Overview
A REST API over Skylark's CFAM detection models. Send an image or a short clip, get back bounding boxes, classes and confidences.
Every model is addressed by a solution id — a task plus a viewpoint, such as
cctv_outdoor_vehicle_detection. The catalog lists them all, along with the
classes each one returns and what a call costs.
Three things worth knowing first
A single image is not always a single inference call. Tracking backbones need a few priming frames before they emit anything, so some solutions cost four frames per image rather than one. The catalog states the cost per model; the playground shows it next to each one.
Clips are asynchronous. Submitting a video returns a job id to poll. The response tells you how long to wait between polls, and that figure rises with the queue — honouring it is how a client shares backpressure instead of adding to it.
Switch on the error code, never the message. The code field is part of
the contract; the prose is not, and will change.
Nothing is stored
Uploads are decoded, analysed and discarded. The service logs no filename and no file content, because an uploaded filename is itself user data.