1 paper · 1 filter
Gangwei Jiang, Shiyao Wang, Tiezheng Ge +3
Recent efforts on scene text erasing have shown promising results. However, existing methods require rich yet costly label annotations to obtain robust models, which limits the use…