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..
OpenBox: Annotate Any Bounding Boxes in 3D
Authors: In-Jae Lee, Mungyeom Kim, Kwonyoung Ryu, Pierre Musacchio, Jaesik Park
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on the Waymo Open Dataset (WOD), the Lyft Level 5 Perception dataset, and the nu Scenes dataset demonstrate improved accuracy and efficiency over baselines. Our project page is available at: https://oliver0922.github.io/Open Box/. |
| Researcher Affiliation | Academia | 1Seoul National University 2POSTECH |
| Pseudocode | No | The paper describes the pipeline overview in Figure 2 and detailed steps in Section 3, but does not provide any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Our code is based on Open PCDet [35] and MMDetection3D [5]. Additional details on training, hyperparameters, and network architecture are provided in Appendix B. ... Code will be released. |
| Open Datasets | Yes | We conduct experiments on Waymo Open Dataset (WOD) [33], Lyft Level 5 Perception Dataset (Lyft) [12], and nu Scenes [2]. |
| Dataset Splits | Yes | We conduct experiments on Waymo Open Dataset (WOD) [33] validation set, Lyft Level 5 Perception Dataset (Lyft) [12] and nu Scenes [2]. ... Table 1: 3D object-detection results on the WOD [33] validation set. ... Table 2: 3D object-detection results on the Lyft [12] validation set. ... Table 3: Annotation performance on Lyft [12] training dataset. ... Table 4: 3D object-detection results on the nu Scenes [2] validation set. |
| Hardware Specification | Yes | We train models [6, 32, 42] on 8 NVIDIA A6000 GPUs (48GB) and 2 AMD EPYC 7763 CPUs. |
| Software Dependencies | No | Our code is based on Open PCDet [35] and MMDetection3D [5]. ... We also employ VDBFusion [36] for SDF. |
| Experiment Setup | Yes | Table 6: Training and network details for experiment configs Voxel R-CNN [6] Point RCNN [32] Center Point [42] optimizer Adam W Adam W Adam W base learning rate 1e-2 1e-2 1e-4 weight decay 1e-3 1e-2 1e-2 momentum 0.9 1e-2 momentum range [0.95, 0.85] [0.95, 0.85] learning rate decay 0.1 0.1 learning rate clip 1e-7 1e-7 gradient norm clip 10 10 35 batch size 16 2 32 epoch 20 60 20 |