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

low result with only 1 GPU #31

Open
Harley-Sun opened this issue Feb 25, 2021 · 1 comment
Open

low result with only 1 GPU #31

Harley-Sun opened this issue Feb 25, 2021 · 1 comment

Comments

@Harley-Sun
Copy link

Hi Yixiao, I have tried to run code with only 1 2080ti, but got a very low result(eg. 20% mAP drop on market1501), so what causes the performance drop? I can't undersand.
Could you please explain for me?

@yxgeee
Copy link
Owner

yxgeee commented Feb 25, 2021

ReID tasks are very sensitive to the number of GPUs, and I think this is caused by the batch size for BN. For example, in default, I use a batch size of 64 with 4 GPUs, indicating that 16 images are used for calculating mean and std in BN. If you use only one GPU and still 64 images in a batch, 64 images are used for BN. Maybe you could try to decrease the batch size from 64 to 16. But I am not sure if the current lr is the optimal setup for one GPU. You need to tune it. For convenience, I recommend using the default settings (4 GPUs, 64 bs) for reproducing the 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

No branches or pull requests

2 participants