OPD-V-Qwen3-VL-4B-Instruct

OPD-V

This model checkpoint is part of OPD-V, presented in the paper OPD-V: Self-Distillation Improves Visual Detail Perception in Vision-Language Models.

Code & Repository: GitHub - OPD-V
Base Model: Qwen/Qwen3-VL-4B-Instruct
Paper: OPD-V: Self-Distillation Improves Visual Detail Perception in Vision-Language Models

Overview

OPD-V is a visual On-Policy Self-Distillation (OPSD) framework for improving multimodal reasoning under Modality Imbalance. Instead of relying on a single privileged teacher, OPD-V contrasts a Positive Teacher conditioned on a Zoom-In Image with a Negative Teacher conditioned on a Mask Image, and distills only the on-policy tokens selected by the resulting Modality-Balance Trust Region.

This checkpoint corresponds to OPD-V-Qwen3-VL-4B-Instruct trained with teacher-box positive images and random-mask negative images.

Citation

@misc{opdv2026,
  title={OPD-V: Self-Distillation Improves Visual Detail Perception in Vision-Language Models},
  author={},
  year={2026},
  eprint={2608.05131},
  archivePrefix={arXiv},
  primaryClass={cs.CV}
}