Hi I just ocred 30 pages of a sanskrit book on Sanskrit OCR. I got WER of 54% and CER of 24 %. Whereas I get WER of 20 % on Indsenz and CER Of 8 %. Have you tried comparing Indsenz with Sanskrit OCR. Which one is better where?
On Tuesday, 21 June 2016 12:36:23 UTC+5:30, rohit saluja wrote: > > Hey thanks a lot. Your replies are really helpful. > > Rohit > > On Saturday, 18 June 2016 23:41:13 UTC+5:30, shree wrote: >> >> I do not know about the training process for cube, it is not documented. >> >> I have uploaded the box/tif pairs generated by text2image under windows >> for sanskrit - there are two versions s21 and s95 - using different fonts >> and exposure levels. Please see >> https://github.com/Shreeshrii/imagessan/tree/master/trainingdata-s21 >> https://github.com/Shreeshrii/imagessan/tree/master/trainingdata-s95 >> >> In s21, each font is used for 3 different exposure levels , -1, 0 and 1. >> tesstrain.sh --lang san --langdata_dir ./langdata --tessdata_dir ./ >> --exposures "-1 0 1" >> >> In s95, each font is used only at 0 exposure level. >> >> ShreeDevi >> ____________________________________________________________ >> भजन - कीर्तन - आरती @ http://bhajans.ramparivar.com >> >> On Tue, Jun 14, 2016 at 3:35 AM, rohit saluja <[email protected]> >> wrote: >> >>> Hey thanks a lot for your reply. This seems to be a great idea to use >>> hin data with sanskrit wordlist. >>> >>> Still I am interested in knowing the things building from scratch. >>> So I used some boxfiles and images I created for sanskrit 2003 font and >>> used the hindi config file from >>> https://github.com/tesseract-ocr/langdata/blob/master/hin/hin.config >>> and I renamed it as san3ds.config. san3ds(3 for 2003 ds for devanagari >>> split) is the new name I am giving for my new training data. >>> >>> I was able to train san3ds without any config file before. >>> >>> I just renamed san3ds.word-dawg as san3ds.cube-word-dawg. Remaining >>> files I kept as it is. >>> I could form san3.traineddata file, but I am getting an error while >>> recognition:- >>> >>> Cube ERROR (CubeRecoContext::Load): unable to read cube language model >>> params from /usr/local/share/tessdata/san3ds.cube.lm >>> Cube ERROR (CubeRecoContext::Create): unable to init CubeRecoContext >>> object >>> init_cube_objects(true, &tessdata_manager):Error:Assert failed:in file >>> tessedit.cpp, line 214 >>> Segmentation fault (core dumped) >>> >>> Any help in this, why this is happening? Is it wrong in renaming >>> word-dawg, I cannot find any separate option for generating cube-word-dawg. >>> >>> Thanks in advance >>> Rohit >>> >>> >>> On Mon, Jun 13, 2016 at 7:04 PM, ShreeDevi Kumar <[email protected]> >>> wrote: >>> >>>> If you look at the readme files in the diff subdirectories starting >>>> with OCR under >>>> https://github.com/Shreeshrii/imagessan/tree/master you will see >>>> results of character and word level accuracy. Depending on the font, >>>> character level accuracy is around 80% and word level accuracy around 60% >>>> >>>> I have not used it for actual OCR of any text because sanskritocr >>>> software by dr. Oliver hellwig gives better results. >>>> >>>> See https://sites.google.com/site/sanskritcode/ocr/1-ocr-ing >>>> >>>> - sent from my phone. excuse the brevity. >>>> On 13-Jun-2016 6:53 pm, "ShreeDevi Kumar" <[email protected]> wrote: >>>> >>>>> Yes, hin traineddata with cube gives better results than san. >>>>> >>>>> I did some rudimentary testing with the new traineddata I made. It >>>>> does not use cube. Look at the config files, it has some options for >>>>> devanagari processing. >>>>> >>>>> You could try to unpack the hin traineddata and then remake the Dawg >>>>> files using sanskrit wordlists and combine them as an experiment. >>>>> >>>>> If you have unicode version of the font used for the docs you want to >>>>> OCR, then train using that. >>>>> >>>>> - sent from my phone. excuse the brevity. >>>>> On 13-Jun-2016 4:47 pm, "rohit saluja" <[email protected]> wrote: >>>>> >>>>>> Thanks again for replying. I will surely check them out. >>>>>> >>>>>> My experience is that OCR on sanskrit data with hin.traineddata gives >>>>>> better results than san.traineddata. I do know know, it is due to cube >>>>>> mode >>>>>> or devanagari preprocessing(segmentation i guess) in devanagari? >>>>>> >>>>>> I wonder why such preprocessing is not applied in san.traineddata. >>>>>> Please let me know whether you are using cube mode in your >>>>>> traineddata or not, and are you using devanagari preprocessing? >>>>>> >>>>>> On Mon, Jun 13, 2016 at 9:18 AM, ShreeDevi Kumar <[email protected]> >>>>>> wrote: >>>>>> >>>>>>> Google has not provided images and box files for San.traineddata >>>>>>> released for 3.04 >>>>>>> >>>>>>> I tried training using text2image with a combination of different >>>>>>> fonts and training text. Results are at >>>>>>> https://github.com/Shreeshrii/imagessan/tree/master/tessdata >>>>>>> >>>>>>> You can give these a try to see if recognition is any better. >>>>>>> >>>>>>> You can unpack any trained data file using -u option with >>>>>>> combine-tessdata to see the config files etc. >>>>>>> >>>>>>> >>>>>>> http://manpages.ubuntu.com/manpages/trusty/man1/combine_tessdata.1.html >>>>>>> >>>>>>> Use the dawg2wordlist to look at the various dictionary word lists >>>>>>> used. >>>>>>> >>>>>>> http://manpages.ubuntu.com/manpages/trusty/man1/dawg2wordlist.1.html >>>>>>> >>>>>>> - sent from my phone. excuse the brevity. >>>>>>> On 12-Jun-2016 11:26 am, "rohit saluja" <[email protected]> wrote: >>>>>>> >>>>>>>> Hey thanks for replying. >>>>>>>> Which options to use with text2image command? Also, is there any >>>>>>>> configuration file and fonts list? >>>>>>>> >>>>>>>> I tried the default option of text2image with tesseract github >>>>>>>> training data with sanskrit 2003, but the recognition results are far >>>>>>>> away >>>>>>>> from san.traineddata file on github. >>>>>>>> Any help in matching san.traineddata results, starting from the >>>>>>>> scratch, would be highly appreciated. >>>>>>>> >>>>>>>> Thanks in advance >>>>>>>> Rohit >>>>>>>> >>>>>>>> On Friday, 6 May 2016 12:59:44 UTC+5:30, rohit saluja wrote: >>>>>>>> >>>>>>>>> Do we have Sanskrit training images and box files available online? >>>>>>>>> >>>>>>>>> Thanks >>>>>>>>> Rohit >>>>>>>>> >>>>>>>> -- >>>>>>>> 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 [email protected]. >>>>>>>> To post to this group, send email to [email protected]. >>>>>>>> Visit this group at https://groups.google.com/group/tesseract-ocr. >>>>>>>> To view this discussion on the web visit >>>>>>>> https://groups.google.com/d/msgid/tesseract-ocr/45767a89-cd11-4f39-9622-3fe7b4d20a4a%40googlegroups.com >>>>>>>> >>>>>>>> <https://groups.google.com/d/msgid/tesseract-ocr/45767a89-cd11-4f39-9622-3fe7b4d20a4a%40googlegroups.com?utm_medium=email&utm_source=footer> >>>>>>>> . >>>>>>>> For more options, visit https://groups.google.com/d/optout. >>>>>>>> >>>>>>> -- >>>>>>> You received this message because you are subscribed to a topic in >>>>>>> the Google Groups "tesseract-ocr" group. >>>>>>> To unsubscribe from this topic, visit >>>>>>> https://groups.google.com/d/topic/tesseract-ocr/apmhpJ3K924/unsubscribe >>>>>>> . >>>>>>> To unsubscribe from this group and all its topics, send an email to >>>>>>> [email protected]. >>>>>>> To post to this group, send email to [email protected]. >>>>>>> Visit this group at https://groups.google.com/group/tesseract-ocr. >>>>>>> To view this discussion on the web visit >>>>>>> https://groups.google.com/d/msgid/tesseract-ocr/CAG2NduXfqoY_BSW9BURAbj_AzdtRykK2ea5e9G2Suq9QCeWMOA%40mail.gmail.com >>>>>>> >>>>>>> <https://groups.google.com/d/msgid/tesseract-ocr/CAG2NduXfqoY_BSW9BURAbj_AzdtRykK2ea5e9G2Suq9QCeWMOA%40mail.gmail.com?utm_medium=email&utm_source=footer> >>>>>>> . >>>>>>> >>>>>>> For more options, visit https://groups.google.com/d/optout. >>>>>>> >>>>>> >>>>>> -- >>>>>> 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 [email protected]. >>>>>> To post to this group, send email to [email protected]. >>>>>> Visit this group at https://groups.google.com/group/tesseract-ocr. >>>>>> To view this discussion on the web visit >>>>>> https://groups.google.com/d/msgid/tesseract-ocr/CAEga%2BsUNCmGHEmPB0fBZjgPmEAXvWNtzzdkkKK%3DRcd_u25f%2B1Q%40mail.gmail.com >>>>>> >>>>>> <https://groups.google.com/d/msgid/tesseract-ocr/CAEga%2BsUNCmGHEmPB0fBZjgPmEAXvWNtzzdkkKK%3DRcd_u25f%2B1Q%40mail.gmail.com?utm_medium=email&utm_source=footer> >>>>>> . >>>>>> For more options, visit https://groups.google.com/d/optout. >>>>>> >>>>> -- >>>> You received this message because you are subscribed to a topic in the >>>> Google Groups "tesseract-ocr" group. >>>> To unsubscribe from this topic, visit >>>> https://groups.google.com/d/topic/tesseract-ocr/apmhpJ3K924/unsubscribe >>>> . >>>> To unsubscribe from this group and all its topics, send an email to >>>> [email protected]. >>>> To post to this group, send email to [email protected]. >>>> Visit this group at https://groups.google.com/group/tesseract-ocr. >>>> To view this discussion on the web visit >>>> https://groups.google.com/d/msgid/tesseract-ocr/CAG2NduWu3-cLcTHi2e%3D0Zr15Do5nawfG93k_dXvBeBwze%2BMHfw%40mail.gmail.com >>>> >>>> <https://groups.google.com/d/msgid/tesseract-ocr/CAG2NduWu3-cLcTHi2e%3D0Zr15Do5nawfG93k_dXvBeBwze%2BMHfw%40mail.gmail.com?utm_medium=email&utm_source=footer> >>>> . >>>> >>>> For more options, visit https://groups.google.com/d/optout. >>>> >>> >>> -- >>> 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 [email protected]. >>> To post to this group, send email to [email protected]. >>> Visit this group at https://groups.google.com/group/tesseract-ocr. >>> To view this discussion on the web visit >>> https://groups.google.com/d/msgid/tesseract-ocr/CAEga%2BsWmbZV-qC1fwKbb%2BmqO3SSaqseZPHu7o1srObOqt%2BPjxw%40mail.gmail.com >>> >>> <https://groups.google.com/d/msgid/tesseract-ocr/CAEga%2BsWmbZV-qC1fwKbb%2BmqO3SSaqseZPHu7o1srObOqt%2BPjxw%40mail.gmail.com?utm_medium=email&utm_source=footer> >>> . >>> >>> For more options, visit https://groups.google.com/d/optout. >>> >> >> -- 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 [email protected]. To post to this group, send email to [email protected]. Visit this group at https://groups.google.com/group/tesseract-ocr. To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/c9edbaa5-fb5d-4c01-87d9-93b1a2308f9f%40googlegroups.com. For more options, visit https://groups.google.com/d/optout.
