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

scrfd 10 false positive #124

Open
MyraBaba opened this issue Jan 10, 2024 · 1 comment
Open

scrfd 10 false positive #124

MyraBaba opened this issue Jan 10, 2024 · 1 comment

Comments

@MyraBaba
Copy link

Hi,

I used 640x640 scrfd_10g_kps.onnx . it gives time to time false positives as I attached.

yours scrfd is scrfd_10g_gnkps.onnx and dynamic heightxweight.

Is there any tuning to prevent false positive (face score 0,77 for below one) or is it make a difference to use your dynamic model ?

@SthPhoenix
Copy link
Owner

SthPhoenix commented May 14, 2024

Hi,

I used 640x640 scrfd_10g_kps.onnx . it gives time to time false positives as I attached.

yours scrfd is scrfd_10g_gnkps.onnx and dynamic heightxweight.

Is there any tuning to prevent false positive (face score 0,77 for below one) or is it make a difference to use your dynamic model ?

Hi! Unfortunately all models available now, as well as models not yet implemented here (yolo_v7 - yolo_v9) tends to give false positives from time to time. You should just try different models and select one which better suits your use case, as different models may have different results on different image types. For now the best supported model in terms of accuracy\recall is yolo_v5l_face.
Also you can try playing with size and norm attributes of results for better filtering. Something like:

if score > 0.7 and size > 30 and norm > 12:
   ...

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