Local Relation Learning for Face Forgery Detection
Authors: Shen Chen, Taiping Yao, Yang Chen, Shouhong Ding, Jilin Li, Rongrong Ji1081-1088
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments show that the proposed method consistently outperforms the state-of-the-arts on widely-used benchmarks. Furthermore, detailed visualization shows the robustness and interpretability of our method.4 Experiments In this section, we first experimentally evaluate the effectiveness of the proposed algorithm against state-of-the-art techniques and investigate its robustness under unseen manipulation methods in Sec. 4.2. Subsequently, we conduct an ablation study to explore the influence of proposed components in Sec. 4.3. Finally, we demonstrate the interpretability of our approach through visualization analysis in Sec. 4.4. |
| Researcher Affiliation | Collaboration | 1 Media Analytics and Computing Lab, Department of Artificial Intelligence, School of Informatics, Xiamen University 2 You Tu Lab, Tencent 3 Institute of Artificial Intelligence, Xiamen University |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper states "We implement the proposed framework via open-source Py Torch (Paszke et al. 2017)", which refers to the framework they used, not their own implementation's source code. It also references GitHub links for external tools (Deepfakes, Face Swap) and datasets (DFD), but no link or explicit statement for their own code. |
| Open Datasets | Yes | Following the convention, we adopt the widely-used benchmark dataset Face Forensics++ (FF++) (R ossler et al. 2019) for training. To evaluate the robustness of our method, we also conduct experiments on the recent proposed large-scale face manipulated dataset, i.e., Deepfake Detection Challenge (DFDC) (Dolhansky et al. 2019), Celeb-DF (Li et al. 2020b) and Deepfake Detection1 (DFD). |
| Dataset Splits | Yes | FF++ is a face forgery detection dataset consisting of 1000 original videos with real faces, in which 720 videos are used for training, 140 videos are reserved for validation and 140 videos for testing. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used to run the experiments, such as GPU models, CPU types, or memory specifications. It only states training parameters and software used. |
| Software Dependencies | No | The paper states: "We implement the proposed framework via open-source Py Torch (Paszke et al. 2017)." While PyTorch is mentioned, a specific version number is not provided. No other software dependencies with version numbers are listed. |
| Experiment Setup | Yes | For the frequency-aware cue, the α in Equ. 3 is empirically set to 0.33. And the number of patches k is set to 5. To enhance the learning of local region relations, we set λ1 and λ2 in Equ. 13 to 10 and 1, respectively. Following Face Forensics++ (R ossler et al. 2019), we resize the input image to 299 × 299, and train the network using Adam optimizer (Kingma and Ba 2015) with a learning rate of 2e-4, a batch size of 32, betas of 0.9 and 0.999, and weight decay equal to 1e-5. The total number of training epochs is set to 50, and the learning rate is reduced to half every 10 epochs. |