Improved generator objectives for GANs
This addresses a fundamental issue in generative modeling for computer vision, though it appears incremental as it builds on existing GAN frameworks.
The paper tackled the problem of poor sample diversity in GAN training by deriving a family of generator objectives that directly target arbitrary f-divergences without minimizing a lower bound, resulting in models that achieve either improved sample quality or greater diversity.
We present a framework to understand GAN training as alternating density ratio estimation and approximate divergence minimization. This provides an interpretation for the mismatched GAN generator and discriminator objectives often used in practice, and explains the problem of poor sample diversity. We also derive a family of generator objectives that target arbitrary $f$-divergences without minimizing a lower bound, and use them to train generative image models that target either improved sample quality or greater sample diversity.