scene-text-recog

Scene character recognition combining of CTPN + CRNN.

code based on bear63/sceneReco commit-ae1731e

CTPN:

Caffe model in frcnn, adapt from tianzhi0549/CTPN of old Caffe.

CRNN:

Thsi repo uses Convolutional recurrent network in PyTorch: meijieru/crnn.pytorch, while the original implementation bgshih/crnn uses Torch.

trained models:

1. pure English set alphabet to English in crnn.pytorch/keys.py.
   CPU model crnn.pth
2. Chinese with English recognition : netCRNN63.pth for GPU, netCRNNcpu.pth for CPU.

copy ctpn_trained_model.caffemodel to ./models/CTPN

copy CRNN model to ./models, CPU model can be used for both CPU/GPU.

CTPN models are tested on Caffe 1.0 CUDA8.0, CRNN models are tested on PyTorch ‘0.4.0’ and ‘0.3.1’.

Run demo

python demo.py