GEditBench v2：一个对齐人类的通用图像编辑基准
GEditBench v2: A Human-Aligned Benchmark for General Image Editing
AI总结：
本文提出GEditBench v2，包含1200个真实用户查询，涵盖23项任务，引入PVC-Judge模型评估视觉一致性，实验表明其在开放源模型中表现优异，揭示当前模型局限性。
摘要：
近期图像编辑技术的进步使模型能处理复杂指令并产生逼真结果。然而，现有评估框架滞后：现有基准任务覆盖狭窄，标准指标无法充分捕捉视觉一致性，即编辑与原始图像之间身份、结构和语义一致性。为解决这些限制，我们引入GEditBench v2，包含1200个真实用户查询，涵盖23项任务，包括一个专门的开放集类别，用于处理超出预定义任务的非约束编辑指令。此外，我们提出PVC-Judge，一个开源的成对评估模型，通过两种新的区域解耦偏好数据合成管道进行训练。此外，我们构建VCReward-Bench，使用专家标注的偏好对来评估PVC-Judge与人类判断在视觉一致性评估上的对齐程度。实验表明，我们的PVC-Judge在开放源模型中实现了最先进的评估性能，甚至在平均上超过了GPT-5.1。最后，通过基准测试16个前沿编辑模型，我们证明GEditBench v2能够实现更符合人类的评估，揭示当前模型的关键限制，并提供一个可靠的基座，以推进精确图像编辑技术的发展。
英文摘要：
Recent advances in image editing have enabled models to handle complex instructions with impressive realism. However, existing evaluation frameworks lag behind: current benchmarks suffer from narrow task coverage, while standard metrics fail to adequately capture visual consistency, i.e., the preservation of identity, structure and semantic coherence between edited and original images. To address these limitations, we introduce GEditBench v2, a comprehensive benchmark with 1,200 real-world user queries spanning 23 tasks, including a dedicated open-set category for unconstrained, out-of-distribution editing instructions beyond predefined tasks. Furthermore, we propose PVC-Judge, an open-source pairwise assessment model for visual consistency, trained via two novel region-decoupled preference data synthesis pipelines. Besides, we construct VCReward-Bench using expert-annotated preference pairs to assess the alignment of PVC-Judge with human judgments on visual consistency evaluation. Experiments show that our PVC-Judge achieves state-of-the-art evaluation performance among open-source models and even surpasses GPT-5.1 on average. Finally, by benchmarking 16 frontier editing models, we show that GEditBench v2 enables more human-aligned evaluation, revealing critical limitations of current models, and providing a reliable foundation for advancing precise image editing.
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。
