Train Status

This tab displays a dashboard where you can inspect the individual losses for each model throughout training (if and when they apply; for example, the unsupervised losses will only be reported for the semi-supervised model).

https://imgur.com/vbbhGKl.png

Important traces are “train_supervised_rmse” (root mean square error between true and predicted keypoints on training data) and “val_supervised_rmse” (rmse for validation data).

The two models we’ve trained are saved as “YYYY-MM-DD/HH-MM-SS”, for example, “2023-12-01/15-30-00” and “2023-12-01/15-30-01”. The earlier one is the supervised model, and the later one is the semi-supervised.

Note

If you don’t see all your models in that tab, hit the refresh button on the top right corner of the screen, and the other models should appear.

Available metrics

The following are the important metrics for all model types (supervised, context, semi-supervised, etc.):

  • train_supervised_loss: this is the same as train_heatmap_mse_loss_weighted, which is the mean square error (MSE) between the true and predicted heatmaps on labeled training data

  • train_supervised_rmse: the root mean square error (RMSE) between the true and predicted (x, y) coordinates on labeled training data; scale is in pixels

  • val_supervised_loss: this is the same as val_heatmap_mse_loss_weighted, which is the MSE between the true and predicted heatmaps on labeled validation data

  • val_supervised_rmse: the RMSE between the true and predicted (x, y) coordinates on labeled validation data; scale is in pixels

The following are important metrics for the semi-supervised models:

  • train_pca_multiview_loss_weighted: the train_pca_multiview_loss (in pixels), which measures multiview consistency, multplied by the loss weight set in the configuration file. This metric is only computed on batches of unlabeled training data.

  • train_pca_singleview_loss_weighted: the train_pca_singleview_loss (in pixels), which measures pose plausibility, multplied by the loss weight set in the configuration file. This metric is only computed on batches of unlabeled training data.

  • train_temporal_loss_weighted: the train_temporal_loss (in pixels), which measures temporal smoothness, multplied by the loss weight set in the configuration file. This metric is only computed on batches of unlabeled training data.

  • total_unsupervised_importance: a weight on all weighted unsupervised losses that linearly increases from 0 to 1 over 100 epochs

  • total_loss: weighted supervised loss (train_heatmap_mse_loss_weighted) plus total_unsupervised_importance times the sum of all applicable weighted unsupervised losses