README.md

tesstrain-akk

Finetune Training and OCR evaluation of Tesseract 5.0.0 Alpha for Akkadian in Cuneiform script with 95% accuracy on the evaluation set

Programs used

• tesseract built from github master
• tesstrain Training workflow for Tesseract 4 as a Makefile
• linegen from kraken
• create_dictdata from pytesstrain
• ISRI Analytic Tools for OCR Evaluation with UTF-8 support
• The ocrevalUAtion tool

Training Input

• training text
• fonts
• unicharset

Training Steps

• conda activate kraken

• nohup bash txt2img.sh > txt2img.log &

• nohup bash img2lstmf.sh > img2lstmf.log &

• nohup bash trainlayer.sh > trainlayer.log &

• nohup bash checkpointeval.sh > reports/checkpointeval.txt &

Traineddata files

• trained tessdata_best
• trained tessdata_fast
• other old models

Evaluation Results

• Summary
• Detail
• Model Level Detail