Industrial Anomaly Detection on MMAD
95.24Anomaly DiscriminationHuman (expert)
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Human (expert)2026.01 | 95.24 | 75 | 92.31 | 83.33 | 94.2 | 86.11 | 80.37 | 86.65 | |
| GPT-4o2026.01 | 68.63 | 65.8 | 55.62 | 73.21 | 83.41 | 94.98 | 82.8 | 74.92 | |
| Reason-IADBackbone=Qwen3-VL, Scale=8B, zero-shot=true2026.02 | 67.73 | 73.12 | 59.74 | 74.69 | 80.65 | 96.78 | 84.24 | 76.71 | |
| MAU-GPTNumber of parameters=4B2026.01 | 67.5 | 39.31 | 41.24 | 66.37 | 77.04 | 67.09 | 71.31 | 61.41 | |
| Qwen3-VLScale=8B, zero-shot=true2026.02 | 67.1 | 57.98 | 58.86 | 66.89 | 77.65 | 92.94 | 83.82 | 72.18 | |
| Reason-IADBackbone=Qwen3-VL, Scale=4B, zero-shot=true2026.02 | 66.58 | 70.62 | 62.82 | 73.28 | 80.88 | 93.8 | 84.88 | 76.12 | |
| Qwen2.5-VLScale=72B, zero-shot=true2026.02 | 66.22 | 57.99 | 62.95 | 72.21 | 81.01 | 93.92 | 86.51 | 74.4 | |
| Qwen3-VLScale=4B, zero-shot=true2026.02 | 65.83 | 53.97 | 58.16 | 66.35 | 77.82 | 91.98 | 83.38 | 71.07 | |
| InternVL3Scale=8B, zero-shot=true2026.02 | 64.67 | 50.91 | 60.06 | 65.49 | 77.16 | 82.01 | 83.58 | 69.13 | |
| GPT-4o-mini2026.01 | 64.33 | 48.58 | 38.75 | 63.68 | 80.4 | 88.56 | 79.74 | 66.29 | |
| InternVL2Scale=76B, zero-shot=true2026.02 | 64.3 | 51.19 | 54.2 | 63.46 | 79.92 | 89.34 | 83.48 | 69.41 | |
| Qwen3-VL thinkingScale=4B, zero-shot=true2026.02 | 64.21 | 39.61 | 42.98 | 48.96 | 38.32 | 78.18 | 64.18 | 53.78 | |
| Qwen3-VLScale=2B, zero-shot=true2026.02 | 63.27 | 47.71 | 52.48 | 63.48 | 77.82 | 92.16 | 80.03 | 68.14 | |
| Qwen3-VL thinkingScale=8B, zero-shot=true2026.02 | 63.07 | 45.81 | 49.56 | 53.52 | 40.63 | 82.51 | 66.25 | 57.34 | |
| AnomalyGPTNumber of parameters=7B2026.01 | 61.53 | 31.39 | 27.2 | 30.76 | 22.24 | 44.48 | 30.23 | 35.41 | |
| Qwen2.5-VLScale=3B, zero-shot=true2026.02 | 60.63 | 47.03 | 54.44 | 63.51 | 79.25 | 87.48 | 82.02 | 67.77 | |
| Qwen2.5-VLScale=7B, zero-shot=true2026.02 | 60.35 | 49.89 | 57.51 | 60.41 | 75.58 | 93.39 | 84.8 | 68.85 | |
| LLaVA-NeXTScale=34B, zero-shot=true2026.02 | 60.31 | 51.4 | 55.47 | 71.62 | 80.49 | 81.45 | 75.08 | 67.97 | |
| InternVL2Number of parameters=34B2026.01 | 59.97 | 43.85 | 47.91 | 57.6 | 78.1 | 74.18 | 80.37 | 63.14 | |
| Reason-IADBackbone=Qwen2.5-VL, Scale=7B, zero-shot=true2026.02 | 59.23 | 61.02 | 55.39 | 66.57 | 78.37 | 97.36 | 85.5 | 71.78 | |
| LLaVA-NeXTNumber of parameters=34B2026.01 | 57.92 | 48.79 | 52.87 | 71.34 | 80.28 | 81.12 | 77.8 | 67.16 | |
| Gemma3Number of parameters=4B2026.01 | 57.37 | 37.07 | 35.71 | 47.49 | 76.32 | 85.11 | 73.65 | 58.96 | |
| Reason-IADBackbone=Qwen3-VL, Scale=2B, zero-shot=true2026.02 | 54.66 | 68.95 | 55.28 | 70.64 | 79.68 | 95.28 | 80.85 | 72.19 | |
| Qwen-VL-ChatNumber of parameters=7B2026.01 | 53.65 | 31.33 | 28.62 | 41.66 | 63.99 | 74.46 | 67.94 | 51.66 | |
| SPHINXNumber of parameters=7B2026.01 | 53.13 | 33.93 | 52.27 | 50.96 | 71.23 | 85.07 | 73.1 | 59.96 | |
| LLaVA-1.5Scale=7B, zero-shot=true2026.02 | 50.81 | 36.82 | 35.94 | 50.96 | 70.45 | 67.33 | 69.12 | 54.49 | |
| Random Chance2026.01 | 50 | 25 | 25 | 25 | 25 | 25 | 25 | 28.57 | |
| LLaVA-1.5Number of parameters=13B2026.01 | 49.96 | 38.78 | 46.17 | 58.17 | 73.09 | 73.62 | 70.98 | 58.68 | |
| Uio2-largeNumber of parameters=1.5B2026.01 | 49.27 | 27.52 | 18.95 | 25.28 | 37.85 | 43.75 | 47.24 | 36.08 | |
| Uio2-XlNumber of parameters=6B2026.01 | 24.52 | 32.23 | 36.61 | 36.99 | 48.64 | 57.85 | 59.21 | 38.62 |