Stable-Pose: Leveraging Transformers for Pose-Guided Text-to-Image Generation

TL;DR

Stable-Pose introduces a transformer-based adapter with coarse-to-fine attention masking to improve pose-guided text-to-image generation, significantly enhancing accuracy in complex human pose scenarios.

Contribution

It proposes a novel adapter model with hierarchical attention masking for better pose guidance in T2I models, leveraging ViT's self-attention for detailed pose representation.

Findings

Achieved 57.1 AP score on LAION-Human, 13% higher than ControlNet.

Effectively handles complex pose conditions like side and rear views.

Demonstrated superior performance across five public datasets.

Abstract

Controllable text-to-image (T2I) diffusion models have shown impressive performance in generating high-quality visual content through the incorporation of various conditions. Current methods, however, exhibit limited performance when guided by skeleton human poses, especially in complex pose conditions such as side or rear perspectives of human figures. To address this issue, we present Stable-Pose, a novel adapter model that introduces a coarse-to-fine attention masking strategy into a vision Transformer (ViT) to gain accurate pose guidance for T2I models. Stable-Pose is designed to adeptly handle pose conditions within pre-trained Stable Diffusion, providing a refined and efficient way of aligning pose representation during image synthesis. We leverage the query-key self-attention mechanism of ViTs to explore the interconnections among different anatomical parts in human pose skeletons. Masked pose images are used to smoothly refine the attention maps based on target pose-related features in a hierarchical manner, transitioning from coarse to fine levels. Additionally, our loss function is formulated to allocate increased emphasis to the pose region, thereby augmenting the model's precision in capturing intricate pose details. We assessed the performance of Stable-Pose across five public datasets under a wide range of indoor and outdoor human pose scenarios. Stable-Pose achieved an AP score of 57.1 in the LAION-Human dataset, marking around 13% improvement over the established technique ControlNet. The project link and code is available at https://github.com/ai-med/StablePose.