Research map / Label-efficient and robust learning / OOD and anomaly detection
G2SF: Geometry-Guided Score Fusion for Multimodal Industrial Anomaly Detection
Chengyu Tao, Xuanming Cao, Juan Du
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 consistency loss L_cns to separation loss (ablation) | G2SF · MVTec-3D AD | I-AUROC: 0.868 -> 0.950 | Improved |
| add margin loss L_mar (ablation) | G2SF · MVTec-3D AD | AUPRO@30%: 0.972 -> 0.975 | Improved |
| add margin loss L_mar (ablation) | G2SF · MVTec-3D AD | I-AUROC: 0.950 -> 0.950 | No clear effect |
| add scaling loss L_sc (ablation) | G2SF · MVTec-3D AD | I-AUROC: 0.950 -> 0.961 | Improved |
| add cross-modal alignment loss L_cma (ablation) | G2SF · MVTec-3D AD | I-AUROC: 0.961 -> 0.971 | Improved |
| use min aggregation over adjacent local spaces instead of single space l_i,0 (ablation) | G2SF · MVTec-3D AD | AUPRO@1%: 0.461 -> 0.468 | Improved |
| use min instead of max aggregation (ablation) | G2SF · MVTec-3D AD | I-AUROC: 0.953 -> 0.971 | Improved |
| use min instead of mean aggregation (ablation) | G2SF · MVTec-3D AD | AUPRO@30%: 0.955 -> 0.979 | Improved |
| use min instead of mean aggregation (ablation) | G2SF · MVTec-3D AD | I-AUROC: 0.972 -> 0.971 | No clear effect |
| learn global scaling factors sigma instead of fixed 0.5 (ablation) | G2SF · Eyecandies | I-AUROC: 0.827 -> 0.902 | Improved |
| learn global scaling factors sigma instead of fixed 0.5 (ablation) | G2SF · MVTec-3D AD | I-AUROC: 0.964 -> 0.971 | Improved |
| fuse scaling factors and Euclidean distances (final score s_i) instead of unimodal image score s^R (ablation) | G2SF · MVTec-3D AD | I-AUROC: 0.849 -> 0.971 | Improved |
Later papers that compared with this method
1 later accepted paper reports G2SF as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Complementary Prototype Mapping for Efficient Multimodal Anomaly Detection (CVPR 2026 · better on 1 of 1 shared result)
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
- Multimodal Industrial Anomaly Detection by Crossmodal Feature Mapping (CVPR 2024 · similarity 0.88)
- Towards Generalizable 3D Anomaly Detection via Relational Inconsistency Modeling (NeurIPS 2026 · similarity 0.87)
- Real-IAD D3: A Real-World 2D/Pseudo-3D/3D Dataset for Industrial Anomaly Detection (CVPR 2025 · similarity 0.86)
- Revisiting Multimodal Fusion for 3D Anomaly Detection from an Architectural Perspective (AAAI 2025 · similarity 0.85)
- Registration is a Powerful Rotation-Invariance Learner for 3D Anomaly Detection (NeurIPS 2025 · similarity 0.85)
- Multi-Sensor Object Anomaly Detection: Unifying Appearance, Geometry, and Internal Properties (CVPR 2025 · 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.
