Exploring Image Regions Not Well Encoded by an INN

Zenan Ling, Fan Zhou, Meng Wei, Quanshi Zhang
Proceedings of The 25th International Conference on Artificial Intelligence and Statistics, PMLR 151:483-509, 2022.

Abstract

This paper proposes a method to clarify image regions that are not well encoded by an invertible neural network (INN), i.e., image regions that significantly decrease the likelihood of the input image. The proposed method can diagnose the limitation of the representation capacity of an INN. Given an input image, our method extracts image regions, which are not well encoded, by maximizing the likelihood of the image. We explicitly model the distribution of not-well-encoded regions. A metric is proposed to evaluate the extraction of the not-well-encoded regions. Finally, we use the proposed method to analyze several state-of-the-art INNs trained on various benchmark datasets.

Cite this Paper

BibTeX
@InProceedings{pmlr-v151-ling22a, title = { Exploring Image Regions Not Well Encoded by an INN }, author = {Ling, Zenan and Zhou, Fan and Wei, Meng and Zhang, Quanshi}, booktitle = {Proceedings of The 25th International Conference on Artificial Intelligence and Statistics}, pages = {483--509}, year = {2022}, editor = {Camps-Valls, Gustau and Ruiz, Francisco J. R. and Valera, Isabel}, volume = {151}, series = {Proceedings of Machine Learning Research}, month = {28--30 Mar}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v151/ling22a/ling22a.pdf}, url = {https://proceedings.mlr.press/v151/ling22a.html}, abstract = { This paper proposes a method to clarify image regions that are not well encoded by an invertible neural network (INN), i.e., image regions that significantly decrease the likelihood of the input image. The proposed method can diagnose the limitation of the representation capacity of an INN. Given an input image, our method extracts image regions, which are not well encoded, by maximizing the likelihood of the image. We explicitly model the distribution of not-well-encoded regions. A metric is proposed to evaluate the extraction of the not-well-encoded regions. Finally, we use the proposed method to analyze several state-of-the-art INNs trained on various benchmark datasets. } }
Endnote
%0 Conference Paper %T Exploring Image Regions Not Well Encoded by an INN %A Zenan Ling %A Fan Zhou %A Meng Wei %A Quanshi Zhang %B Proceedings of The 25th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2022 %E Gustau Camps-Valls %E Francisco J. R. Ruiz %E Isabel Valera %F pmlr-v151-ling22a %I PMLR %P 483--509 %U https://proceedings.mlr.press/v151/ling22a.html %V 151 %X This paper proposes a method to clarify image regions that are not well encoded by an invertible neural network (INN), i.e., image regions that significantly decrease the likelihood of the input image. The proposed method can diagnose the limitation of the representation capacity of an INN. Given an input image, our method extracts image regions, which are not well encoded, by maximizing the likelihood of the image. We explicitly model the distribution of not-well-encoded regions. A metric is proposed to evaluate the extraction of the not-well-encoded regions. Finally, we use the proposed method to analyze several state-of-the-art INNs trained on various benchmark datasets.
APA
Ling, Z., Zhou, F., Wei, M. & Zhang, Q.. (2022). Exploring Image Regions Not Well Encoded by an INN . Proceedings of The 25th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 151:483-509 Available from https://proceedings.mlr.press/v151/ling22a.html.

Related Material