Zero-shot metric monocular depth (Apple Depth Pro): one forward over an Image yields a deploy-time-resolved sharp MetricDepthMap in metres or its greyscale Image visualisation, with no camera metadata needed upstream.
Typical backends
A single uncalibrated photo stream is turned into absolute metric depth and shipped as structured data over HTTP for measurement or back-projection.
The greyscale visualisation arm is streamed from an HTTP image source to an image sink to preview the depth structure of an otherwise uncalibrated scene.
Caveats
- I/O contractThis is a single-output component: the connected sink resolves whether it emits the
MetricDepthMap in metres or its greyscale Image visualisation. Exactly one product is produced per frame. - AccuracyDepth is zero-shot absolute metric depth needing no intrinsics; the
Image arm is the per-frame min-max normalisation of that same metric map to 8-bit greyscale, so it shows relative structure, not absolute metres. - Hard constraintThe Depth Pro checkpoint is bound to the worker as a local artifact and loaded offline at startup; there is no download at runtime, so a cold container needs no outbound network before the first frame is served.
- State lifetimeThe recompute cadence is configurable: unset recomputes every frame, zero computes once then frees the model and replays that result forever, and a positive value recomputes every N seconds, freeing the model between passes.
- CostThe multi-scale ViT strongly prefers a GPU; a cuda device silently falls back to CPU when none is visible, but CPU inference is impractically slow.
- State lifetimeAll configuration is captured once at startup; runtime changes have no effect and require a redeploy.