ADIR: Adaptive Diffusion for Image Reconstruction
Abstract: In recent years, denoising diffusion models have demonstrated outstanding image generation performance.
The information on natural images captured by these models is useful for many image reconstruction applications,
where the task is to restore a clean image from its degraded observation. In this work, we propose a conditional sampling scheme that exploits the prior learned by diffusion models while retaining agreement with the measurements. We then combine it with a novel approach for adapting pre-trained diffusion denoising networks to their input. We perform the adaptation using images that are ``nearest neighbours'' to the degraded image, retrieved from a diverse dataset using an off-the-shelf visual-language model. To evaluate our method, we test it on two state-of-the-art publicly available diffusion models, Stable Diffusion and Guided Diffusion. We show that our proposed \textbf{A}daptive \textbf{D}iffusion for \textbf{I}mage \textbf{R}econstruction (\textbf{ADIR}) approach achieves significant improvement in image reconstruction tasks. Our code will be available online upon publication.
Submission Length: Long submission (more than 12 pages of main content)
Previous TMLR Submission Url: https://openreview.net/forum?id=ATVqgUofSr
Changes Since Last Submission: Dear Editor and Reviewers,
We would like to express our gratitude to the reviewers for their insightful comments and suggestions on our manuscript titled ''ADIR: Adaptive Diffusion for Image Reconstruction''.
We have carefully considered their feedback and have made significant revisions to address all the concerns raised.
1) In response to the reviewers' comments regarding the zoom in figures, we updated them with zoom boxes for clearer comparison.
2) We also included the works mentioned by the reviewers.
3) Table 6 now includes runtime comparison of the different adaptation approaches.
4) Following a reviewer's comment, we improved the presentation of the diffusion noise scheduler.
We sincerely hope that the reviewers and the editor find these revisions satisfactory and consider our manuscript for publication in TMLR.
We look forward to hearing from you regarding the outcome of the re-evaluation process.
Sincerely,
The authors
Assigned Action Editor: ~Jia-Bin_Huang1
Submission Number: 2302
Loading