arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2610.39664cs.CV
收藏

EditAgent:重新思考智能体图像编辑

EditAgent: Rethinking Agentic Image Editing

Adobe Research(奥多比研究院);UC San Diego(加利福尼亚大学圣迭戈分校)

Haiyang Xu, Tianyu Wang, Jialu Li, Jiteng Mu, Yuanjun Xiong, Jianming Zhang, Zhuowen Tu, Mingze Xu

AI总结:

EditAgent扩展智能体图像编辑的动作与观察空间,通过路由多类方法、并行生成候选并借助VLM评判,提升编辑性能并可作为训练数据引擎。

摘要:

智能体图像编辑器会将复杂请求分解为多个步骤,但当某一步失败时,它们大多会用相同的有限工具集重试,且评判器会直接通过视觉对每个输出打分。我们认为这类智能体受限于两个空间:智能体可执行的动作空间,以及其评判器可观测的观察空间。在我们经专家标注的基准测试中,通用编辑器在超过40%的案例中无法产出可用结果,且没有任何评判器能在各类编辑任务中保持可靠。我们提出EditAgent,它扩展了上述两个空间:它将每一步路由至通用编辑器与专用方法,并为每个候选结果附上验证记录。借助这些记录,现成的VLM(视觉语言模型)评判器的表现优于微调后的奖励模型。候选结果并行生成,并通过一次列表调用完成评判:在已确定的对比中,78.5%的专家更偏好这种方式而非顺序优化,且评判调用次数不超过后者的一半。EditAgent在三个基准测试中提升了两款开放权重编辑器的性能,并可作为训练更强编辑器的数据引擎。

英文摘要:

Agentic image editors decompose complex requests into steps, but when a step fails, they mostly retry with the same narrow set of tools and a judge that scores each output by eye directly. We argue that such agents are bounded by two spaces: the action space of what the agent can do, and the observation space of what its judge can see. On our expert-annotated benchmark, a general-purpose editor yields no usable result in over 40% of the cases, and no judge is reliable across edit types. We present EditAgent, which expands both spaces. It routes each step to general editors and specialized methods, and attaches to each candidate a verification record. With these records, off-the-shelf VLM judges outperform finetuned reward models. Candidates are generated in parallel and judged in one listwise call: experts prefer this to sequential refinement in 78.5% of decided comparisons, at no more than half the judge calls. EditAgent improves two open-weight editors on three benchmarks and serves as a data engine for training stronger editors.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑