HeatmapMSELoss

class lightning_pose.losses.losses.HeatmapMSELoss[source]

Bases: HeatmapLoss

MSE loss between heatmaps.

Attributes Summary

loss_name

Methods Summary

compute_loss(targets, predictions)

Attributes Documentation

loss_name = 'heatmap_mse'

Methods Documentation

compute_loss(targets: Tensor, {'__torchtyping__': True, 'details': ('batch_x_num_keypoints', 'heatmap_height', 'heatmap_width'), 'cls_name': 'TensorType'}], predictions: Tensor, {'__torchtyping__': True, 'details': ('batch_x_num_keypoints', 'heatmap_height', 'heatmap_width'), 'cls_name': 'TensorType'}]) → Tensor, {'__torchtyping__': True, 'details': ('batch_x_num_keypoints', 'heatmap_height', 'heatmap_width',), 'cls_name': 'TensorType'}][source]
__init__(data_module: BaseDataModule | UnlabeledDataModule | None = None, log_weight: float = 0.0, **kwargs) → None[source]
Parameters:

  • data_module – give losses access to data for computing data-specific loss params

  • epsilon – loss values below epsilon will be zeroed out

  • log_weight – natural log of the weight in front of the loss term in the final objective function

__new__(**kwargs)