Your library can do 3D convolutional auto keras.
Input images are 3D images, such as 3D CT scan images.
If you add this feature, it will benefit many people.
3D-MINST
It could also be interesting to explore separable convolution. See https://github.com/tensorflow/hub/issues/49.
@bhack Thanks! Very good suggestion.
Check also https://arxiv.org/abs/1807.11195
And https://arxiv.org/abs/1807.11473