I2AM: Interpreting Image-to-Image Latent Diffusion Models via Attribution Maps

2024-07-17

arXiv.org

AI Summary

This work addresses the limited interpretability of cross-attention mechanisms in image-to-image (I2I) diffusion models. We propose the first bidirectional attribution mapping method for I2I diffusion, uncovering cross-image feature transfer pathways between reference and generated images. Our approach aggregates cross-attention scores across timesteps, layers, and attention heads to construct bidirectional attribution maps. We further introduce IMACS—a novel quantitative evaluation metric that measures alignment between attribution maps and semantic masks—enabling the first rigorous, semantics-aware assessment of attribution fidelity. Experiments across image editing, inpainting, and super-resolution tasks demonstrate that our method accurately localizes generation-critical regions. Moreover, IMACS exhibits strong statistical correlation with standard perceptual metrics (e.g., FID, LPIPS), confirming its utility for model diagnostics and performance optimization.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Interpretability, Analysis, and Evaluation of NLP Models

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Large-scale diffusion models have made significant advances in image generation, particularly through cross-attention mechanisms. While cross-attention has been well-studied in text-to-image tasks, their interpretability in image-to-image (I2I) diffusion models remains underexplored. This paper introduces Image-to-Image Attribution Maps (I2AM), a method that enhances the interpretability of I2I models by visualizing bidirectional attribution maps, from the reference image to the generated image and vice versa. I2AM aggregates cross-attention scores across time steps, attention heads, and layers, offering insights into how critical features are transferred between images. We demonstrate the effectiveness of I2AM across object detection, inpainting, and super-resolution tasks. Our results demonstrate that I2AM successfully identifies key regions responsible for generating the output, even in complex scenes. Additionally, we introduce the Inpainting Mask Attention Consistency Score (IMACS) as a novel evaluation metric to assess the alignment between attribution maps and inpainting masks, which correlates strongly with existing performance metrics. Through extensive experiments, we show that I2AM enables model debugging and refinement, providing practical tools for improving I2I model's performance and interpretability.

Problem

• Enhances interpretability of image-to-image diffusion models.
• Visualizes bidirectional attribution maps for feature transfer analysis.
• Introduces IMACS for evaluating inpainting mask alignment.

Innovation

• Bidirectional attribution maps enhance I2I model interpretability.
• Aggregates cross-attention scores across time, heads, layers.
• Introduces IMACS for inpainting mask alignment evaluation.