Research map / Point clouds / Segmentation and detection
Instance-Aware Embedding for Point Cloud Instance Segmentation
Tong He, Yifan Liu, Chunhua Shen, Xinlong Wang, Changming Sun
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 focal loss to the semantic branch (ablation) | PointNet++ two-branch baseline · ScanNetV2 val | AP50: 22.0 -> 24.0 | Improved |
| add focal loss to the semantic branch (ablation) | PointNet++ two-branch baseline · ScanNetV2 val | AP25: 45.2 -> 45.5 | No clear effect |
| add instance-aware context (IAM) to the instance decoder (ablation) | Baseline + FL · ScanNetV2 val | AP50: 24.0 -> 27.6 | Improved |
| add instance-aware context (IAM) to the instance decoder (ablation) | Baseline + FL · ScanNetV2 val | AP25: 45.5 -> 48.2 | Improved |
| add centroid constraint loss Lcen (ablation) | Baseline + FL + InsContext · ScanNetV2 val | AP50: 27.6 -> 28.9 | Improved |
| add geometric embedding (predicted centroids concatenated to embedding) (ablation) | Baseline + FL + InsContext + Lcen · ScanNetV2 val | AP50: 28.9 -> 31.5 | Improved |
| add geometric embedding (predicted centroids concatenated to embedding) (ablation) | Baseline + FL + InsContext + Lcen · ScanNetV2 val | AP25: 48.9 -> 50.4 | Improved |
| use more representative points than K=18 (ablation) | IAM · ScanNetV2 val | AP50 | No clear effect |
| enlarge the embedding dimension (ablation) | discriminative embedding · ScanNetV2 val | AP50 | No clear effect |
| apply K-NN only at the bottleneck layer (IAM) instead of every point (ASIS) | instance segmentation network · training memory per sample | memory: 700M -> 400M | Improved |
| apply K-NN only at the bottleneck layer (IAM) instead of every point (ASIS) | instance segmentation network · inference time per block | time: 60ms -> 42ms | Improved |
| IAM versus ASIS on ScanNetV2 validation | PointNet++ backbone · ScanNetV2 val | mAP@0.25: 47.4 -> 50.4 | 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.
- JSNet — JSNet: Joint Instance and Semantic Segmentation of 3D Point Clouds (AAAI 2020) · better on 3 of 4 shared results
- GSPN — GSPN: Generative Shape Proposal Network for 3D Instance Segmentation in Point Cloud (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
- Associatively Segmenting Instances and Semantics in Point Clouds (CVPR 2019 · similarity 0.92)
- Insightful Instance Features for 3D Instance Segmentation (CVPR 2025 · similarity 0.89)
- FreePoint: Unsupervised Point Cloud Instance Segmentation (CVPR 2024 · similarity 0.89)
- Edge-Aware 3D Instance Segmentation Network with Intelligent Semantic Prior (CVPR 2024 · similarity 0.89)
- Relation3D : Enhancing Relation Modeling for Point Cloud Instance Segmentation (CVPR 2025 · similarity 0.88)
- Open3DIS: Open-Vocabulary 3D Instance Segmentation with 2D Mask Guidance (CVPR 2024 · 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.
