Research map / Point clouds / Segmentation and detection
Novel Class Discovery for 3D Point Cloud Semantic Segmentation
Read the paperCheck 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 |
|---|---|---|---|
| Use NOPS instead of EUMS† | MinkowskiUNet-34C for novel class discovery · SemanticPOSS, average across four splits | novel mIoU: 14.94 -> 21.40 | Helped |
| Use NOPS instead of EUMS† | MinkowskiUNet-34C for novel class discovery · SemanticKITTI, average across four splits | novel mIoU: 17.05 -> 22.84 | Helped |
| Use NOPS instead of EUMS† | MinkowskiUNet-34C for novel class discovery · POSS-4⁰ | novel mIoU: 17.44 -> 35.70 | Helped |
| Use NOPS instead of EUMS† | MinkowskiUNet-34C for novel class discovery · POSS-3³ | novel mIoU: 13.01 -> 10.94 | Hurt |
| Train from scratch rather than pre-train on base classes (ablation) | NOPS on SemanticPOSS · SemanticPOSS | novel mIoU | Helped |
| Add the class-feature queue (ablation) | NOPS on SemanticPOSS · SemanticPOSS | novel mIoU | Helped |
| Add feature selection for the queue and training (ablation) | NOPS on SemanticPOSS · SemanticPOSS | novel mIoU: 20.26 -> 20.63 | Helped |
| Apply adaptive thresholding to features used to derive pseudo-labels (ablation) | NOPS on SemanticPOSS · SemanticPOSS | novel mIoU: 20.63 -> 20.90 | Helped |
| Set feature-selection percentile p to 0.3 rather than 0.1 (ablation) | NOPS · SemanticPOSS, average across four splits | novel mIoU: 19.44 -> 21.40 | Helped |
| Set feature-selection percentile p to 0.9 rather than 0.3 (ablation) | NOPS · POSS-3³ | novel mIoU: 10.94 -> 14.38 | Helped |
Closest papers by meaning
- Novel Class Discovery for Point Cloud Segmentation via Joint Learning of Causal Representation and Reasoning (NeurIPS 2025 · similarity 0.92)
- LogoSP: Local-global Grouping of Superpoints for Unsupervised Semantic Segmentation of 3D Point Clouds (CVPR 2025 · similarity 0.89)
- PDF: A Probability-Driven Framework for Open World 3D Point Cloud Semantic Segmentation (CVPR 2024 · similarity 0.88)
- Rethinking Few-shot 3D Point Cloud Semantic Segmentation (CVPR 2024 · similarity 0.88)
- Weakly Supervised Point Cloud Semantic Segmentation via Artificial Oracle (CVPR 2024 · similarity 0.88)
- Generated and Pseudo Content guided Prototype Refinement for Few-shot Point Cloud Segmentation (NeurIPS 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.