CRNN for English Handwriting Recognition with CTC Loss

Dataset

• IAM Dataset: http://www.fki.inf.unibe.ch/databases/iam-handwriting-database

Command

export PYTHONPATH=path_to_root_folder

cp src/example_config.py src/config.py

Folder Structure

├── data_gen.py
├── LICENSE
├── logs
├── models
├── notebooks
├── README.md
├── requirements.txt
├── src
│   ├── config.py
│   ├── data_generator.py
│   ├── log.py
│   ├── train.py
│   └── utils.py

Reference

CTC Loss

• https://www.dlology.com/blog/how-to-train-a-keras-model-to-recognize-variable-length-text/
• https://hackernoon.com/latest-deep-learning-ocr-with-keras-and-supervisely-in-15-minutes-34aecd630ed8
• https://github.com/SeanNaren/deepspeech.pytorch/blob/master/model.py

Kaggle ctc loss

• https://dinantdatascientist.blogspot.com/2018/02/kaggle-tensorflow-speech-recognition.html

Explained ctc loss:

• https://gab41.lab41.org/speech-recognition-you-down-with-ctc-8d3b558943f0
• https://distill.pub/2017/ctc/
• https://stats.stackexchange.com/questions/320868/what-is-connectionist-temporal-classification-ctc

CTC loss params

• https://kur.deepgram.com/specification.html#using-ctc-loss