Your Local GAN: Designing Two Dimensional Local Attention Mechanisms for Generative Models

TL;DR

This paper introduces a local sparse attention layer for GANs that preserves 2D geometry, significantly improves image quality metrics, and provides new visualization techniques for attention heads.

Contribution

The paper proposes a novel 2D local sparse attention mechanism for GANs, improving image generation quality and enabling visualization of attention-based features.

Findings

FID score improved from 18.65 to 15.94 on ImageNet

Attention heads capture meaningful 2D geometric features

New inversion method visualizes attention saliency maps

Abstract

We introduce a new local sparse attention layer that preserves two-dimensional geometry and locality. We show that by just replacing the dense attention layer of SAGAN with our construction, we obtain very significant FID, Inception score and pure visual improvements. FID score is improved from $18.65$ to $15.94$ on ImageNet, keeping all other parameters the same. The sparse attention patterns that we propose for our new layer are designed using a novel information theoretic criterion that uses information flow graphs. We also present a novel way to invert Generative Adversarial Networks with attention. Our method extracts from the attention layer of the discriminator a saliency map, which we use to construct a new loss function for the inversion. This allows us to visualize the newly introduced attention heads and show that they indeed capture interesting aspects of two-dimensional geometry of real images.