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..
Towards 3D Objectness Learning in an Open World
Authors: Taichi Liu, Zhenyu Wang, Ruofeng Liu, Guang Wang, Desheng Zhang
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments demonstrate the extraordinary performance of OP3Det, which significantly surpasses existing open-world 3D detectors by up to 16.0% in AR and achieves a 13.5% improvement compared to closed-world 3D detectors. |
| Researcher Affiliation | Academia | Taichi Liu1, Zhenyu Wang2, Ruofeng Liu3, Guang Wang4, Desheng Zhang1 1Rutgers University 2Tsinghua University 3Michigan State University 4Florida State University |
| Pseudocode | Yes | We summarize our method in Algorithm 1 and 2. Specifically, OP3Det utilizes both point clouds and RGB images for multi-modal training to detect in the 3D open world. |
| Open Source Code | No | While the code and data will be released after publication, the paper and supplementary material provide detailed descriptions of the datasets, model architecture, training procedure, and evaluation protocols, which are sufficient to reproduce the main results. |
| Open Datasets | Yes | For indoor scenes, we utilize SUN RGB-D [21] and Scan Net V2 [22] datasets. For outdoor 3D detection, we mainly conduct experiments on the KITTI [23] dataset. |
| Dataset Splits | No | For indoor scenes, we utilize SUN RGB-D [21] and Scan Net V2 [22] datasets. For outdoor 3D detection, we mainly conduct experiments on the KITTI [23] dataset. We mainly follow the setting of [7] for category splitting. |
| Hardware Specification | Yes | We provide details on the GPU type (e.g., NVIDIA A100), batch size, and number of training epochs. |
| Software Dependencies | No | We implement with mmdetection3D [62], and train with the Adam W [63] optimizer. We use Res Net50 [64] and FPN [65] for the image feature extractor, and sparse 3D Res Net for the voxel feature extractor. |
| Experiment Setup | Yes | We implement with mmdetection3D [62], and train with the Adam W [63] optimizer. We use Res Net50 [64] and FPN [65] for the image feature extractor, and sparse 3D Res Net for the voxel feature extractor. We use the multi-scale of δ = (0.2, 0.5, 1, 2). Npoint is set to the half number of the total points. We utilize the 0.6 threshold to filter low-quality discovered 3D objects. |