Zotero OCR plugin errors

nathan.hopson
Zotero 7.0.26
Zotero OCR 0.9.2

Can anyone help explain or troubleshoot this discrepancy?

Zotero OCR settings:
https://s3.amazonaws.com/zotero.org/images/forums/u3279/tfp3610qi4k91ilswf1j.png

Results:
https://s3.amazonaws.com/zotero.org/images/forums/u3279/jr4f99fz2kmtasmftng8.png

From the plugin's GitHub, this doesn't appear to be a known issue.
(I'm posting here because it's a bigger community and has always been super helpful, but if nobody has a solution I will post there. I hope that's not bad etiquette)
  • aborel
    It is in principle better to ask for advice about plugins where you have the best chance to contact the relevant developer, i.e. on Github rather than here. It's an efficiency consideration rather than an etiquette problem. But you're in luck, some developers are frequent visitors on this forum ;-)

    You mention a discrepancy, can you be more explicit? I can guess a few things based on the screenshots, but a description in your own words will always facilitate the discussion.
  • nathan.hopson
    @aborel thanks for the reply

    The discrepancy is that the settings and output are not matched up.
    I have unchecked "Save output as note" and "Save the intermediate images" but they are output anyway.

    In the interim, I have additionally noticed:
    1. The resulting PDF balloons in size, from 19MB to 650 in the case of one 55-page B/W scan, for instance. Makes Zotero super heavy, takes forever to open, etc.
    2. The text layer is nonsense machine text when viewed in some PDF viewers but not others. Acrobat and Preview render correctly, but Skim gives me a lot of 蝗ｽ髫帶黒魃ｨ蟋泌藤莨. Don't know if that's relevant, but this is related to the issue that caused me to install the plugin to begin with: Acrobat (fully updated) newly gives that junk machine text for many Japanese files, which I assume is a font embedding issue but can't be certain.

    In any case, I assume 1 and 2 are Tesseract issues, not Zotero OCR issues, but they make the plugin basically unusable for me unless there's a workaround.
  • aborel
    Thanks for the explanation!
    Does this happen only with this file, or do you have other Japanese PDFs where similar problems happen? Can you share a file that could be used for tests?
  • nathan.hopson
    edited 3 days ago
    The first issue (note and images) has happened with every Japanese PDF I've tried since installing -- until this morning, of course. Now it's just file size and the note being saved separately despite my settings, the latter of which I can live with.

    The file size and garbage text was new yesterday with every PDF I tried. This was true for both PDFs directly downloaded from Japan's national library (NDL) -- flattened or not -- and also for PDFs compiled from scanned images (TIFF).

    Here's the file that blew up to 650MB:
    *deleted*
    Thanks for you help!
  • aborel
    Dropbox tells me that the access has failed for an unknown reason :-/
  • aborel
    The error is gone, but I'm seeing a folder, not a specific file. And I can't see a filename that would match the screenshots from yesterday.
    Another file might probably work just as well for testing, but I'd like to be sure we agree on the same file.
  • nathan.hopson
    Hi again @aborel

    That was very strange! Dropbox killed my link sharing for several hours and then the link was corrupted. I don't know what happened, but I tested the link below and it should lead to the right file:
    https://www.dropbox.com/scl/fi/b375aes9hdol0voas3r3m/icr2-flat.pdf?rlkey=ki50ncjwkf57qoe2rdn609lvy&dl=0
  • aborel
    Thanks, I've got it this time! I will investigate.
  • aborel
    edited 3 days ago
    - On my machine, both the image removal and the note option seem to work normally, but I will continue to investigate possible causes.

    - OCR result: I can't read Japanese so I'm not the best judge, but overall I don't see a general problem with the file you provided. In some pages the text layer certainly has errors, some others look OK in my untrained eye. If a specific PDF reader has a problem with the file, maybe it should be reported there? If you can point out a specific page that exemplifies this problem, I can take a closer look, of course.

    - file size: this is certainly a weakness in Zotero-OCR, but for me it wasn't quite as bad as you report. My OCRed PDF was 217MB, not 650? Looking at the original PDF, I'd say that it was produced with a 150 to 200 DPI resolution. You can try 150 instead of 300 in your preferences: the output PDF will be larger than the original one, but far less than with the default of 300, and the apparent quality should not be affected. In my test at 150, the OCRed PDF is 80MB.
  • aborel
    I'm suspecting a possible scenario, I need a few more details to try and reproduce:
    - I see the Attanger plugin in your Zotero preferences, are you using Zotero-OCR on linked files managed by Attanger?
    - have you OCRed several files that are stored in the same folder?
    - did you first try to use the plugin with the note and images options activated?
    - are the residual images (and image-list.txt) perhaps older files that were generated in one of you first tests?

    I would also recommend trying again with the new Zotero-OCR 0.9.3 release. We have introduced several features that will tell you more about the execution of the plugin, and make diagnostics much easier.
Sign In or Register to comment.