Document Visual Question Answering on DeepForm
50.8AccuracyVisual-SR1
Evaluation results
| Method | Links | |
|---|---|---|
| Visual-SR1Backbone=Qwen2.5-VL-7B2025.09 | 50.8 | |
| DeFactoBackbone=Qwen2.5-VL-7B2025.09 | 45.6 | |
| Qwen2.5-VLBackbone=Qwen2.5-VL-7B2025.09 | 27.9 | |
| ViCropBackbone=LLaVA-1.5 (Vicuna-7B)2025.09 | 20.3 | |
| GRITBackbone=Qwen2.5-VL-3B2025.09 | 17.9 |