Skip to content
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.

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

Add support for STL10 at resolutions 32, 48, and 96 #25

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

Conversation

ilyakava
Copy link

Adds WideResnet at sizes 48, 96.
Adds dataloading for entire STL10 dataset.
Moves transforms to dataset/ code.
Adds results and instructions to README.md.

class WideResNet96(nn.Module):
def __init__(self, num_classes, depth=28, widen_factor=2, dropRate=0.0):
super(WideResNet96, self).__init__()
nChannels = [16, 16*widen_factor, 32*widen_factor, 64*widen_factor, 64*widen_factor]

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

should this line be "nChannels = [16, 16widen_factor, 32widen_factor, 64widen_factor, 128widen_factor]"?

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This change doubles the parameter count from 2.67M to 5.93M. I started training the larger model now since I don't remember the performance difference from a few months ago. I'll post the results soon.

Copy link
Author

@ilyakava ilyakava Mar 25, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@chencjGene It doesn't provide any gains. Last layer 64 = max top1 91.8 vs last layer 128 = max top1 91.76. The larger network trains slightly slower and starts to overfit sooner. wider results, narrower results

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants