Document Visual Question Answering on KLC
38.9AccuracyDeFacto
Evaluation results
| Method | Links | |
|---|---|---|
| DeFactoBackbone=Qwen2.5-VL-7B2025.09 | 38.9 | |
| Visual-SR1Backbone=Qwen2.5-VL-7B2025.09 | 36.6 | |
| Qwen2.5-VLBackbone=Qwen2.5-VL-7B2025.09 | 36.5 | |
| DeepEyesBackbone=Qwen2.5-VL-7B2025.09 | 32.2 | |
| ViCropBackbone=LLaVA-1.5 (Vicuna-7B)2025.09 | 29.1 | |
| GRITBackbone=Qwen2.5-VL-3B2025.09 | 17.8 |