Community Wishlist Survey 2019/Categories/Automatic category suggestion for Commons images

Automatic category suggestion for Commons images

  • Problem: Hell lot of uncategorised / ill-categorised pages
  • Who would benefit: Whole Wikimedia community
  • Proposed solution: Implement AI to facilitate face recognition, auto-categorisation in commons.
  • More comments: This will benefit both wikiproject and decentralisation of AI. Wikipedia is one of the largest repositories of training data right now. If wikipedia open-source the trained JSON (i.e. features-label pair), this will help fast deployment of face-recognition, landmark-recognition, animal-recognition, plant-recognition and object-oriented besides other. This will help Wikipedia strengthen the goal of open-data.
  • Phabricator tickets: T192444 T155538 T155848
  • Proposer: Capankajsmilyo (talk) 09:55, 31 October 2018 (UTC)Reply[reply]


phab:T192444 is about identifying faces in images on commons. --AKlapper (WMF) (talk) 15:06, 31 October 2018 (UTC)Reply[reply]

  • Hi Capankajsmilyo. I think it would be good to rename this wish to reflect the problem we are trying to get fixed here (maybe Automatic category suggestion for Commons images). That way people can understand better what it is and vote it up. -- NKohli (WMF) (talk) 18:47, 31 October 2018 (UTC)Reply[reply]
I'm okay with both. Capankajsmilyo (talk) 18:50, 31 October 2018 (UTC)Reply[reply]

I like the idea. But before starting with heavy image recognition one could already use meta data like title, description, location for suggesting appropriate categories. Technically, i also wonder if it would make sense to build a general service for it so that other programms like different upload tools or Commons Android app could use it. --Arnd (talk) 12:02, 1 November 2018 (UTC)Reply[reply]

As a very active Commons categorizer, I would like something like this as it would make our jobs that much easier, at least in sorting by locations. I would add two things that could really grease the skids here, and, I think, wouldn't be difficult to code and implement:
  • Use geotagging, where available, to determine where the picture is (putatively, anyway) and make location suggestions based on that (assuming the appropriate subcategory exists).
  • If the uploader has assigned the image to a location subcategory, and one exists for another category the uploader has chosen, suggest it.

Daniel Case (talk) 03:45, 4 November 2018 (UTC)Reply[reply]

Structured data should make this significantly easier to implement. Wait a year or two more. --Izno (talk) 13:42, 5 November 2018 (UTC)Reply[reply]

This work is really good and fun if this is possible. (Mostafameraji (talk) 13:57, 7 November 2018 (UTC))Reply[reply]