AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

FaceForensics++: Learning to Detect Manipulated Facial Images

Andreas Rossler, Davide Cozzolino, Luisa Verdoliva, Christian Riess, Justus Thies, Matthias Niessner

ICCV 2019 · Image generation › Autoregressive and latent generation · approach: Generalizable fake image detection · 3,635 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
crop the tracked face region instead of feeding the full image (ablation)XceptionNet · FaceForensics++ test, HQaccuracy (%): 74.78 -> 95.73Improved
crop the tracked face region instead of feeding the full image (ablation)XceptionNet · FaceForensics++ test, Rawaccuracy (%): 82.01 -> 99.26Improved
crop the tracked face region instead of feeding the full image (ablation)XceptionNet · FaceForensics++ test, LQaccuracy (%): 70.52 -> 81.00Improved
increase training corpus from 10 to 300 videos (ablation)XceptionNet · FaceForensics++ test, Compressed 40, all methodsaccuracy (%): 60.55 -> 76.01Improved
increase training corpus from 10 to 300 videos (ablation)XceptionNet · FaceForensics++ test, Compressed 23, all methodsaccuracy (%): 67.71 -> 92.65Improved
increase training corpus from 10 to 300 videos (ablation)XceptionNet · FaceForensics++ test, Raw, all methodsaccuracy (%): 92.81 -> 98.88Improved
use video compression (raw to LQ) on the test dataSteg. Features + SVM · FaceForensics++ testaccuracy (%): 97.63 -> 55.98Hurt
use video compression (raw to LQ) on the test dataXceptionNet · FaceForensics++ testaccuracy (%): 99.26 -> 81.00Hurt
use ImageNet-pretrained deep XceptionNet instead of shallow CNNsXceptionNet vs MesoNet · FaceForensics++ test, LQaccuracy (%): 70.47 -> 81.00Improved
test on NeuralTextures instead of DeepFakesXceptionNet · FaceForensics++ manipulation-specific, Compressed 40accuracy (%): 94.28 -> 82.11Hurt
lower video quality for human observers (raw to LQ)Human observers · user studyaccuracy (%): 68.69 -> 58.73Hurt
use automatic detector instead of human observersXceptionNet · FaceForensics++ LQaccuracy (%): 58.73 -> 81.00Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy