InferShape cannot decide shapes when adding a Dropout layers

[gregoirerllr]

Framework version:
keras-mxnet==2.2.4.1
mxnet-cu90mkl==1.4.0

I am creating a functional model with Keras. Each time I try to add a Dropout layer, I get the following error:
RuntimeError: simple_bind error. Arguments: [14:28:22] src/executor/../common/exec_utils.h:392: InferShape pass cannot decide shapes for the following arguments (0s means unknown dimensions). Please consider providing them as inputs: /input_511: [0,299,299,3],

I tried to force the input_shape of the Dropout layers, but it still fails.

The model compiles when all Dropout layers are removed.

[WeberJulian]

I have the same problem with mxnet-mkl, and my model works on both the tensorflow and the plaidml backend