crnn-tensorflow

This is fork from this repository.

This software implements the Convolutional Recurrent Neural Network (CRNN) in tensorflow.Origin software could be found in crnn

run demo

A demo program can be found in inference.sh. Before running the demo,you should change paths. The demo reads an example image and recognizes its text content.

Train a new model

1. Create .txt file with

path_to_jpg_image path_to_txt_file_iwth_label_for_image

contents(see dataset/train_list.txt example). Each image contains 1 word and text file with label contains 1 string.

2. Use dataset/create_dataset.py to convert your dataset to tfrecord (dont forget change alphabet)

3. Change model parameters in train.py

4. Train your model with train.py

5. Use export_model.sh to freeze your graph.