Attention_OCR_KOR

The progress was used to Korean OCR based on Google Attention OCR.

Modify Google's attention model for Korean text recognition.

More details can be found in this paper - "Attention-based Extraction of Structured Information from Street View Imagery"

And i can reference Chinse "Attention-OCR-CN"

This project can run on Windows10, using the python3 environment and The network is built using tensorflow

I generated FSNS format tfrecord for Korean text recognition and a dictionary of 2,445 korean characters.

How to use your own image data to train the model

cd python
python train.py --dataset_name=newtextdataset

Loss Function

Verify your own model

Verify your own model

python eval.py

The results can be view used tensorboard

tensorboard  --logdir=/tmp/attention_ocr/eval

Accuracy

Use a trained model

python demo_inference.py --batch_size=32 \
  --checkpoint=model.ckpt-192211\
  --image_path_pattern=./datasets/data/fsns/temp/~~~.png