List of the 10 training fonts?

[koch-aai]

Thanks for uploading this trained model - could you possibly provide some info about the training data?

Specifically the fonts used and the average string length. Has this been tested on SVHN by any chance?

Thanks!

[Shreeshrii]

1. It has NOT been tested at all. It is a proof of concept finetune training. Users are encouraged to finetune specific to their own user case, fonts etc.

2. The training text used for last training version with commas is at https://github.com/Shreeshrii/tessdata_shreetest/blob/master/eng.digits.training_text

3. Modified versions of eng.punc and eng.numbers have been used. These could be further modified based on user requirements. They might cause minor improvements in recognition. Files used can be compared with the ones in langdata/eng and are made available at https://github.com/Shreeshrii/tessdata_shreetest/blob/master/eng.digits.punc https://github.com/Shreeshrii/tessdata_shreetest/blob/master/eng.digits.numbers

4. Fonts used for the last training were those listed in src/training/language_specific.sh for Latin script (with minor modifications). Again, finetuning with fonts used in images to be OCRed will lead to better accuracy. Here is the list used

./digits/eng.Arial_Bold.exp0.lstmf
./digits/eng.Arial_Bold_Italic.exp0.lstmf
./digits/eng.Arial.exp0.lstmf
./digits/eng.Arial_Italic.exp0.lstmf
./digits/eng.Courier_New_Bold.exp0.lstmf
./digits/eng.Courier_New_Bold_Italic.exp0.lstmf
./digits/eng.Courier_New.exp0.lstmf
./digits/eng.Courier_New_Italic.exp0.lstmf
./digits/eng.FreeMono.exp0.lstmf
./digits/eng.FreeSans.exp0.lstmf
./digits/eng.FreeSerif.exp0.lstmf
./digits/eng.Georgia_Bold.exp0.lstmf
./digits/eng.Georgia_Bold_Italic.exp0.lstmf
./digits/eng.Georgia.exp0.lstmf
./digits/eng.Georgia_Italic.exp0.lstmf
./digits/eng.Times_New_Roman_Bold.exp0.lstmf
./digits/eng.Times_New_Roman_Bold_Italic.exp0.lstmf
./digits/eng.Times_New_Roman.exp0.lstmf
./digits/eng.Times_New_Roman_Italic.exp0.lstmf
./digits/eng.Trebuchet_MS_Bold.exp0.lstmf
./digits/eng.Trebuchet_MS_Bold_Italic.exp0.lstmf
./digits/eng.Trebuchet_MS.exp0.lstmf
./digits/eng.Trebuchet_MS_Italic.exp0.lstmf
./digits/eng.Verdana_Bold.exp0.lstmf
./digits/eng.Verdana_Bold_Italic.exp0.lstmf
./digits/eng.Verdana.exp0.lstmf
./digits/eng.Verdana_Italic.exp0.lstmf

An earlier training with 10 fonts used only the non-italic version of the fonts and did not include the freefonts - FreeMono, FreeSans, FreeSerif.

[Bech007]

Hi @Shreeshrii
i have a dataset .txt i want train tesseract at them but i don't know how i can do that? thank

[Shreeshrii]

Please see https://github.com/tesseract-ocr/tesseract/wiki/TrainingTesseract-4.00 and https://github.com/Shreeshrii/tess4training

[myjun1124]

This digit traindata was useful to me. It worked well after preprocessing that I used for recognizing temperature on the screen from thermal camera.

[Shreeshrii]

Thanks for the comment @myjun1124. Glad to know it worked for you.

[nikhilcms]

Hi @Shreeshrii , I found that tesseract 4.1.1 works good for extraction of words, but many times in failed to extract digits ( specifically bold ), how can i solve this issue ?