I have a layer called data that takes input image 318x318. When it passes through my convolutional layers, the final size will be 300x300. I would like to combine the 300x300 version of input image to this 300x300 output image from convolutional layers using Elementwise layer in Caffe. Is there any way, I can create this 300x300 image version from 318x318 input image on the fly in caffe?
Or can anybody suggest me the right way of doing this in Caffe ? my approach may be wrong.
I suppose your spatial dimensions are reduced because you are not padding your convolution layers. If you want to precisely fit the input size and the output size, padding the convolutions is probably a better strategy than "resizing on the fly"...