I have finished the detection branch and am still training the model to verify its correctness. All the features will be published to develop branch, and keep master stable.

• ICDAR Dataset
• SynthText 800K Dataset
• detection branch
• eval
• multi-gpu training

Questions

• Should I fix weights of the backbone network, resnet50 ?

  for param in self.backbone.parameters():
      param.requires_grad = False

Introduction

This is a PyTorch implementation of FOTS.