Unet can only classify images of size proportional to training input image size

How can I get my trained U-net to classify my entire image? I have trained the U-net on small training images (128x128 pixels) taken from the image. Now when I use semanticseg() to classify the image, it only works on areas proportional in size to the training image. This is a problem because my entire image is very large and rectangular, not square.

Answer 1

[Srivardhan Gadila]

You can refer to the following Workflow. It uses overlap-tile strategy for the 3D U-Net network and the same can be followed for the U-net network.