DiffusionFake: Enhancing Generalization in Deepfake Detection via Guided Stable Diffusion

Authors: ke sun, Shen Chen, Taiping Yao, Hong Liu, Xiaoshuai Sun, Shouhong Ding, Rongrong Ji

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 4 Experiment
Researcher Affiliation Collaboration 1 Key Laboratory of Multimedia Trusted Perception and Efficient Computing, Ministry of Education of China, Xiamen University, 361005, P.R. China. 2 Youtu Lab, Tencent, P.R. China. 3 Osaka University, Japan.
Pseudocode No The paper does not contain any sections or figures explicitly labeled 'Pseudocode' or 'Algorithm'.
Open Source Code Yes Code are available in https://github.com/skJack/DiffusionFake.git.
Open Datasets Yes Dataset. To evaluate the generalization ability of Diffusion Fake, we conduct experiments on several challenging datasets: (1) Face Forensics++ (FF++) [31]: (2) Celeb-DF [23]: (3) Deep Fake Detection (DFD): (4) DFDC Preview (DFDC-P) [8]: (5) Wild Deepfake [54]: (6) Diff Swap [6]:
Dataset Splits No The paper mentions following 'the data split strategy used in Face Forensics++ [31]' but does not explicitly state the train/validation/test percentages or sample counts for reproduction within its own text.
Hardware Specification No The paper does not specify the exact hardware used for training or inference, such as specific GPU or CPU models, or details about the computing environment.
Software Dependencies Yes During training, we utilize a pre-trained Stable Diffusion 1.5 model with frozen parameters.
Experiment Setup Yes Input images are resized to 224x224 pixels. We employ the Adam optimizer with a learning rate of 1e-5 and a batch size of 32. The model is trained for 20 epochs. The hyperparameters λs and λt are set to 0.7 and 1, respectively.