Research map / Point clouds / Segmentation and detection
Open-World Semantic Segmentation for LIDAR Point Clouds
Jun Cen, Peng Yun, Shiwei Zhang, Junhao Cai, Di Luan, Mingqian Tang, Ming Liu, Michael Yu Wang
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 the calibration loss L_cal (second-largest score on unknown class) (ablation) | REAL on Cylinder3D · SemanticKITTI open-set (val) | AUPR: 0 -> 10.0 | Improved |
| add the calibration loss L_cal (ablation) | REAL on Cylinder3D · SemanticKITTI open-set (val) | AUROC: 0 -> 77.5 | Improved |
| add unknown object synthesis loss L_syn by resizing objects (ablation) | REAL on Cylinder3D · SemanticKITTI open-set (val) | AUPR: 10.0 -> 20.8 | Improved |
| add unknown object synthesis loss L_syn (ablation) | REAL on Cylinder3D · SemanticKITTI open-set (val) | AUROC: 77.5 -> 84.9 | Improved |
| add unknown object synthesis loss L_syn (ablation) | REAL on Cylinder3D · SemanticKITTI open-set (val) | mIoU_old: 58.1 -> 57.8 | No clear effect |
| use REAL instead of MC-Dropout for open-set scoring | Cylinder3D · nuScenes open-set (val) | AUPR: 14.9 -> 21.2 | Improved |
| use REAL instead of MSP for open-set scoring | Cylinder3D · SemanticKITTI open-set (val) | AUPR: 6.7 -> 20.8 | Improved |
| use REAL instead of MaxLogit for open-set scoring | Cylinder3D · SemanticKITTI open-set (val) | AUROC: 70.5 -> 84.9 | Improved |
| open-set finetuning with REAL on closed-set accuracy | Cylinder3D · nuScenes open-set (val) | mIoU_old: 58.7 -> 56.8 | Hurt |
| use pseudo labels from the open-set model on unlabeled old-class points in incremental learning | REAL IL · SemanticKITTI 18+1 incremental (val) | mIoU: 21.6 -> 64.3 | Improved |
| use pseudo labels instead of finetuning only on novel labels | REAL IL · SemanticKITTI 18+1 incremental (val) | mIoU_old: 0 -> 65.0 | Improved |
| use pseudo labels vs Feature extraction | REAL IL · nuScenes 12+4 incremental (val) | mIoU: 5.5 -> 74.9 | Improved |
Closest papers by meaning
- PolarNet: An Improved Grid Representation for Online LiDAR Point Clouds Semantic Segmentation (CVPR 2020 · similarity 0.91)
- Lidar Panoptic Segmentation in an Open World (IJCV 2024 · similarity 0.90)
- SFPNet: Sparse Focal Point Network for Semantic Segmentation on General LiDAR Point Clouds (ECCV 2024 · similarity 0.90)
- On-the-fly Category Discovery for LiDAR Semantic Segmentation (ECCV 2024 · similarity 0.90)
- LESS: Label-Efficient Semantic Segmentation for LiDAR Point Clouds (ECCV 2022 · similarity 0.89)
- PDF: A Probability-Driven Framework for Open World 3D Point Cloud Semantic Segmentation (CVPR 2024 · similarity 0.88)
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.
