Semantic Image Synthesis via Class-Adaptive Cross-Attention

TL;DR

This paper introduces a novel semantic image synthesis method that replaces SPADE layers with class-adaptive cross-attention, improving global consistency and style transfer while maintaining high-quality image generation.

Contribution

The paper proposes a new architecture using cross-attention layers instead of SPADE for better shape-style correlation learning in semantic image synthesis.

Findings

Achieves state-of-the-art generation quality

Improves global style consistency and transfer

Enables shape manipulation without manual masks

Abstract

In semantic image synthesis the state of the art is dominated by methods that use customized variants of the SPatially-Adaptive DE-normalization (SPADE) layers, which allow for good visual generation quality and editing versatility. By design, such layers learn pixel-wise modulation parameters to de-normalize the generator activations based on the semantic class each pixel belongs to. Thus, they tend to overlook global image statistics, ultimately leading to unconvincing local style editing and causing global inconsistencies such as color or illumination distribution shifts. Also, SPADE layers require the semantic segmentation mask for mapping styles in the generator, preventing shape manipulations without manual intervention. In response, we designed a novel architecture where cross-attention layers are used in place of SPADE for learning shape-style correlations and so conditioning the image generation process. Our model inherits the versatility of SPADE, at the same time obtaining state-of-the-art generation quality, as well as improved global and local style transfer. Code and models available at https://github.com/TFonta/CA2SIS.