Implement paper https://arxiv.org/pdf/1704.03549.pdf

Some change while implement

1, Pass one view instead of each of the four views through the same CNN feature extractor.

Generate text in image

python generate_images.py

Train

config params in file config.py
python run.py

Predict

python predict.py --image path_image