Representations

Lilian Ngweta, Subha Maity, Alex Gittens, Yuekai Sun, Mikhail Yurochkin
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:26063-26086, 2023.

Abstract

Learning visual representations with interpretable features, i.e., disentangled representations, remains a challenging problem. Existing methods demonstrate some success but are hard to apply to large-scale vision datasets like ImageNet. In this work, we propose a simple post-processing framework to disentangle content and style in learned representations from pre-trained vision models. We model the pre-trained features probabilistically as linearly entangled combinations of the latent content and style factors and develop a simple disentanglement algorithm based on the probabilistic model. We show that the method provably disentangles content and style features and verify its efficacy empirically. Our post-processed features yield significant domain generalization performance improvements when the distribution shift occurs due to style changes or style-related spurious correlations.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-ngweta23a, title = {Simple Disentanglement of Style and Content in Visual Representations}, author = {Ngweta, Lilian and Maity, Subha and Gittens, Alex and Sun, Yuekai and Yurochkin, Mikhail}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {26063--26086}, 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/ngweta23a/ngweta23a.pdf}, url = {https://proceedings.mlr.press/v202/ngweta23a.html}, abstract = {Learning visual representations with interpretable features, i.e., disentangled representations, remains a challenging problem. Existing methods demonstrate some success but are hard to apply to large-scale vision datasets like ImageNet. In this work, we propose a simple post-processing framework to disentangle content and style in learned representations from pre-trained vision models. We model the pre-trained features probabilistically as linearly entangled combinations of the latent content and style factors and develop a simple disentanglement algorithm based on the probabilistic model. We show that the method provably disentangles content and style features and verify its efficacy empirically. Our post-processed features yield significant domain generalization performance improvements when the distribution shift occurs due to style changes or style-related spurious correlations.} }
Endnote
%0 Conference Paper %T Simple Disentanglement of Style and Content in Visual Representations %A Lilian Ngweta %A Subha Maity %A Alex Gittens %A Yuekai Sun %A Mikhail Yurochkin %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-ngweta23a %I PMLR %P 26063--26086 %U https://proceedings.mlr.press/v202/ngweta23a.html %V 202 %X Learning visual representations with interpretable features, i.e., disentangled representations, remains a challenging problem. Existing methods demonstrate some success but are hard to apply to large-scale vision datasets like ImageNet. In this work, we propose a simple post-processing framework to disentangle content and style in learned representations from pre-trained vision models. We model the pre-trained features probabilistically as linearly entangled combinations of the latent content and style factors and develop a simple disentanglement algorithm based on the probabilistic model. We show that the method provably disentangles content and style features and verify its efficacy empirically. Our post-processed features yield significant domain generalization performance improvements when the distribution shift occurs due to style changes or style-related spurious correlations.
APA
Ngweta, L., Maity, S., Gittens, A., Sun, Y. & Yurochkin, M.. (2023). Simple Disentanglement of Style and Content in Visual Representations. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:26063-26086 Available from https://proceedings.mlr.press/v202/ngweta23a.html.

Related Material