Skip to content

EXJUSTICE/StyleGAN_MesoscopicNeuralNetworks

Repository files navigation

Mesoscopic Neural Networks for Synthetic Image Identification

Inception-powered Synthetic Image Detection based on Mesoscopic Neural Network Approach. Based on Afchar's apublication "MesoNet: a Compact Facial Video Forgery Detection Network".

Detailed tutorial, "Seeing is Believing — Mesoscopic Neural Networks for Synthetic Image Detection", available on GradientCrescent.

Structure

The are twoIpython Notebooks in this repository, covering the Meso4 and Meso4-Inception architectures, as well as the InceptionV3 architectures. These two notebooks are completely independent. Note that the InceptionV3 model has been commented out in both books - to run a benchmark, uncomment these and comment out the mesoscopic architecture instead.

Dataset

500 images of FFHQ and StyleGAN datasets, representing true and synthetic data classes, respectively, were used to evaluate our models. Data is temporarily stored on a Google Drive, but is also available on the NVIDIA StyleGAN repository

About

Synthetic Image Detection based on Mesoscopic Neural Network Approach by Afchar et al.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors