RepMix: Representation Mixing for Robust Attribution of Synthesized Images

TL;DR

RepMix introduces a novel GAN fingerprinting method that effectively attributes synthetic images to their source, maintaining accuracy despite semantic variations and common image transformations, and is validated on a new challenging benchmark.

Contribution

The paper presents RepMix, a new representation mixing technique with a novel loss for robust GAN image attribution, outperforming existing methods in generalization and robustness.

Findings

RepMix achieves high accuracy in semantic generalization.

RepMix demonstrates robustness to common image transformations.

The method outperforms existing GAN fingerprinting techniques.

Abstract

Rapid advances in Generative Adversarial Networks (GANs) raise new challenges for image attribution; detecting whether an image is synthetic and, if so, determining which GAN architecture created it. Uniquely, we present a solution to this task capable of 1) matching images invariant to their semantic content; 2) robust to benign transformations (changes in quality, resolution, shape, etc.) commonly encountered as images are re-shared online. In order to formalize our research, a challenging benchmark, Attribution88, is collected for robust and practical image attribution. We then propose RepMix, our GAN fingerprinting technique based on representation mixing and a novel loss. We validate its capability of tracing the provenance of GAN-generated images invariant to the semantic content of the image and also robust to perturbations. We show our approach improves significantly from existing GAN fingerprinting works on both semantic generalization and robustness. Data and code are available at https://github.com/TuBui/image_attribution.