load_model_from_checkpoint

lightning_pose.utils.predictions.load_model_from_checkpoint(cfg: DictConfig, ckpt_file: str, eval: bool = False, data_module: BaseDataModule | UnlabeledDataModule | None = None, skip_data_module: bool = False) → HeatmapTracker | SemiSupervisedHeatmapTracker | HeatmapTrackerMHCRNN | SemiSupervisedHeatmapTrackerMHCRNN | HeatmapTrackerMultiviewTransformer | SemiSupervisedHeatmapTrackerMultiviewTransformer | RegressionTracker | SemiSupervisedRegressionTracker[source]

Load Lightning Pose model from checkpoint file.

Parameters:

  • cfg – model config

  • ckpt_file – absolute path to model checkpoint

  • eval – True for eval mode, False for train mode

  • data_module – used to initialize unsupervised losses

  • skip_data_module – if data_module is not None this is ignored. If False and data_module=None, a data module is created from the config file and unsupervised losses are accessible in the model. If True and data_module=None, the unsupervised losses are not accessible in the model; this is recommended for running inference on new videos

Returns:

model as a Lightning Module