Continuous prediction values for semantic segmentation pixels

[NoaMillsUSDA-ARS]

I'm running a semantic segmentation model that predicts two classes. I am able to construct a prediction raster of pixels with values 0 and 1. I know that internally, in order for backprop to work, the algorithm creates a prediction vector for each pixel specifying the pseudo-probability that the pixel is each class value (ie [0.3, 0.7] if the algorithm is 70% certain the pixel is class 2). Is it possible to extract this information from the model, instead of just getting binary values for each pixel?

[lewfish]

If you pass smooth_output = True to SemanticSegmentationLabelStoreConfig, you should get the probabilities in the output. See the docs