I am working in training a CNN in fourier domain. To speed up training, I thought of taking the fft of the entire dataset before training and feeding this data to the sequential model. But inside the first layer of the model, which is a custom Keras layer, the training data is shown to have float32 data type. Does the sequential model take in only real input data? Thanks.
Adding an InputLayer(... dtype='complex64') layer, i.e. an InputLayer() with data type specified as 'complex64' as the first layer of the sequential model allowed me to pass complex64 data to the model.