Convolutional Recurrent Neural Network

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

Train a new model

1. Construct dataset and alphabeth following generate_lmdb_dataset.ipynb
2. Execute PYTHONIOENCODING=utf-8 python3.6 train.py --adadelta --trainRoot train_sample --valRoot test_sample --cuda --random_sample -alphabeth alphabeth. Explore train.py for details.

Run demo

Put the downloaded model file crnn.pth into directory data/. Then launch the demo by:

python demo.py

The demo reads an example image and recognizes its text content.

Example image:

Expected output: loading pretrained model from ./data/crnn.pth a-----v--a-i-l-a-bb-l-ee-- => available