Research map / Point clouds / Segmentation and detection
GaPro: Box-Supervised 3D Point Cloud Instance Segmentation Using Gaussian Processes as Pseudo Labelers
Tuan Duc Ngo, Binh-Son Hua, Khoi Nguyen
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 GP-generated pseudo labels instead of ignoring overlapped regions (ablation) | ISBNet box-supervised · ScanNetV2 val | AP: 38.1 -> 48.9 | Improved |
| use GP instead of assign-to-smaller-box heuristic (ablation) | pseudo labeling · ScanNetV2 val | AP: 41.8 -> 48.9 | Improved |
| use GP classification rather than linear classifier (ablation) | pseudo labeling · ScanNetV2 val | AP: 44.2 -> 48.9 | Improved |
| run GP on superpoints rather than points (ablation) | GaPro · ScanNetV2 val | AP: 45.7 -> 48.9 | Improved |
| GP classification instead of GP regression (ablation) | GaPro · ScanNetV2 val | AP: 47.8 -> 48.9 | Improved |
| add KL divergence loss on mean and variance (ablation) | GaPro + ISBNet · ScanNetV2 val | AP: 48.9 -> 49.2 | Improved |
| add self-training with deep features (ablation) | GaPro + ISBNet · ScanNetV2 val | AP: 48.9 -> 50.0 | Improved |
| use MSE loss for uncertainty instead of no loss (ablation) | GaPro + ISBNet · ScanNetV2 val | AP: 50.0 -> 49.9 | No clear effect |
| learnable GP length and output scale (ablation) | GaPro · ScanNetV2 val | AP: 48.0 -> 50.6 | Improved |
| use deep features in GP instead of position (ablation) | GaPro · ScanNetV2 val | AP: 48.5 -> 50.6 | Improved |
| add noise to GT box corners (ablation) | GaPro + ISBNet · ScanNetV2 val | AP: 48.3 (2cm) -> 21.0 (20% dim) | Hurt |
| drop GT boxes (ablation) | GaPro + ISBNet · ScanNetV2 val | AP: 49.6 (5%) -> 30.6 (80%) | Hurt |
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.
- FCAF3D — FCAF3D: Fully Convolutional Anchor-Free 3D Object Detection (ECCV 2022) · 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
- FreePoint: Unsupervised Point Cloud Instance Segmentation (CVPR 2024 · similarity 0.91)
- Sketchy Bounding-box Supervision for 3D Instance Segmentation (CVPR 2025 · similarity 0.90)
- PointInst3D: Segmenting 3D Instances by Points (ECCV 2022 · similarity 0.90)
- Don’t Starve the Boundaries: Boundary-Constrained Label Propagation for Weakly Supervised 3D Segmentation (ECCV 2026 · similarity 0.90)
- DyCo3D: Robust Instance Segmentation of 3D Point Clouds Through Dynamic Convolution (CVPR 2021 · similarity 0.90)
- BSNet: Box-Supervised Simulation-assisted Mean Teacher for 3D Instance Segmentation (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.
