Seeing What a GAN Cannot Generate
This work addresses the issue of mode collapse for researchers and practitioners using GANs, providing tools to understand and diagnose failures, though it is incremental as it builds on existing GAN analysis methods.
The paper tackles the problem of mode collapse in GANs by developing methods to visualize and quantify which object classes are omitted during training, using semantic segmentation and approximate inversion techniques to analyze failures in recent GANs across multiple datasets.
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.