Visual Reasoning on VStar (Accuracy)
80.11VStar AccuracyProLaViT (Full)
Evaluation Results
| Method | Links | |
|---|---|---|
| ProLaViT (Full)Reasoning Paradigm=Progressive Latent Visual Thought, Reasoning Space=Latent, Training Strategy=Distance-Weighted Diversity Loss2026.07 | 80.11 | |
| CoVTReasoning Paradigm=Latent-Space Reasoning, Reasoning Space=Latent2026.07 | 79.05 | |
| ProLaViT (Base)Reasoning Paradigm=Progressive Latent Visual Thought, Reasoning Space=Latent2026.07 | 79.05 | |
| ProLaViT + L_div^marginReasoning Paradigm=Progressive Latent Visual Thought, Reasoning Space=Latent2026.07 | 79.05 | |
| SFTReasoning Paradigm=Text-Space Reasoning, Reasoning Space=Text2026.07 | 78.01 | |
| One-step Latent Pred.Reasoning Paradigm=Latent-Space Reasoning, Reasoning Space=Latent2026.07 | 78.01 | |
| Qwen2.5-VL-InstructReasoning Paradigm=Base Model, Reasoning Space=N/A2026.07 | 76.44 | |
| LVRReasoning Paradigm=Latent-Space Reasoning, Reasoning Space=Latent2026.07 | 76.43 | |
| CoT SFTReasoning Paradigm=Text-Space Reasoning, Reasoning Space=Text2026.07 | 75.91 |