GA-DAN: Geometry-Aware Domain Adaptation Network for Scene Text Detection and Recognition

The contributions

  • 在geometry 和 appearance 两个维度上进行建模，消除domain shifts
  • 设计了一种新颖的多模态sptial learning mechanism同时引入了一种spatial transformation discriminator 去实现multi-modal adaptation in geometry space
  • 设计了一个disentangled cycle-consistency loss 可以平衡appearance和geometry 的一致性。

Method