Research map / Detection and segmentation / 3D detection and driving
Segment-Fusion: Hierarchical Context Fusion for Robust 3D Semantic Segmentation
Anirud Thyagharajan, Benjamin Ummenhofer, Prashant Laddha, Om Ji Omer, Sreenivas Subramoney
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 Segment-Fusion to SparseConvNet | SparseConvNet · ScanNet test | mIoU: 66.6 -> 71.9 | Improved |
| Add Segment-Fusion to PointConv | PointConv · ScanNet test | mIoU: 58.2 -> 62.0 | Improved |
| Add Segment-Fusion to MinkNet42 | MinkNet42 · ScanNet test | mIoU: 72.4 -> 74.7 | Improved |
| Add Segment-Fusion to VMNet | VMNet · ScanNet test | mIoU: 74.2 -> 76.7 | Improved |
| Add Segment-Fusion to MinkNet18 | MinkNet18 · S3DIS Area 5 | mIoU: 63.8 -> 65.3 | Improved |
| Add Segment-Fusion to KPConv | KPConv · S3DIS Area 5 | mIoU: 64.6 -> 65.2 | Improved |
| Add Segment-Fusion to ASIS | ASIS · S3DIS Area 5 | mIoU: 32.4 -> 33.2 | Improved |
| Use J-GSV segment majority voting without the fusion network (ablation) | SparseConvNet · ScanNet validation | mIoU: 67.1 -> 70.4 | Improved |
| Add segment loss to instance-loss training (ablation) | SparseConvNet + Segment-Fusion · ScanNet validation | mIoU: 69.7 -> 71.5 | Improved |
| Use instance loss only, omitting segment loss (ablation) | SparseConvNet + Segment-Fusion · ScanNet validation | mIoU: 71.5 -> 69.7 | Hurt |
| Use J-GSV segment majority voting without the fusion network (ablation) | PointConv · ScanNet validation | mIoU: 58.3 -> 61.5 | Improved |
| Add segment loss to instance-loss training (ablation) | PointConv + Segment-Fusion · ScanNet validation | mIoU: 61.4 -> 63.6 | Improved |
Closest papers by meaning
- SGFormer: Semantic-Geometry Fusion Transformer for Multi-modal 3D Panoptic Segmentation (AAAI 2025 · similarity 0.88)
- Edge-Aware 3D Instance Segmentation Network with Intelligent Semantic Prior (CVPR 2024 · similarity 0.87)
- Segment Any 3D Object with Language (ICLR 2025 · similarity 0.87)
- BFANet: Revisiting 3D Semantic Segmentation with Boundary Feature Analysis (CVPR 2025 · similarity 0.87)
- A Unified Framework for 3D Scene Understanding (NeurIPS 2024 · similarity 0.87)
- SAI3D: Segment Any Instance in 3D Scenes (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.
