Loss Function for Semantic Segmentation

What is the loss function for semantic segmentation or U-Net algorithm?

This is from the Image_segmentation_Unet assignment.

“as many masks as object classes”… what is a mask?

I believe the mask here refers to an image mask (ie. a mask of colors, with each color representing a different class).

You can think of the image mask as the color coded image you see in the predictions in the assignments.

1 Like