Research map / 3D vision / 3D generation
Virtual Multi-view Fusion for 3D Semantic Segmentation
Abhijit Kundu, Xiaoqi Yin, Alireza Fathi, David Ross, Brian Brewington, Thomas Funkhouser, Caroline Pantofaru
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 |
|---|---|---|---|
| render virtual views instead of using real photographic images (ablation) | multiview fusion · ScanNet val | 3D mIoU: 60.1 -> 63.2 | Improved |
| add normal and global coordinate channels to the rendered input (ablation) | virtual-view model · ScanNet val | 3D mIoU: 63.2 -> 66.1 | Improved |
| increase field of view to 2x (ablation) | virtual-view model · ScanNet val | 3D mIoU: 66.1 -> 67.9 | Improved |
| add virtual view sampling with backface culling (ablation) | virtual-view model · ScanNet val | 3D mIoU: 67.9 -> 70.1 | Improved |
| add multiview fusion after single-view prediction (real views) (ablation) | 2D segmentation model · ScanNet val | 2D image IoU: 60.1 -> 68.5 | Improved |
| add multiview fusion after single-view prediction (full method) (ablation) | 2D segmentation model · ScanNet val | 2D image IoU: 67.0 -> 74.9 | Improved |
| use virtual multiview fusion instead of real-view fusion with small training sets (ablation) | multiview fusion · ScanNet (low resolution, fewer training scenes) | 3D mIoU | Improved |
| use all ~2000 virtual views instead of ~10 at inference (ablation) | Ours · ScanNet val | 3D mIoU: 61.7 -> 70.1 | Improved |
| use all ~2000 virtual views instead of ~40 at inference (ablation) | Ours · ScanNet val | 3D mIoU: 68.2 -> 70.1 | Improved |
| average features across views instead of picking max-probability category (ablation) | fusion function · ScanNet val | 3D mIoU | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- OccuSeg — OccuSeg: Occupancy-Aware 3D Instance Segmentation (CVPR 2020) · better on 0 of 1 shared result
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
- 3DMV: Joint 3D-Multi-View Prediction for 3D Semantic Scene Segmentation (ECCV 2018 · similarity 0.91)
- Learning Multi-View Aggregation in the Wild for Large-Scale 3D Semantic Segmentation (CVPR 2022 · similarity 0.90)
- VMNet: Voxel-Mesh Network for Geodesic-Aware 3D Semantic Segmentation (ICCV 2021 · similarity 0.89)
- BFANet: Revisiting 3D Semantic Segmentation with Boundary Feature Analysis (CVPR 2025 · similarity 0.87)
- Segment Any 3D Object with Language (ICLR 2025 · similarity 0.87)
- ODIN: A Single Model for 2D and 3D Segmentation (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.
