UI Grounding on ScreenSpot Web
76.8RecallRT-DETRv2 + ScreenParse
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| RT-DETRv2 + ScreenParseSize=43M, Training Data=ScreenParse2026.02 | 76.8 | 85.7 | |
| OmniParser V2 + ScreenParseSize=20M, Training Data=ScreenParse2026.02 | 72.7 | 80 | |
| YOLO + ScreenParseSize=25M, Training Data=ScreenParse2026.02 | 68.3 | 79.8 | |
| ScreenVLMSize=316M2026.02 | 55.7 | 74.6 | |
| OmniParser V2Size=20M2026.02 | 54.1 | 62.9 | |
| Qwen3-VL-2B-Instruct + ScreenParseSize=2B, Training Data=ScreenParse2026.02 | 47.7 | 72 | |
| Qwen3-VL-2B-InstructSize=2B2026.02 | 23.2 | 29.2 | |
| Qwen3-VL-8B-InstructSize=8B2026.02 | 22.9 | 34.6 | |
| InternVL3-2B + ScreenParseSize=2B, Training Data=ScreenParse2026.02 | 17.2 | 59.2 | |
| InternVL3-2BSize=2B2026.02 | 0.2 | 5.7 |