Research map / Label-efficient and robust learning / Domain adaptation and generalization
Detecting Deepfakes With Self-Blended Images
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 |
|---|---|---|---|
| train with self-blended images rather than BI | EFNB4 deepfake detector · CDF | AUC (%): 93.18 vs. 69.40 | Helped |
| train with self-blended images rather than I2G | ResNet-34 detector · DFDC | AUC (%): 66.41 vs. 51.72 | Helped |
| add source augmentation (ablation) | SBI generation · CDF | AUC (%): 76.61 -> 93.18 | Helped |
| add target augmentation (ablation) | SBI generation · CDF | AUC (%): 70.68 -> 93.18 | Helped |
| add source-target augmentation (ablation) | SBI generation · FF++ | AUC (%): 89.18 -> 99.64 | Helped |
| add resize and translation (ablation) | SBI generation · DFDCP | AUC (%): 81.04 -> 86.15 | Helped |
| use the top 20 landmark-similar BI source candidates instead of all 100 (ablation) | BI training data · CDF | AUC (%): 69.40 -> 71.44 | Helped |
| jointly train on SBIs and BI (ablation) | EFNB4 detector · CDF | AUC (%): 93.18 -> 89.36 | Hurt |
| add source-target augmentation to BI (ablation) | BI synthetic training data · CDF | AUC (%): 69.40 -> 78.99 | Helped |
| add source-target augmentation to BI (ablation) | BI synthetic training data · DFD | AUC (%): 97.50 -> 99.04 | Helped |
| use a larger-capacity EfficientNet-b4 rather than ResNet-50 (ablation) | SBI-trained detector · FFIW | AUC (%): 79.30 -> 84.83 | Helped |
| train with pristine videos from each dataset rather than FF++ only (ablation) | SBI-trained detector · DFDCP | AUC (%): 86.15 (FF++ training); 88.70 (DFDCP training) | Mixed |
Closest papers by meaning
- Can We Leave Deepfake Data Behind in Training Deepfake Detector? (NeurIPS 2024 · similarity 0.89)
- DF40: Toward Next-Generation Deepfake Detection (NeurIPS 2024 · similarity 0.88)
- Generalizing Deepfake Video Detection with Plug-and-Play: Video-Level Blending and Spatiotemporal Adapter Tuning (CVPR 2025 · similarity 0.88)
- From Specificity to Generality: Revisiting Generalizable Artifacts in Detecting Face Deepfakes (NeurIPS 2025 · similarity 0.88)
- FreqBlender: Enhancing DeepFake Detection by Blending Frequency Knowledge (NeurIPS 2024 · similarity 0.87)
- Exposing the Deception: Uncovering More Forgery Clues for Deepfake Detection (AAAI 2024 · similarity 0.87)
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.