Research map / Point clouds / Segmentation and detection
PointInst3D: Segmenting 3D Instances by Points
Tong He, Wei Yin, Chunhua Shen, Anton van den Hengel
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 center prior tailored to 3D points (ablation) | baseline · ScanNetV2 val | mAP: 33.7 -> 34.1 | Improved |
| add dynamic target assignment via optimal transport (without AUX head) (ablation) | baseline · ScanNetV2 val | mAP: 33.7 -> 36.8 | Improved |
| add auxiliary instance head with static assignment (ablation) | baseline · ScanNetV2 val | mAP: 33.7 -> 36.5 | Improved |
| combine dynamic targets with auxiliary head (ablation) | baseline · ScanNetV2 val | mAP: 33.7 -> 39.6 | Improved |
| combine dynamic targets with auxiliary head (ablation) | baseline · ScanNetV2 val | AP@50: 52.4 -> 59.2 | Improved |
| add dynamic targets on top of auxiliary head (ablation) | baseline + AUX · ScanNetV2 val | mAP: 36.5 -> 39.6 | Improved |
| use PointInst3D instead of DyCo3D with same K | DyCo3D · ScanNetV2 val | mAP: +1.8% mAP and 26% faster inference | Improved |
| use PointInst3D instead of HAIS | HAIS · S3DIS 6-fold | mRec: 69.4 -> 74.0 | Improved |
| use PointInst3D instead of HAIS on Area-5 | HAIS · S3DIS Area-5 | mCov: 64.3 -> 64.3 | No clear effect |
| scale backbone from small to large | PointInst3D · ScanNetV2 val | mAP: 39.6 -> 45.6 | Improved |
| use PointInst3D instead of HAIS on test set | HAIS · ScanNetV2 test | mAP: 45.7 -> 43.8 | Hurt |
| vary number of sampled points K (ablation) | PointInst3D · ScanNetV2 val | mAP: K set to 256 for highest mAP | Improved |
Closest papers by meaning
- DyCo3D: Robust Instance Segmentation of 3D Point Clouds Through Dynamic Convolution (CVPR 2021 · similarity 0.94)
- FreePoint: Unsupervised Point Cloud Instance Segmentation (CVPR 2024 · similarity 0.91)
- Spherical Mask: Coarse-to-Fine 3D Point Cloud Instance Segmentation with Spherical Representation (CVPR 2024 · similarity 0.90)
- OneFormer3D: One Transformer for Unified Point Cloud Segmentation (CVPR 2024 · similarity 0.90)
- Open3DIS: Open-Vocabulary 3D Instance Segmentation with 2D Mask Guidance (CVPR 2024 · similarity 0.89)
- Relation3D : Enhancing Relation Modeling for Point Cloud Instance Segmentation (CVPR 2025 · similarity 0.88)
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.
