search

tSchlegl / f-AnoGAN

Code for reproducing f-AnoGAN training and anomaly scoring
MIT License
246 stars 63 forks source link

How can I detect anomaly at pixel-level? #16

Open nnajeh opened 3 years ago

nickk124 commented 3 years ago

The paper describes using pixel-wise residuals between the input/ground truth and the reconstruction.

nnajeh commented 3 years ago

@nickk124 Do you mean the same as the image-level residual MSE(fake_img, real_img) because the function not cleat at the paper?

nickk124 commented 3 years ago

Yes, I think so. See item (4) in section 3.1 of the f-AnoGAN paper, as well as equation (5).

nnajeh commented 3 years ago

Thank you for your help.

nnajeh commented 2 years ago

@nickk124 I think the equation of,pixel_level anomaly detection is the loss L1 NOT MSE, l1(fake_imgs, real_imgs) because in the paper A(x) =|x- G(E(x))|. What do you think please, Thank you.