about training

[Sh1p-wy]

hi,
Are the two networks (image manipulation tract feature extractor and local anomaly detection network) trained separately?

[RonyAbecidan]

Hello again,

After re-reading the article, you're right, they have been trained separately on different datasets before being merged.

[Sh1p-wy]

@RonyAbecidan lol, thanks again.
I would like to know the model details, do you know what is the full model for both networks?
I was able to successfully train the mantranet using your code, but the results are not so good, and I would like to train the two models correctly and separately.

[Sh1p-wy]

after re-reading the paper, I guess that the image manipulation tract feature extractor is the same as VGG net.
but, I still do not know the structure of the second network.