Hi,
There is a case like this: I firstly tuned my segmentation model with shape of input tensor to be (64, 3, 512, 512), but I finally used an input with shape (128, 3, 256, 256). There is no other changes in this case. Should I re-tune my model from scratch in this way ?