Skip to content

PoseKit Labeler

PoseKit is now organized under hydra_suite.posekit and launched via posekit-labeler.

Full UI Control Reference

For complete PoseKit interface documentation (every form control, question labels, defaults, value-selection guidance, and failure modes), see:

  • /reference/ui-components-posekit/

Capabilities

  • Project setup wizard (classes/keypoints/skeleton)
  • Frame-by-frame keypoint annotation
  • Autosave with safer write semantics
  • Metadata/tagging and smart frame selection
  • Split generation and training/evaluation dialogs

SLEAP Backend Setup

For SLEAP env creation and ONNX/TensorRT integration setup, see:

  • /getting-started/integrations/

Key UX Concepts

  • Keypoint order is semantic and must remain stable.
  • Visibility flags encode present/occluded/missing states.
  • Project settings changes can require label migration.

Core Hotkeys

  • A/D: previous/next frame
  • Q/E: previous/next keypoint
  • Ctrl+S: save
  • V/O/N: visible/occluded/missing mode

Output Expectations

  • YOLO pose labels are normalized text files per image.
  • Project metadata and optional posekit artifacts are stored in project output directories.
  1. Finalize keypoint spec before large-scale labeling.
  2. Label a pilot subset and run sanity checks.
  3. Generate split files and train a baseline.
  4. Use model-assisted passes and active learning to iterate.