Anomaly Detection on MVTec AD (I-AUROC, P-AUROC, P-AUPRO)
0.993P-AUROCGLASS-j
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GLASS-jThroughput (img/s)=13272024.07 | 0.993 | 0.999 | 0.968 | |
| GLASS-mThroughput (img/s)=13272024.07 | 0.991 | 0.997 | 0.964 | |
| GLASS-hThroughput (img/s)=13272024.07 | 0.99 | 0.998 | 0.959 | |
| Dinomaly ViT-Bbackbone=ViT-B2026.03 | 0.984 | 0.996 | 0.948 | |
| RD++Throughput (img/s)=6232024.07 | 0.983 | 0.994 | 0.95 | |
| RD++2026.03 | 0.983 | 0.994 | 0.95 | |
| Final Modeldescription=Incorporates FMAS and WDAM2026.03 | 0.983 | 0.99 | 0.977 | |
| BGADThroughput (img/s)=2062024.07 | 0.982 | 0.979 | 0.963 | |
| CFABackbone=WideResNet-502026.02 | 0.982 | 0.992 | — | |
| StructCore (Base)Backbone=DINOv2 ViT-B/14 (skip-layer), Coreset ratio=1%2026.02 | 0.982 | 0.987 | — | |
| THFR2026.03 | 0.982 | 0.992 | 0.95 | |
| PatchCore Ensemble2026.03 | 0.982 | 0.995 | 0.976 | |
| PatchCoreThroughput (img/s)=312024.07 | 0.981 | 0.991 | 0.928 | |
| SimpleNetThroughput (img/s)=13062024.07 | 0.981 | 0.995 | 0.9 | |
| PatchCore (max pooling)Backbone=WideResNet-50, Coreset ratio=1%2026.02 | 0.981 | 0.989 | — | |
| PatchCore + StructCore (struct only)Backbone=WideResNet-50, Coreset ratio=1%2026.02 | 0.981 | 0.991 | — | |
| PatchCore + StructCore (hybrid)Backbone=WideResNet-50, Coreset ratio=1%2026.02 | 0.981 | 0.992 | — | |
| StructCore (struct only)Backbone=DINOv2 ViT-B/14 (skip-layer), Coreset ratio=1%2026.02 | 0.981 | 0.996 | — | |
| StructCore (hybrid, auto-λ)Backbone=DINOv2 ViT-B/14 (skip-layer), Coreset ratio=1%2026.02 | 0.981 | 0.996 | — | |
| PatchCore2026.03 | 0.981 | 0.991 | 0.974 | |
| IDEALFew-shot setup=N8A4, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.978 | 0.986 | — | |
| LeMO-FProtocol=Online evaluation, Variant=F2023.05 | 0.976 | 0.972 | 0.917 | |
| CFAProtocol=Online evaluation2023.05 | 0.975 | 0.966 | 0.912 | |
| IDEALFew-shot setup=N2A1, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.975 | 0.974 | — | |
| IDEALFew-shot setup=N4A1, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.975 | 0.982 | — | |
| PaDiMBackbone=WideResNet-502026.02 | 0.973 | 0.955 | — | |
| LeMO-LProtocol=Online evaluation, Variant=L2023.05 | 0.971 | 0.967 | 0.909 | |
| NAGLFew-shot setup=N8A4, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.971 | 0.972 | — | |
| NAGLFew-shot setup=N4A1, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.97 | 0.969 | — | |
| IDEALFew-shot setup=N1A1, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.968 | 0.963 | — | |
| FastFlowProtocol=Online evaluation2023.05 | 0.967 | 0.927 | 0.896 | |
| NAGLFew-shot setup=N2A1, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.966 | 0.96 | — | |
| NAGLFew-shot setup=N1A1, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.961 | 0.951 | — | |
| DSRThroughput (img/s)=5822024.07 | 0.958 | 0.982 | 0.917 | |
| RegADFew-shot setup=N8, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.958 | 0.95 | — | |
| RegADFew-shot setup=N4, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.951 | 0.94 | — | |
| PromptADFew-shot setup=N8, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.948 | 0.96 | — | |
| PromptADFew-shot setup=N4, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.946 | 0.949 | — | |
| AHLFew-shot setup=A4, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.945 | 0.929 | — | |
| InCTRLFew-shot setup=N8, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.944 | 0.958 | — | |
| WinCLIPFew-shot setup=N8, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.942 | 0.955 | — | |
| InCTRLFew-shot setup=N4, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.94 | 0.943 | — | |
| DRAFew-shot setup=A4, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.94 | 0.937 | — | |
| PromptADFew-shot setup=N2, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.938 | 0.935 | — | |
| WinCLIPFew-shot setup=N4, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.938 | 0.939 | — | |
| InCTRLFew-shot setup=N2, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.937 | 0.927 | — | |
| AHLFew-shot setup=A1, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.936 | 0.917 | — | |
| WinCLIPFew-shot setup=N2, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.936 | 0.934 | — | |
| RegADFew-shot setup=N2, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.935 | 0.926 | — | |
| FoundADFew-shot setup=N8, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.935 | 0.932 | — | |
| PatchCoreProtocol=Online evaluation2023.05 | 0.934 | 0.872 | 0.929 | |
| PromptADFew-shot setup=N1, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.934 | 0.928 | — | |
| WinCLIPFew-shot setup=N1, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.934 | 0.931 | — | |
| InCTRLFew-shot setup=N1, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.934 | 0.925 | — | |
| FoundADFew-shot setup=N4, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.932 | 0.929 | — | |
| RegADFew-shot setup=N1, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.931 | 0.916 | — | |
| DRAFew-shot setup=A1, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.929 | 0.922 | — | |
| FoundADFew-shot setup=N2, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.928 | 0.903 | — | |
| AA-CLIPAvailable training shots=full2025.03 | 0.919 | — | — | |
| AA-CLIPAvailable training shots=642025.03 | 0.916 | — | — | |
| AA-CLIPAvailable training shots=162025.03 | 0.912 | — | — | |
| AnomalyCLIPSource=ICLR 2024, Available training shots=full2025.03 | 0.911 | — | — | |
| AA-CLIPAvailable training shots=22025.03 | 0.91 | — | — | |
| FoundADFew-shot setup=N1, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.907 | 0.899 | — | |
| AdaCLIPSource=ECCV 2024, Available training shots=full2025.03 | 0.899 | — | — | |
| ResADFew-shot setup=N8, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.878 | 0.894 | — | |
| VANDSource=CVPRW 2023, Available training shots=full2025.03 | 0.876 | — | — | |
| ResADFew-shot setup=N4, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.869 | 0.871 | — | |
| ResADFew-shot setup=N2, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.863 | 0.857 | — | |
| WinCLIPSource=CVPR 20232025.03 | 0.851 | — | — | |
| MVFA-ADSource=CVPR 2024, Available training shots=full2025.03 | 0.849 | — | — | |
| ResADFew-shot setup=N1, Backbone=ViT-S/14, Resolution=448 x 4482026.05 | 0.843 | 0.807 | — | |
| DRAEMProtocol=Online evaluation2023.05 | 0.73 | 0.791 | 0.462 | |
| CLIPSource=OpenCLIP2025.03 | 0.384 | — | — |