Skip to content

docs: clarify normalization statistics in Knowledge Distillation tutorial #3928

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Charuhasini30 wants to merge 1 commit into
base: main
Choose a base branch
from
Open

docs: clarify normalization statistics in Knowledge Distillation tutorial #3928

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 3 additions & 3 deletions beginner_source/knowledge_distillation_tutorial.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -56,9 +56,9 @@
# A common practice in neural networks is to normalize the input, which is done for multiple reasons,
# including avoiding saturation in commonly used activation functions and increasing numerical stability.
# Our normalization process consists of subtracting the mean and dividing by the standard deviation along each channel.
# The tensors "mean=[0.485, 0.456, 0.406]" and "std=[0.229, 0.224, 0.225]" were already computed,
# and they represent the mean and standard deviation of each channel in the
# predefined subset of CIFAR-10 intended to be the training set.
# The tensors "mean=[0.485, 0.456, 0.406]" and "std=[0.229, 0.224, 0.225]" are commonly used
# normalization values. Note that these values correspond to ImageNet statistics rather than
# statistics computed directly from the CIFAR-10 training set.
# Notice how we use these values for the test set as well, without recomputing the mean and standard deviation from scratch.
# This is because the network was trained on features produced by subtracting and dividing the numbers above, and we want to maintain consistency.
# Furthermore, in real life, we would not be able to compute the mean and standard deviation of the test set since,
Expand Down