Recognition corrections

Recognition runs on the server, but the desktop app is a great place to turn it on for an album and to fix up the results while you cull. The pipelines and corrections here mirror the web — for the full picture of what each pipeline does, see the recognition chapters in the Photographer track.

Turning pipelines on

Open an album's edit modal and find Recognition pipelines. The parent toggle is 👤 People detection; enabling it reveals the sub-options:

• 🎯 Read numbers (jersey, bib, car, sail, saddle, bike)
• 🏁 Match athletes to their team
• 🚗 Vehicle detection
• ⛑ Helmet-paint recognition

Running and re-running recognition

The album hero has a recognition pill with a ▾ menu. Depending on which pipelines are on, you'll see actions like:

• Detect faces / Re-detect faces
• Re-match faces against your registry
• Re-run helmet detection
• Re-run subject / livery clustering
• Resume identifier OCR
• Derive teams — re-aggregate team labels from your clusters and number bindings

Actions your plan doesn't include show an upgrade prompt instead of running.

Fixing clusters

On an album's People tab, recognition groups similar faces (and subjects) into clusters:

• Name a cluster to register that person — they then match across every album (see Athlete registry).
• Team pins a cluster to a team.
• You can also rename and merge clusters when recognition splits one person into two.

In the loupe, on-screen tools let you confirm or reject a suggested identity (Y / N), tag a subject, or ignore a cluster — handy for correcting as you cull rather than in a separate pass.