Dual Contrastive Loss and Attention for GANs
arXiv:2103.16748
Abstract
Generative Adversarial Networks (GANs) produce impressive results on unconditional image generation when powered with large-scale image datasets. Yet generated images are still easy to spot especially on datasets with high variance (e.g. bedroom, church). In this paper, we propose various improvements to further push the boundaries in image generation. Specifically, we propose a novel dual contrastive loss and show that, with this loss, discriminator learns more generalized and distinguishable representations to incentivize generation. In addition, we revisit attention and extensively experiment with different attention blocks in the generator. We find attention to be still an important module for successful image generation even though it was not used in the recent state-of-the-art models. Lastly, we study different attention architectures in the discriminator, and propose a reference attention mechanism. By combining the strengths of these remedies, we improve the compelling state-of-the-art Fréchet Inception Distance (FID) by at least 17.5% on several benchmark datasets. We obtain even more significant improvements on compositional synthetic scenes (up to 47.5% in FID). Code and models are available at https://github.com/ningyu1991/AttentionDualContrastGAN .
Accepted to ICCV'21
References in corpus (7)
- GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium
- Pay Less Attention with Lightweight and Dynamic Convolutions
- Amortised MAP Inference for Image Super-resolution
- Image Augmentations for GAN Training
- Learning to Predict Layout-to-image Conditional Convolutions for Semantic Image Synthesis
- Training GANs with Stronger Augmentations via Contrastive Discriminator
- Beyond the Spectrum: Detecting Deepfakes via Re-Synthesis