Task Detection on ProAct-75 1.0
28.33mAccProAct-Helper
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| ProAct-HelperVariant=HBM2026.02 | 28.33 | 36.72 | 51.03 | 67.58 | |
| ProAct-HelperVariant=HBM, Backbone=Qwen2.5-VL-7B-Instruct2026.02 | 27.07 | 34.79 | 52.91 | 69.2 | |
| ProAct-HelperVariant=plain, Backbone=Qwen2.5-VL-7B-Instruct2026.02 | 25.49 | 34.24 | 48.94 | 65.71 | |
| ProAct-HelperVariant=plain2026.02 | 24.02 | 31.85 | 48.21 | 65.05 | |
| Gemini-2.5-ProCategory=Closed-Source MLLMs2026.02 | 22.95 | 32.39 | 35.23 | 52.11 | |
| Qwen3-VL-30B-A3B-InstructCategory=Open-Source MLLMs2026.02 | 22.8 | 31.7 | 34.66 | 51.48 | |
| Gemini-2.5-FlashCategory=Closed-Source MLLMs2026.02 | 21.78 | 30.66 | 31.14 | 47.49 | |
| Qwen3-Omni-30BCategory=Open-Source MLLMs2026.02 | 18.96 | 26.87 | 17.55 | 29.87 | |
| Qwen3-VL-FlashCategory=Closed-Source MLLMs2026.02 | 17.36 | 24.84 | 17.91 | 30.37 | |
| GPT-4oCategory=Closed-Source MLLMs2026.02 | 16.67 | 24.9 | 20.27 | 33.71 | |
| Qwen2.5-VL-32B-InstructCategory=Open-Source MLLMs2026.02 | 16.62 | 23.96 | 29.45 | 45.5 | |
| Qwen3-VL-PlusCategory=Closed-Source MLLMs2026.02 | 14.88 | 21.52 | 17.39 | 29.63 | |
| Qwen2.5-VL-3B-InstructCategory=Open-Source MLLMs2026.02 | 12.53 | 18.92 | 23.18 | 37.64 |