bobleono / doagan_clean

DOA-GAN: Dual-Order Attentive Generative Adversarial Network for Image Copy-move Forgery Detection and Localization

DOA-GAN

Requirements

• PyTorch-1.4+

Install the required packages by:

pip install -r requirements.txt

Pretrained models

Pretrained models can be downloaded from drive link

Test on USC-ISI

python main.py --dataset usc --ckpt ./ckpt/three_channel.pkl [--plot]

--plot flag will save the output images in fig/ directory.

Test on CASIA/COMO

python main.py --dataset [casia/como] --ckpt ./ckpt/single_channel.pkl [--plot]

Test on Custom folder

Put forged images in images/ folder, and run

python run_on_folder.py --ckpt [model_weight_file] --out-channel [1 or 3]

e.g.,

python run_on_folder.py --ckpt ./ckpt/three_channel.pkl --out-channel 3

The output masks will be saved in fig_test_folder folder.

Citation

@InProceedings{Islam_2020_CVPR,
author = {Islam, Ashraful and Long, Chengjiang and Basharat, Arslan and Hoogs, Anthony},
title = {DOA-GAN: Dual-Order Attentive Generative Adversarial Network for Image Copy-Move Forgery Detection and Localization},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2020}
}