Research map / Point clouds / Segmentation and detection
G2P: Gaussian-to-Point Attribute Alignment for Boundary-Aware 3D Segmentation
Hojun Song, Chae-yeong Song, Jeong-hun Hong, chaewon moon, Soo Ye Kim, Yiyi Liao, Jaehyup Lee, Sang-hyo Park
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 |
|---|---|---|---|
| transfer Gaussian attributes to points (G2P) over PT v3 | Point Transformer V3 · ScanNet v2 val | mIoU: 77.0 -> 78.4 | Improved |
| G2P vs BFANet on hidden test | segmentation · ScanNet200 hidden test | mIoU: 36.0 -> 35.7 | No clear effect |
| G2P on appearance-challenging classes | Point Transformer V3 · ScanNet v2 appearance-challenging | avg IoU: 66.8 -> 69.2 | Improved |
| G2P instance segmentation vs PT v3 | Point Transformer V3 · ScanNet200 instance val | mAP: 23.1 -> 23.2 | No clear effect |
| add boundary guidance only (ablation) | G2P · ScanNet v2 val | mIoU: 77.0 -> 77.8 | Improved |
| add appearance distillation only (ablation) | G2P · ScanNet v2 val | mIoU: 77.0 -> 77.8 | Improved |
| combine boundary guidance and distillation (ablation) | G2P · ScanNet v2 val | mIoU: 77.8 -> 78.4 | Improved |
| use Mahalanobis distance with k=20 instead of Euclidean (ablation) | G2P · ScanNet v2 val | mIoU: 77.0 -> 78.4 | Improved |
| semantic-only boundary supervision (ablation) | G2P · ScanNet v2 val | mIoU: 77.0 -> 76.9 | No clear effect |
| add scale-based boundary pruning to semantic boundary (ablation) | G2P · ScanNet v2 val | mIoU: 76.9 -> 77.8 | Improved |
| increase distillation weight lambda_d (ablation) | G2P · ScanNet v2 geometry-dependent classes | avg IoU: 82.41 -> 80.58 | Mixed |
| use GS opacity and color inputs in distillation instead of normals (ablation) | appearance encoder · ScanNet v2 val | mIoU: 76.5 -> 77.8 | 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.
- BFANet — BFANet: Revisiting 3D Semantic Segmentation with Boundary Feature Analysis (CVPR 2025) · better on 2 of 3 shared results
- UniPre3D — UniPre3D: Unified Pre-training of 3D Point Cloud Models with Cross-Modal Gaussian Splatting (CVPR 2025) · better on 2 of 2 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
- PointGS: Semantic-Consistent Unsupervised 3D Point Cloud Segmentation with 3D Gaussian Splatting (CVPR 2026 · similarity 0.88)
- Gaussian Prompter: Linking 2D Prompts for 3D Gaussian Segmentation (TPAMI 2025 · similarity 0.88)
- Trace3D: Consistent Segmentation Lifting via Gaussian Instance Tracing (ICCV 2025 · similarity 0.87)
- Bootstraping Clustering of Gaussians for View-consistent 3D Scene Understanding (AAAI 2025 · similarity 0.86)
- FreePoint: Unsupervised Point Cloud Instance Segmentation (CVPR 2024 · similarity 0.86)
- VA-GS: Enhancing the Geometric Representation of Gaussian Splatting via View Alignment (NeurIPS 2025 · similarity 0.86)
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.
