Python ChannelNormalize have different parameter order with scala ChannelNormalize

[jenniew]

Python ChannelNormalize initializer:
def init(self, mean_r, mean_b, mean_g, std_r=1.0, std_g=1.0, std_b=1.0, bigdl_type="float"):
super(ChannelNormalize, self).init(bigdl_type, mean_r, mean_g, mean_b, std_r, std_g, std_b)
The order is: mean_r, mean_b, mean_g, std_r, std_g, std_b
Scala ChannelNormalize initializer:
def apply(meanR: Float, meanG: Float, meanB: Float,
stdR: Float = 1, stdG: Float = 1, stdB: Float = 1)
The order is: meanR, meanG, meanB, stdR, stdG, stdB

[yiheng]

Do you mean we invoke the python ChannelNormalize incorrectly in the example?

[jenniew]

Yes

[wzhongyuan]

@jenniew

yes, the order is different.... not sure why it's done like this...

I will run another with this fixed to see if we have a better result once there is available cluster resource