Pose estimation (human)

Body keypoints are returned per person, giving posture, gesture and orientation without identifying anyone. That makes it suitable for safety work, ergonomics and activity analysis in settings where recognising individuals would be neither necessary nor permitted. Keypoints are returned per frame with a confidence value, so unreliable estimates can be discarded rather than acted on. Multiple people in a frame are handled together.

Capacity can be permanent or temporary: a one-off archive can run on burst capacity and release it afterwards, while a continuous workload sits on a cluster reserved for you. Our AI advisors help pick the model and the hardware before anything is committed.