Research map / Video / Video segmentation and tracking
Preserving Knowledge across Space and Time for Continual Video Deepfake Detection
Taehoon Kim, Jongwook Choi, Heejae Jo, Byungmin Park, Jongwon Choi
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 |
|---|---|---|---|
| add naive modality-wise frequency distillation (MSFD) (ablation) | iCaRL-FC baseline · Protocol 1 Dataset-Incremental (6 datasets) | AUC_all: 89.84 -> 92.65 | Improved |
| add raw-feature multi-layer distillation (ablation) | iCaRL-FC baseline · Protocol 1 Dataset-Incremental (6 datasets) | AUC_all: 89.84 -> 90.64 | Improved |
| add Modality-Specific Adaptor (MSA) (ablation) | MSFD + CDL · Protocol 1 Dataset-Incremental (6 datasets) | AUC_all: 92.77 -> 93.02 | Improved |
| add Cross-Modality Decorrelation Loss (CDL) (ablation) | MSFD + MSA · Protocol 1 Dataset-Incremental (6 datasets) | AUC_all: 92.14 -> 93.02 | Improved |
| add CDL (ablation) | MSFD + MSA · DF40 face reenactment (continual) | AUC: 75.6 -> 85.6 | Improved |
| decompose features into spatial, temporal and spatiotemporal spectra (ablation) | multi-layer distillation · Protocol 1 Dataset-Incremental (6 datasets) | AUC_all: 90.64 -> 93.02 | Improved |
| add modality-specific reweighting (MSR) without masking (ablation) | MSA · Protocol 1 Dataset-Incremental (6 datasets) | AUC_all: 92.77 -> 92.97 | Improved |
| add distillation masking (MSDM) without reweighting (ablation) | MSA · Protocol 1 Dataset-Incremental (6 datasets) | AUC_all: 92.77 -> 91.91 | Hurt |
| use learnable Gumbel-Softmax mask instead of fixed low-pass mask (ablation) | MSDM · Protocol 1 Dataset-Incremental (6 datasets) | AUC_all: 92.38 -> 93.02 | Improved |
| use magnitude instead of phase spectrum for distillation (ablation) | MSFD · Protocol 1 Dataset-Incremental (6 datasets) | AUC_all: 90.04 -> 93.02 | Improved |
| use FFT-based decomposition instead of removing FFT (ablation) | MSFD · Protocol 1 Dataset-Incremental (6 datasets) | AUC_all: 92.21 -> 93.02 | Improved |
| use center-based memory update instead of random (ablation) | MSFD · Protocol 1 Dataset-Incremental (6 datasets) | F_avg: 3.89 -> 2.29 | Improved |
Closest papers by meaning
- Generalizing Deepfake Video Detection with Plug-and-Play: Video-Level Blending and Spatiotemporal Adapter Tuning (CVPR 2025 · similarity 0.91)
- Delving into the Local: Dynamic Inconsistency Learning for DeepFake Video Detection (AAAI 2022 · similarity 0.89)
- Generalization-Preserved Learning: Closing the Backdoor to Catastrophic Forgetting in Continual Deepfake Detection (ICCV 2025 · similarity 0.89)
- Towards More General Video-based Deepfake Detection through Facial Component Guided Adaptation for Foundation Model (CVPR 2025 · similarity 0.87)
- Exploiting Style Latent Flows for Generalizing Deepfake Video Detection (CVPR 2024 · similarity 0.86)
- AVFF: Audio-Visual Feature Fusion for Video Deepfake Detection (CVPR 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.
