Research map / Point clouds / Point cloud classification
Multimodal Industrial Anomaly Detection via Hybrid Fusion
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use hybrid fusion with UFF, DLF, and RGB, 3D, and fused memory banks | M3DM · MVTec-3D AD, RGB + 3D | I-AUROC: 0.937 -> 0.945 | Helped |
| Add UFF patch-wise contrastive fusion (ablation) | M3DM with a fused single memory bank · MVTec-3D AD | I-AUROC: 0.857 -> 0.898 | Helped |
| Add UFF patch-wise contrastive fusion (ablation) | M3DM with a fused single memory bank · MVTec-3D AD | AUPRO: 0.944 -> 0.956 | Helped |
| Add DLF decision layer fusion (ablation) | M3DM with separate RGB and 3D memory banks · MVTec-3D AD | I-AUROC: 0.929 -> 0.932 | Helped |
| Add DLF decision layer fusion (ablation) | M3DM with separate RGB and 3D memory banks · MVTec-3D AD | AUPRO: 0.953 -> 0.959 | Helped |
| Add a fused-feature memory bank alongside the RGB and 3D banks (ablation) | M3DM with DLF · MVTec-3D AD | I-AUROC: 0.932 -> 0.945 | Helped |
| Add a fused-feature memory bank alongside the RGB and 3D banks (ablation) | M3DM with DLF · MVTec-3D AD | AUPRO: 0.959 -> 0.964 | Helped |
| Use 128 points per group and 1024 groups instead of 64 points and 784 groups (ablation) | Point Transformer with point-group sampling · MVTec-3D AD, 3D | I-AUROC: 0.793 -> 0.841 | Helped |
| Project point features to smaller 4×4 patches rather than retaining point-group features (ablation) | Point Transformer with PFA · MVTec-3D AD, 3D | I-AUROC: 0.841 -> 0.874 | Helped |
| Use 5-shot training rather than the full training dataset (ablation) | M3DM · MVTec-3D AD | AUPRO: 0.964 -> 0.927 | Hurt |
| Use 10-shot training rather than the full training dataset (ablation) | M3DM · MVTec-3D AD | AUPRO: 0.964 -> 0.939 | Hurt |
| Replace supervised ImageNet pretraining with DINO self-supervised pretraining (ablation) | RGB ViT feature extractor · MVTec-3D AD, RGB | I-AUROC: 0.670 -> 0.850 | Helped |
Closest papers by meaning
- Multimodal Industrial Anomaly Detection by Crossmodal Feature Mapping (CVPR 2024 · similarity 0.92)
- Real-IAD D3: A Real-World 2D/Pseudo-3D/3D Dataset for Industrial Anomaly Detection (CVPR 2025 · similarity 0.89)
- Revisiting Multimodal Fusion for 3D Anomaly Detection from an Architectural Perspective (AAAI 2025 · similarity 0.86)
- 3CAD: A Large-Scale Real-World 3C Product Dataset for Unsupervised Anomaly Detection (AAAI 2025 · similarity 0.85)
- Rethinking Reverse Distillation for Multi-Modal Anomaly Detection (AAAI 2024 · similarity 0.84)
- Registration is a Powerful Rotation-Invariance Learner for 3D Anomaly Detection (NeurIPS 2025 · similarity 0.84)
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.