hi, i’m using pretrained resnet34 for my unet encoder, for input images resolution, does it have to be 224x224 or is it okay to use 256x256? thanks!
Classification models provided by
torchvision.models accept inputs with variable input shapes as long as the spatial size is not too small.
Assuming you don’t see any functionality issues, I would guess your approach should work.
Yes, but i have this issue
U-Net segmentation result