OPRO：面向面板感知上下文图像生成的正交面板相对算子
OPRO: Orthogonal Panel-Relative Operators for Panel-Aware In-Context Image Generation
把可学习的面板相对正交算子叠加到冻结的位置编码上，在保持等距性与同面板不变性的同时实现面板感知的上下文条件注入，可直接复用于基于扩散 Transformer 的图像编辑与多面板布局生成管线。
原文依据 compose learnable, panel-specific orthogonal operators onto the backbone's frozen positional encodings
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论本文提出一种参数高效的适配方法，用于基于预训练扩散 Transformer 的面板感知上下文图像生成。其核心思想是在骨干网络冻结的位置编码上叠加可学习的、面板专属的正交算子。该设计带来两个理想性质：(1) 等距性，即保持内部特征的几何结构；(2) 同面板不变性，即维持模型预训练时已有的面板内合成行为。通过受控实验，我们证明该适配方法的有效性并不依赖于特定的位置编码设计，而是可泛化到多种位置编码机制。通过实现有效的面板相对条件注入，所提方法能够稳定提升上下文图像指令编辑流程，包括最优的现有方法。
英文原文摘要 871 字符
We introduce a parameter-efficient adaptation method for panel-aware in-context image generation with pre-trained diffusion transformers. The key idea is to compose learnable, panel-specific orthogonal operators onto the backbone's frozen positional encodings. This design provides two desirable properties: (1) isometry, which preserves the geometry of internal features, and (2) same-panel invariance, which maintains the model's pre-trained intra-panel synthesis behavior. Through controlled experiments, we demonstrate that the effectiveness of our adaptation method is not tied to a specific positional encoding design but generalizes across diverse positional encoding regimes. By enabling effective panel-relative conditioning, the proposed method consistently improves in-context image-based instructional editing pipelines, including state-of-the-art approaches.