It says in the lab description of U-Net that U-Net can accommodate for various input sizes. Why is this the case?
great question. Because it is an end-to-end fully convolutional network (FCN), i.e. it only contains Convolutional layers and does not contain any Dense layer, because of which it can accept image of any size.
I recommend that you read this article.
In TensorFlow we need to create an Input layer and pass image size as an argument. So, in terms of implementation, how can we use variably sized training / test images without resizing them?
Rosa, many thanks for the article link. It is very well written and turned out to be quite useful!