Visual Reasoning Robustness on VisArgs specific masking 1.0 (test)
58.78Answer Flip RateGemini-1.5-Flash
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Gemini-1.5-FlashMasking Type=Specific Masking2025.12 | 58.78 | 79.08 | 30.6 | |
| GPT-4oMasking Type=Specific Masking2025.12 | 74.74 | 92.86 | 35.51 | |
| Llama-3.2-90B-Vision-InstructMasking Type=Specific Masking2025.12 | 75.72 | 93.73 | 49.97 | |
| Qwen-2.5-VL-7B-InstructMasking Type=Specific Masking2025.12 | 85.72 | 95.59 | 26.57 |