Pixel-level Anomaly Detection on Real-IAD
99.1AUROCUniADet
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| UniADetLanguage-free=true, Few-shot=2-shot, Backbone=DINOv2-R ViT-L/162026.01 | 99.1 | 46.7 | |
| UniADetReference Image=Used, Shots=1, Backbone=DINOv2 ViT-L/142026.01 | 98.9 | 48.4 | |
| UniADetLanguage-free=true, Few-shot=2-shot, Backbone=DINOv3 ViT-L/162026.01 | 98.9 | 47 | |
| UniADetReference Image=Used, Shots=1, Backbone=DINOv3 (ViT-L/16)2026.01 | 98.6 | 47.6 | |
| UniADetLanguage-free=true, Few-shot=2-shot, Backbone=CLIP ViT-L/14@336px2026.01 | 98.2 | 36.4 | |
| UniADetReference Image=Used, Shots=1, Backbone=CLIP ViT-L/14@336px2026.01 | 98 | 35.3 | |
| UniADetEvaluation Protocol=Zero-shot, Language-free=true, Backbone=DINOv2-R ViT-L/162026.01 | 97.7 | 43.1 | |
| UniADetEvaluation Protocol=Zero-shot, Language-free=true, Backbone=DINOv3 ViT-L/162026.01 | 97.5 | 41.6 | |
| AdaptCLIPLanguage-free=false, Few-shot=2-shot2026.01 | 97.3 | 37.8 | |
| AdaptCLIPReference Image=Not Used, Shots=02026.01 | 97.1 | 36.6 | |
| AnomalyCLIP+Language-free=false, Few-shot=2-shot2026.01 | 96.6 | 28.1 | |
| UniADetEvaluation Protocol=Zero-shot, Language-free=true, Backbone=CLIP ViT-L/14@336px2026.01 | 96.6 | 33.6 | |
| AnomalyCLIP+Reference Image=Not Used, Shots=02026.01 | 96.5 | 27.9 | |
| Bayes-PFLEvaluation Protocol=Zero-shot, Language-free=false, Backbone=CLIP ViT-L/14@336px2026.01 | 96.5 | 27.6 | |
| AdaCLIPEvaluation Protocol=Zero-shot, Language-free=false, Backbone=CLIP ViT-L/14@336px2026.01 | 96.1 | 30.5 | |
| InCtrlLanguage-free=false, Few-shot=2-shot2026.01 | 96 | 20.1 | |
| InCtrlReference Image=Not Used, Shots=02026.01 | 95.4 | 19.1 | |
| WinCLIP+Language-free=false, Few-shot=2-shot2026.01 | 95.3 | 14.8 | |
| AnomalyCLIPEvaluation Protocol=Zero-shot, Language-free=false, Backbone=CLIP ViT-L/14@336px2026.01 | 95.1 | 26.7 | |
| WinCLIP+Reference Image=Not Used, Shots=02026.01 | 95 | 13.9 | |
| AdaptCLIPEvaluation Protocol=Zero-shot, Language-free=false, Backbone=CLIP ViT-L/14@336px2026.01 | 94.9 | 28.2 | |
| WinCLIPEvaluation Protocol=Zero-shot, Language-free=false, Backbone=CLIP ViT-L/14@336px2026.01 | 84.5 | 3.3 |