How can I get the statistics automatically for Normalize()

#1

Since last year, I’ve been using means and standard deviations that are far from my dataset and would like to know how to compute them in pytorch.

#2

Hi @Omar_Zayed,
you can just iterate over your dataset, collect data and then compute statistics.

Here an example with MNIST dataset:

from torchvision.datasets import MNIST
from torchvision.transforms import ToTensor

from torch.utils.data import DataLoader
import torch

dataset = MNIST(root=".", download=True, transform=ToTensor())
dt = DataLoader(dataset, batch_size=8)

data = []

for sample, label in dt:
  data.append(sample)

data_stacked = torch.stack(data, dim=0)

# stats per channel:
mean = data_stacked.mean(dim=[0, 1, 3, 4])
std = data_stacked.std(dim=[0, 1, 3, 4])

print(f"mean: {mean}")
print(f"std: {std}")