Recognize Action Skeleton (PoseC3D)

@kennymckormick1 VersionClassify

Buffers a sliding window of per-frame Landmarks2d.Human22 skeletons, renders them into pseudo-heatmap volumes, and runs a PoseC3D 3D-CNN to emit a ranked [Classification] or a single top-1 Classification of NTU RGB+D 60 actions.

How it fits

Typical backends

Live action telemetry. Label a body-landmark stream into the expected skeleton layout, recognise the action, and publish each result over HTTP.

Action-triggered alert. Recognise actions from a landmark stream and gate them so only the actions of interest pass downstream.

Caveats

State lifetimeRecognition is temporal — it consumes one person's skeleton per fire and holds an internal sliding window that must fill before any real action is produced; the buffer persists across fires until the container restarts.
FallbackBefore the window fills, and on any frame with too few usable keypoints, a defined no-action placeholder is emitted so a single-output vertex always produces exactly one message per fire.
AccuracyAction labels are drawn from the NTU RGB+D 60 taxonomy, so an action outside those classes is reported as the closest in-vocabulary label; confirm the taxonomy covers the deployment before trusting the top label.
Hard constraintThe checkpoint was trained on roughly frontal indoor capture, so extreme camera angles or heavy body truncation degrade accuracy.

Versions

19f8b7c4latestdefault
Automated release
6/25/2026