Research map / Point clouds / Segmentation and detection
Dynamic Convolution for 3D Point Cloud Instance Segmentation
Tong He, 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 |
|---|---|---|---|
| group homogeneous points before generating filters (ablation) | DyCo3D baseline · ScanNetV2 val | mAP: 24.8 -> 29.4 | Improved |
| add position embeddings to grouped filters (ablation) | DyCo3D · ScanNetV2 val | mAP: 29.4 -> 31.8 | Improved |
| add category-aware decoding (ablation) | DyCo3D · ScanNetV2 val | mAP: 31.8 -> 34.1 | Improved |
| add light-weight transformer on bottleneck (ablation) | DyCo3D · ScanNetV2 val | mAP: 34.1 -> 34.8 | Improved |
| add position embedding to per-point filters (CondInst in 3D) (ablation) | baseline · ScanNetV2 val | mAP: 24.8 -> 27.0 | Improved |
| restrict CondInst positives to central areas (ablation) | CondInst · ScanNetV2 val | mAP: 27.0 -> 28.1 | Improved |
| replace detect-then-segment MLP/CNN with dynamic filters (ablation) | DyCo3D · ScanNetV2 val | mAP: 16.7 -> 34.8 | Improved |
| add extra ScoreNet (ablation) | DyCo3D · ScanNetV2 val | mAP: 0.3% improvement | No clear effect |
| add self-supervised pre-training (ablation) | DyCo3D Minkowski · ScanNetV2 val | mAP: 38.7 -> 39.3 | Improved |
| reduce PointGroup clustering radius from 3cm to 1cm (ablation) | PointGroup · ScanNetV2 val | mAP: drops 23.9% (3cm to 1cm) | Hurt |
| vary clustering radius 3cm to 30cm (ablation) | DyCo3D vs PointGroup · ScanNetV2 val | mAP: drops by 2.8% vs about 9% | Mixed |
| increase mask feature dimension D' from 2 to 32 (ablation) | DyCo3D · ScanNetV2 val | mAP: best at D'=16 | 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.
- PointGroup — PointGroup: Dual-Set Point Grouping for 3D Instance Segmentation (CVPR 2020) · better on 4 of 4 shared results
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
- DyCo3D: Robust Instance Segmentation of 3D Point Clouds Through Dynamic Convolution (CVPR 2021 · similarity 0.96)
- PointInst3D: Segmenting 3D Instances by Points (ECCV 2022 · similarity 0.93)
- 3D Instance Segmentation via Enhanced Spatial and Semantic Supervision (ICCV 2023 · similarity 0.92)
- OneFormer3D: One Transformer for Unified Point Cloud Segmentation (CVPR 2024 · similarity 0.90)
- FreePoint: Unsupervised Point Cloud Instance Segmentation (CVPR 2024 · similarity 0.90)
- Spherical Mask: Coarse-to-Fine 3D Point Cloud Instance Segmentation with Spherical Representation (CVPR 2024 · similarity 0.89)
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.
