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

PaDiM Shared Performance #7

Open
ingbeeedd opened this issue Dec 15, 2021 · 4 comments
Open

PaDiM Shared Performance #7

ingbeeedd opened this issue Dec 15, 2021 · 4 comments

Comments

@ingbeeedd
Copy link

How much is the difference in performance compare to original PaDiM?

@Pangoraw
Copy link
Owner

I think it really depends on the kind of images you are dealing with.

For texture images where the normal data has a small number of modalities then a shared variant may be good enough.

For objects where the location is the patches are very different based on their location on the image then it won't work as well.

Doing a quick test on 2 mvtec datasets:

Dataset PRO Score ROCAUC
Tile (texture) 86% 84%
Pill (objects) 82% 44%

The main advantage being the much smaller memory footprint and not really performance.

@ingbeeedd
Copy link
Author

How much did you test batch size? Wouldn't the performance be close if it was a little large test batch size?

@Pangoraw
Copy link
Owner

I won't expect the batch size to have any impact on the performance since batch samples don't interact with each other. What impact do you think it could have ?

@ingbeeedd
Copy link
Author

I'll leave it after a test

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