Diverse Image Generation via Self-Conditioned GANs
arXiv:2006.10728
Abstract
We introduce a simple but effective unsupervised method for generating realistic and diverse images. We train a class-conditional GAN model without using manually annotated class labels. Instead, our model is conditional on labels automatically derived from clustering in the discriminator's feature space. Our clustering step automatically discovers diverse modes, and explicitly requires the generator to cover them. Experiments on standard mode collapse benchmarks show that our method outperforms several competing methods when addressing mode collapse. Our method also performs well on large-scale datasets such as ImageNet and Places365, improving both image diversity and standard quality metrics, compared to previous methods.
arXiv commentCVPR 2020. Code: https://github.com/stevliu/self-conditioned-gan. Webpage: http://selfcondgan.csail.mit.edu/
References in corpus (12)
- Conditional Generative Adversarial Nets
- Large Scale GAN Training for High Fidelity Natural Image Synthesis
- Spectral Normalization for Generative Adversarial Networks
- NIPS 2016 Tutorial: Generative Adversarial Networks
- The Unreasonable Effectiveness of Deep Features as a Perceptual Metric
- Which Training Methods for GANs do actually Converge?
- VEEGAN: Reducing Mode Collapse in GANs using Implicit Variational Learning
- Unrolled Generative Adversarial Networks
- Multimodal Unsupervised Image-to-Image Translation
- GAN Dissection: Visualizing and Understanding Generative Adversarial Networks
- High-Fidelity Image Generation With Fewer Labels
- Disconnected Manifold Learning for Generative Adversarial Networks