CIFAR 10 mean and standard deviation per channel for inputs

[Chrisatmachine (Chris)]

Hello everyone,

I was wondering if someone could take the time to explain what Jeremy is doing here to find the mean and standard deviation I think its going right over my head.

If someone could show me the math to come to those numbers as a conclusion it would be much appreciated.

[ramesh (Ramesh Sampath)]

This thread might help you -

Image normalization in PyTorch

Hi, I’m working on an image classification problem. I’m working in PyTorch and I need to normalize the images so that they have a mean 0.0 and a standard deviation of 1.0 (reference: https://cs231n.github.io/neural-networks-2/#datapre ). I implemented the mean subtraction and std division, but I stumbled as the network’s behaviour is strange and I’m not sure with my implementation. Now, this is my transforms settings: transform = transforms.Compose([ transforms.ToPILImage(), transform…

[xyz]

Hi, probably there are better ways, but what I often do is something like that: (Not tested, out of my head)

1. create a ImageClassifierData object md with some stats

s = np.asarray( [md.trn_ds.get_x(ind) for ind in range(md.trn_ds.n)])
means = [s[..., channel].mean() for channel in [0, 1, 2]]
stds = [s[..., channel].std() for channel in [0, 1, 2]]

If running out of memory take random subset of indices.