Research map / Point clouds / Segmentation and detection
PrimitiveNet: Primitive Instance Segmentation With Local Primitive Embedding Under Adversarial Metric
Jingwei Huang, Yanfeng Zhang, Mingwei Sun
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 |
|---|---|---|---|
| use local point-pair supervision instead of global random pairs (ablation) | PrimitiveNet · ABC | AP: 25.9% -> 57.7% | Improved |
| use local point-pair supervision instead of semi-local pairs (ablation) | PrimitiveNet · ABC | AP: 38.7% -> 57.7% | Improved |
| use local point-pair supervision instead of global random pairs (ablation) | PrimitiveNet · ABC | seg mIOU: 73.6% -> 85.7% | Improved |
| use local point-pair supervision instead of global random pairs (ablation) | PrimitiveNet · Self-collected real scenes | seg mIOU: 37.0% -> 68.3% | Improved |
| use PointNet plus SpConv high-resolution backbone instead of SpConv alone (ablation) | PrimitiveNet · ABC | AP: 53.1% -> 57.7% | Improved |
| use PointNet plus SpConv backbone instead of SPVCNN (ablation) | PrimitiveNet · ABC | AP: 50.5% -> 57.7% | Improved |
| use PointNet plus SpConv backbone instead of PointNet++ (ablation) | PrimitiveNet · ABC | AP: 12.7% -> 57.7% | Improved |
| use PointNet plus SpConv backbone instead of SpConv alone (ablation) | PrimitiveNet · ABC | label mIOU: 91.8% -> 91.3% | No clear effect |
| use adversarial metric instead of Euclidean distance (ablation) | PrimitiveNet · ABC | Difference IOU: 71.1% -> 88.8% | Improved |
| use adversarial metric instead of Mahalanobis distance (ablation) | PrimitiveNet · ABC | Difference IOU: 69.8% -> 88.8% | Improved |
| add explicit tangent space prediction (ablation) | PrimitiveNet · ABC with Gaussian noise | AP | Improved |
| use PrimitiveNet segmentation in place of region growing [51] for scan abstraction | Scan abstraction pipeline [43] · Scene A | triangles: 163k -> 52k | Improved |
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.
- PointGroup — PointGroup: Dual-Set Point Grouping for 3D Instance Segmentation (CVPR 2020) · better on 6 of 6 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
- HPNet: Deep Primitive Segmentation Using Hybrid Representations (ICCV 2021 · similarity 0.89)
- FreePoint: Unsupervised Point Cloud Instance Segmentation (CVPR 2024 · similarity 0.87)
- Instance-Aware Embedding for Point Cloud Instance Segmentation (ECCV 2020 · similarity 0.87)
- InSeGAN: A Generative Approach to Segmenting Identical Instances in Depth Images (ICCV 2021 · similarity 0.87)
- CPFN: Cascaded Primitive Fitting Networks for High-Resolution Point Clouds (ICCV 2021 · similarity 0.87)
- PrimitiveUDF: Primitive-Based Unsigned Distance Fields for Surface Reconstruction from Point Clouds (ECCV 2026 · 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.
