Real In, Real Out: What If We Only Use Real Data for Scene Text Editing?

One-line takeaway

Fudan studies scene text editing trained only on real data: instead of synthetic paired data and complex disentanglement, it disrupts each real image (crop, shuffle, mask and more) so the sample provides its own paired supervision.

复旦大学研究场景文字编辑能否只用真实数据训练：不再依赖合成配对数据和复杂的解耦设计，而是对真实图片做裁剪、打乱、遮挡等破坏，让每张真实样本自己构成配对监督。