Pose and depth
Attach a model to the camera and its output appears as a track: pose as JSON keypoints, depth as a second video. Each attachment is placed on a node that serves that model while the camera is published once. Detach and the track stops. Nothing is deployed in between.
LIVE
or publish a camera into this session
Record the camera with 5 s of pre-roll; it appears on the recordings page.
import rgb # RGB_API_KEY, RGB_ORIGIN from the environment
session = rgb.sessions.create(name="pose-demo")
cam = session.source("cam0")
pose = session.attach("rtmpose-v1", to=cam, as_="pose")
depth = session.attach("depth-anything-v2-small", to=cam, as_="depth")
pose.watch(draw_skeleton) # JSON keypoints on the camera's clock
session.watch_video("cam0/depth", draw) # a model output that is video
Session graph
Model outputs
pose not attached
depth not attached