Visual Question Answering on SV-QA HRBench-8K
65.1Accuracy (Q1)DeepEyes
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DeepEyesReasoning Paradigm=Text + Visual Cropping2026.05 | 65.1 | 72.4 | 49 | |
| LVRReasoning Paradigm=Visual Latent2026.05 | 62.9 | 71.1 | 46.6 | |
| SLVR-7BReasoning Paradigm=Visual + Semantic Latent2026.05 | 62.5 | 73.6 | 50.6 | |
| Visual-SR1Reasoning Paradigm=Text CoT2026.05 | 56.3 | 67.4 | 44.4 | |
| Qwen2.5-VL-7BReasoning Paradigm=Text-only2026.05 | 53.5 | 66.1 | 37.5 | |
| GRITReasoning Paradigm=Text CoT2026.05 | 49.1 | 61.9 | 38.6 |