I2AM: Interpreting Image-to-Image Latent Diffusion Models via Bi-Attribution Maps
This work addresses the interpretability gap in image-to-image diffusion models for researchers and practitioners, offering tools for debugging and refinement, though it is incremental as it builds on existing cross-attention mechanisms.
The paper tackles the problem of interpreting image-to-image latent diffusion models by introducing I2AM, a method that visualizes bidirectional attribution maps to show how features transfer between images, and demonstrates its effectiveness in tasks like object detection and inpainting with a new evaluation metric IMACS that correlates with performance metrics.
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.