I’m building a segmentation model by following the tutorial. Is it possible to create an U-Net that doesn’t output a mask but instead the masked image itself?
1 Like
If you want to do this from the model itself you need to use an additional output where you will multiply the original input image with the binarized output mask.
If you don’t need to do this in the model you can do the same as a post-processing step.
2 Likes
Thank you @Bhack, I was thinking that is possible to train it using the complete image and the segmented one avoiding the use of the binarized mask. I will follow your tip!
Yes It is possible but I don’t think It Is convenient