DeRainGS: Gaussian Splatting for Enhanced Scene Reconstruction in Rainy Environments

Authors: Shuhong Liu, Xiang Chen, Hongming Chen, Quanfeng Xu, Mingrui Li

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments across a wide range of rain scenarios demonstrate that our method delivers state-of-the-art performance, remarkably outperforming existing occlusion-free methods by a large margin. We evaluate our method on the waterdrop scenes from Derain Ne RF (Li et al. 2024) and our Hydro Views dataset... To assess the quality of the reconstruction, we adhere to common metrics including PSNR, SSIM, and LPIPS... Ablation Study We conduct ablation studies on both the raindrop and rain streak scenes for each component of De Rain GS and present the average results for quantitative evaluation.
Researcher Affiliation Academia 1The University of Tokyo, 7-3-1 Hongo, Tokyo, Japan 2Nanjing University of Science and Technology, 200 Xiaolingwei Rd, Nanjing, China 3Dalian Maritime University, 1 Linghai Rd, Dalian, China 4Shanghai Astronomical Observatory, 80 Nandan Rd, Shanghai, China 5University of Chinese Academy of Sciences, 1 Yanqihu East Rd, Beijing, China 6Dalian University of Technology, 2 Linggong Rd, Dalian, China
Pseudocode No The paper describes the methodology using text, mathematical formulas, and block diagrams (Figure 3), but it does not contain any explicitly labeled 'Pseudocode' or 'Algorithm' blocks.
Open Source Code No The paper provides a 'Home Page and Datasets https://deraings.github.io'. While this links to a project homepage, it is not a direct link to a specific source-code repository as required by the prompt's criteria for a 'Yes' answer. It is a high-level project overview page.
Open Datasets Yes To benchmark this task, we construct the Hydro Views dataset that comprises a diverse collection of both synthesized and real-world scene images... Home Page and Datasets https://deraings.github.io... We synthesize data derived from real-world scenes captured by Mip Ne RF-360 (Barron et al. 2022) and Tanks-and-Temples (Knapitsch et al. 2017), along with real-world rainy environments that we have independently collected. We evaluate our method on the waterdrop scenes from Derain Ne RF (Li et al. 2024) and our Hydro Views dataset.
Dataset Splits No The paper states: 'We evaluate our method on the waterdrop scenes from Derain Ne RF (Li et al. 2024) and our Hydro Views dataset, where each synthesized scene features three distinct patterns of raindrops and rain streaks.' and 'Results are averaged over three different types of raindrops or rain streaks within each scene.' However, it does not provide specific training/validation/test splits (e.g., percentages or counts) for these datasets, nor does it refer to standard splits with citations.
Hardware Specification Yes All experiments were conducted on a single NVIDIA A100-80GB GPU.
Software Dependencies No We implement De Rain GS in Py Torch using Adam optimizer... The paper mentions software components like PyTorch and Adam optimizer but does not provide specific version numbers for any of them.
Experiment Setup Yes We implement De Rain GS in Py Torch using Adam optimizer with learning rates of 1.6e-4, 5e-4, and 2.5e-3 for G s means, scaling, and SH features, and 1e-3 for tuning occlusion masking modules. We minimize L over 35,000 iterations.