描述而非命令：基于自然语言意图的语义图像编辑
Describe, Don't Dictate: Semantic Image Editing with Natural Language Intent
AI总结：
本文提出 DescriptiveEdit，将指令式图像编辑转化为参考图像条件下的文本到图像生成，并通过 Cross-Attentive UNet 注入参考特征，在 Emu Edit 上提升编辑准确性与一致性。
摘要：
尽管文本到图像生成已取得进展，语义图像编辑仍是一项挑战。基于反演的算法不可避免地引入重建误差，而基于指令的模型主要受限于数据集质量和规模。为解决这些问题，我们提出一种基于描述性提示的编辑框架，名为 DescriptiveEdit。其核心思想是将“基于指令的图像编辑”重新构建为“基于参考图像的文本到图像生成”，从而在不修改架构或进行反演的情况下，保留训练良好的 Text-to-Image 模型的生成能力。具体而言，该方法以参考图像和提示为输入，引入 Cross-Attentive UNet，通过新增注意力桥接，将参考图像特征注入提示到编辑图像的生成过程。得益于其文本到图像的本质，DescriptiveEdit 克服了指令数据集质量方面的限制，可与 ControlNet、IP-Adapter 及其他扩展无缝集成，并具有更强的可扩展性。在 Emu Edit 基准上的实验表明，该方法提升了编辑准确性和一致性。
英文摘要：
Despite the progress in text-to-image generation, semantic image editing remains a challenge. Inversion-based algorithms unavoidably introduce reconstruction errors, while instruction-based models mainly suffer from limited dataset quality and scale. To address these problems, we propose a descriptive-prompt-based editing framework, named DescriptiveEdit. The core idea is to re-frame `instruction-based image editing' as `reference-image-based text-to-image generation', which preserves the generative power of well-trained Text-to-Image models without architectural modifications or inversion. Specifically, taking the reference image and a prompt as input, we introduce a Cross-Attentive UNet, which newly adds attention bridges to inject reference image features into the prompt-to-edit-image generation process. Owing to its text-to-image nature, DescriptiveEdit overcomes limitations in instruction dataset quality, integrates seamlessly with ControlNet, IP-Adapter, and other extensions, and is more scalable. Experiments on the Emu Edit benchmark show it improves editing accuracy and consistency.
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。
