I want to build a customized convolutional layer in Keras that does convolution for two kernels and one image, I want it to be implemented in my network so I finally have a fully connected network and can be learning end-to-end.
In other words, I just want a trainable layer that takes two filters and one image as inputs and give one image as output... the output image is equal to the convolution of the 3 inputs ... that is everything I want.
Thanks.