Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Mixture-of-Attack-Experts with Class Regularization for Unified Physical-Digital Face Attack Detection
Authors: Shunxin Chen, Ajian Liu, Junze Zheng, Jun Wan, Kailai Peng, Sergio Escalera, Zhen Lei
AAAI 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on two unified physical-digital attack datasets demonstrate the State-of-The-Art (So TA) performance of the proposed method. |
| Researcher Affiliation | Academia | 1Nanjing University of Posts and Telecommunications, Nanjing, China 2Nanjing Artificial Intelligence Research of IA, Nanjing, China 3University of Chinese Academy of Sciences, Nanjing, China 4MAIS, Institute of Automation, Chinese Academy of Sciences, Beijing, China 5Macau University of Science and Technology (MUST), Macau, China 6Purple Mountain Laboratory, China 7Computer Vision Center (CVC), Barcelona, Catalonia, Spain 8School of Artificial Intelligence, University of Chinese Academy of Sciences, China 9CAIR, HKISI, Chinese Academy of Sciences, Hong Kong, China |
| Pseudocode | No | The paper describes the methodology using textual descriptions, mathematical equations (Eq. 1-14), and detailed architectural diagrams (Fig. 3, Fig. 4) but does not include any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide an explicit statement about the release of source code, nor does it include any links to a code repository. |
| Open Datasets | Yes | To evaluate the performance of our proposed method in comparison to existing methods, we employed two publicly available datasets, Uni Attack Data (Fang et al. 2024) and JFSFDB (Yu et al. 2024), for face forgery detection. |
| Dataset Splits | No | Uni Attack Data serves as the primary evaluation dataset due to its advantage of ID consistency. Our method exhibited superior performance and generalization capabilities across both datasets. Additionally, to substantiate the effectiveness of each proposed module, we conducted comprehensive ablation studies. Two protocols are defined: Protocol 1 evaluates unified attack detection with all attack types in training, validation, and test sets, while Protocol 2 tests generalization to unseen attacks using a leave-one-type-out approach, divided into P2.1 (unseen physical attacks) and P2.2 (unseen digital attacks). |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running the experiments. |
| Software Dependencies | No | The paper mentions the use of the Adam optimizer, Vi T-B/16, and CLIP, but does not provide specific version numbers for any software dependencies or libraries. |
| Experiment Setup | Yes | We configured Vi T-B/16 as the image encoder, with the number of experts and heads in the Mo AE set to 4 and 2, respectively. The Adam optimizer was employed, with a learning rate of 1e-6 and a weight decay of 5e-4. The model was trained for 300 iterations. |