Extension source: https://gerrit.wikimedia.org/r/plugins/gitiles/mediawiki/extensions/MachineVision
Description of extension
The MachineVision extension is being developed to support on-wiki usage of AI-generated image metadata. Specifically, it handles:
- Requesting AI-generated image metadata from machine vision providers
- Providing storage for AI-generated image metadata
- Serving AI-generated data to users for verification (and promotion to Structured Data on Commons) and recording verification results
Though the expectation is that we will be working with third-party providers in the near term, it is designed to work with any number of internal or external machine vision providers.
The initial use case for the MachineVision extension is computer-aided tagging. Label suggestions will be requested from a third-party provider for a pool of existing, high-quality Commons images, the provided labels will be translated into Wikidata IDs, and users will have the opportunity to confirm or reject the suggested labels. Suggested labels will also be requested on upload as images are uploaded, and the uploaders will be notified of the opportunity to review suggested labels. SDC depicts statements will be added to images for approved labels.
Target date for enabling in production: week of October 28, 2019
Basic checklist:
- Brief description of what it does as perceived by end-users (e.g. display X, offer X to use when doing Z)
- Analysis of backend system performance and ensuring metrics/monitoring/grafana is in place. Assigned to: @aaron.
- Analysis of perceived performance by end-users (e.g. load/save time, general responsiveness). Assigned to: @Krinkle.
- Analysis of impact (if any) on site-wide metrics (startup bundle size, Wikipedia page load time). Assigned to: TBD.