Multimodal Facial Understanding on FaceFocalDesc (test)
74.38Accuracy (Cls)Focal-RegionFace
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Focal-RegionFaceJudge Model=GPT-4o, Evaluation Protocol=Global Ranking2026.01 | 74.38 | 83.86 | 84.72 | 81.33 | 79.51 | 57.71 | 1 | |
| Focal-RegionFaceJudge Model=Gemini-2.5-Pro, Evaluation Protocol=Global Ranking2026.01 | 70.46 | 82.91 | 93.83 | 91.81 | 74.7 | 67.56 | 1 | |
| Gemma3Judge Model=GPT-4o, Evaluation Protocol=Global Ranking2026.01 | 67.35 | 60.1 | 72.15 | 71.7 | 68.16 | 19.47 | 2 | |
| Qwen2.5-VLJudge Model=GPT-4o, Evaluation Protocol=Global Ranking2026.01 | 67.28 | 64.62 | 78.2 | 74.73 | 69.68 | 14.57 | 3 | |
| Llama3.2-VisionJudge Model=GPT-4o, Evaluation Protocol=Global Ranking2026.01 | 63.61 | 51.16 | 60.78 | 67.85 | 61.53 | 3.7 | 5 | |
| Gemma3Judge Model=Gemini-2.5-Pro, Evaluation Protocol=Global Ranking2026.01 | 59.04 | 47.69 | 71.4 | 76.53 | 58.01 | 12.4 | 3 | |
| Deepseek-Janus-ProJudge Model=GPT-4o, Evaluation Protocol=Global Ranking2026.01 | 55.2 | 39.91 | 73.41 | 68.19 | 55.98 | 7.55 | 4 | |
| Qwen2.5-VLJudge Model=Gemini-2.5-Pro, Evaluation Protocol=Global Ranking2026.01 | 52.69 | 47.35 | 74.49 | 73.22 | 51.88 | 13.51 | 2 | |
| Llama3.2-VisionJudge Model=Gemini-2.5-Pro, Evaluation Protocol=Global Ranking2026.01 | 51.01 | 33.18 | 74.33 | 68.7 | 45.96 | 4.87 | 4 | |
| Deepseek-Janus-ProJudge Model=Gemini-2.5-Pro, Evaluation Protocol=Global Ranking2026.01 | 44.33 | 13.76 | 79.83 | 80.11 | 43.78 | 1.66 | 5 |