Research map / Detection and segmentation / CNN features
Multi-Attentional Deepfake Detection
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| use the multi-attentional framework with EfficientNet-B4 rather than the EfficientNet-B4 binary-classification baseline | EfficientNet-B4 deepfake detection · FaceForensics++ HQ | ACC (%): 96.63 -> 97.60 | Helped |
| use the multi-attentional framework with EfficientNet-B4 rather than the EfficientNet-B4 binary-classification baseline | EfficientNet-B4 deepfake detection · FaceForensics++ LQ | ACC (%): 86.67 -> 88.69 | Helped |
| use the multi-attentional framework rather than the EfficientNet-B4 baseline | EfficientNet-B4 deepfake detection · FaceForensics++ LQ | AUC (%): 88.20 -> 90.40 | Helped |
| use the multi-attentional framework rather than F^3-Net | deepfake detection · FaceForensics++ LQ | ACC (%): 90.43 -> 88.69 | Hurt |
| use the multi-attentional framework rather than NTechLab | DFDC detection · DFDC testing dataset | logloss: 0.1703 -> 0.1679 | Helped |
| use the multi-attentional framework rather than Two Branch | deepfake detection trained on FF++ · Celeb-DF | AUC (%): 73.41 -> 67.44 | Hurt |
| use four attention maps instead of one (ablation) | multi-attentional deepfake detection · FF++(HQ) | ACC (%): 97.26 -> 97.60 | Helped |
| use four attention maps instead of one (ablation) | multi-attentional deepfake detection · Celeb-DF | AUC (%): 67.30 -> 67.44 | No clear effect |
| use two attention maps instead of one (ablation) | multi-attentional deepfake detection · Celeb-DF | AUC (%): 67.30 -> 65.74 | Hurt |
| use regional independence loss rather than no auxiliary loss (ablation) | multi-attentional deepfake detection without AGDA · FF++(HQ) | ACC: 96.74 -> 97.38 | Helped |
| use regional independence loss rather than no auxiliary loss (ablation) | multi-attentional deepfake detection without AGDA · Celeb-DF | AUC: 64.86 -> 65.85 | Helped |
| use regional independence loss instead of AMS while using soft AGDA (ablation) | multi-attentional deepfake detection · FF++(HQ) | ACC: 96.78 -> 97.60 | Helped |
Closest papers by meaning
- DF40: Toward Next-Generation Deepfake Detection (NeurIPS 2024 · similarity 0.91)
- Rethinking Vision-Language Model in Face Forensics: Multi-Modal Interpretable Forged Face Detector (CVPR 2025 · similarity 0.91)
- Exposing the Deception: Uncovering More Forgery Clues for Deepfake Detection (AAAI 2024 · similarity 0.91)
- From Specificity to Generality: Revisiting Generalizable Artifacts in Detecting Face Deepfakes (NeurIPS 2025 · similarity 0.90)
- Critical Forgetting-Based Multi-Scale Disentanglement for Deepfake Detection (AAAI 2025 · similarity 0.89)
- FreqBlender: Enhancing DeepFake Detection by Blending Frequency Knowledge (NeurIPS 2024 · similarity 0.89)
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.