HuggingFace Daily Papers（社区热门论文）· HuggingFace Daily Papers（社区热门论文）·· 8 天前AI 评分38
MT-OPSD：面向多轮图像编辑的在线策略自蒸馏框架
AI 导读
针对现有图像编辑模型在递归编辑中快速退化的问题，研究者提出 MT-OPSD 在线策略自蒸馏框架，让模型在自生成的条件下状态上训练，并由干净条件教师模型提供编辑监督，无需多轮标注。该工作同时发布 LME-Bench，包含 100 组十轮编辑会话，用于评估长程鲁棒性。在三个编辑骨干上的实验显示，MT-OPSD 显著提升长程编辑成功率、减少多轮崩溃，并基本保持单轮编辑质量。
正文
Abstract:Instruction-based image editing has achieved strong performance in single-turn settings, yet practical editing is often iterative, with each instruction applied to the output of the previous turn. We find that existing editing models degrade rapidly under recursive editing and attribute this failure to a train-test mismatch in the conditioning distribution: models are trained on clean source images but must repeatedly condition on their own imperfect outputs at inference time. To address this, we propose MT-OPSD, an on-policy self-distillation framework that trains the model on self-generated conditioning states with editing supervision from a clean-conditioned teacher, without requiring multi-turn annotations. We further introduce LME-Bench, a benchmark of 100 ten-turn editing sessions for evaluating long-horizon robustness. Experiments across three editing backbones show that MT-OPSD substantially improves long-horizon editing success and reduces multi-turn collapse while largely preserving single-turn editing quality.
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG) |
| Cite as: | arXiv:2609.35611 [cs.CV] |
| (or arXiv:2609.35611v1 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2609.35611 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Liangbing Zhao [view email]
[v1]
Mon, 28 Sep 2026 16:55:24 UTC (41,129 KB)
来源：HuggingFace Daily Papers（社区热门论文） · arxiv.org