Beyond the Spectrum: Detecting Deepfakes via Re-Synthesis
Authors: Yang He, Ning Yu, Margret Keuper, Mario Fritz
IJCAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate the improved effectiveness, cross-GAN generalization, and robustness against perturbations of our approach in a variety of detection scenarios involving multiple generators over Celeb A-HQ, FFHQ, and LSUN datasets. Source code is available at https://github.com/SSAW14/Beyondthe Spectrum. |
| Researcher Affiliation | Academia | Yang He1, Ning Yu2,3, Margret Keuper4, Mario Fritz1 1CISPA Helmholtz Center for Information Security, Saarbr ucken, Germany 2Max Planck Institute for Informatics, Saarbr ucken, Germany 3University of Maryland, College Park, United States 4University of Mannheim, Mannheim, Germany yang.he@cispa.saarland, ningyu@mpi-inf.mpg.de, keuper@uni-mannheim.de, fritz@cispa.saarland |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Source code is available at https://github.com/SSAW14/Beyondthe Spectrum. |
| Open Datasets | Yes | Celeb A-HQ is a 1024 1024 facial image dataset released by [Karras et al., 2018]. FFHQ [Karras et al., 2019] provides another 70k 1024 1024 facial images. LSUN is large-scale scene dataset [Yu et al., 2015]. |
| Dataset Splits | No | For Celeb A-HQ, we prepare 25k/25k real and fake images as the training set, and 2.5k/2.5k images as the testing set. For LSUN, we prepare 50k/50k real and fake images for training, and 10k/10k images for testing. While training and testing set sizes are given, there is no explicit mention of a separate validation set split or how such a split was used. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for running the experiments (e.g., GPU model, CPU model, memory specifications). |
| Software Dependencies | No | The paper mentions using ResNet-50 and VGG-based perceptual loss, but it does not specify the version numbers for any software, libraries, or frameworks used in the implementation. |
| Experiment Setup | Yes | We set the loss weight for each feature from pixel to stage 5 as [1, 1/32, 1/16, 1/8, 1/4, 1]. Second, we train the detectors on pixel artifacts and stage5 artifacts from the VGG network. For each detector, we train a Res Net-50 [He et al., 2016] from scratch for 20 epochs using the SGD optimizer with momentum. The initial learning rate is 0.01 and we reduce it to 0.001 at the 10th epoch. |