Research map / Label-efficient and robust learning / Recognition and long-tail
R2M: Real-Aware Residual Model Merging for Robust and Generalizable Deepfake Detection
Jinhee Park, Guisik Kim, Choongsang Cho, Junseok Kwon
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 common-alignment calibration to residual routing (ablation) | R2M · DF40 CDF (rank r0=0, rA=1) | mean AUC: 0.398 -> 0.830 | Improved |
| add common-alignment calibration at best ranks (ablation) | R2M · DF40 FF (r0=2, rA=3) | mean AUC: 0.980 -> 0.992 | Improved |
| increase residual rank rA from 1 to 3 and r0 from 0 to 2 (ablation) | R2M · DF40 Unseen | mean AUC: 0.792 -> 0.818 | Improved |
| remove dominant Real-sensitive directions (r0>0) (ablation) | R2M · DF40 CDF | mean AUC (calibrated): 0.880 -> 0.901 | Improved |
| use R2M over best training-free baseline CART | R2M · DF40 Protocol 3 | mean AUC: 0.766 -> 0.818 | Improved |
| use R2M over CART | R2M · DF40 Protocol 2 | mean AUC: 0.877 -> 0.901 | Improved |
| use R2M over CART | R2M · DF40 Protocol 1 | mean AUC: 0.988 -> 0.992 | Improved |
| merge specialists instead of training all-in-one model | Weight Averaging · DF40 Protocol 3 | mean AUC: 0.665 -> 0.724 | Improved |
| R2M vs Task Arithmetic on CDF EFS | R2M · DF40 CDF EFS | AUC: 0.926 -> 0.898 | Hurt |
| add domain grouping before routing six heterogeneous specialists (ablation) | R2M · six specialists, FF FS | AUC: 0.778 -> 0.929 | Improved |
Closest papers by meaning
- D^3: Scaling Up Deepfake Detection by Learning from Discrepancy (CVPR 2025 · similarity 0.87)
- Standing on the Shoulders of Giants: Reprogramming Visual-Language Model for General Deepfake Detection (AAAI 2025 · similarity 0.85)
- Veritas: Generalizable Deepfake Detection via Pattern-Aware Reasoning (ICLR 2026 · similarity 0.84)
- Pixels Don't Lie (But Your Detector Might): Bootstrapping MLLM-as-a-Judge for Trustworthy Deepfake Detection and Reasoning Supervision (CVPR 2026 · similarity 0.84)
- DiffusionFake: Enhancing Generalization in Deepfake Detection via Guided Stable Diffusion (NeurIPS 2024 · similarity 0.84)
- Towards More General Video-based Deepfake Detection through Facial Component Guided Adaptation for Foundation Model (CVPR 2025 · similarity 0.83)
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.
