Attenion-OCR README.

---

This markdown is for "Attention-based Extraction of Structured Information from Street View Imagery" or "Attention-OCR" project.

If you want to use your own images stored in fsns format, just follow the following steps:

Step 1.

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.

Step 2.

Then use tfrecord_gen.py to transfer the images made by step 1 to fsns dataset.

Step 3.

Move the binary files to train/test/validation folders and modify the size (number of images) in DEFAULT_CONFIG in fsns.py.

Step 4.

Run train.py or evil.py.

For further information, check these out:

1. [https://github.com/OmmmmooooO/models/tree/master/attention_ocr]
2. [https://stackoverflow.com/questions/44430310/how-to-create-dataset-in-the-same-format-as-the-fsns-dataset?rq=1]