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..
NopeRoomGS: Indoor 3D Gaussian Splatting Optimization without Camera Pose Input
Authors: Wenbo Li, Yan Xu, Mingde Yao, Fengjie Liang, Jiankai Sun, Menglu Wang, Guofeng Zhang, Linjiang Huang, Hongsheng Li
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on Replica, Scan Net and Tanks & Temples demonstrate the state-of-the-art performance of our method in both camera pose estimation and novel view synthesis. |
| Researcher Affiliation | Academia | 1MMLab, The Chinese University of Hong Kong 2CPII 3University of Michigan 4Hong Kong Polytechnic University 5Stanford 6USTC 7Zhejiang University 8BUAA EMAIL, EMAIL, EMAIL, EMAIL, EMAIL |
| Pseudocode | No | The paper describes the methodology in narrative text within sections 3.1 and 3.2, outlining the local and global stages, but does not contain any clearly labeled "Pseudocode" or "Algorithm" blocks or structured code-like formatting. |
| Open Source Code | No | Although no code is released, it includes detailed instructions, data preparation, training, and evaluation. These are sufficient to faithfully reproduce the main experimental results and we will open source upon the paper acceptance. |
| Open Datasets | Yes | Extensive experiments on Replica [44], Scan Net [9] and Tanks & Temples [27] demonstrate the state-of-the-art performance of our method in both camera pose estimation and novel view synthesis. |
| Dataset Splits | No | For camera pose estimation, the optimized camera poses are aligned with Procrustes Analysis, as described in prior works [5, 54], and compared against the ground-truth poses from training views. For novel view synthesis, we adopt the evaluation protocol of CF-3DGS [10] and Ne RFmm [54]. The optimized 3DGS model, trained exclusively on the training views, is kept fixed, while the camera poses of the test views are refined by minimizing the photometric reconstruction error between synthesized and groundtruth images. |
| Hardware Specification | No | For detailed information, please refer to the supplementary materials. |
| Software Dependencies | No | Our method is implemented using Py Torch [39], building on the optimization settings from 3DGS [25] with necessary adjustments. |
| Experiment Setup | No | Our method is implemented using Py Torch [39], building on the optimization settings from 3DGS [25] with necessary adjustments. A key feature is synchronizing new frame additions with point densification intervals to ensure steady scene expansion. For detailed information, please refer to the supplementary materials. |