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

bugs here #27

Closed
bertjiazheng opened this issue Nov 6, 2017 · 3 comments
Closed

bugs here #27

bertjiazheng opened this issue Nov 6, 2017 · 3 comments

Comments

@bertjiazheng
Copy link

It seems a bug in color_normalization

t.sub_(m).div_(s)

@bearpaw
Copy link
Owner

bearpaw commented Nov 6, 2017

Hi @bertjiazheng . It seems that original hourglass network didn't subtract mean.

I follow the implementation of convolutional pose machine to subtract mean only: https://github.com/shihenw/convolutional-pose-machines-release/blob/afda94fe0d53eb77c5fe948e607f199791f4073c/testing/src/applyModel.m#L94

You may also try to divide the std and see whether it helps improve the performance. Feedbacks are always welcome!

@wangziren1
Copy link

wangziren1 commented Nov 13, 2017

Hi @bearpaw, why don't you use original hourglass network that didn't subtract mean? I also see your hg-8
training log and find your accuracy increase more slowly than original paper. Does this result from nomalization i.e.image subtracting mean?

@bearpaw
Copy link
Owner

bearpaw commented Nov 13, 2017

@wangziren1 Please try so if you are interested and give us a feedback. Thanks very much!

@bearpaw bearpaw closed this as completed Sep 12, 2018
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

No branches or pull requests

3 participants