Can't load ManTraNet pretrained weight into the model?

[ryanleetakhim]

I tried to run the evaluator.py after downloaded the ManTraNet pretrained_weights at https://github.com/ISICV/ManTraNet/tree/master/pretrained_weights. However, an error appeared as followings:
ValueError: Layer #0 (named "Featex" in the current model) was found to correspond to layer Featex in the save file. However the new layer Featex expects 27 weights, but the saved weights have 26 elements.

[siweic0818]

You need to use some specific tf+keras versions to successfully load ManTraNet weights. Versions that work for me: tf 1.8.0 + keras 2.2.0 (the same as ManTraNet github), tf 1.10.0 + keras 2.2.0, tf 1.13.1 + keras 2.2.4.