I believe the implementation for U-Net in this repository uses padded convolution. The U-Net paper however mentions the use of Overlap-Tile Strategy and Cropping.
Is there a possibility of adding Overlap-Tile Strategy to this code?
What would be the benefits of using this over padded convolutions?
I believe the implementation for U-Net in this repository uses padded convolution. The U-Net paper however mentions the use of Overlap-Tile Strategy and Cropping.
Is there a possibility of adding Overlap-Tile Strategy to this code? What would be the benefits of using this over padded convolutions?