I have about 16k datapoints each consisting of 4 images that are closely related to each other. I want to run each image through a pre-trained CNN like ResNet and run each output through an LSTM. So for each datapoint ResNet and LSTM would run 4 times and produce one output that gets passed on to some dense layers.
How would I implement this in Keras? Since the dataset is fairly large in size I would like to use something like flow_from_directory so I don't have to load the whole thing into memory.
I'm not sure how to put these pieces together. Thanks