Metrics

ATOMMIC provides a number of metrics for each task to evaluate the performance of the models. The metrics are implemented as classes that can be instantiated and called with the desired inputs. Depending on the chosen task, the corresponding metrics will be also logged on the selected logger.

In tools, you can find scripts that allows you to evaluate the performance of a model on a dataset. The scripts take as input the ground truth and the predictions of the model and compute the metrics for each task.

The metrics are implemented in the following modules:

• mse():

  Mean Squared Error (MSE) metric for reconstruction, quantitative, and multitask tasks.

• nmse():

  Normalized Mean Squared Error (NMSE) metric for reconstruction, quantitative, and multitask tasks.

• psnr():

  Peak Signal-to-Noise Ratio (PSNR) metric for reconstruction, quantitative, and multitask tasks.

• ssim():

  Structural Similarity Index (SSIM) metric for reconstruction, quantitative, and multitask tasks.

• ReconstructionMetrics:

  Class that wraps all the metrics for reconstruction, quantitative, and multitask tasks.

• asd():

  Average Surface Distance (ASD) metric for segmentation and multitask tasks.

• binary_cross_entropy_with_logits_metric():

  Binary Cross Entropy with Logits (BCE) metric for segmentation and multitask tasks.

• dice_metric():

  Dice metric for segmentation and multitask tasks.

• f1_per_class_metric():

  F1 per class metric for segmentation and multitask tasks.

• hausdorff_distance_metric():

  Hausdorff Distance (HD) metric for segmentation and multitask tasks.

• hausdorff_distance_95_metric():

  95th percentile of the Hausdorff Distance (HD95) metric for segmentation and multitask tasks.

• iou_metric():

  Intersection over Union (IoU) metric for segmentation and multitask tasks.

• precision_metric():

  Precision metric for segmentation and multitask tasks.

• recall_metric():

  Recall metric for segmentation and multitask tasks.

• surface_distances():

  Surface Distances (SD) metric for segmentation and multitask tasks.

• SegmentationMetrics:

  Class that wraps all the metrics for segmentation and multitask tasks.