UI Grounding on ScreenSpot PC
59RecallRT-DETRv2 + ScreenParse
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| RT-DETRv2 + ScreenParseSize=43M, Training Data=ScreenParse2026.02 | 59 | 69.9 | |
| OmniParser V2 + ScreenParseSize=20M, Training Data=ScreenParse2026.02 | 53.6 | 62.4 | |
| YOLO + ScreenParseSize=25M, Training Data=ScreenParse2026.02 | 52.1 | 59.7 | |
| OmniParser V2Size=20M2026.02 | 48.3 | 55.7 | |
| ScreenVLMSize=316M2026.02 | 22.2 | 83.9 | |
| Qwen3-VL-8B-InstructSize=8B2026.02 | 20.1 | 30 | |
| Qwen3-VL-2B-Instruct + ScreenParseSize=2B, Training Data=ScreenParse2026.02 | 20.1 | 44.3 | |
| Qwen3-VL-2B-InstructSize=2B2026.02 | 17.1 | 21.8 | |
| InternVL3-2B + ScreenParseSize=2B, Training Data=ScreenParse2026.02 | 8.1 | 62.8 | |
| InternVL3-2BSize=2B2026.02 | 1.8 | 10.9 |