Research map / Image generation / Autoregressive and latent generation
Geometry-Aligned and Anomaly-Aware Reconstruction for 3D Anomaly Detection
Linchun Wu, Qin Zou, Yuanhao Yue, Zhongyuan 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 geometry noise rectification (ablation) | AARD baseline · Anomaly-ShapeNet | O-AUROC: 0.772 -> 0.802 | Improved |
| add vertex noise rectification (ablation) | AARD baseline · Anomaly-ShapeNet | O-AUROC: 0.772 -> 0.796 | Improved |
| combine geometry and vertex noise rectification (ablation) | AARD baseline · Anomaly-ShapeNet | F-Score: 0.719 -> 0.832 | Improved |
| add normal reference and anomaly mask on top of noise rectification (ablation) | AARD · Anomaly-ShapeNet | O-AUROC: 0.825 -> 0.915 | Improved |
| add input reference and anomaly mask on top of noise rectification (ablation) | AARD · Anomaly-ShapeNet | O-AUROC: 0.825 -> 0.893 | Improved |
| add input reference to normal reference and mask (full model) (ablation) | AARD · Anomaly-ShapeNet | O-AUROC: 0.915 -> 0.931 | Improved |
| add normal reference to input reference and mask (ablation) | AARD · Anomaly-ShapeNet | F-Score: 0.947 -> 0.953 | No clear effect |
| add NR, IR and AM to noise rectification (ablation) | AARD · Anomaly-ShapeNet | F-Score: 0.832 -> 0.953 | Improved |
| vary grid number g from 4 to 2 or 10 (ablation) | geometry rectification · Anomaly-ShapeNet | O-AUROC: 0.931 -> 0.857 (g=2); 0.931 -> 0.898 (g=10) | Hurt |
| vary vertex group size k from 16 to 8 or 40 (ablation) | vertex rectification · Anomaly-ShapeNet | O-AUROC: 0.931 -> 0.891 (k=8); 0.931 -> 0.886 (k=40) | Hurt |
| vary gamma from 0.5 to 0.1 or 0.9 (ablation) | vertex rectification · Anomaly-ShapeNet | O-AUROC: 0.931 -> 0.889 (0.1); 0.931 -> 0.908 (0.9) | Hurt |
| AARD versus PASDF on Car and Fish categories | AARD · Real3D-AD | O-AUROC: Car 0.725 vs 0.959; Fish 0.891 vs 0.989 (PASDF) | Mixed |
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 0.91)
- Registration is a Powerful Rotation-Invariance Learner for 3D Anomaly Detection (NeurIPS 2025 · similarity 0.89)
- Towards Generalizable 3D Anomaly Detection via Relational Inconsistency Modeling (NeurIPS 2026 · similarity 0.88)
- FAST: Foreground‑aware Diffusion with Accelerated Sampling Trajectory for Segmentation‑oriented Anomaly Synthesis (NeurIPS 2025 · similarity 0.87)
- Looking 3D: Anomaly Detection with 2D-3D Alignment (CVPR 2024 · similarity 0.87)
- AnomalyDiffusion: Few-Shot Anomaly Image Generation with Diffusion Model (AAAI 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.
