从规模到速度：面向图像编辑的自适应测试时扩展
From Scale to Speed: Adaptive Test-Time Scaling for Image Editing
把测试时扩展（Image-CoT）从 T2I 迁移到图像编辑时，可用「编辑难度估计→动态采样预算」+「区域定位与描述一致性做早期剪枝」+「实例级验证器驱动的深度优先提前停止」三件套，在 Step1X-Edit、BAGEL、FLUX.1 Kontext 上以同等采样预算换 2 倍以上加速，是可直接复用的推理期加速与筛选框架。
原文依据 a difficulty-aware resource allocation that assigns dynamic budgets based on estimated edit difficulty
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论图像思维链（Image Chain-of-Thought，Image-CoT）是一种通过延长推理时间来提升图像生成的测试时扩展范式。多数 Image-CoT 方法聚焦于文本到图像（T2I）生成。与 T2I 生成不同，图像编辑是目标导向的：其解空间受源图像与指令约束。这种不匹配使得将 Image-CoT 应用于编辑时出现三个挑战：固定采样预算导致的资源分配低效、使用通用 MLLM 分数进行早期验证不可靠、以及大规模采样带来的冗余编辑结果。为此，我们提出 ADaptive Edit-CoT（ADE-CoT），一个按需的测试时扩展框架，用以提升编辑的效率与性能。它包含三项关键策略：（1）难度感知的资源分配，依据估计的编辑难度分配动态预算；（2）早期剪枝中的编辑专用验证，利用区域定位与描述一致性来挑选有前景的候选；（3）由实例特定验证器引导的深度优先机会式停止，在找到与意图对齐的结果时即终止。在三个基准上对三个 SOTA 编辑模型（Step1X-Edit、BAGEL、FLUX.1 Kontext）的大量实验表明，ADE-CoT 取得了更优的性能—效率权衡。在相当的采样预算下，ADE-CoT 获得更好的性能，且相比 Best-of-N 实现超过 2 倍的加速。
英文原文摘要 1429 字符
Image Chain-of-Thought (Image-CoT) is a test-time scaling paradigm that improves image generation by extending inference time. Most Image-CoT methods focus on text-to-image (T2I) generation. Unlike T2I generation, image editing is goal-directed: the solution space is constrained by the source image and instruction. This mismatch causes three challenges when applying Image-CoT to editing: inefficient resource allocation with fixed sampling budgets, unreliable early-stage verification using general MLLM scores, and redundant edited results from large-scale sampling. To address this, we propose ADaptive Edit-CoT (ADE-CoT), an on-demand test-time scaling framework to enhance editing efficiency and performance. It incorporates three key strategies: (1) a difficulty-aware resource allocation that assigns dynamic budgets based on estimated edit difficulty; (2) edit-specific verification in early pruning that uses region localization and caption consistency to select promising candidates; and (3) depth-first opportunistic stopping, guided by an instance-specific verifier, that terminates when intent-aligned results are found. Extensive experiments on three SOTA editing models (Step1X-Edit, BAGEL, FLUX.1 Kontext) across three benchmarks show that ADE-CoT achieves superior performance-efficiency trade-offs. With comparable sampling budgets, ADE-CoT obtains better performance with more than 2x speedup over Best-of-N.