Hi, I am a beginner in deep learning. For the past week I have been wandering how to train a patch-based neural network, since my images are quite big, my basic idea to divide it into patches to utilize the high resolution. My current thought is to design an end to end net, take the patches from one image, go through a feature extractor part then feed into a classifier part, finally output the label of this whole image without patch label.
What I am asking is that how do I design the net? Say I cut one image into 35 patches, should I just concatenate the features from feature extractor part every 35 times and then feed into the classifier part? Can you provide some examples? I can’t find any on google. Thanks a lot in advance.