CompBench：面向复杂指令引导图像编辑的基准
CompBench: Benchmarking Complex Instruction-guided Image Editing
做指令引导图像编辑的研究者可以直接用 CompBench 的细粒度指令与四维解耦（位置/外观/动态/物体）来诊断模型在复杂场景下的精确操控短板，也可借鉴其 MLLM-人工协同构建流程来造自己的评测数据。
原文依据 we introduce CompBench, a large-scale benchmark specifically designed for complex instruction-guided image editing
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论尽管现实应用日益需要复杂的场景操控，现有的指令引导图像编辑基准往往简化了任务复杂度，且缺乏全面、细粒度的指令。为弥补这一空白，我们提出 CompBench，一个专为复杂指令引导图像编辑设计的大规模基准。CompBench 包含具有挑战性的编辑场景，涵盖细粒度指令遵循、空间与上下文推理，从而能够全面评估图像编辑模型的精确操控能力。为构建 CompBench，我们提出了一个多模态大模型与人工协同的框架，并配有定制的任务流水线。此外，我们提出一种指令解耦策略，将编辑意图拆解为位置、外观、动态和物体四个关键维度，确保指令与复杂编辑需求之间更紧密的对齐。大量评估表明，CompBench 揭示了当前图像编辑模型的基本局限，并为下一代指令引导图像编辑系统的发展提供了关键洞见。项目主页见 https://comp-bench.github.io/。
英文原文摘要 1224 字符
While real-world applications increasingly demand intricate scene manipulation, existing instruction-guided image editing benchmarks often oversimplify task complexity and lack comprehensive, fine-grained instructions. To bridge this gap, we introduce CompBench, a large-scale benchmark specifically designed for complex instruction-guided image editing. CompBench features challenging editing scenarios that incorporate fine-grained instruction following, spatial and contextual reasoning, thereby enabling comprehensive evaluation of image editing models' precise manipulation capabilities. To construct CompBench, we propose an MLLM-human collaborative framework with tailored task pipelines. Furthermore, we propose an instruction decoupling strategy that disentangles editing intents into four key dimensions: location, appearance, dynamics, and objects, ensuring closer alignment between instructions and complex editing requirements. Extensive evaluations reveal that CompBench exposes fundamental limitations of current image editing models and provides critical insights for the development of next-generation instruction-guided image editing systems. Our project page is available at https://comp-bench.github.io/.