Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Enhanced Control for Diffusion Bridge in Image Restoration

Conghan Yue, Zhengwei Peng, Junlong Ma, Dongyu Zhang

TL;DR

This paper addresses the limited conditional control in diffusion‑bridge image restoration by proposing ECDB, which injects low‑quality inputs as rich conditioning through four modules (DM, CHM, DFM, CM) and a Conditional Fusion Schedule. DM is kept pretrained and frozen to preserve denoising capability, while CHM, DFM, and CM introduce flexible, learnable conditioning paths that are fused dynamically via a time‑aware schedule. The approach yields state‑of‑the‑art results across deraining, inpainting, and 4× super‑resolution on standard benchmarks, validating both improved performance and training efficiency. The work offers a practical advance for diffusion‑based restoration with enhanced controllability and broad applicability to real‑world degradation scenarios, and it provides publicly available code.

Abstract

Image restoration refers to the process of restoring a damaged low-quality image back to its corresponding high-quality image. Typically, we use convolutional neural networks to directly learn the mapping from low-quality images to high-quality images achieving image restoration. Recently, a special type of diffusion bridge model has achieved more advanced results in image restoration. It can transform the direct mapping from low-quality to high-quality images into a diffusion process, restoring low-quality images through a reverse process. However, the current diffusion bridge restoration models do not emphasize the idea of conditional control, which may affect performance. This paper introduces the ECDB model enhancing the control of the diffusion bridge with low-quality images as conditions. Moreover, in response to the characteristic of diffusion models having low denoising level at larger values of \(\bm t \), we also propose a Conditional Fusion Schedule, which more effectively handles the conditional feature information of various modules. Experimental results prove that the ECDB model has achieved state-of-the-art results in many image restoration tasks, including deraining, inpainting and super-resolution. Code is avaliable at https://github.com/Hammour-steak/ECDB.

Enhanced Control for Diffusion Bridge in Image Restoration

TL;DR

This paper addresses the limited conditional control in diffusion‑bridge image restoration by proposing ECDB, which injects low‑quality inputs as rich conditioning through four modules (DM, CHM, DFM, CM) and a Conditional Fusion Schedule. DM is kept pretrained and frozen to preserve denoising capability, while CHM, DFM, and CM introduce flexible, learnable conditioning paths that are fused dynamically via a time‑aware schedule. The approach yields state‑of‑the‑art results across deraining, inpainting, and 4× super‑resolution on standard benchmarks, validating both improved performance and training efficiency. The work offers a practical advance for diffusion‑based restoration with enhanced controllability and broad applicability to real‑world degradation scenarios, and it provides publicly available code.

Abstract

Image restoration refers to the process of restoring a damaged low-quality image back to its corresponding high-quality image. Typically, we use convolutional neural networks to directly learn the mapping from low-quality images to high-quality images achieving image restoration. Recently, a special type of diffusion bridge model has achieved more advanced results in image restoration. It can transform the direct mapping from low-quality to high-quality images into a diffusion process, restoring low-quality images through a reverse process. However, the current diffusion bridge restoration models do not emphasize the idea of conditional control, which may affect performance. This paper introduces the ECDB model enhancing the control of the diffusion bridge with low-quality images as conditions. Moreover, in response to the characteristic of diffusion models having low denoising level at larger values of , we also propose a Conditional Fusion Schedule, which more effectively handles the conditional feature information of various modules. Experimental results prove that the ECDB model has achieved state-of-the-art results in many image restoration tasks, including deraining, inpainting and super-resolution. Code is avaliable at https://github.com/Hammour-steak/ECDB.
Paper Structure (13 sections, 6 equations, 4 figures, 5 tables)

This paper contains 13 sections, 6 equations, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Preliminaries
  4. ECDB
  5. Architectures
  6. Conditional Fusion Schedule
  7. Experiments
  8. Specific Tasks
  9. Image Deraining
  10. Image Super-Resolution
  11. Image Inpainting
  12. Ablation Study
  13. Conclusion

Figures (4)

  • Figure 1: The overview of the proposed ECDB
  • Figure 2: Graph of Conditional Fusion Schedule ablation study on the validation dataset.
  • Figure 3: Qualitative comparison of the visual results of different deraining methods on the Rain100L (Left) and Rain100H (Right) dataset.
  • Figure 4: Qualitative comparison of the visual results from different super-resolution methods on the DIV2K dataset (Left) and inpainting methods on the CelebA-HQ dataset (Right).