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 [1].
StructSR: Refuse Spurious Details in Real-World Image Super-Resolution
Authors: Yachao Li, Dong Liang, Tianyu Ding, Sheng-Jun Huang
AAAI 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our experimental results demonstrate that Struct SR significantly improves the fidelity of structure and texture, improving the PSNR and SSIM metrics by an average of 5.27% and 9.36% on a synthetic dataset (DIV2K-Val) and 4.13% and 8.64% on two real-world datasets (Real SR and DReal SR) when integrated with four state-of-the-art diffusion-based Real-ISR methods. |
| Researcher Affiliation | Collaboration | 1College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics, MIIT Key Laboratory of Pattern Analysis and Machine Intelligence, Collaborative Innovation Center of Novel Software Technology and Industrialization, Nanjing 211106, China. 2Microsoft, Washington, USA |
| Pseudocode | Yes | Algorithm 1: Struct SR Inference Process |
| Open Source Code | Yes | Code https://github.com/LYCEXE/Struct SR |
| Open Datasets | Yes | Test Datasets. We use one synthetic dataset and two real-world datasets to comprehensively evaluate Struct SR. Following the pipeline in Stable SR (Wang et al. 2023a), first, we randomly crop 3K patches (resolution: 512 × 512) from the DIV2K validation set (Agustsson and Timofte 2017) and degrade them as synthetic images, named DIV2K-Val. Then, we use two real-world datasets, Real SR (Cai et al. 2019) and DReal SR (Wei et al. 2020), to center-crop LR images (128 × 128) as real-world images. |
| Dataset Splits | Yes | Test Datasets. We use one synthetic dataset and two real-world datasets to comprehensively evaluate Struct SR. Following the pipeline in Stable SR (Wang et al. 2023a), first, we randomly crop 3K patches (resolution: 512 × 512) from the DIV2K validation set (Agustsson and Timofte 2017) and degrade them as synthetic images, named DIV2K-Val. Then, we use two real-world datasets, Real SR (Cai et al. 2019) and DReal SR (Wei et al. 2020), to center-crop LR images (128 × 128) as real-world images. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used to run its experiments. |
| Software Dependencies | No | The paper mentions using publicly released codes and models for state-of-the-art diffusion-based Real-ISR methods (Stable SR, Diff BIR, PASD, See SR) and GAN-based methods (BSRGAN, Real-ESRGAN, LDL, Fe Ma SR), but it does not specify version numbers for any underlying software dependencies like Python, PyTorch, or CUDA. |
| Experiment Setup | Yes | For Struct SR, we set TSAS = 0.3 T as the inference timesteps for screening and intervene in the inference process according to the sampler. We present the ablation study on TSAS, SCE, and IDE in the supplementary material. ... The model parameters and samplers are set according to the original paper. |