Research map / 3D vision / 3D generation
VMNet: Voxel-Mesh Network for Geodesic-Aware 3D Semantic Segmentation
Zeyu Hu, Xuyang Bai, Jiaxiang Shang, Runze Zhang, Jiayu Dong, Xin Wang, Guangyuan Sun, Hongbo Fu, Chiew-Lan Tai
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 geodesic mesh branch to the Euclidean-only network (ablation) | VMNet · ScanNet v2 val | mIoU: 71.0 -> 73.3 | Improved |
| add intra-domain attentive aggregation (ablation) | Euclidean baseline · ScanNet v2 val | mIoU: 70.2 -> 72.1 | Improved |
| add inter-domain attentive fusion (ablation) | Baseline + Intra · ScanNet v2 val | mIoU: 72.1 -> 73.3 | Improved |
| use scalar attention instead of vector attention (ablation) | VMNet · ScanNet v2 val | mIoU: 72.3 -> 73.3 | Improved |
| use scalar attention instead of EdgeConv (ablation) | VMNet · ScanNet v2 val | mIoU: 72.6 -> 73.3 | Improved |
| use VC then QEM instead of VC only for mesh simplification (ablation) | VMNet · ScanNet v2 val | mIoU: 72.3 -> 73.3 | Improved |
| apply VC on first two mesh levels before QEM instead of QEM only (ablation) | VMNet · ScanNet v2 val | mIoU: 72.9 -> 73.3 | Improved |
| use primal instead of dual inter-domain attention (ablation) | VMNet · ScanNet v2 val | mIoU: 72.8 -> 73.3 | Improved |
| decode in the geodesic domain instead of projecting back to voxels (Variant 1) (ablation) | VMNet · ScanNet v2 val | mIoU: 72.4 -> 73.3 | Improved |
| add mesh branch to the encoder too (Variant 2) (ablation) | VMNet · ScanNet v2 val | mIoU: 73.3 -> 73.4 | No clear effect |
| add more mesh levels to the geodesic branch (ablation) | VMNet · ScanNet v2 val | mIoU | Improved |
| add geodesic information to sparse voxel network (VMNet vs MinkowskiNet) | VMNet · ScanNet v2 test | mIoU: 73.6 -> 74.6 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- TextureNet — TextureNet: Consistent Local Parametrizations for Learning From High-Resolution Signals on Meshes (CVPR 2019) · better on 1 of 1 shared result
- ScanComplete — ScanComplete: Large-Scale Scene Completion and Semantic Segmentation for 3D Scans (CVPR 2018) · better on 1 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
- Voxel-Mesh Network for Geodesic-Aware 3D Semantic Segmentation of Indoor Scenes (TPAMI 2022 · similarity 0.98)
- Voxel Proposal Network via Multi-Frame Knowledge Distillation for Semantic Scene Completion (NeurIPS 2024 · similarity 0.87)
- Not All Voxels Are Equal: Hardness-Aware Semantic Scene Completion with Self-Distillation (CVPR 2024 · similarity 0.86)
- Geometric Exploitation for Indoor Panoramic Semantic Segmentation (NeurIPS 2024 · similarity 0.86)
- OA-CNNs: Omni-Adaptive Sparse CNNs for 3D Semantic Segmentation (CVPR 2024 · similarity 0.86)
- Edge-Aware 3D Instance Segmentation Network with Intelligent Semantic Prior (CVPR 2024 · similarity 0.86)
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.
