Industrial Anomaly Detection on DAGM (test)
97.48AccuracyGLLS
Evaluation Results
| Method | Links | |
|---|---|---|
| GLLSBackbone=Qwen3-VL, Scale=8B, Shots=1-shot2026.07 | 97.48 | |
| GLLSBackbone=Qwen3-VL, Scale=8B, Shots=0-shot2026.07 | 97.15 | |
| GLLSBackbone=Qwen3-VL, Scale=4B, Shots=1-shot2026.07 | 96.65 | |
| GLLSBackbone=Qwen3-VL, Scale=4B, Shots=0-shot2026.07 | 96.32 | |
| GLLSBackbone=Qwen2.5-VL, Scale=7B, Shots=1-shot2026.07 | 96.12 | |
| AD-FMScale=7B, Backbone=Qwen2.5-VL2026.07 | 95.51 | |
| GLLSBackbone=Qwen2.5-VL, Scale=7B, Shots=0-shot2026.07 | 95.43 | |
| AnomalyClipThresholding Strategy=Youden’s J statistic2026.07 | 95.36 | |
| Qwen2.5-VLScale=7B2026.07 | 94.83 | |
| IAD-R1Scale=7B, Backbone=LLaVA-OV2026.07 | 94.8 | |
| InternVL2Scale=76B2026.07 | 94.52 | |
| AdaptCLIPThresholding Strategy=Youden’s J statistic2026.07 | 94.37 | |
| GPT4o-mini2026.07 | 90.8 | |
| GPT4o2026.07 | 90.17 | |
| GPT-4.12026.07 | 89.93 | |
| InternVL3Scale=38B2026.07 | 89.87 | |
| WinClipThresholding Strategy=Youden’s J statistic2026.07 | 88.9 | |
| WinClipThresholding Strategy=τ = 0.52026.07 | 86.9 | |
| AdaptCLIPThresholding Strategy=τ = 0.52026.07 | 82.17 | |
| InternVL3Scale=8B2026.07 | 81.33 | |
| LLaVA-NeXTScale=34B2026.07 | 23.15 | |
| AnomalyClipThresholding Strategy=τ = 0.52026.07 | 22.19 |