While automated descriptions are not yet of a quality that can replace human generated ones, they may have a logical place for search or related content scenarios. Using machine generated descriptions, and tags to inform search results and related content results could benefit users even if they are not surfaced to users directly.
Possible resources
http://deeplearning.cs.toronto.edu/i2t
https://github.com/Visu4link/Pastec
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Open | None | T49492 Automatically propose/suggest a category for images | |||
| Open | None | T76886 Investigate computer vision image classification and description tools for shadow tags and search descriptions |
Even if the description is not usable directly, this could be also interesting to suggest categories for the user to add.
Face detection could be interesting when combined with T58666, to invite people to add annotations.
http://www.pastec.io/ claims to have a database built from Commons images, might be interesting to test.
Might be also interesting for visual captchas (T64960) although using a publicly available algorithm is always a weakness there.
Seems related to https://commons.wikimedia.org/wiki/User:Multichill/Using_OpenCV_to_categorize_files (see talk page too). Good hackathon project for Prague?