search

kouprlabs / voltaserve

⚡️ Cloud Storage for Creators
https://voltaserve.com
Other
244 stars 13 forks source link

Ability to OCR images for search, single and batch, with a dialog that asks to set the language #120

Open cheney-yan-ifl opened 2 months ago

cheney-yan-ifl commented 2 months ago

Currently the app only supports search in documents. Expecting OCR support for images.

bouassaba commented 2 months ago

@cheney-yan-ifl to make the images searchable, you have to enable the "Insights" feature, here in the demo video it shows how to do it (I already moved the video to the exact location): https://youtu.be/Uf3EWb2hDfs?t=352 Just make sure you choose the correct language when enabling the "Insights". Give it a try and let me know if it works for you.

cheney-yan-ifl commented 2 months ago

Thanks. It works. It will be convenient if there's a global setting for automatically turn on insights for images.

bouassaba commented 2 months ago

@cheney-yan-ifl the problem is that there is no efficient way to automatically detect: 1) if an image has text 2) if yes - what's the language of that text?

EDIT: That would involve training ML models on a massive amount of images to be able to get this working at an acceptable success rate.

bouassaba commented 2 months ago

I will rename this GitHub to "automatically detect images with text and their language", mark it as a "feature" and keep this open for research.