Closed Shreeshrii closed 3 years ago
PR #214, #215 and #217 are needed to be applied for this to work.
see https://github.com/Shreeshrii/tesstrain/tree/ben/data/ben for training results and reports for Bengali using groundtruth provided in https://groups.google.com/g/tesseract-ocr/c/kpF4jmik6W0/m/bo4HL0ejBgAJ
I have changed the plots to also display the best fit curve for the CER values. New plots look like the following:
I will try and convert the bash scripts creating ocreval reports to make.
Improvements over current version:
See sample plots below:
Training from Scratch - Sanskrit in Devanagari + IAST + English
Training with START_MODEL=san, PlusMinus - Sanskrit in Devanagari