Forensics Adapter: Adapting CLIP for Generalizable Face Forgery Detection (CVPR 2025)

👥 Authors: Xinjie Cui, Yuezun Li (corresponding author), Ao Luo, Jiaran Zhou, Junyu Dong

---

📚 Resources

Section	Content
📄 Paper	arXiv Preprint
⚖️ Model Weights	Google Drive | Baidu Netdisk

---

📊 Benchmark Comparison

🖼️ Frame-Level Comparison

🏆 Champion Method Alert: Our approach establishes new state-of-the-art on all frame-level benchmarks!

Method	Venue	CDF-v1	CDF-v2	DFDC	DFDCP	DFD	Avg. 📈
SPSL	CVPR'21	0.815	0.765	0.704	0.741	0.812	0.767
SRM	CVPR'21	0.793	0.755	0.700	0.741	0.812	0.760
Reece	CVPR'22	0.768	0.732	0.713	0.734	0.812	0.752
SBI	CVPR'22	-	0.813	-	0.799	0.774	-
UCF	ICCV'23	0.779	0.753	0.719	0.759	0.807	0.763
ED	AAAI'24	0.818	0.864	0.721	0.851	-	-
LSDA	CVPR'24	0.867	0.830	0.736	0.815	0.880	0.826
CFM	TIFS'24	-	0.828	-	0.758	0.915	-
Ours	CVPR'25	🥇 0.914	🥇 0.900	🥇 0.843	🥇 0.890	🥇 0.933	🥇 0.896

🎥 Video-Level Comparison

🏆 Champion Method Alert: Our approach achieves new SOTA performance across all video-level benchmarks!

Method	Venue	CDF-v2	DFDC	DFDCP
TALL	ICCV'23	0.908	0.768	-
AltFreezing	CVPR'23	0.895	-	-
SeeABLE	ICCV'23	0.873	0.759	0.863
IID	CVPR'23	0.838	-	0.812
TALL++	IJCV'24	0.920	0.785	-
SAM	CVPR'24	0.890	-	-
SBI	CVPR'22	0.932	0.724	0.862
CADDM	CVPR'23	0.939	0.739	-
SFDG	CVPR'23	0.758	0.736	-
LAA-NET	CVPR'24	0.954	-	0.869
LSDA	CVPR'24	0.911	0.770	-
CFM	TIFS'24	0.897	-	0.802
Ours	CVPR'25	🥇 0.957	🥇 0.872	🥇 0.929

---

🚀 Start

• ⏳ Environment Setup
• 📂 Dataset
• 🏋️ Training
• 🧪 Testing
• 📝 Citation

⏳ Environment Setup

Ensure your environment meets the following requirements:

• 🐍 Python 3.9
• 🔥 PyTorch 1.11
• 🚀 CUDA 11.3

Install dependencies:

git clone https://github.com/OUC-VAS/ForensicsAdapter.git
cd ForensicsAdapter
conda create -n FA python=3.9
conda activate FA
sh install.sh

📂 Dataset

We use multiple datasets for training and evaluation:

• FF++
• DFDC
• DFDCP
• DFD
• CD1/CD2

The dataset downloading and processing procedures can be referred to the implementation provided in DeepfakeBench .

🏋️ Training

Make sure to modify the relevant configurations in the train.yaml file before training.

Start training with the following command:

python train.py 

🧪 Testing

Make sure to modify the relevant configurations in the test.yaml file before testing.

To test the model, you can directly load our pre-trained weights and run a command like the following:

python /data/cuixinjie/FA/test.py 

📝 Citation

If our work is useful for your research, please cite it as follows:

@InProceedings{Cui_2025_CVPR,
    author    = {Cui, Xinjie and Li, Yuezun and Luo, Ao and Zhou, Jiaran and Dong, Junyu},
    title     = {Forensics Adapter: Adapting CLIP for Generalizable Face Forgery Detection},
    booktitle = {Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR)},
    month     = {June},
    year      = {2025},
    pages     = {19207-19217}
}