Research map / Point clouds / Segmentation and detection
ForestFormer3D: A Unified Framework for End-to-End Segmentation of Forest LiDAR 3D Point Clouds
Binbin Xiang, Maciej Wielgosz, Stefano Puliti, Kamil Král, Martin Krůček, Azim Missarov, Rasmus Astrup
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 ISA-guided query point selection to OneFormer3D (ablation) | OneFormer3D · FOR-instanceV2 test | F1 (pp): 73.1 -> 76.5 | Improved |
| replace overlap-based block merging with score-based merging (BM1) (ablation) | OneFormer3D · FOR-instanceV2 test | F1 (pp): 73.1 -> 76.2 | Improved |
| add discarding of masks near crop boundary (BM2) over BM1 with ISA (ablation) | ISA + score-based merging · FOR-instanceV2 test | F1: 78.5 -> 80.7 | Improved |
| replace one-to-one Hungarian matching with one-to-many association (ablation) | ForestFormer3D · FOR-instanceV2 test | F1: 80.7 -> 81.5 | Improved |
| add all three components to OneFormer3D baseline (ablation) | OneFormer3D · FOR-instanceV2 test | F1: 73.1 -> 81.5 | Improved |
| add all three components to OneFormer3D baseline (ablation) | OneFormer3D · FOR-instanceV2 val | semantic mIoU: 84.1 -> 83.6 | Hurt |
| ISA-guided FPS in embedding space instead of plain FPS for queries (ablation) | query selection · FOR-instanceV2 val | coverage rate: 79.9% -> 92.7% | Improved |
| unified instance model versus ForAINetV2_R16 on semantic segmentation | ForestFormer3D · FOR-instanceV2 test | semantic mIoU: 87.5 -> 86.2 | Hurt |
| increase input radius and number of query points (ablation) | ForestFormer3D · FOR-instanceV2 test | F1 | Improved |
| raise the score threshold for mask removal (ablation) | ForestFormer3D · FOR-instanceV2 | Prec vs Rec/Cov | Mixed |
| lower point cloud density below 500 pts/m2 (ablation) | ForestFormer3D · FOR-instanceV2 test | Rec, Cov, F1 | Hurt |
| add multi-density augmentation training (ablation) | ForestFormer3D · FOR-instanceV2 test | F1 | Mixed |
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.
- OneFormer3D — OneFormer3D: One Transformer for Unified Point Cloud Segmentation (CVPR 2024) · 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
- Scaling Up Forest Vision with Synthetic Data (IJCV 2026 · similarity 0.90)
- ForestLPR: LiDAR Place Recognition in Forests Attentioning Multiple BEV Density Images (CVPR 2025 · similarity 0.88)
- OneFormer3D: One Transformer for Unified Point Cloud Segmentation (CVPR 2024 · similarity 0.87)
- Bringing SAM to new heights: leveraging elevation data for tree crown segmentation from drone imagery (NeurIPS 2025 · similarity 0.86)
- HOTFormerLoc: Hierarchical Octree Transformer for Versatile Lidar Place Recognition Across Ground and Aerial Views (CVPR 2025 · similarity 0.86)
- 3D-AVS: LiDAR-based 3D Auto-Vocabulary Segmentation (CVPR 2025 · similarity 0.85)
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.
