Research map / Point clouds / Point cloud classification
Learning to Segment 3D Point Clouds in 2D Image Space
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Map point clouds to 2D grids with graph drawing, then segment with multi-scale U-Net | ShapeNet point-cloud segmentation · ShapeNet | class mIoU (%): 84.6 -> 86.5 | Helped |
| Map point clouds to 2D grids with graph drawing, then segment with multi-scale U-Net | ShapeNet point-cloud segmentation · ShapeNet | instance mIoU (%): 87.4 -> 88.0 | Helped |
| Use the graph-drawing and U-Net pipeline | PartNet point-cloud segmentation · PartNet average across levels | part-category mIoU (%): 58.1 -> 63.0 | Helped |
| Use multi-scale Inception convolutions instead of a single convolution scale (ablation) | U-Net on ShapeNet · ShapeNet | instance mIoU (%): 83.1 -> 88.0; 82.5 -> 88.0 | Helped |
| Use U-Net skip connections and its architecture instead of the tested CNN alternatives (ablation) | ShapeNet image segmentation · ShapeNet | mIoU (%): 81.6 -> 88.0; 78.1 -> 88.0; 86.9 -> 88.0 | Helped |
| Replace Delaunay triangulation with KNN (K=20) (ablation) | Graph construction and graph drawing · ShapeNet | instance mIoU (%): 88.0 -> 87.1 | Hurt |
| Use 3D point distance instead of graph shortest-path distance (ablation) | Graph-to-grid layout objective · ShapeNet | instance mIoU (%): 88.0 -> 86.4 | Hurt |
| Reduce graph-drawing grid size from 16x16 to 10x10 (ablation) | Point-cloud image representation · ShapeNet | mIoU (%): 88.0 -> 82.4 | Hurt |
| Increase graph-drawing grid size from 16x16 to 24x24 (ablation) | Point-cloud image representation · ShapeNet | mIoU (%): 88.0 -> 87.5 | Hurt |
| Increase the number of clusters from 32 to 64 or 128 (ablation) | Hierarchical point-cloud representation · ShapeNet | mIoU (%): 88.0 -> 86.7; 88.0 -> 85.9 | Hurt |
| Increase the number of clusters from 32 to 64 or 128 (ablation) | Hierarchical point-cloud representation · ShapeNet | Time (ms): 821.2 -> 775.4; 821.2 -> 1164.5 | Mixed |
| Increase image grid size from 16x16 to 24x24 (ablation) | Point-cloud image representation · ShapeNet | Time (ms): 18.6 -> 63.3 | Hurt |
Closest papers by meaning
- Convolution in the Cloud: Learning Deformable Kernels in 3D Graph Convolution Networks for Point Cloud Analysis (CVPR 2020 · similarity 0.90)
- Learning Localized Generative Models for 3D Point Clouds via Graph Convolution (ICLR 2019 · similarity 0.90)
- Mining Point Cloud Local Structures by Kernel Correlation and Graph Pooling (CVPR 2018 · similarity 0.89)
- Point-SAM: Promptable 3D Segmentation Model for Point Clouds (ICLR 2025 · similarity 0.88)
- FreePoint: Unsupervised Point Cloud Instance Segmentation (CVPR 2024 · similarity 0.87)
- X-3D: Explicit 3D Structure Modeling for Point Cloud Recognition (CVPR 2024 · similarity 0.87)
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.