Semantic segmentation when the labels are in a matrix

I'm trying to perform semantic segmentation with segnet using

lgraph = segnetLayers(imageSize,numClasses,encoderDepth);

And my dataset is in the form of matrixes and not of images.

I keep getting an error message from trainNetwork:

"Invalid training data. Y must be a vector of categorical responses."

I don't understand how to format the labels as vectors when the segmentation is pixel-wise! The output has as many pixels as the input and not a single catagory. All the examples are for reading from files..

Thanks in advance to anyone who can help..