The program is very helpful for data of the same size. But when the size of the images is not constant, it could be interesting to have a layer which computes fixed length representation. Do you plan to add it in a next version? If yes, when? You can find a link for GitHub lib (keras) and the paper.
This looks useful and we will explore the support once DLS is released with Keras 2.X support. Currently Keras 2.X support is on hold while we deliver other high priority features.