Dataset Masks

Masks complement labels by removing unwanted pixels before dataset samples are generated. They are useful for clouds, saturation, sensor errors, outliers, water, shadows, or other pixels that should not train or validate a model.

Datasets apply source image masks by default during finalization. Use the dataset's Masks toggle only when you want finalization to ignore image masks for that dataset version.

Manage mask usage, effect, and type on the source image in Spectral Explorer before finalizing the dataset.

Dataset finalization follows the same source pipeline used by other downstream products:

image data -> image masks used in processing -> dataset samples

V2 layer masks are derived mask layers. They can be displayed and used as inputs to processing operations, but dataset finalization uses image masks unless a workflow explicitly turns a derived mask into a source-level image mask.