V-Zero is an answer-label-free framework designed to improve fine-grained visual reasoning in multimodal large language models (MLLMs). It bypasses the need for costly external answer labels or manual verification rules by utilizing on-policy distillation combined with contrastive evidence gating. During training, the student model samples trajectories on the full image, while a teacher model replays those trajectories under paired positive (task-relevant) and negative (task-irrelevant) crops to evaluate student-sampled reasoning paths.
1@article{sun2026vzero,
2 title={V-Zero: Answer-Label-Free On-Policy Distillation with Contrastive Evidence Gating for Fine-Grained Visual Reasoning},
3 author={Sun, Haoxiang and Yi, Zhihang and Deng, Langxuan and Zhou, Yuhao and Jia, Peiqi and Zhao, Jian and Yuan, Li and Lv, Jiancheng and Wang, Tao},
4 journal={arXiv preprint arXiv:2606.25319},
5 year={2026}
6}
