Privacy-Preserving Face Recognition in the Frequency Domain

Authors: Yinggui Wang, Jian Liu, Man Luo, Le Yang, Li Wang2558-2566

AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments In this section, we first evaluate the proposed analysis network for trade-off analysis between privacy and accuracy. Performance comparisons of different algorithms over standard face datasets are carried out, following by attacking experiments and discussions for PPFR-FD.
Researcher Affiliation Collaboration Yinggui Wang 1, Jian Liu1, Man Luo1, Le Yang2, Li Wang1 1Ant Group, 2University of Canterbury
Pseudocode No The paper describes methods in textual paragraphs and uses schematic diagrams (e.g., Fig. 2) to illustrate processes, but it does not contain any formal pseudocode or algorithm blocks.
Open Source Code No The paper does not contain any explicit statements about releasing source code or provide links to a code repository.
Open Datasets Yes We use the MS-Celeb-1M dataset with 3,648,176 images from 79,891 subjects as the training set.
Dataset Splits No The paper mentions using MS-Celeb-1M, CASIA, and LFW as training and test datasets, and 7 benchmarks for evaluation, but it does not explicitly define or specify a validation set or its split details.
Hardware Specification No The paper does not provide specific details regarding the hardware used for running experiments, such as GPU models, CPU specifications, or cloud computing instances.
Software Dependencies No The paper mentions using Mobile Net V2, Arc Face loss, Res Net50, SE-blocks, and SGD optimizer, but it does not provide specific version numbers for any of the software dependencies or libraries used.
Experiment Setup Yes All models are trained for 50 epochs using the SGD optimizer with the momentum of 0.9, weight decay of 0.0001. For the threshold γ in (1), we set it to 0.3. λ in (2) is set to 1. We train the baseline model on Res Net50 backbone with SE-blocks and batch size of 512.