Research map / Image generation / GANs
Global Texture Enhancement for Fake Face Detection in the Wild
Zhengzhe Liu, Xiaojuan Qi, Philip H.S. Torr
Check 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 changed | Where | Measured | Effect |
|---|---|---|---|
| Add Gram Blocks that encode global texture features | ResNet-18 fake-face detector · StyleGAN vs. CelebA-HQ, average across in-domain, edits, and cross-GAN tests | accuracy: 88.20 -> 95.51 | Improved |
| Add Gram Blocks that encode global texture features | ResNet-18 fake-face detector · PGGAN vs. CelebA-HQ, average across settings | accuracy: 89.45 -> 92.28 | Improved |
| Add Gram Blocks that encode global texture features | ResNet-18 fake-face detector · Low-resolution GANs (DCGAN, DRAGAN, StarGAN) vs. CelebA | accuracy: 75.11 ± 8.10 -> 81.65 ± 3.51; 65.53 ± 8.20 -> 76.40 ± 6.06; 64.04 ± 7.69 -> 74.96 ± 4.90 | Improved |
| Add Gram Blocks that encode global texture features | ResNet-18 detector trained on faces · ImageNet vs. BigGAN | accuracy: 71.93 ± 2.09 -> 80.29 ± 3.20 | Improved |
| Use only cheek skin rather than the full image (ablation) | ResNet-18 · In-domain StyleGAN vs. CelebA-HQ | accuracy: 99.99% -> 99.93% | No clear effect |
| Convert the cheek skin region to grayscale (ablation) | ResNet-18 · In-domain StyleGAN vs. CelebA-HQ | accuracy: 99.93% -> 99.76% | No clear effect |
| Smooth small skin textures with an L0 filter (ablation) | ResNet-18 · In-domain StyleGAN vs. CelebA-HQ | accuracy: 99.93% -> 78.64% | Hurt |
| Add Gram Blocks to the StyleGAN discriminator | Gram-Net fake-face detector · StyleGAN with Gram-Block discriminator vs. CelebA-HQ | accuracy: 80.55 ± 6.37 -> 93.35 ± 2.25 | Improved |
Closest papers by meaning
- Towards Universal Fake Image Detectors That Generalize Across Generative Models (CVPR 2023 · similarity 0.90)
- Exploring Adversarial Fake Images on Face Manifold (CVPR 2021 · similarity 0.88)
- Leveraging Frequency Analysis for Deep Fake Image Recognition (ICML 2020 · similarity 0.88)
- Transfer Learning of Real Image Features with Soft Contrastive Loss for Fake Image Detection (AAAI 2025 · similarity 0.86)
- FerretNet: Efficient Synthetic Image Detection via Local Pixel Dependencies (NeurIPS 2025 · similarity 0.85)
- Secret Lies in Color: Enhancing AI-Generated Images Detection with Color Distribution Analysis (CVPR 2025 · similarity 0.85)
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.
