S2VOPD-Qwen3.5-4B

S²VOPD-Qwen3.5-4B

Official model release for Self-Supervised Visual On-Policy Distillation.

Paper · Project page · Code

A fine-grained visual understanding model built on Qwen/Qwen3.5-4B and trained with on-policy self-distillation -- no reward model, no human annotation, and no larger teacher.

Results

Fine-grained perception

Greedy decoding, 4096 tokens, seed 42.

Mathematical reasoning

24576 tokens, T=0.3, top-p 0.95, top-k 20, presence penalty 1.5, seed 42; judge = Qwen2.5-72B-Instruct.

The long generation budget is required for the math suite: at 16k, 8-11% of generations still hit the cap, so math numbers produced under a greedy/4096 protocol are not comparable to these.

Method

On-policy self-distillation. The student generates on a degraded view of an image while an EMA teacher scores the same prefix on the clean original. The disagreement between the two is the entire training signal -- it is large exactly on the tokens that depend on fine visual detail, and near zero elsewhere. The loss is a generalized JSD. No reward model, no human annotation, and no larger teacher.

Augmentation

Every training image receives exactly one degradation, drawn from three families at 30% each; the remaining 10% is left untouched.

One degradation per image is deliberate. Stacking them sequentially destroys the image, and once the student can no longer recover the content the teacher-student gap stops carrying information about visual detail.

Training data

Vision-OPD-6K -- 6,241 samples.

Usage

from transformers import AutoProcessor, AutoModelForImageTextToText
import torch

model = AutoModelForImageTextToText.from_pretrained(
    "s2vopd/S2VOPD-Qwen3.5-4B", dtype=torch.bfloat16,
    attn_implementation="flash_attention_2", device_map="cuda")
processor = AutoProcessor.from_pretrained("s2vopd/S2VOPD-Qwen3.5-4B")

Citation

@misc{li2026selfsupervisedvisualonpolicydistillation,
  title={Self-Supervised Visual On-Policy Distillation},
  author={Yijiang Li and Yijun Liang and Yunjie Tian and Bingyang Wang and Ke Zhang and Zhenfei Yin and Di Fu and Philip Torr and Nuno Vasconcelos},
  year={2026},
  eprint={2608.14144},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2608.14144},
}