marcinnawrocki / fake-faces-detector Goto Github PK

Make functions to fast 2DFT transformation and visualize effects.

Test 2D DFT on CelebA and FFHQ datasets and compare them to gathered databases for nvidia GANs.

https://arxiv.org/pdf/2005.04945.pdf
In this paper authors try to subtract image from feature. They used own structures. Używają w chuj iteracji tego generalnie ale ciekawa sprawa imho.

PGGAN:
https://research.nvidia.com/sites/default/files/pubs/2017-10_Progressive-Growing-of/karras2018iclr-paper.pdf
StyleGanv1:
https://arxiv.org/pdf/1812.04948.pdf

Done:

1. https://arxiv.org/abs/1809.08754
2. https://www.researchgate.net/publication/338382561_Deep_Fake_Image_Detection_Based_on_Pairwise_Learning

ToDo:
0.https://www.researchgate.net/publication/324466507_Constrained_Convolutional_Neural_Networks_A_New_Approach_Towards_General_Purpose_Image_Manipulation_Detection

1. https://arxiv.org/pdf/2005.04945.pdf
2. https://ieeexplore.ieee.org/document/8553251
3. https://www.cv-foundation.org/openaccess/content_iccv_2015/papers/Simo-Serra_Discriminative_Learning_of_ICCV_2015_paper.pdf
4. https://www.techscience.com/jihpp/v2n2/40538
5. https://www.researchgate.net/publication/327350438_Fake_Face_Detection_Methods_Can_They_Be_Generalized

Probably start with Xception and DenseNet

Readed:

• https://arxiv.org/pdf/1907.06515.pdf (Detecting and Simulating Artifacts in GAN Fake
  Images)
• https://distill.pub/2016/deconv-checkerboard/
• https://arxiv.org/abs/1903.06836 (Detecting GAN generated Fake Images using Co-occurrence Matrices)
• https://www.researchgate.net/publication/326053461_Detection_of_GAN-Generated_Fake_Images_over_Social_Networks
• https://ieeexplore.ieee.org/document/6115849 (histograms)
  Related papers:
• http://dde.binghamton.edu/kodovsky/pdf/TIFS2012-SRM.pdf Rich Models for Steganalysis of Digital Images
• https://arxiv.org/pdf/1703.04615.pdf residual forgery detection

IEE
https://ieeexplore.ieee.org/document/7026072 (residual)
https://ieeexplore.ieee.org/document/9017530 (CNN based on DFT)
https://ieeexplore.ieee.org/document/8489301 (MLVNM on DFT)

Based on this repo:
https://github.com/calmevtime/DCTNet

Collect database of real/fake for all three GAN networks:

• PGGAN
• StyleGANv1
• StyleGANv2

https://arxiv.org/pdf/1808.07276.pdf

In one of the papers:
Effect of JPEG compression: In [36], the authors investigated
the sensitivity of their methods on compression. Using a compression method similar to Twitter, they trained their detection
methods on original uncompressed images and tested on JPEG
compressed images. The objective of this study was to test the
robustness of the detection techniques when images are posted
in social networks such as Twitter. In the second scenario, they
train and test on the JPEG compressed images. We performed
both of these experiments on the cycleGAN dataset. Since we are
not aware of the exact JPEG quantization tables used in Twitter,
our approach was similar but we tested on three different JPEG
quality factors (QF): 95, 85 and 75. We used 50% of the data for
training, 25% for validation and 25% for testing. The results are
reported on the 25% testing data of the cycleGAN dataset (9,076
images). As shown in Tab. 3, the accuracy progressively drops
as the QF decreases from 95-75, when trained on the original images. But when the JPEG compressed images are used for training, the accuracy shows a substantial increase. Even at a QF of
75, the accuracy is still 87.31%. This is also consistent with the
results reported in [36], where they report close to a 10% drop in
accuracy on Twitter-like compressed images.