AI Paper Atlas

Research map / Vision-language and multimodal / Cross-modal learning

UMCL: Unimodal-generated Multimodal Contrastive Learning for Cross-compression-rate Deepfake Detection

Ching-Yi Lai, Chih-Yu Jian, Pei-Cheng Chuang, Chia-Ming Lee, Chih–Chung Hsu, Hsu Chiou-Ting, Chia‐Wen Lin

IJCV 2026 · Vision-language and multimodal › Cross-modal learning · approach: Multimodal fusion with missing modalities · 3 citations (OpenAlex)

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 changedWhereMeasuredEffect
replace CPML attention fusion with ASA and add CQSL (UMCL vs CPML)CPML · FF++ raw -> c23AUC (%): 91.27 -> 95.69Improved
UMCL vs CPML under training on c23, testing on rawCPML · FF++ c23 -> rawAUC (%): 90.13 -> 96.46Improved
UMCL vs CPML in-compression on c23CPML · FF++ c23 -> c23AUC (%): 97.42 -> 96.59Hurt
UMCL vs CPML in-compression on c40CPML · FF++ c40 -> c40AUC (%): 91.36 -> 90.70Hurt
UMCL vs CPML on DFDC to FF++ raw transferCPML · DFDC -> FF++ (raw)AUC (%): 80.77 -> 89.11Improved
UMCL vs CPML trained on F2F, cross-manipulation averageCPML · FF++ c23 trained on F2FAUC (%): 94.16 -> 96.79Improved
UMCL vs CPML trained on NT, cross-manipulation averageCPML · FF++ c23 trained on NTAUC (%): 94.85 -> 94.53No clear effect
UMCL vs CPML with rPPG sampling ratio 30%CPML · FF++ (raw) train and testAUC (%): 54.22 -> 86.49Improved
UMCL vs CPML with landmark perturbation sigma=0.05CPML · FF++ (raw) train and testAUC (%): 34.40 -> 76.23Improved
UMCL vs CPML with no degradation on c40 testCPML · FF++ (raw) -> c40AUC (%): 63.03 -> 79.72Improved
UMCL vs CPML in-domain DFDCCPML · DFDCAUC (%): 88.76 -> 92.17Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy