CRNN

tensorflow 1.8

You can use TextRecognitionDataGenerator to generate text image.

Before building your vocab, you should specify charset in vocab.py, then,

python vocab.py \
--operation build --output_dir datasets --name vocab

Convert images to tf record file, you can find the script in my another Repo.

Specify DEFAULT_CONFIG in dataset file (eg. datasets/text.py).

cd python
bash run_train.sh

watch curve of loss:

tensorboard --logdir=your_train_logdir --port=your_port

Run eval repeatedly, it will check your ckpt-dir automatically

bash run_eval_loop.sh

watch curve of acc:

tensorboard --logdir=eval_dir --port=your_port

or eval once

bash run_eval_once.sh

run_test.sh

An End-to-End Trainable Neural Network for Image-based Sequence Recognition and Its Application to Scene Text Recognition