i have been reading through deep learning with pytorch book, and i have a small question.
in this code we are trying to calculate the mean and std of the pixels to normalize them
n_channels = batch.shape for c in range(n_channels): mean = torch.mean(batch[:, c]) std = torch.std(batch[:, c]) batch[:, c] = (batch[:, c] - mean) / std
what is the meaning of batch[:, c] given that the channels have the 2nd dimension, if the channel dimension has another position like the last dimension how can i calculate the mean across the channels??