search

qurator-spk / sbb_binarization

Document Image Binarization
Apache License 2.0
67 stars 14 forks source link

ocrd-tool.json: register SavedModel resources #53

Closed bertsky closed 1 year ago

bertsky commented 1 year ago

This is how decentralized resource locations would look like now. I have used @apacha's release URLs for now. (Notice there is a deviation in how the 2021-03-09 archive is substructured. Also, the MacOS directories may get in the way.)

bertsky commented 1 year ago

@apacha we still have not uploaded the converted models to qurator-data.de, and the Huggingface model repo does not support producing archives (only URLs for single file names or programmatic downloads via huggingface_hub.snapshot_download), so currently the release archive on your fork is the only resource we can use within ocrd resmgr.

Could you please remove the subdirectories __MACOSX and model_2021_03_09/.DS_Store from the zip file and re-release?

Otherwise, @vahidrezanezhad can we add the fixed zip file to a new release here on the main repo? (Would be as simple as gh release -R qurator-spk/sbb_binarization upload v0.0.11 default-2021-03-09.zip ...)

cneud commented 1 year ago

Sorry for the delays! If it is of immediate help for OCR-D I can add the cleaned Saved Models here to the release.

bertsky commented 1 year ago

If it is of immediate help for OCR-D I can add the cleaned Saved Models here to the release.

Yes, that would really help! (In the current state, the model does not resolve after downloading because of the macos directories getting in the way...)

cneud commented 1 year ago

Now added here https://github.com/qurator-spk/sbb_binarization/releases/download/v0.0.11/saved_model_2021_03_09.zip

cneud commented 1 year ago

For consistency, I now also added a clean version of https://github.com/qurator-spk/sbb_binarization/releases/download/v0.0.11/saved_model_2020_01_16.zip

apacha commented 1 year ago

@bertsky I tried to clean the archives and tried to remove the .DS_Store files and __MACOSX subdirectories. The new versions are uploaded to https://github.com/apacha/sbb_binarization/releases/tag/pre-trained-models. Not sure if it still is necessary though, now that you have them stored in the upstream repo.

BTW: I'm happy that my proposed approach of storing the models in a github release finally made it ;-).