AI Paper Atlas

Research map / Probabilistic and scientific modeling / Molecules and proteins

Breaking Manifold Continuity: Vector Quantized Modeling for Real-Centric Deepfake Detection

Changshuo Wang, Jiangming Wang, Ke-Yue Zhang, Taiping Yao, Shouhong Ding, Ran Yi, Lizhuang Ma

ICML 2026 · Probabilistic and scientific modeling › Molecules and proteins · approach: Sparse graphical and tensor models

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
add discrete real codebook (ablation)Effort baseline · DFDC and DFDCP avg.video-level AUC: 0.876 -> 0.890Improved
add masked ViT (ablation)codebook model · DFDC and DFDCP avg.video-level AUC: 0.890 -> 0.894Improved
add adaptive tangent space projection (TSP) (ablation)codebook + masked ViT · DFDC and DFDCP avg.video-level AUC: 0.894 -> 0.900Improved
replace static averaging with hypernetwork-based fusion (ablation)full model without fusion · DFDC and DFDCP avg.video-level AUC: 0.900 -> 0.906Improved
add codebook on DFDC (ablation)Effort baseline · DFDCvideo-level AUC: 0.843 -> 0.861Improved
add all components on DFDCP (ablation)Effort baseline · DFDCPvideo-level AUC: 0.909 -> 0.932Improved
raise codebook size from 128 to 1024 (ablation)codebook · FSh, DFDC, DFDCP avg.video-level AUC: 0.860 -> 0.910Improved
raise codebook size from 1024 to 2048 (ablation)codebook · FSh, DFDC, DFDCP avg.video-level AUC: 0.910 -> 0.909No clear effect
raise codebook size from 1024 to 4096 (ablation)codebook · FSh, DFDC, DFDCP avg.video-level AUC: 0.910 -> 0.896Hurt
add full pipeline vs Effort baseline cross-datasetEffort · Cross-dataset avg.video-level AUC: 0.898 -> 0.929Improved
add full pipeline vs Effort baseline cross-methodEffort · Cross-method avg.video-level AUC: 0.930 -> 0.954Improved
add modules to Effort in inference latencyEffort · Inference time per 100 batchesseconds: 16.24s -> 18.22sHurt

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