Soumik, I have uploaded the bash scripts and the generated reports and graphs to `ben` branch in my fork of tesstrain repo. See
https://github.com/Shreeshrii/tesstrain/tree/ben and https://github.com/Shreeshrii/tesstrain/commit/a6474ef2dbbac47803d13b6f92fdcf8c9dc3107b Results for the validation data (not seen by lstmtraining either for training or eval, shows an improvement over both ben and script/Bengali. To improve results further, check groundtruth transcription for any missing words, normalize the text and try with some more training data. On Fri, Jan 1, 2021 at 6:41 PM Shree Devi Kumar <shreesh...@gmail.com> wrote: > > nohup make MODEL_NAME=ben START_MODEL=ben LANG_TYPE=Indic > GROUND_TRUTH_DIR=data/ben-ground-truth TESSDATA=$HOME/tessdata_best > DEBUG_INTERVAL=-1 training MAX_ITERATIONS=50000 >> data/ben.log & > > Graphs are created using the training log file as well as validation log > files. Some of these require using PRs which have not yet been merged in > tesstrain repo. > > See > https://github.com/tesseract-ocr/tesstrain/pulls > > For Evaluation reports, I used > https://github.com/eddieantonio/ocreval > > > > On Fri, Jan 1, 2021 at 12:09 PM Soumik Ranjan Dasgupta < > ranjansou...@gmail.com> wrote: > >> Hi Shreeshrii, >> >> Can you please tell me the training command used? Also, how can I create >> the graphs and these other documents? >> >> On Sat, 26 Dec 2020, 18:37 Shree Devi Kumar, <shreesh...@gmail.com> >> wrote: >> >>> Soumik, >>> >>> I used your groundtruth and trained using ben as the START_MODEL. I got >>> best results on the validation set of images at around 5000 iterations. see >>> attached Accuracy report and CER graph. >>> >>> >>> >>> On Thu, Dec 24, 2020 at 8:36 PM Soumik Ranjan Dasgupta < >>> ranjansou...@gmail.com> wrote: >>> >>>> Hi everyone, >>>> I wanted to do fine-tune the ben.traineddata model by using some >>>> ancient text that were supposedly printed with typeset. I have roughly >>>> around 1k lines of text and tried the normal fine-tuning approach with >>>> around 25k iterations. >>>> The thing that surprised me the most was even after packing the >>>> traineddata (character error was around 4%) and testing an unseen image, >>>> the performance was exactly the same. Not a single character was different! >>>> You can find the traineddata, training data, the logs and the source >>>> code at this link: >>>> https://github.com/srdg/unarchived_ben_tess/releases/tag/v0.0.4-alpha >>>> >>>> Can anyone tell me exactly what I am doing wrong here? Do I need to >>>> change any training parameter, increase my training data, or anything else >>>> completely? >>>> >>>> Best regards, >>>> Soumik >>>> >>>> -- >>>> You received this message because you are subscribed to the Google >>>> Groups "tesseract-ocr" group. >>>> To unsubscribe from this group and stop receiving emails from it, send >>>> an email to tesseract-ocr+unsubscr...@googlegroups.com. >>>> To view this discussion on the web visit >>>> https://groups.google.com/d/msgid/tesseract-ocr/1fc044d1-b0ae-45d5-9041-e6fbf8ec5089n%40googlegroups.com >>>> <https://groups.google.com/d/msgid/tesseract-ocr/1fc044d1-b0ae-45d5-9041-e6fbf8ec5089n%40googlegroups.com?utm_medium=email&utm_source=footer> >>>> . >>>> >>> >>> >>> -- >>> >>> ____________________________________________________________ >>> भजन - कीर्तन - आरती @ http://bhajans.ramparivar.com >>> >>> -- >>> You received this message because you are subscribed to the Google >>> Groups "tesseract-ocr" group. >>> To unsubscribe from this group and stop receiving emails from it, send >>> an email to tesseract-ocr+unsubscr...@googlegroups.com. >>> To view this discussion on the web visit >>> https://groups.google.com/d/msgid/tesseract-ocr/CAG2NduVZ3A7CUEqw29Gxu6r1-cLHPTLFt%3D%3D0C0109D_6x6C7Kw%40mail.gmail.com >>> <https://groups.google.com/d/msgid/tesseract-ocr/CAG2NduVZ3A7CUEqw29Gxu6r1-cLHPTLFt%3D%3D0C0109D_6x6C7Kw%40mail.gmail.com?utm_medium=email&utm_source=footer> >>> . >>> >> -- >> You received this message because you are subscribed to the Google Groups >> "tesseract-ocr" group. >> To unsubscribe from this group and stop receiving emails from it, send an >> email to tesseract-ocr+unsubscr...@googlegroups.com. >> To view this discussion on the web visit >> https://groups.google.com/d/msgid/tesseract-ocr/CAM-%2BFN%3DZggnH4wV5vUhY9nsSqjKg9xZ5TQDoCMwSqf7H0oPogQ%40mail.gmail.com >> <https://groups.google.com/d/msgid/tesseract-ocr/CAM-%2BFN%3DZggnH4wV5vUhY9nsSqjKg9xZ5TQDoCMwSqf7H0oPogQ%40mail.gmail.com?utm_medium=email&utm_source=footer> >> . >> > > > -- > > ____________________________________________________________ > भजन - कीर्तन - आरती @ http://bhajans.ramparivar.com > -- ____________________________________________________________ भजन - कीर्तन - आरती @ http://bhajans.ramparivar.com -- You received this message because you are subscribed to the Google Groups "tesseract-ocr" group. To unsubscribe from this group and stop receiving emails from it, send an email to tesseract-ocr+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/CAG2NduUm6emCJpYh3sj4PyuFtO0sVLS5V-FMOvkXVmyhuT%3D9%3DQ%40mail.gmail.com.
