tesseract-ocr / tessdata_fast Goto Github PK

I maintain the Fedora tesseract-tessdata package. I'm unsure how to deal with the tessconfigs folder when packaging. The release archive [1] contains the empty submodule folder, and some broken symlinks which point to files inside the submodule. Is it safe to just ignore the folder and these files? Is there any worth pulling the tessconfigs folder separately and packaging it?

[1] https://github.com/tesseract-ocr/tessdata_fast/archive/4.1.0/tessdata_fast-4.1.0.tar.gz

The chinese models chi_sim, chi_sim_vert, chi_tra and chi_tra_vert use chi instead of zho which is the ISO 639-3 standard.

Since all other files are named according to this standard I dont see a reason why this should not be the case for chinese.

super@super-lap:/home/tesseract-lstm$ bin/tesseract --psm 0 ./tessinput.tif stdout
Page 1
Segmentation fault (core dumped)

I am using tesseract 4.0 .net wrapper dll and tying to use tesseract_fast trained data, but it is not gettng instialized and it throws an unhandled exception (read write) . I tried with engine mode "Lstmonly" and "default" but couldn't intialize. Kindly help me whether i am using in a wrong way or i should not used this fast traineddata.
Error is as follows

I downloaded por.traineddata (for portuguese language) to the StreamingAssets/tessdata folder and I get a "TessAPIInit failed. Output: -1". Should this be enough? I notice there are a bunch of different files for starting with eng like "eng.cube.params", "eng.cube.nn" and so many others? Should there be the same for portuguese?

I'm asking this to clarify, as README was last modified before the 5.0.0 release.

How to actually use these tessdata files?

Please provide a hint in the README.

Thank you

On windows 7,I am trying to use Fast Traineddata files in my java project. But I am getting Invalid Memory access when using it even after setting datapath.

I have tried to use best data files but it also gives same error. Default data files are working but it is huge file so I was going for fast files.

Tesseract tess = new Tesseract();

tess.setDatapath("C:\\Users\\U6070534\\Downloads\\tess4j\\tessdata");
tess.setLanguage("eng");

String inputFilePath = "C:\\Users\\U6070534\\IdeaProjects\\ocrsample\\screenshot\\craft0.png";
    try {
        textpath.add(tess.doOCR(new File(inputFilePath)));
    } catch (TesseractException e1) {
        e1.printStackTrace();
    }

Exception in thread "main" java.lang.Error: Invalid memory access
at com.sun.jna.Native.invokePointer(Native Method)
at com.sun.jna.Function.invokePointer(Function.java:470)
at com.sun.jna.Function.invoke(Function.java:404)
at com.sun.jna.Function.invoke(Function.java:315)
at com.sun.jna.Library$Handler.invoke(Library.java:212)
at com.sun.proxy.$Proxy0.TessBaseAPIGetUTF8Text(Unknown Source)
at net.sourceforge.tess4j.Tesseract.getOCRText(Tesseract.java:437)
at net.sourceforge.tess4j.Tesseract.doOCR(Tesseract.java:292)
at net.sourceforge.tess4j.Tesseract.doOCR(Tesseract.java:213)
at net.sourceforge.tess4j.Tesseract.doOCR(Tesseract.java:197)
at OcrReader.main(OcrReader.java:25)
Failed loading language 'eng'
Tesseract couldn't load any languages!
Process finished with exit code 1

It would be useful to have an official 4.0.0 tag for the data as well.

Hi,
I see "network" in some your description, i'm witting an app running offline, there is no network. Does tesseract 4.0 can be used offline?

As emailed to the mailing list, does it make sense to tag another release?

> Hi,
>
> With Tesseract now switching to regular (alpha) releases of 5.0.0; does
> it make sense to consider some versioning for language files as well?
>
> The Internet Archive has switched to using Tesseract for all our OCR,
> and I'm hoping that we can record exactly what version of language files
> was used for a specific OCR job. Currently, the answer is simple, since
> we're using the default packages from Ubuntu focal, but I am working on
> switching to Tesseract release/tag 5.0.0-20201231.
>
> But the tessdata_fast (or tessdata_best, for that matter) do not seem to
> have any recent 5.x releases:
> https://github.com/tesseract-ocr/tessdata_fast/releases
>
> Are there plans to create a release/tag for the tessdata_* repositories?
>
> Cheers,
> Merlijn

And the follow-up:

On 27/01/2021 12:42, Shree Devi Kumar wrote:
>> The Internet Archive has switched to using Tesseract for all our OCR,
> 
> I am so happy to hear this. It will be great to have the Indic languages
> that were marked as non-ocrable so far be converted to text correctly on
> Internet Archive.
> 
> Is there any page with instructions to do this? Can a language be specified
> while OCRing? eg. Better results are many times received using
> script/Devanagari instead of san for Sanskrit.
> 
> Regarding your question about tessdata, there have only been minor changes
> to tessdata files but adding a tag is a good idea. I suggest you post this
> as a feature request in the repo.

Installations on case insensitive filesystems (macOS, Windows, ...) cannot install both Lao.traineddata and lao.traineddata. It might also be confusing for users to know the difference between -l Lao and -l lao.

I suggest to rename the first one. Which name would be fine?

Fast (integer version) trained LSTM models

According to the wiki, equ and osd trained data will reuse the 3.x data file. The weird thing is that osd is copied but equ is not. Is there any reason? e.g. equ is deprecated in 4.x

Please see details at

tesseract-ocr/tessdata#88 (comment)

tesseract-ocr/langdata#116

tesseract-ocr/tessdata_best#23

@jbreiden @AlexanderP - FYI - regarding problem with packaged traineddata for kur_ara.

what does "vert" traineddata do?