Research map / 3D vision / 3D generation
3D Instance Segmentation via Multi-Task Metric Learning
Jean Lahoud, Bernard Ghanem, Marc Pollefeys, Martin R. Oswald
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 directional loss (multi-task) to feature embedding (ablation) | 3D instance segmentation network · ScanNet v2 val | AP50: 0.333 -> 0.402 | Improved |
| add directional loss (multi-task) to feature embedding (ablation) | 3D instance segmentation network · ScanNet v2 val | AP: 0.171 -> 0.203 | Improved |
| add directional loss (multi-task) to feature embedding (ablation) | 3D instance segmentation network · ScanNet v2 val | AP25: 0.492 -> 0.554 | Improved |
| add directional loss (multi-task) to feature embedding (ablation) | 3D instance segmentation network · Synthetic toy dataset, Obj1 | AP50: 97.3 -> 98.0 | Improved |
| add directional loss (multi-task) to feature embedding (ablation) | 3D instance segmentation network · Synthetic toy dataset, Obj2 | AP50: 92.7 -> 93.5 | Improved |
| add directional loss (multi-task) to feature embedding (ablation) | 3D instance segmentation network · Synthetic toy dataset, Obj5 | AP50: 95.2 -> 95.3 | No clear effect |
| add directional loss on the counter class (ablation) | 3D instance segmentation network · ScanNet v2 val, counter | AP50: 0.004 -> 0.008 | No clear effect |
| replace segmentation-as-instance with learned feature embedding (FE only) | 3D instance segmentation · ScanNet v2 val | AP50: 0.145 -> 0.333 | Improved |
| replace connected components with learned feature embedding (FE only) | 3D instance segmentation · ScanNet v2 val | AP50: 0.230 -> 0.333 | Improved |
| apply connected components to semantic segmentation labels | semantic segmentation as instances · ScanNet v2 val | AP50: 0.145 -> 0.230 | Improved |
| learned embedding vs connected components on ground-truth labels | toy data · Synthetic toy dataset, Obj5 | AP50: 79.9 -> 95.3 | 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.
- 3D-SIS — 3D-SIS: 3D Semantic Instance Segmentation of RGB-D Scans (CVPR 2019) · 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
- VoxDet: Rethinking 3D Semantic Scene Completion as Dense Object Detection (NeurIPS 2025 · similarity 0.90)
- 3D Instance Segmentation via Enhanced Spatial and Semantic Supervision (ICCV 2023 · similarity 0.90)
- Edge-Aware 3D Instance Segmentation Network with Intelligent Semantic Prior (CVPR 2024 · similarity 0.90)
- UniC-Lift: Unified 3D Instance Segmentation via Contrastive Learning (AAAI 2026 · similarity 0.89)
- Sketchy Bounding-box Supervision for 3D Instance Segmentation (CVPR 2025 · similarity 0.88)
- Insightful Instance Features for 3D Instance Segmentation (CVPR 2025 · 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.
