Mask TextSpotter 结构
- a feature pyramid network (FPN) as backbone.
- a region proposal network (RPN) for generating text proposals.
- a Fast R-CNN for bounding boxes regression.
- a mask branch for text instance segmentation, character segmentation, and text sequence recognition.
数据流动
训练阶段, 由 RPN 产生文本建议(text proposals), 然后文本建议的 RoI 特征喂给 Fast R-CNN, 并且 mask branch 生成精准的文本候选框(candidate boxes), 文本实例分割图(text instance segmentation maps), 字符分割图(character segmentation maps), 和文本序列(text sequence)