Research map / Point clouds / Segmentation and detection
Counterfactual Occlusion-Aware Learning via Visibility Intervention for LiDAR Anomaly Detection
Longyu Yang, Jun Liu, Yap-Peng Tan, Fumin Shen, Heng Tao Shen, Xiaofeng Zhu, Ping Hu
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 physics-conformed anomaly construction (ablation) | MinkNet baseline · STU val (MinkNet) | AP: 0.91 -> 78.49 | Improved |
| add anomaly class logit scoring (ablation) | COVAL MinkNet · STU val (MinkNet) | AP: 78.49 -> 87.09 | Improved |
| add visibility-variant counterfactual reconstruction (ablation) | COVAL MinkNet · STU val (MinkNet) | AP: 87.09 -> 88.52 | Improved |
| add visibility-invariant counterfactual consistency (ablation) | COVAL MinkNet · STU val (MinkNet) | AP: 88.52 -> 89.00 | Improved |
| use physics-conformed intervention instead of random masking (ablation) | COVAL MinkNet · STU val (MinkNet) | AP: 87.64 -> 89.00 | Improved |
| vary reconstruction weight lambda_rec from 0.1 to 1.0 (no consistency) (ablation) | COVAL MinkNet · STU val (MinkNet) | AP: 88.77 -> 88.52 | No clear effect |
| raise reconstruction weight to 5.0 (ablation) | COVAL MinkNet · STU val (MinkNet) | AP: 88.52 -> 87.21 | Hurt |
| vary consistency weight lambda_cons from 1.0 to 2.0 (ablation) | COVAL MinkNet · STU val (MinkNet) | AP: 89.00 -> 88.48 | No clear effect |
| use only highest-resolution feature scale instead of all scales (ablation) | COVAL MinkNet · STU val (MinkNet) | FPR@95: 0.34 -> 0.13 | Improved |
| use all scales instead of top-4 scales (ablation) | COVAL MinkNet · STU val (MinkNet) | AP: 89.45 -> 89.00 | No clear effect |
| use all thing classes instead of a cube as synthetic anomaly (ablation) | COVAL MinkNet · STU val (MinkNet) | AP: 71.13 -> 87.09 | Improved |
| use larger inserted object size 0.1-0.3 instead of 0.5-0.7 (ablation) | COVAL MinkNet · STU val (MinkNet) | AP: 75.20 -> 87.09 | Improved |
Closest papers by meaning
- Learning to Identify Out-of-Distribution Objects for 3D LiDAR Anomaly Segmentation (CVPR 2026 · similarity 0.88)
- OccAM's Laser: Occlusion-Based Attribution Maps for 3D Object Detectors on LiDAR Data (CVPR 2022 · similarity 0.88)
- LiON: Learning Point-Wise Abstaining Penalty for LiDAR Outlier DetectioN Using Diverse Synthetic Data (AAAI 2025 · similarity 0.87)
- Spotting the Unexpected (STU): A 3D LiDAR Dataset for Anomaly Segmentation in Autonomous Driving (CVPR 2025 · similarity 0.86)
- MOS: Model Synergy for Test-Time Adaptation on LiDAR-Based 3D Object Detection (ICLR 2025 · similarity 0.85)
- An Empirical Study of the Generalization Ability of Lidar 3D Object Detectors to Unseen Domains (CVPR 2024 · 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.
