Seeing What a GAN Cannot Generate

TL;DR

This paper investigates mode collapse in GANs by visualizing omitted object classes and analyzing specific failure cases, providing insights into what GANs fail to generate.

Contribution

It introduces a method to visualize and quantify mode collapse at both distribution and instance levels, revealing common failure modes of recent GANs.

Findings

Identified object classes frequently omitted by GANs.

Visualized specific instances where GANs fail to generate accurate images.

Analyzed multiple datasets and GAN architectures to find common failure patterns.

Abstract

Despite the success of Generative Adversarial Networks (GANs), mode collapse remains a serious issue during GAN training. To date, little work has focused on understanding and quantifying which modes have been dropped by a model. In this work, we visualize mode collapse at both the distribution level and the instance level. First, we deploy a semantic segmentation network to compare the distribution of segmented objects in the generated images with the target distribution in the training set. Differences in statistics reveal object classes that are omitted by a GAN. Second, given the identified omitted object classes, we visualize the GAN's omissions directly. In particular, we compare specific differences between individual photos and their approximate inversions by a GAN. To this end, we relax the problem of inversion and solve the tractable problem of inverting a GAN layer instead of the entire generator. Finally, we use this framework to analyze several recent GANs trained on multiple datasets and identify their typical failure cases.