Model with loss function over multiple inputs

Question

I want to build a convolutional network which performs regression using multiple images. The network should first work on each image separately and produce a number. The num