This repository contains software to detect computer-generated images, such as those created by Generative Adversarial Network (GAN) algorithms, distinguishing them from typical photos captured by cameras. A computer generated image is a strong indication that the content is fabricated, which is vital to ascertain for digital image forensics.
The slides provide an overview of the approach and the detection results on NVIDIA's StyleGAN3 image generator. This algorithm was one of the detectors evaluated on StyleGAN3 images, see the details here. Overall, this detector was able to achieve a strong performance of 0.92 Area Under the ROC Curve (AUC) metric (1.0 being the perfect score) on this test data.
The model file (resnet101_v3.pt) is available at https://drive.google.com/file/d/1iWjIj_2YEVCHDcE3uPw7c09hz7jRX5un/view?usp=sharing
Command line example:
python ./generated_image_detection/common/generated_image_classification.py --run_test --tile=224 --test_dir $pdata --model_arch resnet101 --batch_size 1 --use_gpu --test_model_name ./models/resnet101_v3.pt --mean 0.485,0.456,0.406 --std 0.229,0.224,0.225
For further information please contact Dr. Arslan Basharat (arslan.basharat@kitware.com).
This software is provided the uses a permissive BSD License.
This research was developed with funding from the Defense Advanced Research Projects Agency (DARPA) under Contract No. HR001120C0123. The views, opinions and/or findings expressed are those of the authors and should not be interpreted as representing the official views or policies of the Department of Defense or the U.S. Government.
Distribution A: Approved for public release: distribution unlimited.
