Research map / 3D vision / 3D generation
Segment Anything in 3D with Radiance Fields
Jiazhong Cen, Jiemin Fang, Zanwei Zhou, Chen Yang, Lingxi Xie, Xiaopeng Zhang, Wei Shen, Qi Chuan Tian
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add 3D distance based confidence decay to self-prompting (ablation) | SA3D-TensoRF · NVOS | mIoU (%): 87.9 -> 90.3 | Improved |
| add confidence decay on hard scene LLFF-trex (ablation) | SA3D-TensoRF · NVOS Trex | IoU (%): 64.0 -> 82.0 | Improved |
| add confidence decay on easy scenes (ablation) | SA3D-TensoRF · NVOS other scenes | IoU (%): Fern 82.9 -> 82.9 | No clear effect |
| use SAM instead of SEEM as 2D segmenter (ablation) | SA3D-TensoRF · NVOS | mIoU (%): 86.0 -> 90.3 | Improved |
| use SAM instead of RITM as 2D segmenter (ablation) | SA3D-TensoRF · NVOS | mIoU (%): 81.2 -> 90.3 | Improved |
| increase mask grid resolution from 40^3 to 320^3 (ablation) | SA3D-TensoRF · NVOS | mIoU (%): 88.0 -> 90.3 | Improved |
| use explicit mask grids instead of an MLP mask (ablation) | SA3D-TensoRF · NVOS | mIoU (%): 80.9 -> 90.3 | Improved |
| add feature cache to SA3D-TensoRF | SA3D-TensoRF · NVOS | mIoU (%) / time: 90.3 -> 90.1 mIoU; 45 s -> 30 s + 15 | Mixed |
| add feature cache to SA3D-GS | SA3D-GS · Horns self-prompting time | ms per step: 508.1 -> 40.7 | Improved |
| replace TensoRF/NeRF with 3D-GS as radiance field | SA3D · NVOS | mIoU (%): 90.1 -> 92.2 | Improved |
| replace TensoRF with 3D-GS as radiance field | SA3D · Replica mean | mIoU (%): 82.7 -> 79.1 | Hurt |
| add iterative multi-view SA3D over single-view mask inversion | SA3D-GS · SPIn-NeRF mean | IoU (%): 74.6 -> 93.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- OmniSeg3D — OmniSeg3D: Omniversal 3D Segmentation via Hierarchical Contrastive Learning (CVPR 2024) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Any3DIS: Class-Agnostic 3D Instance Segmentation by 2D Mask Tracking (CVPR 2025 · similarity 0.89)
- SAI3D: Segment Any Instance in 3D Scenes (CVPR 2024 · similarity 0.89)
- SA3DIP: Segment Any 3D Instance with Potential 3D Priors (NeurIPS 2024 · similarity 0.88)
- SAM2Object: Consolidating View Consistency via SAM2 for Zero-Shot 3D Instance Segmentation (CVPR 2025 · similarity 0.88)
- Point-SAM: Promptable 3D Segmentation Model for Point Clouds (ICLR 2025 · similarity 0.87)
- SANeRF-HQ: Segment Anything for NeRF in High Quality (CVPR 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
