Metrics

Metrics are used to evaluate model performance. Frogbox comes with a wide range of metrics through torchmetrics.

For instance, to use SSIM to evaluate a supervised model, one can add the SSIM metric from torchmetrics like this:

metrics:
  SSIM:
    object: torchmetrics.image.StructuralSimilarityIndexMeasure
    kwargs:
      data_range: 1.0
      kernel_size: 11
      sigma: 1.5

Implementing custom metrics

Custom metrics must subclass torchmetrics.Metric.

See this guide for a detailed description of how to implement custom metrics.