Sylheti-Nagri-OCR

Execution

• run scripts/resource_gen.ipynb with proper variable paths to generate:

  • Synthetic Grapheme based dictionary
  • Single line vocab file (such file are needed for synthtiger,EasyOCR etc)
  • Multi line vocab file (such file are needed for synthindic)
  • Train data unique word based dictionary
  • Separates folds (train and test) and creates data.txt where absolute path of the image and label are tab separated

• Synthtiger: follow advanced usage section

  • needed scripts are stored under scripts/synthtiger

  • Font Customization:

    python scripts/synthtiger/extract_font_charset.py -w 1 /home/apsisdev/OCR/SylhetiNagri/fonts/

System Info:

OS          : Ubuntu 22.04.2 LTS x86_64 
Host        : Z490 GAMING X AX -CF 
Kernel      : 5.19.0-38-generic 
Shell       : bash 5.1.16 
DE          : GNOME 42.5 
CPU         : Intel i9-10900K (20) @ 5.300GHz 
GPU         : NVIDIA GeForce RTX 3090 
Memory      : 13503MiB / 32022MiB 

TODO

• github installation and configuration
• resource_gen documentation
• synthtiger generation
• synthindic generation
• tfrs2 records
• training notebook
• inference notebook
• evaluation on real data