https://github.com/norlandz/ocr-sentence-ui

Handwriting ocr, CRNN + CTC
https://github.com/norlandz/ocr-sentence-ui

crnn machine-learning ocr vue3

Last synced: 5 months ago
JSON representation

Handwriting ocr, CRNN + CTC

• Host: GitHub
• URL: https://github.com/norlandz/ocr-sentence-ui
• Owner: Norlandz
• Created: 2024-12-19T10:24:59.000Z (10 months ago)
• Default Branch: main
• Last Pushed: 2024-12-19T17:12:31.000Z (10 months ago)
• Last Synced: 2025-02-18T01:42:09.130Z (8 months ago)
• Topics: crnn, machine-learning, ocr, vue3
• Language: TypeScript
• Homepage:
• Size: 479 KB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: docs/readme.md

Awesome Lists containing this project

README

          
- 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:

  ![demo](<./demo - ocrSentence - 20241219_172540567.png>)

- 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