Project Information
- Name of project: MachineVision
- Project home page: https://www.mediawiki.org/wiki/Wikimedia_Product/Machine_vision_middleware
- Name of team which owns the project: Reading Infrastructure
- Primary contact for the project: Michael Holloway
- Target date for deployment: August 30, 2019
- Link to code repository: https://gerrit.wikimedia.org/r/plugins/gitiles/mediawiki/extensions/MachineVision
- Is this a brand-new project: Yes
- Has this project ever been reviewed before: (Phab tasks, etc.): No
- Has any risk assessment (STRIDE, etc.) been performed: No
- Is there an existing RFC or has this been presented to the community: No
- Is this project tied to a team quarterly goal: Yes
- Does this project require its own privacy policy: No
Description of the project and how it will be used
This is a project to support the incorporation of machine vision (MV) generated metadata into Foundation products. Specifically, the project will support:
- Requesting MV-generated image metadata from machine vision providers (both internal and third-party external)
- Providing temporary storage for MV results pending human editor verification
- Serving MV data to Commons users for verification and promotion to Structured Data on Commons
- Providing the results of human editor verification back to third-party MV providers for model refinement`
For full discussion, see the project page at https://www.mediawiki.org/wiki/Wikimedia_Product/Machine_vision_middleware (including linked Phab tickets).
Description of any sensitive data to be collected or exposed
None
Technologies employed
MediaWiki, MySQL, possibly (<50% likelihood) a Node.js service similar to cxserver
Dependencies and vendor code
- https://github.com/rahiel/open_nsfw--
- Third-party machine vision providers (e.g., Google Cloud Vision, Clarifai)
Working test environment
This project in an early stage, but a couple of dev/testing APIs are set up in WMCS:
- An instance of the proposed NSFW image scoring service is running at https://nsfw.wmflabs.org. (See build and usage instructions at https://github.com/rahiel/open_nsfw--.)
- A dev version of an API providing an image labeling response, illustrating the kinds of data we'll be working with, is available at https://visionoid.wmflabs.org.