Roy Ganz • Michael Elad

BIGRoC: Boosting Image Generation via a Robust Classifier
Roy Ganz, Michael Elad

Abstract: The interest of the machine learning community in image synthesis has grown significantly in recent years, with the introduction of a wide range of deep generative models and means for training them. In this work, we propose a general model-agnostic technique for improving the image quality and the distribution fidelity of generated images, obtained by any generative model. Our method, termed BIGRoC (Boosting Image Generation via a Robust Classifier), is based on a post-processing procedure via the guidance of a given robust classifier and without a need for additional training of the generative model. Given a synthesized image, we propose to update it through projected gradient steps over the robust classifier, in an attempt to refine its recognition. We demonstrate this post-processing algorithm on various image synthesis methods and show a significant improvement of the generated images, both quantitatively and qualitatively, on CIFAR-10 and ImageNet. Specifically, BIGRoC improves the image synthesis state of the art on ImageNet 128x128 by 14.81%, attaining an FID score of 2.53 and on 256x256 by 7.87%, achieving an FID of 3.63.

Ganz, Roy, and Michael Elad. "BIGRoC: Boosting Image Generation via a Robust Classifier." arXiv preprint arXiv:2108.03702 (2021).

@article{DBLP:journals/corr/abs-2108-03702,
author    = {Roy Ganz and Michael Elad},
title     = {BIGRoC: Boosting Image Generation via a Robust Classifier},
journal   = {CoRR},
year      = {2021}
}

The entire code is contained in Colab jupiter notebook to facilitate the environment installation. Each such notebook is self-contained with the relevant package installation and explanations.

• Robustness Package - Code.
• FID is calculated natively in PyTorch using Seitzer implementation - Code
• Mimicry - Code, Paper
• Guided-Diffusion - Code, Paper