missing normalization before inference

[goteguru]

Shouldn't we normalize image data before inference?
It seems in facial_detection_recog_emotion.ipynb In [23] we are feeding uin8 grayscale data into the network whereas the network was trained for float [0:1] normalized data.