Dense Parsing on ScreenParse (test)
60.6Page IoUScreenVLM
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ScreenVLMSize=316M, Fine-tuned=true2026.02 | 60.6 | 19.7 | 30.3 | |
| RT-DETRv2Size=43M, Fine-tuned=true, Fine-tuned on=ScreenParse2026.02 | 60 | 17.2 | 36.2 | |
| Qwen3-VL-2B-InstructSize=2B, Fine-tuned=true, Fine-tuned on=ScreenParse2026.02 | 58.5 | 16.6 | 15.2 | |
| YOLOSize=25M, Fine-tuned=true, Fine-tuned on=ScreenParse2026.02 | 53.3 | 13.3 | 29.9 | |
| InternVL3-2BSize=2B, Fine-tuned=true, Fine-tuned on=ScreenParse2026.02 | 50.9 | 17.4 | 7.2 | |
| OmniParser V2Size=20M, Fine-tuned=true, Fine-tuned on=ScreenParse2026.02 | 50.3 | 14.1 | 25.1 | |
| Qwen3-VL-8B-InstructSize=8B, Fine-tuned=false2026.02 | 29.4 | — | — | |
| OmniParser V2Size=20M, Fine-tuned=false2026.02 | 27 | — | — | |
| Qwen3-VL-2B-InstructSize=2B, Fine-tuned=false2026.02 | 22.8 | 5.1 | 2.3 | |
| InternVL3-2BSize=2B, Fine-tuned=false2026.02 | 11.1 | 3 | 0 |