Research map / Label-efficient and robust learning / OOD and anomaly detection
Hierarchical Point-Patch Fusion with Adaptive Patch Codebook for 3D Shape Anomaly Detection
Xueyang Kang, Zizhao Li, Tian Lan, Dong Gong, Kourosh Khoshelham, Liangliang Nan
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 pseudo-anomaly negative augmentation during training (ablation) | patch-point fusion model · Real3D-AD (object-level) | AUC-ROC (%): 68.2 -> 84.2 | Improved |
| add point-to-patch cross-attention fusion (ablation) | patch-point fusion model · Real3D-AD (object-level) | AUC-ROC (%): 72.6 -> 84.2 | Improved |
| use RoPE cross-attention instead of vanilla cross-attention (ablation) | patch-point fusion model · Real3D-AD (object-level) | AUC-ROC (%): 79.1 -> 84.2 | Improved |
| add patch score modulation of normal and anomalous features (ablation) | patch-point fusion model · Real3D-AD (object-level) | AUC-ROC (%): 80.3 -> 84.2 | Improved |
| select the max-similarity scale instead of fusing all scales (ablation) | patch codebook retrieval · Real3D-AD (object-level) | AUC-ROC (%): 81.2 -> 84.2 | Improved |
| use multi-scale sphere patches instead of semantic-part patches (ablation) | patchification · Real3D-AD (object-level) | AUC-ROC (%): 75.1 -> 84.2 | Improved |
| use multi-scale sphere patches instead of single-scale FPS spheres (ablation) | patchification · Real3D-AD (object-level) | AUC-ROC (%): 81.9 -> 84.2 | Improved |
| use multi-scale spheres instead of 3D grid patches (ablation) | patchification · Real3D-AD (object-level) | AUC-ROC (%): 80.3 -> 84.2 | Improved |
| use mean point coordinates queried by the encoder instead of direct feature pooling for patch features (ablation) | patch feature extraction · Real3D-AD (object-level) | AUC-ROC (%): 77.8 -> 79.4 | Improved |
| increase patch size beyond 64 (ablation) | patchification · Real3D-AD | AUC-ROC | Hurt |
| increase voxel resolution beyond 128^3 (ablation) | preprocessing · Real3D-AD | AUC-ROC | No clear effect |
| increase patch number beyond 128 (ablation) | patchification · Real3D-AD | AUC-ROC | No clear effect |
Closest papers by meaning
- Towards Scalable 3D Anomaly Detection and Localization: A Benchmark via 3D Anomaly Synthesis and A Self-Supervised Learning Network (CVPR 2024 · similarity 0.90)
- Towards Generalizable 3D Anomaly Detection via Relational Inconsistency Modeling (NeurIPS 2026 · similarity 0.90)
- Registration is a Powerful Rotation-Invariance Learner for 3D Anomaly Detection (NeurIPS 2025 · similarity 0.90)
- Looking 3D: Anomaly Detection with 2D-3D Alignment (CVPR 2024 · similarity 0.87)
- 3CAD: A Large-Scale Real-World 3C Product Dataset for Unsupervised Anomaly Detection (AAAI 2025 · similarity 0.86)
- Multimodal Industrial Anomaly Detection by Crossmodal Feature Mapping (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.
