Unsupervised Diffusion-Based Degradation Modeling for Real-World Super-Resolution
Authors: Yuying Chen, Mingde Yao, Wenbo Li, Renjing Pei, Jinjing Zhao, Wenqi Ren
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments across multiple real-world datasets demonstrate that our framework achieves state-of-the-art performance in both qualitative and quantitative comparison. |
| Researcher Affiliation | Collaboration | Yuying Chen1, Mingde Yao2, Wenbo Li3, Renjing Pei3, Jinjing Zhao4, Wenqi Ren1* 1Shenzhen Campus of Sun Yat-sen University 2The Chinese University of Hong Kong 3Huawei Noah s Ark Lab 4National Key Laboratory of Science and Technology on Information System Security, China |
| Pseudocode | No | The paper describes the methodology using prose and diagrams (Figure 2, Figure 3) but does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper states, "We adopt Stable SR1 to implement DDM, with minor modifications on loading the training dataset. ... 1https://github.com/Ice Clear/Stable SR". This links to a third-party project (Stable SR) used by the authors, but does not provide access to their specific modifications or the full implementation of their proposed UDDM framework. |
| Open Datasets | Yes | Training Datasets. The HR dataset consists of DIV2K (Agustsson and Timofte 2017), Flickr2K (Timofte et al. 2017) and Outdoor Scene Training (Wang et al. 2018) datasets. The LR dataset is made up of the training sets from Real SR (Cai et al. 2019) and DReal SR (Wei et al. 2020). Testing Datasets. We evaluate our approach on the testing sets of Real SR (Cai et al. 2019) and DReal SR (Wei et al. 2020). |
| Dataset Splits | Yes | Training Datasets. The HR dataset consists of DIV2K (Agustsson and Timofte 2017), Flickr2K (Timofte et al. 2017) and Outdoor Scene Training (Wang et al. 2018) datasets. The LR dataset is made up of the training sets from Real SR (Cai et al. 2019) and DReal SR (Wei et al. 2020). Testing Datasets. We evaluate our approach on the testing sets of Real SR (Cai et al. 2019) and DReal SR (Wei et al. 2020). |
| Hardware Specification | No | The paper mentions the training process resolution (e.g., "images cropped to 128 128 resolution") but does not provide any specific details about the hardware used (e.g., GPU models, CPU types, memory). |
| Software Dependencies | No | The paper mentions using "Stable SR" and "DDPM sampling" without providing specific version numbers for these tools or any other software dependencies like Python, PyTorch, or TensorFlow. |
| Experiment Setup | Yes | To generate degraded images, we finetune Stable SR (Wang et al. 2023) for 223 epochs with a batch size of 192, and the prompt is fixed as null. The training process is conducted on images cropped to 128 128 resolution. We adopt DDPM sampling (Ho, Jain, and Abbeel 2020) with 200 timesteps for inference. ... To super-resolve images, we finetune Stable SR for 119 epochs with a batch size of 24, and the prompt is fixed as null. The training process is conducted on images cropped to 512 512 resolution. |