Official Pytorch implementation of CVPR 2022 paper "Proactive Image Manipulation Detection ".

Vishal Asnani, Xi Yin, Tal Hassner, Sijia Liu, Xiaoming Liu

The paper and supplementary can be found at:

• PyTorch 1.5.0
• Numpy 1.14.2
• Scikit-learn 0.22.2

• Every GM is used with different datasets they are trained on. Please refer to Table 2 of the supplementary for GM-dataset information. Download the dataset for the corressponding GMs from https://drive.google.com/file/d/1fAS7Sj3FhS6v31Z2hb9mp9gaGavgnLu5/view?usp=sharing
• The training data is used as CELEBA-HQ which is provided in the above link as CELEBA_HQ_TRAIN folder.

The pre-trained model trained on STGAN can be downloaded from: https://drive.google.com/file/d/1p9zETa9rCU0wx8wD5Ige2TbCL8WciV7o/view?usp=sharing

• Go to the folder STGAN
• Download the STGAN repository files and pre-trained model from https://github.com/csmliu/STGAN
• Provide the train and test path in respective codes as sepecified below.
• Provide the model path to resume training
• Run the code as shown below:

python train.py

• Download the repository files and pre-trained model of GMs in the respective folder, StarGAN: https://github.com/yunjey/stargan , CycleGAN: https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix , GauGAN: https://github.com/NVlabs/SPADE
• Download the pre-trained model for our template from https://drive.google.com/file/d/1p9zETa9rCU0wx8wD5Ige2TbCL8WciV7o/view?usp=sharing
• Provide the model path in the code
• Run the code as shown below for StarGAN:

python test_stargan.py

• Run the code as shown below for CycleGAN:

python test_cyclegan.py

• Run the code as shown below for GauGAN:

python test_gaugan.py

If you would like to use our work, please cite:

@inproceedings{asnani2022proactive
      title={Proactive Image Manipulation Detection}, 
      author={Asnani, Vishal and Yin, Xi and Hassner, Tal and Liu, Sijia and Liu, Xiaoming},
      booktitle={IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
      year={2022}
      
}