Research map / Video / Video segmentation and tracking
Quality-Agnostic Deepfake Detection with Intra-model Collaborative Learning
Binh M. Le, Simon S. Woo
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 |
|---|---|---|---|
| replace HSIC with soft-label collaborative loss (ablation) | QAD (ResNet50) · NeuralTextures | AUC (%): 88.2 -> 84.0 | Hurt |
| add HSIC dependence maximization at intermediate layers (ablation) | ResNet50 baseline · NeuralTextures | AUC (%): 88.2 -> 90.1 | Improved |
| use pairwise loss instead of HSIC (ablation) | QAD (ResNet50) · NeuralTextures | AUC (%): 88.2 -> 89.1 | Mixed |
| add Adversarial Weight Perturbation (cross-entropy) (ablation) | ResNet50 baseline · NeuralTextures | AUC (%): 88.2 -> 90.7 | Improved |
| use KL divergence instead of cross-entropy in AWP (ablation) | QAD (ResNet50) · NeuralTextures | AUC (%): 90.7 -> 89.4 | Hurt |
| combine HSIC and AWP (ablation) | ResNet50 baseline · NeuralTextures | AUC (%): 88.2 -> 91.3 | Improved |
| apply QAD training to different backbones (ablation) | ResNet18/34, EfficientNet-B0 · 7 datasets avg | AUC (%): +0.86 to +1.3 points | Improved |
| one QAD model instead of separate per-quality models | EfficientNet-B1 · Video compression, quality-aware | Avg AUC (%): 96.39 -> 97.79 | Improved |
| QAD vs best baseline under random JPEG compression on NeuralTextures | QAD-E · NeuralTextures JPEG | AUC (%): 79.00 -> 76.27 | Hurt |
| train QAD on FF++ and test cross-dataset on DFDC | QAD-E · DFDC | AUC (%): 71.9 -> 65.3 | Hurt |
| increase alpha above 0.002 (ablation) | QAD (ResNet50) · NeuralTextures | AUC | Improved |
Closest papers by meaning
- D^3: Scaling Up Deepfake Detection by Learning from Discrepancy (CVPR 2025 · similarity 0.87)
- FreqDebias: Towards Generalizable Deepfake Detection via Consistency-Driven Frequency Debiasing (CVPR 2025 · similarity 0.86)
- Standing on the Shoulders of Giants: Reprogramming Visual-Language Model for General Deepfake Detection (AAAI 2025 · similarity 0.86)
- Exposing the Deception: Uncovering More Forgery Clues for Deepfake Detection (AAAI 2024 · similarity 0.86)
- AVFF: Audio-Visual Feature Fusion for Video Deepfake Detection (CVPR 2024 · similarity 0.86)
- DF40: Toward Next-Generation Deepfake Detection (NeurIPS 2024 · 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.
