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..
3D Gaussian Splatting based Scene-independent Relocalization with Unidirectional and Bidirectional Feature Fusion
Authors: Junyi Wang, Yuze Wang, Wantong Duan, Meng Wang, Yue Qi
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results on public 7 Scenes, Cambridge Landmarks, TUM RGB-D and Bonn demonstrate state-of-the-art performance. |
| Researcher Affiliation | Academia | 1. School of Computer Science and Technology, Shandong University, Qingdao, Shandong, China. 2. State Key Laboratory of Virtual Reality Technology and Systems, Beihang University, Beijing, China. 3. Qingdao Research Institute of Beihang University, Qingdao, Shandong, China. Corresponding Author: Yue Qi. |
| Pseudocode | No | The paper describes methods and processes in structured text sections such as "Unidirectional feature fusion module" and "Confidence matrix regression" but does not include any explicitly labeled pseudocode or algorithm blocks (e.g., Algorithm 1). |
| Open Source Code | Yes | Question: Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [Yes] Justification: [Yes] |
| Open Datasets | Yes | We conduct experiments on indoor 7 Scenes Shotton et al. (2013), TUM RGB-D Sturm et al. (2012), Bonn Palazzolo et al. (2019), Scan Net Dai et al. (2017), outdoor Mega Depth Li & Snavely (2018) and Cambridge Landmarks Kendall et al. (2015). |
| Dataset Splits | Yes | In scene-independent setting, we train GS-Reloc Net on Scan Net and test it on 7 Scenes, TUM RGB-D and Bonn. For outdoor scene-independent setting, GS-Reloc Net is trained on Mega Depth Li & Snavely (2018) and tested on Cambridge Landmarks. When performing the relocalization task on in a scene-dependent manner, GS-Reloc Net is trained on 7 Scenes and Cambridge Landmarks respectively. |
| Hardware Specification | Yes | On average, it processes testing images at 65 ms (15.4 FPS) on an Nvidia 4090 GPU across 7 Scenes and Cambridge Landmarks. |
| Software Dependencies | No | The Pn P with RANSAC uses Open CV implementation with following parameters. The paper mentions OpenCV but does not specify a version number for it or any other software component used. |
| Experiment Setup | Yes | GS-Reloc Net leverages an ADAM W optimizer with learning rates 2 10 4. The loss coefficient α in Eq. (4) is set to 0.3, the variables T, Q are initially set to 0.0. 3D GS training details: iterations: 30000. position_lr_init: 0.00016. position_lr_final: 0.0000016. position_lr_delay_mult: 0.01. position_lr_max_steps: 30000. feature_lr: 0.0025. opacity_lr: 0.05. scaling_lr: 0.005. rotation_lr: 0.001. densify_from_iter: 500. densify_until_iter: 15000. densification_interval: 100. opacity_prune_threshold: 0.005. densify_grad_threshold: 0.0002. |