Query regarding Data Normalization

Helo everyone,

I have queries regarding datanormalization in CNN. Can you please address my queries?

1)When normalizing image data, Should i use the complete dataset (train + test + validation) to calculate the mean and std ? Or, should i split the dataset(train,test,val) and calculate the mean and std on trainset and use this mean and std to normalize the test and val data?

  1. When using optical folow data to train a convolution autoencoder, should we normalize the optical flow data? If so, can we use the imagenet normalization parameters(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])? Or, do we have to calculate the normalization parameters of the optical flow dataset?

Thank you

© Copyright 2013-2020 Analytics Vidhya