Page MenuHomePhabricator

Refactor face detection to be done on-demand
Closed, ResolvedPublic2 Estimated Story Points

Description

New image caching work (T95350) has obsoleted the old disk storage, which effectively breaks centering lead images w/ faces. This is because the current implementation relies on a side-effect of saving data to disk which is no longer being triggered (since we're not writing data to the data store).

Acceptance criteria

  • Images in article "preview" gallery w/ faces are centered
  • Face "feature" bounds are cached as unit rectangles and keyed on WMF commons filename

Details

Related Changes in Gerrit:

Event Timeline

BGerstle-WMF raised the priority of this task from to Needs Triage.
BGerstle-WMF updated the task description. (Show Details)
BGerstle-WMF moved this task to Needs Triage on the Wikipedia-iOS-App board.
BGerstle-WMF subscribed.

This sounds like a regression. Why recalculate it each time?

@Mhurd I think we discussed this in person already, but just reiterating here:

Use existing architecture to calculate focal offsets on demand (if not stored on the MWKImage already)

Change 226083 merged by jenkins-bot:
async detect faces & center w/ content offset

https://gerrit.wikimedia.org/r/226083