• This program can do handwriting ocr.

• Model architecture is based on CRNN + CTC. Model is built with Pytorch. UI is written in Vue. Server is built with FastApi & Uvicorn

• You will need to start up those 2 services in order to draw and ocr your handwriting.

• Demo:

• Model architecture specifically: CNN + MLP + ResNet > CNN > 2 bidirectional GRU > MLP > CTC loss

  • see: https://github.com/Norlandz/ocrSentenceModel .\demo-ipynb\ocrSentence.ipynb .\src\mlModel\Crnn.py

  • (tried to add Attention mechanism, but only made it worse... idk how)

• Note:

  • The model was only trained for 30min on a relatively small dataset. So the accuracy is not high.

  • The time for conversion is long. Because I didnt do any optimization on the process.

• Repos for the same project:

  • https://github.com/Norlandz/ocr-sentence-ui
  • https://github.com/Norlandz/ocrSentenceModel