Research map / Point clouds / Segmentation and detection
Weakly Supervised 3D Scene Segmentation with Region-Level Boundary Awareness and Instance Discrimination
Kangcheng Liu, Yuzhi Zhao, Qiang Nie, Zhi Gao, Ben M. Chen
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 boundary labels w_ij from boundary prediction network (ablation) | WS3D · ScanNet instance seg val (5% labels) | AP@50: 39.9 -> 45.6 | Improved |
| add pairwise term to the energy loss (ablation) | WS3D · ScanNet instance seg val (5% labels) | AP@50: 37.0 -> 45.6 | Improved |
| add confidence indicator H_ij in the energy function (ablation) | WS3D · ScanNet instance seg val (5% labels) | AP@50: 40.1 -> 45.6 | Improved |
| add confidence indicator H_ij in the contrastive loss (ablation) | WS3D · ScanNet instance seg val (5% labels) | AP@50: 40.7 -> 45.6 | Improved |
| apply contrastive learning at multiple stages instead of the fifth only (ablation) | WS3D · ScanNet instance seg val (5% labels) | AP@50: 42.2 -> 45.6 | Improved |
| add unsupervised region-level contrastive learning (ablation) | WS3D · ScanNet instance seg val (5% labels) | AP@50: 38.1 -> 45.6 | Improved |
| add supervised cross-entropy branch (ablation) | WS3D · ScanNet instance seg val (5% labels) | AP@50: 42.9 -> 45.6 | Improved |
| add unsupervised branch to supervised-only training (ablation) | WS3D · ScanNet instance seg val (5% labels) | AP@50: 34.8 -> 45.6 | Improved |
| add WS3D unlabeled-data losses to supervised-only GPC at 1% labels | Sup-only-GPC · ScanNet val (1% labels) | mIoU: 40.9 -> 49.9 | Improved |
| add WS3D losses to supervised-only PointGroup at 1% labels | Sup-only-GPC · ScanNet instance seg val (1% labels) | AP@50: 10.8 -> 32.5 | Improved |
| replace GPC with WS3D at 1% labels | GPC · S3DIS Area5 val (1% labels) | mIoU: 38.2 -> 45.3 | Improved |
| add test set as unlabeled data (transductive learning) | WS3D · ScanNet val (20% labels) | mIoU: 68.5 -> 71.4 | Improved |
Closest papers by meaning
- General 3D Vision-Language Model With Fast Rendering and Pre-Training Vision-Language Alignment (TPAMI 2025 · similarity 0.92)
- Exploring Data-Efficient 3D Scene Understanding With Contrastive Scene Contexts (CVPR 2021 · similarity 0.91)
- Associatively Segmenting Instances and Semantics in Point Clouds (CVPR 2019 · similarity 0.90)
- JSIS3D: Joint Semantic-Instance Segmentation of 3D Point Clouds With Multi-Task Pointwise Networks and Multi-Value Conditional Random Fields (CVPR 2019 · similarity 0.90)
- LogoSP: Local-global Grouping of Superpoints for Unsupervised Semantic Segmentation of 3D Point Clouds (CVPR 2025 · similarity 0.89)
- Multi-Modality Affinity Inference for Weakly Supervised 3D Semantic Segmentation (AAAI 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.
