arXiv cs.AI· Kangjie Chen, Xiangyu Li, Jinhao Deng, Dongbin Zhang, Chaoda Zheng, Shijia Chen, Yuqi Zhang, Choo Sin Wai, Boyi Sun, Puyi Wang, Yu Zhang, Xianming Liu, Boyang Wang·· 10 小时前AI 评分31
H3Remover:无需配对数据强化视频物体移除
H3Remover: Reinforcing Video Object Removal without Paired Data
AI 导读
H3Remover 通过强化 MiniMax-H3 Base Ref2VA 实现无需配对无物体训练目标的视频物体移除。该方法用目标移除奖励、背景几何奖励和外观奖励约束编辑行为,并以冻结的 HPSv3 提供感知质量反馈,DiffusionART 在冻结主干下优化 LoRA 适配器。实验显示其重建效果较原始 H3 显著提升,定性对比中物体移除更有效且周围场景内容保留更好。
正文 · 原文
Authors:Kangjie Chen, Xiangyu Li, Jinhao Deng, Dongbin Zhang, Chaoda Zheng, Shijia Chen, Yuqi Zhang, Choo Sin Wai, Boyi Sun, Puyi Wang, Yu Zhang, Xianming Liu, Boyang Wang
Abstract:Pretrained video editors possess rich removal and completion priors, yet may retain targets or alter unrelated content. We present H3Remover, which reinforces MiniMax-H3 Base Ref2VA for video object removal without paired object-free training targets. To discourage superficial edits such as recoloring, a target-removal reward favors calibrated depth changes consistent with revealing farther content. To limit unintended scene changes, a background-geometry reward encourages preservation of relative depth ordering. To preserve background appearance without retaining unwanted shadows or reflections, an appearance reward constrains RGB changes in source-derived protected regions while exempting targets and estimated effects. Frozen HPSv3 complements these constraints with perceptual-quality feedback for synthesized content. Evaluation regions remain fixed across candidates, and DiffusionART optimizes LoRA adapters with the backbone frozen. Experiments show substantial reconstruction gains over original H3, with qualitative comparisons demonstrating more effective object removal and better preservation of surrounding scene content.
| Comments: | Project page: this https URL , code at: this https URL |
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2611.24633 [cs.CV] |
| (or arXiv:2611.24633v1 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2611.24633 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Kangjie Chen [view email]
[v1]
Wed, 18 Nov 2026 15:45:18 UTC (28,749 KB)
来源:arXiv cs.AI · arxiv.org