Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
DeblurDiff: Real-Word Image Deblurring with Generative Diffusion Models
Authors: Lingshun Kong, Jiawei Zhang, Dongqing Zou, Fu Lee Wang, Jimmy S. REN, Xiaohe Wu, Jiangxin Dong, Jinshan Pan
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experimental results demonstrate that the proposed method outperforms state-of-the-art image deblurring methods on both benchmark and real-world images. |
| Researcher Affiliation | Collaboration | 1Nanjing University of Science and Technology 2Sense Time Research 3PBVR 4Hong Kong Metropolitan University 5Harbin Institute of Technology |
| Pseudocode | No | The paper describes the methods narratively and using mathematical equations (e.g., Eq. 1, 2, 3, 4, 5, 6, 7) and architectural diagrams (Figure 3), but does not contain a distinct pseudocode or algorithm block. |
| Open Source Code | No | If the paper is accepted, we will make all training and testing code, as well as the datasets used, publicly available. |
| Open Datasets | Yes | Current deblurring datasets are generally small in scale and low in resolution, which is insufficient for effectively training diffusion models. Therefore, we do not use common deblurring datasets such as Go Pro [14] as our training set. Instead, we have collected and created a large-scale dataset containing approximately 500,000 data pairs. Our training dataset consists of three parts: (1) Existing deblurring datasets (including MC-Blur [34] and RSBlur [20]). (2) We capture some high-definition video clips, generating blurred and clear data pairs using the same strategy as REDS [15]. (3) We collect a large number of high-definition images and generate various motion blur kernels to synthesize corresponding blurred images. We provide more details about the training dataset in the supplementary material. Test Datasets. We evaluate the proposed Deblur Diff on commonly used image deblurring datasets, including synthetic datasets (Go Pro [14], DVD [23]) and Real Blurry Images [4], Real Blur [19], RWBI [33]. |
| Dataset Splits | No | The paper lists various training and test datasets but does not explicitly provide details on how these datasets were split into training, validation, and test sets, nor does it specify proportions or sample counts for such splits for its custom combined dataset. |
| Hardware Specification | Yes | The model is trained for 100,000 iterations using 8 NVIDIA 80G-A100 GPUs. |
| Software Dependencies | Yes | We use SD2.1 as the base model. |
| Experiment Setup | Yes | Implementation Details. We use SD2.1 as the base model. We employ the Adam optimizer [9] to train Deblur Diff with a batch size of 128. The learning rate is set to a fixed value of 5 × 10−5. The model is trained for 100,000 iterations using 8 NVIDIA 80G-A100 GPUs. |