← Egocentric datasets
Food ServiceIn build-out

Commercial Kitchen & Food Service

Line cooking, plating and service under real time pressure, where the same action repeats hundreds of times a shift.

IN BUILD-OUT

Capture for this environment has not started, so there are no frames to show yet. The rule set and record schema are the same as the programmes already running.

What this environment asks for

Commercial kitchens differ from home kitchens in tempo rather than in task. The action vocabulary overlaps heavily, but the same action repeats hundreds of times per shift at a speed that pushes most clips towards the lower boundary limit.

This environment is in build-out. The rule set, review model and schema are the same as the household and retail programmes already running, so capture can start against a known standard rather than a new one.

How the work runs

  • Applying the published rule set unchanged, so records are interoperable with the household set.
  • Segmenting high-tempo repeated actions at the pause between covers.
  • Describing portions and garnishes with the attributes needed to disambiguate near-identical plates.
  • Capturing station hand-offs, which involve two operators rather than one.
  • Recording equipment operation as distinct from the food handling around it.
  • Running the same four-pass review before any batch is released.

What makes it hard

  • Service tempo pushes most actions close to the half-second floor.
  • Near-identical plates make object reference ambiguous without position or state.
  • Steam and heat haze obscure the operating hand at the pass.
  • Hand-offs cross between operators, which the single-operator schema must handle explicitly.
  • Confined line space produces constant large head motion.
  • Shift-length recordings make reviewer consistency the limiting factor.

Result

Programme defined against the existing rule set and schema, with capture partners being onboarded.

What comes out

Same record structure as the live programmes: clip index, timecodes, hand enum, per-hand descriptions, camera motion, scene and review state.