Research map / Image generation / Autoregressive and latent generation
μFlow: Leveraging Average Images for Improving Generalisation of Deepfake Faces Detectors
Orazio Pontorno, Mattia Litrico, Luca Guarnera, Mario Valerio Giuffrida, Sebastiano Battiato
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 changed | Where | Measured | Effect |
|---|---|---|---|
| train only on real images instead of Real+GANs fake training | μFlow vs ODDN · Real+GANs training, avg OOD accuracy | accuracy: 81.8 -> 96.8 | Improved |
| real-only training instead of Real+DM-CS training | μFlow vs D3 · DM-OS test, accuracy | accuracy: +2.6 | Improved |
| real-only training instead of Real+DM-OS training | μFlow vs D3 · Real+DM-OS training, avg accuracy | accuracy: 85.4 -> 93.5 | Improved |
| compare against detectors trained on mixed GAN and DM fakes | μFlow vs D3 · Real+Mix*, avg accuracy | accuracy: 92.6 -> 94.7 | Improved |
| compare against real-only baselines | μFlow vs LAA-Net · Real-only baselines, avg accuracy | accuracy: 83.0 -> 94.7 | Improved |
| map features to average-image distribution instead of standard normal Gaussian (ablation) | μFlow · WILD OOD | accuracy | Improved |
| use fixed threshold tau instead of LOF or OC-SVM (ablation) | μFlow · WILD OOD | accuracy | No clear effect |
| vary significance level alpha, choose alpha=0.01 (ablation) | μFlow · WILD OOD | accuracy | No clear effect |
| increase number K of averaged images from 500 to 700 (ablation) | μFlow · Silhouette score | Silhouette score: 0.72 -> 0.74 | Improved |
| test under resize, flip, Gaussian noise, salt-pepper without training on them | μFlow vs D3 · content-preserving transformations (avg) | accuracy: 89.4 -> 90.9 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ODDN — ODDN: Addressing Unpaired Data Challenges in Open-World Deepfake Detection on Online Social Networks (AAAI 2025) · better on 6 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- From Specificity to Generality: Revisiting Generalizable Artifacts in Detecting Face Deepfakes (NeurIPS 2025 · similarity 0.89)
- DiffusionFake: Enhancing Generalization in Deepfake Detection via Guided Stable Diffusion (NeurIPS 2024 · similarity 0.88)
- D^3: Scaling Up Deepfake Detection by Learning from Discrepancy (CVPR 2025 · similarity 0.88)
- DF40: Toward Next-Generation Deepfake Detection (NeurIPS 2024 · similarity 0.88)
- Standing on the Shoulders of Giants: Reprogramming Visual-Language Model for General Deepfake Detection (AAAI 2025 · similarity 0.88)
- Towards More General Video-based Deepfake Detection through Facial Component Guided Adaptation for Foundation Model (CVPR 2025 · similarity 0.86)
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.
