FreqBlender: Enhancing DeepFake Detection by Blending Frequency Knowledge

Authors: hanzhe li, Jiaran Zhou, Yuezun Li, Baoyuan Wu, Bin Li, Junyu Dong

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results demonstrate the effectiveness of our method in enhancing Deep Fake detection, making it a potential plug-and-play strategy for other methods.
Researcher Affiliation Academia Hanzhe Li1, Jiaran Zhou1, Yuezun Li1, Baoyuan Wu2 Bin Li3 Junyu Dong1 1 School of Computer Science and Technology, Ocean University of China 2 School of Data Science, The Chinese University of Hong Kong, Shenzhen, China 3 Guangdong Provincial Key Laboratory of Intelligent Information Processing, Shenzhen University, Shenzhen, China
Pseudocode No The paper describes the architecture of the Frequency Parsing Network (FPNet) with diagrams in Figure 5, but it does not contain any formal pseudocode or algorithm blocks.
Open Source Code No The code will be thoroughly organized and released after the paper is accepted.
Open Datasets Yes Our method is evaluated using several standard datasets, including Face Forensics++ [5] (FF++), Celeb-DF (CDF) [6], Deep Fake Detection Challenge (DFDC) [8], Deep Fake Detection Challenge Preview (DFDCP) [7], and FFIW-10k (FFIW) [9] datasets.
Dataset Splits Yes We follow the original training and testing split provided by the datasets for experiments.
Hardware Specification Yes Our method is implemented using Py Torch 2.0.1 [36] with a Nvidia 3090ti.
Software Dependencies Yes Our method is implemented using Py Torch 2.0.1 [36] with a Nvidia 3090ti.
Experiment Setup Yes In the training stage of FPNet, the image size is set to 400 400. The batch size is set to 8 and the Adam optimizer is utilized with an initial learning rate of 1e 4. The training epoch is set to 200. The hyperparameters in the objective function in Eq. (5) are set as follows: "λ1 = 1/12, λ2 = 1, λ3 = 1e 3, λ4 = 1/4".