Skip to content

Pose Training

End-to-end pipeline for training custom pose estimation models.

Capabilities

  • Converters: CVAT XML, Lightning Pose, COCO keypoints/points/localizer formats
  • Training: YOLO pose, POLO point-detection, PyTorch localizer heatmap models
  • Inference: Evaluate trained models on video, export to DataFrames
  • Prep: Dataset splitting, label filtering/simplification, data.yaml generation

pose_training

Custom model training pipeline for pose estimation, point detection, and localization.

Converters: transform annotation formats to YOLO pose / POLO point / localizer labels. Training: train YOLO pose, POLO point-detection, or localizer heatmap models. Inference: test trained models on video (non-production; production uses TRex).

Requires optional dependency

Pose: pip install mosaic-behavior[pose] POLO: pip install mosaic-behavior[polo] Localizer: pip install mosaic-behavior[localizer]