CoChat
Sign Up
Cited by 5

The DOI matches its Crossref record; 4 of the 4 references it cites were found in a registry or an index.

Aggregating Layers for Deepfake Detection

Jevnisek, Amir, Avidan, Shai·2022·2022 26th International Conference on Pattern Recognition (ICPR)
DOI 10.1109/icpr56361.2022.9956052
Open this paper in CoChat — free account

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1109/icpr56361.2022.9956052

Cites

4 of 4 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Deepfakes FaceForensics++ fake images examples. Top to bottom and left to right: Deepfakes, FaceSwap, Face2Face and NeuralTextures.
Fig. 1 Deepfakes FaceForensics++ fake images examples. Top to bottom and left to right: Deepfakes, FaceSwap, Face2Face and NeuralTextures.CC-BY · From the PDF
Fig. 2 Synthetic Images GAN generated images, top to bottom, right to left: SGAN, Glow and GMM trained with CelebA-HQ, PGAN, SGAN and SGAN2 trained with FFHQ.
Fig. 2 Synthetic Images GAN generated images, top to bottom, right to left: SGAN, Glow and GMM trained with CelebA-HQ, PGAN, SGAN and SGAN2 trained with FFHQ.CC-BY · From the PDF
Image 3
CC-BY · From the PDF
Fig. 4 Deepfake Detection: Layer Importance . Layers are indexed on the horizontal axis and importance (computed as the dot product of the weight of the Linear Regression and features of each layer) on the vertical axis. The average importance of ∼ 1000 real (fake) images is shown with blue (orange) bars. The graph on top (bottom) shows the average importance of a network trained with NeuralTextures ( Deepfakes ) models. As can be seen, the network 'automatically' calibrated itself to examine different features for different models. Namely, different features contribute to the detection of different manipulations.
Fig. 4 Deepfake Detection: Layer Importance . Layers are indexed on the horizontal axis and importance (computed as the dot product of the weight of the Linear Regression and features of each layer) on the vertical axis. The average importance of ∼ 1000 real (fake) images is shown with blue (orange) bars. The graph on top (bottom) shows the average importance of a network trained with NeuralTextures ( Deepfakes ) models. As can be seen, the network 'automatically' calibrated itself to examine different features for different models. Namely, different features contribute to the detection of different manipulations.CC-BY · From the PDF
Image 5
CC-BY · From the PDF
Image 6
CC-BY · From the PDF
Image 7
CC-BY · From the PDF
Image 8
CC-BY · From the PDF
Image 9
CC-BY · From the PDF
Image 10
CC-BY · From the PDF
Image 11
CC-BY · From the PDF
Fig. 7 Deepfakes Network Trimming Effect . Average Precision deviation vs. an experiment. The experiments are described as ( D train , D test ) pairs. The model is trained on D train and tested on D test . The colors code different values of the Primitive-Projection blocks budget N pp -blocks .
Fig. 7 Deepfakes Network Trimming Effect . Average Precision deviation vs. an experiment. The experiments are described as ( D train , D test ) pairs. The model is trained on D train and tested on D test . The colors code different values of the Primitive-Projection blocks budget N pp -blocks .CC-BY · From the PDF
Fig. 8 Deepfakes Trimming . AP score degradation as a function of the number of analysis network parameters (The xaxis is presented in logarithmic scale). For example, choosing the best three Primitive Projection blocks requires less than 6% of the total number of parameters in the analysis network and leads to a degradation of less than 0 . 25% , compared to using the entire analysis network of 16 M parameters.
Fig. 8 Deepfakes Trimming . AP score degradation as a function of the number of analysis network parameters (The xaxis is presented in logarithmic scale). For example, choosing the best three Primitive Projection blocks requires less than 6% of the total number of parameters in the analysis network and leads to a degradation of less than 0 . 25% , compared to using the entire analysis network of 16 M parameters.CC-BY · From the PDF

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Research works printed but not linked: 11

  1. 10.1109/CVPR46437.2021.00639 · DOI · Paper page
  2. 10.1109/ICCV.2015.425 · DOI · Paper page
  3. 10.1109/ICCV.2019.00009 · DOI · Paper page
  4. 10.1109/CVPR.2018.00255 · DOI
  5. 10.1007/978-3-030-58574-7_7 · DOI · Paper page
  6. 10.1109/CVPRW.2017.229 · DOI · Paper page
  7. chai, 2020, What makes fake images detectable? understanding properties that generalize, European Conference on Computer Vision, 103
  8. haliassos, 2021, Lips don’t lie: A generalisable and robust approach to face forgery detection, Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 5039
  9. 10.1109/CVPR.2018.00474 · DOI · Paper page
  10. 10.1109/CVPR46437.2021.00361 · DOI · Paper page
  11. 10.1145/3306346.3323035 · DOI

Cited by

Sorted by citations

See all 5

Metadata

Authors
Jevnisek, A., & Avidan, S.
Affiliations
Tel-Aviv University,School of Electrical Engineering,Tel-Aviv,Israel
Venue
2022 26th International Conference on Pattern Recognition (ICPR)
Publisher
IEEE
Published
2022-08-21
Pages
2027-2033
DOI
10.1109/icpr56361.2022.9956052
Cited by
5
References
4
Licence

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review