Object-centric Typographic Robustness (Binary Choice) on Typo-D (Clean)
97.4Robustness (Character)IT-RT
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| IT-RTBackbone=Qwen2.5-VL-7B, Training Strategy=IT-RT2025.12 | 97.4 | 85.5 | 91.1 | 99.6 | 93.4 | |
| RIO-RTBackbone=Llama-3.2-11B-Vision, Training Strategy=RIO-RT2025.12 | 97.4 | 92.7 | 97.4 | 99.8 | 96.8 | |
| pretrainedBackbone=LLaVA-1.5-7B, Training Strategy=pretrained2025.12 | 96.4 | 90.5 | 94.7 | 97.9 | 94.9 | |
| CoTBackbone=Llama-3.2-11B-Vision, Training Strategy=CoT2025.12 | 95.9 | 92.1 | 94.2 | 99.4 | 95.4 | |
| CoTBackbone=Qwen2.5-VL-7B, Training Strategy=CoT2025.12 | 95.4 | 85.7 | 93.9 | 99.4 | 93.6 | |
| RIO-RTBackbone=Qwen2.5-VL-7B, Training Strategy=RIO-RT2025.12 | 95.4 | 88.7 | 94.2 | 99.8 | 94.5 | |
| IT-RTBackbone=Llama-3.2-11B-Vision, Training Strategy=IT-RT2025.12 | 95.4 | 89.5 | 95.3 | 97.5 | 94.4 | |
| CoTBackbone=LLaVA-1.5-13B, Training Strategy=CoT2025.12 | 93.8 | 87.7 | 92.1 | 98.7 | 93.1 | |
| pretrainedBackbone=Llama-3.2-11B-Vision, Training Strategy=pretrained2025.12 | 93.3 | 87.9 | 93.2 | 92.6 | 91.7 | |
| RIO-RTBackbone=LLaVA-1.5-7B, Training Strategy=RIO-RT2025.12 | 91.8 | 80.8 | 75.8 | 96.6 | 86.3 | |
| pretrainedBackbone=Qwen2.5-VL-7B, Training Strategy=pretrained2025.12 | 85.1 | 83 | 80.8 | 98.3 | 86.8 | |
| CoTBackbone=LLaVA-1.5-7B, Training Strategy=CoT2025.12 | 81 | 76 | 84.5 | 96.8 | 84.6 | |
| pretrainedBackbone=LLaVA-1.5-13B, Training Strategy=pretrained2025.12 | 81 | 88.3 | 80.3 | 98.1 | 86.9 | |
| RIO-RTBackbone=LLaVA-1.5-13B, Training Strategy=RIO-RT2025.12 | 79.5 | 86.3 | 78.7 | 98.9 | 85.8 | |
| IT-RTBackbone=LLaVA-1.5-13B, Training Strategy=IT-RT2025.12 | 71.3 | 84.2 | 60 | 98.7 | 78.6 | |
| IT-RTBackbone=LLaVA-1.5-7B, Training Strategy=IT-RT2025.12 | 55.4 | 51.1 | 47.9 | 28 | 45.6 |