UI Grounding on ScreenSpot Mobile
58.4RecallRT-DETRv2 + ScreenParse
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| RT-DETRv2 + ScreenParseSize=43M, Training Data=ScreenParse2026.02 | 58.4 | 73.6 | |
| OmniParser V2 + ScreenParseSize=20M, Training Data=ScreenParse2026.02 | 55.2 | 73.5 | |
| YOLO + ScreenParseSize=25M, Training Data=ScreenParse2026.02 | 50.4 | 68.2 | |
| OmniParser V2Size=20M2026.02 | 48.9 | 52.1 | |
| Qwen3-VL-8B-InstructSize=8B2026.02 | 19.3 | 31.1 | |
| Qwen3-VL-2B-InstructSize=2B2026.02 | 14.3 | 25 | |
| Qwen3-VL-2B-Instruct + ScreenParseSize=2B, Training Data=ScreenParse2026.02 | 10.8 | 47.7 | |
| InternVL3-2B + ScreenParseSize=2B, Training Data=ScreenParse2026.02 | 10 | 52.7 | |
| InternVL3-2BSize=2B2026.02 | 6.8 | 32.1 | |
| ScreenVLMSize=316M2026.02 | 6.6 | 84.7 |