I am working on a problem, wherein I am provided with the training images and each image has two masks for two different categories. How should I segment the images and then do pixel-wise classification to get the output image with 0 for background, 1 and 2 representing the two categories?

I tried combining the masks and then feed the mask and the corresponding image to the segmentation model, but somehow I failed in doing so. Should I even go for combining the masks, since we need to classify the pixels as well in the output.