Research map / Point clouds / Point cloud classification
Anomaly Factory 3D: A Modular Framework for Diverse Pseudo-Anomaly Synthesis in Unsupervised 3D Anomaly Detection
Ali Balapour, Faraz Hach
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 |
|---|---|---|---|
| replace Norm-AS with AF3AD synthesis in PO3AD | PO3AD · AnomalyShapeNet | O-AUROC: 83.9 -> 91.5 | Improved |
| replace Norm-AS with AF3AD synthesis in PO3AD | PO3AD · Real3D-AD | O-AUROC: 76.6 -> 85.2 | Improved |
| replace Patch-Gen with AF3AD in R3D-AD | R3D-AD · AnomalyShapeNet (15 classes) | O-AUROC: 62.7 -> 67.2 | Improved |
| R3D-AD with AF3AD versus released R3D-AD without Patch-Gen | R3D-AD · AnomalyShapeNet (15 classes) | O-AUROC: 58.7 -> 67.2 | Improved |
| replace Patch-Gen with AF3AD in R3D-AD | R3D-AD · AnomalyShapeNet (15 classes) | P-AUROC: 60.7 -> 60.7 | No clear effect |
| add multi-head offset prediction to AF3AD training (ablation) | PO3AD+AF3AD · AnomalyShapeNet (15 classes, h=128) | O-AUROC: 90.3 -> 97.0 | Improved |
| AF3AD instead of Norm-AS synthesis under multi-head h=128 (ablation) | PO3AD multi-head · AnomalyShapeNet (15 classes) | O-AUROC: 91.6 -> 97.0 | Improved |
| AF3AD instead of Norm-AS with weak baseline head h=32 (ablation) | PO3AD baseline head · AnomalyShapeNet (15 classes, h=32) | O-AUROC: 90.9 -> 88.4 | Hurt |
| combine preset groups A, B, C, D versus group A alone (ablation) | AF3AD · AnomalyShapeNet (15 classes, index-0 split) | O-AUROC: 62.5 -> 88.9 | Improved |
| add groups A, C, D to best single group B (ablation) | AF3AD · AnomalyShapeNet (15 classes, index-0 split) | O-AUROC: 74.0 -> 88.9 | Improved |
| add Gaussian noise sigma 0.001 to test data (ablation) | AF3AD · AnomalyShapeNet (15 classes) | O-AUROC: 93.3 -> 93.3 | No clear effect |
| add Gaussian noise sigma 0.003 to test data (ablation) | AF3AD · AnomalyShapeNet (15 classes) | P-AUPR: 59.0 -> 29.6 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- PO3AD — PO3AD: Predicting Point Offsets toward Better 3D Point Cloud Anomaly Detection (CVPR 2025) · better on 3 of 3 shared results
- R3D-AD — R3D-AD: Reconstruction via Diffusion for 3D Anomaly Detection (ECCV 2024) · better on 2 of 2 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
- Towards Scalable 3D Anomaly Detection and Localization: A Benchmark via 3D Anomaly Synthesis and A Self-Supervised Learning Network (CVPR 2024 · similarity 0.92)
- PO3AD: Predicting Point Offsets toward Better 3D Point Cloud Anomaly Detection (CVPR 2025 · similarity 0.92)
- Registration is a Powerful Rotation-Invariance Learner for 3D Anomaly Detection (NeurIPS 2025 · similarity 0.90)
- PointAD: Comprehending 3D Anomalies from Points and Pixels for Zero-shot 3D Anomaly Detection (NeurIPS 2024 · similarity 0.89)
- Towards Generalizable 3D Anomaly Detection via Relational Inconsistency Modeling (NeurIPS 2026 · similarity 0.89)
- Looking 3D: Anomaly Detection with 2D-3D Alignment (CVPR 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.
