This repository is official implementation for Contrastive Pseudo Learning for Open-World DeepFake Attribution, ICCV 2023 and Rethinking Open-World DeepFake Attribution with Multi-perspective Sensory Learning, IJCV 2025
The challenge in sourcing attribution for forgery faces has gained widespread attention due to the rapid development of generative techniques. While many recent works have taken essential steps on GAN-generated faces, more threatening attacks related to identity swapping or diffusion models are still overlooked. And the forgery traces hidden in unknown attacks from the open-world unlabeled faces remain under-explored. To push the related frontier research, we introduce a novel task named Open-World DeepFake Attribution, and the corresponding benchmark OW-DFA and OW-DFA++, which aims to evaluate attribution performance against various types of fake faces in open-world scenarios.
-
Prepare Deepfake Detection datasets
Dataset Paper Link FaceForensics++ FaceForensics++: Learning to Detect Manipulated Facial Images Paper Code Celeb-DF Celeb-DF: A Large-scale Challenging Dataset for DeepFake Forensics Paper Code ForgeryNet ForgeryNet: A Versatile Benchmark for Comprehensive Forgery Analysis Paper Home DFFD On the Detection of Digital Face Manipulation Paper Home ForgeryNIR ForgeryNIR: Deep Face Forgery and Detection in Near-Infrared Scenario Paper Code DF^3 GLFF: Global and Local Feature Fusion for AI-synthesized Image Detection Paper Code -
Download dataset and unzip data under the directory of
/Datasets/deepfakes_detection_datasets/ -
Process dataset with script
scripts/preprocess/create_academic_meta.ipynb, and you will get the following structure:data/release ├── AttributeManipulation │ ├── FaceAPP │ │ └── DFFD │ ├── MaskGAN │ │ └── ForgeryNet │ ├── SC-FEGAN │ │ └── ForgeryNet │ ├── StarGAN │ │ └── DFFD │ └── StarGAN2 │ └── ForgeryNet ├── EntireFaceSyncthesis │ ├── CycleGAN │ │ └── ForgeryNIR │ ├── PGGAN │ │ └── DFFD │ ├── StyleGAN │ │ └── DFFD │ └── StyleGAN2 │ ├── ForgeryNet │ └── ForgeryNIR ├── ExpressionTransfer │ ├── ATVG-Net │ │ └── ForgeryNet │ ├── Face2Face │ │ └── faceforensics │ ├── FOMM │ │ └── ForgeryNet │ ├── NeuralTextures │ │ └── faceforensics │ └── Talking-Head-Video │ └── ForgeryNet ├── IdentitySwap │ ├── DeepFaceLab │ │ └── ForgeryNet │ ├── Deepfakes │ │ └── faceforensics │ ├── FaceShifter │ │ └── ForgeryNet │ ├── FaceSwap │ │ └── faceforensics │ └── FSGAN │ └── ForgeryNet ├── RealFace │ └── Real │ ├── CelebDF │ └── faceforensics ├── meta_data │ ├── Protocol1_openset_fake_large_merge_meta.csv │ ├── Protocol1_openset_fake_val_merge_meta.csv │ ├── Protocol2_openset_real_fake_large_merge_meta.csv │ └── Protocol2_openset_real_fake_val_merge_meta.csv └── shape_predictor_68_face_landmarks.dat
Step1. Create a conda environment and activate it.
conda create --name owdfa python=3.9 -y
conda activate owdfaStep2. Install the required python libraries.
cd OW-DFA
pip3 install -r requirements.txt
wandb offlineStep3. Train MPSL model on OW-DFA dataset. (IJCV 2025)
python3 -u -m torch.distributed.launch --nproc_per_node=1 --master_port 12345 \
train.py -c configs/train_mpsl.yamlOR Train CPL model on OW-DFA dataset. (ICCV 2023)
python3 -u -m torch.distributed.launch --nproc_per_node=1 --master_port 12345 \
train.py -c configs/train_cpl.yamlIf you find this project useful in your research, please consider cite:
@inproceedings{sun2023contrastive,
title={Contrastive pseudo learning for open-world deepfake attribution},
author={Sun, Zhimin and Chen, Shen and Yao, Taiping and Yin, Bangjie and Yi, Ran and Ding, Shouhong and Ma, Lizhuang},
booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
pages={20882--20892},
year={2023}
}
@article{sun2025rethinking,
title={Rethinking open-world deepfake attribution with multi-perspective sensory learning},
author={Sun, Zhimin and Chen, Shen and Yao, Taiping and Yi, Ran and Ding, Shouhong and Ma, Lizhuang},
journal={International Journal of Computer Vision},
volume={133},
number={2},
pages={628--651},
year={2025},
publisher={Springer}
}