How GAN Generators can Invert Networks in Real-Time

Rudolf Herdt, Maximilian Schmidt, Daniel Otero Baguer, Jean Le’Clerc Arrastia, Peter Maaß
Proceedings of the 15th Asian Conference on Machine Learning, PMLR 222:422-437, 2024.

Abstract

In this work, we propose a fast and accurate method to reconstruct activations of classification and semantic segmentation networks by stitching them with a GAN generator utilizing a 1x1 convolution. We test our approach on images of animals from the AFHQ wild dataset, ImageNet1K, and real-world digital pathology scans of stained tissue samples. Our results show comparable performance to established gradient descent methods but with a processing time that is two orders of magnitude faster, making this approach promising for practical applications.

Cite this Paper

BibTeX
@InProceedings{pmlr-v222-herdt24a, title = {How {GAN} Generators can Invert Networks in Real-Time}, author = {Herdt, Rudolf and Schmidt, Maximilian and Otero Baguer, Daniel and Le'Clerc Arrastia, Jean and Maa{\ss}, Peter}, booktitle = {Proceedings of the 15th Asian Conference on Machine Learning}, pages = {422--437}, year = {2024}, editor = {Yanıkoğlu, Berrin and Buntine, Wray}, volume = {222}, series = {Proceedings of Machine Learning Research}, month = {11--14 Nov}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v222/herdt24a/herdt24a.pdf}, url = {https://proceedings.mlr.press/v222/herdt24a.html}, abstract = {In this work, we propose a fast and accurate method to reconstruct activations of classification and semantic segmentation networks by stitching them with a GAN generator utilizing a 1x1 convolution. We test our approach on images of animals from the AFHQ wild dataset, ImageNet1K, and real-world digital pathology scans of stained tissue samples. Our results show comparable performance to established gradient descent methods but with a processing time that is two orders of magnitude faster, making this approach promising for practical applications.} }
Endnote
%0 Conference Paper %T How GAN Generators can Invert Networks in Real-Time %A Rudolf Herdt %A Maximilian Schmidt %A Daniel Otero Baguer %A Jean Le’Clerc Arrastia %A Peter Maaß %B Proceedings of the 15th Asian Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2024 %E Berrin Yanıkoğlu %E Wray Buntine %F pmlr-v222-herdt24a %I PMLR %P 422--437 %U https://proceedings.mlr.press/v222/herdt24a.html %V 222 %X In this work, we propose a fast and accurate method to reconstruct activations of classification and semantic segmentation networks by stitching them with a GAN generator utilizing a 1x1 convolution. We test our approach on images of animals from the AFHQ wild dataset, ImageNet1K, and real-world digital pathology scans of stained tissue samples. Our results show comparable performance to established gradient descent methods but with a processing time that is two orders of magnitude faster, making this approach promising for practical applications.
APA
Herdt, R., Schmidt, M., Otero Baguer, D., Le’Clerc Arrastia, J. & Maaß, P.. (2024). How GAN Generators can Invert Networks in Real-Time. Proceedings of the 15th Asian Conference on Machine Learning, in Proceedings of Machine Learning Research 222:422-437 Available from https://proceedings.mlr.press/v222/herdt24a.html.

Related Material