search

ISICV / ManTraNet

ManTra-Net: Manipulation Tracing Network For Detection And Localization of Image Forgeries With Anomalous Features
269 stars 70 forks source link

How to use for classification #7

Closed chris-clem closed 3 years ago

chris-clem commented 4 years ago

Hi,

How can I use ManTraNet for fake image classification? I was thinking about using a threshold for the sum of the ManTraNet output, but the output for real images show quite some noise:

image

Is there a better way to do it?

22wei22 commented 3 years ago

How to use ManTraNet for fake image classification? Do you have ideas?

rex-yue-wu commented 3 years ago

I don't know what is your application. But it is clear that the first image is out of the scope that the ManTraNet can handle -- it is already processed by concatenating two images together.

For fake image classification, I think you need to build additional modules to take the ManTraNet intermediate feature (not the averaged feature, but the deviations) as the input for a classification network.