ELBO-T2IAlign: A Generic ELBO-Based Method for Calibrating Pixel-level Text-Image Alignment in Diffusion Models

TL;DR

This paper introduces ELBO-T2IAlign, a training-free, generic method to calibrate pixel-level text-image alignment in diffusion models, improving their performance in segmentation and image editing tasks by addressing misalignment issues.

Contribution

We propose a novel ELBO-based calibration method for diffusion models that does not require training and effectively improves pixel-text alignment across architectures.

Findings

Effective calibration of pixel-text alignment demonstrated on benchmark datasets.

Addresses misalignment caused by data bias in diffusion models.

Works across various diffusion model architectures.

Abstract

Diffusion models excel at image generation. Recent studies have shown that these models not only generate high-quality images but also encode text-image alignment information through attention maps or loss functions. This information is valuable for various downstream tasks, including segmentation, text-guided image editing, and compositional image generation. However, current methods heavily rely on the assumption of perfect text-image alignment in diffusion models, which is not the case. In this paper, we propose using zero-shot referring image segmentation as a proxy task to evaluate the pixel-level image and class-level text alignment of popular diffusion models. We conduct an in-depth analysis of pixel-text misalignment in diffusion models from the perspective of training data bias. We find that misalignment occurs in images with small sized, occluded, or rare object classes. Therefore, we propose ELBO-T2IAlign, a simple yet effective method to calibrate pixel-text alignment in diffusion models based on the evidence lower bound (ELBO) of likelihood. Our method is training-free and generic, eliminating the need to identify the specific cause of misalignment and works well across various diffusion model architectures. Extensive experiments on commonly used benchmark datasets on image segmentation and generation have verified the effectiveness of our proposed calibration approach.