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..
Quantifying and Alleviating Co-Adaptation in Sparse-View 3D Gaussian Splatting
Authors: Kangjie Chen, Yingji Zhong, Zhihao Li, Jiaqi Lin, Youyu Chen, Minghan Qin, Haoqian Wang
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct experiments on three datasets: LLFF [47], DTU [51], and Blender dataset [52]. ... We evaluate five sparse-view 3DGS-based methods with and without our proposed co-adaptation suppression strategies... We report PSNR, SSIM, LPIPS, and Co-Adaptation scores (CA) on both training and novel views to assess reconstruction quality and co-adaptation reduction. |
| Researcher Affiliation | Collaboration | 1Tsinghua University 2HKUST 3Huawei Noah s Ark Lab 4Harbin Institute of Technology |
| Pseudocode | No | The paper only provides equations and descriptive text for its methods and calculations, such as in Section 3.1 and Appendix A.1, but does not contain a clearly labeled pseudocode or algorithm block. |
| Open Source Code | No | We plan to release code and instructions upon paper acceptance to ensure faithful reproduction of our results. |
| Open Datasets | Yes | We conduct experiments on three datasets: LLFF [47], DTU [51], and Blender dataset [52]. Following prior works [10, 53, 54, 55], we use 3 training views for LLFF and DTU, and 8 views for Blender. |
| Dataset Splits | Yes | Following prior works [10, 53, 54, 55], we use 3 training views for LLFF and DTU, and 8 views for Blender. Test views follow prior settings [10, 53, 54, 55]. |
| Hardware Specification | Yes | We did not specify compute resources in the experimental section, but all experiments were conducted on 8 NVIDIA RTX 3090 GPUs. |
| Software Dependencies | No | The paper mentions following official implementations for baselines (e.g., 3DGS [1], DNGaussian [2], FSGS [3], Co R-GS [22], and Binocular3DGS [10]) but does not specify version numbers for any underlying software libraries (e.g., Python, PyTorch, CUDA). |
| Experiment Setup | Yes | To ensure fair comparisons, we adopt unified parameter settings across all scenes within each dataset. ... For co-adaptation suppression, we use a fixed dropout probability of 0.2 across all methods and datasets, based on ablations conducted on Binocular3DGS. ... we tune the opacity noise scale individually for each method within [0.05, 0.8] and fix it across all scenes in the same dataset. The number of training iterations for each baseline follows its official implementation. |