---
title: "视觉对比自蒸馏：无需外部教师、特权答案或视觉证据的在线自蒸馏方法"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.21556v1"
permalink: "https://feed.astralor.com/items/itm_dd2199f2d68391496fa993af"
published: "2026-07-23T17:37:37.000Z"
score: 78
channel: "ai"
tags: ["Qwen3-VL","ViRL39K","多模态大模型","自蒸馏","视觉对比学习"]
---

# 视觉对比自蒸馏：无需外部教师、特权答案或视觉证据的在线自蒸馏方法

## AI 点评

> 这篇论文巧妙地将输入条件化差异转化为蒸馏信号，彻底去掉了传统自蒸馏中需要的外部教师和特权答案，让自蒸馏变得更‘纯粹’。对于正在训练多模态模型的工程师和研究者，VCSD提供了一个即插即用、效果显著的训练技巧——尤其适合算力有限但希望提升视觉理解质量的团队。我推荐重点关注其对比锐化机制的实际效果。

## 速览

本文提出视觉对比自蒸馏（VCSD），一种简化版在线自蒸馏方法，通过输入条件化生成不对称信号：EMA教师在同一提示下分别基于原始图像和内容擦除控制生成两种分布，取token级对数概率差来突出视觉内容带来的增益，并用此对比来锐化教师分布，蒸馏给学生。在ViRL39K数据集上，VCSD在Qwen3-VL和Qwen3.5模型上一致超越匹配的在线自蒸馏基线，例如Qwen3-VL 2B模型在7个基准上从62.27%提升至67.04%。VCSD无需外部教师、特权答案、视觉证据、推理轨迹或额外推理开销。

## 关键点

- 提出视觉对比自蒸馏（VCSD），通过输入条件化（原始图像 vs 内容擦除）产生不对称信号，无需外部教师或特权答案
- 利用EMA教师生成两种分布，取token级对数概率差来识别视觉内容特有的增益，并用于锐化教师原始图像分布
- 在ViRL39K数据集上，VCSD显著提升Qwen3-VL和Qwen3.5系列模型在7个基准上的性能，最高提升达4.75个百分点
- VCSD无需额外推理开销，不依赖推理轨迹或视觉证据信号，训练流程更简洁

## 全文

On-policy self-distillation (OPSD) is promising as it removes the external teacher required by on-policy distillation (OPD), yet it still needs asymmetric information between teacher and student to ensure that the self-teacher provides a stronger learning signal than the student. Existing methods create this asymmetry either through privileged answers or visual evidence. We ask whether both can be removed, yielding a simpler form of OPSD driven purely by input conditioning. For this purpose, we propose Visual Contrastive Self-Distillation, namely VCSD, which converts image-content removal into an on-policy self-distillation signal. At each student-generated response prefix, the EMA teacher produces two next-token distributions under the same prompt and prefix -- one conditioned on the original image and the other on a content-erased control. Their token-wise log-probability difference highlights candidates whose likelihood is specifically increased by the instance-level visual content. We use this contrast to sharpen the teacher's original-image distribution within its plausible support, and distill the resulting full-distribution target into the student. Using ViRL39K dataset, VCSD consistently outperforms matched OPSD across Qwen3-VL and Qwen3.5 models. For example, on Qwen3-VL, it improves the seven-benchmark aggregate from $62.27\% \rightarrow 67.04\%$ at 2B, $71.30\% \rightarrow 73.16\%$ at 4B, and $72.51\% \rightarrow 76.26\%$ at 8B. Furthermore, VCSD requires no external teacher, privileged answers, visual evidence signals, reasoning traces, or additional inference-time cost.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_dd2199f2d68391496fa993af