Attenion-OCR README.
This markdown is for "Attention-based Extraction of Structured Information from Street View Imagery" or "Attention-OCR" project.
Use image_to_fsns_format.ipynb
to transfer and/or augment your image to [H, W, Channel]=[150, 600, 3] included 4 different/same views per image.
Then use tfrecord_gen.py
to transfer the images made by step 1 to fsns dataset.
Move the binary files to train/test/validation folders and modify the size
(number of images) in DEFAULT_CONFIG
in fsns.py
.
Run train.py
or evil.py
.
For further information, check these out: