Researching

[patrickphat]

Do a literature on Layout + OCR + KV https://docs.google.com/spreadsheets/d/1bBCp9b9eI6BoiDnjxNMkd0EexMPGJNJ2ZRinzG3E0J4/edit?usp=sharing

Question: Tại sao không theo hướng spectral mà theo hướng spatial

[patrickphat]

Research for OCR task:

Learn about CTC:

• https://towardsdatascience.com/intuitively-understanding-connectionist-temporal-classification-3797e43a86c

CTCModel implementation:

• https://towardsdatascience.com/intuitively-understanding-connectionist-temporal-classification-3797e43a86c

CRNN implementation:

• https://github.com/bgshih/crnn

Tensorflow version of CRNN:

• https://github.com/Belval/CRNN

Keras CRNN:

• https://github.com/qjadud1994/CRNN-Keras/blob/master/Model.py
• https://github.com/kurapan/CRNN (use this)

CTC loss in keras:

• https://chadrick-kwag.net/tf-keras-rnn-ctc-example/

Prepare data for CTC Loss:

• https://stackoverflow.com/questions/40779479/preparing-target-label-in-tensorflow-python-for-ctc-loss
• https://stackoverflow.com/questions/42578532/how-to-generate-read-sparse-sequence-labels-for-ctc-loss-within-tensorflow

[patrickphat]

Research for KV task

GCNN:

• https://towardsdatascience.com/using-graph-convolutional-neural-networks-on-structured-documents-for-information-extraction-c1088dcd2b8f