Visual Question Answering on SV-QA HRBench-4K
71.3Q1 AccuracyDeepEyes
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DeepEyesReasoning Paradigm=Text + Visual Cropping2026.05 | 71.3 | 79.3 | 60.6 | |
| SLVR-7BReasoning Paradigm=Visual + Semantic Latent2026.05 | 70.4 | 82.8 | 61.1 | |
| LVRReasoning Paradigm=Visual Latent2026.05 | 70.1 | 80.4 | 57.9 | |
| Visual-SR1Reasoning Paradigm=Text CoT2026.05 | 63.9 | 76.4 | 54.6 | |
| Qwen2.5-VL-7BReasoning Paradigm=Text-only2026.05 | 60.5 | 74.1 | 45.8 | |
| GRITReasoning Paradigm=Text CoT2026.05 | 58.4 | 70.4 | 48.6 |