AI Paper Atlas

Research map / Benchmarks / Celeb-DF

Celeb-DF leaderboard

30 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on Celeb-DF. Best AUC: 99.97 (Patch + RFM, CVPR 2021).

Numbers are as each paper reports them, read by AI from its results tables and not checked by hand. Papers can differ in setting (training data, model size, extra data); the topic column shows what each paper is about. Open the paper before you compare or quote a number.

Celeb-DF · AUC

26 papers, each with its best reported number for its own method; higher is better.

#AUC ↑MethodPaperTopic
199.97Patch + RFM
Patch
Representative Forgery Mining for Fake Face Detection
CVPR 2021 · Table: Comparison of RFM with
CNN features
299.96SFDG (Ours)
EfficientNet-b4
Dynamic Graph Learning With Content-Guided Spatial-Frequency Relation Reasoning for Deepfake Detection
CVPR 2023 · Table 1
CNN features
399.96Ours (SIA)An Information Theoretic Approach for Attention-Driven Face Forgery Detection
ECCV 2022 · Table 3
CNN features
499.95FakeDiffer
Xception
FakeDiffer: Distributional Disparity Learning on Differentiated Reconstruction for Face Forgery Detection
AAAI 2025 · Table 1
OOD and anomaly detection
599.94RECCE
Xception
End-to-End Reconstruction-Classification Learning for Face Forgery Detection
CVPR 2022 · Table 1
CNN features
699.92M2F2-Det
CLIP/ViT-L-patch14-336
Rethinking Vision-Language Model in Face Forensics: Multi-Modal Interpretable Forged Face Detector
CVPR 2025 · Table 2
Human and embodied modeling
795.32CS-DFD
EfficientNet
Suppressing Gradient Conflict for Generalizable Deepfake Detection
TPAMI 2026 · Table 1
Domain adaptation and generalization
893.88Ours (Efficient-b4)
EfficientNet-b4
Implicit Identity Leakage: The Stumbling Block to Improving Deepfake Detection Generalization
CVPR 2023 · Table 2
OOD and anomaly detection
993.3FInferFInfer: Frame Inference-Based Deepfake Detection for High-Visual-Quality Videos
AAAI 2022 · Table 3
Video segmentation and tracking
1089.5NACO with SPM and TCMLearning Natural Consistency Representation for Face Forgery Video Detection
ECCV 2024 · Table 5
Video segmentation and tracking
16 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

Celeb-DF · accuracy

10 papers, each with its best reported number for its own method; higher is better.

#accuracy ↑MethodPaperTopic
199.81HCIL
ResNet-50
Hierarchical Contrastive Inconsistency Learning for Deepfake Video Detection
ECCV 2022 · Table 2
Video segmentation and tracking
299.61Ours
ResNet-50
Delving into the Local: Dynamic Inconsistency Learning for DeepFake Video Detection
AAAI 2022 · Table 2
Video segmentation and tracking
399.35FakeDiffer
Xception
FakeDiffer: Distributional Disparity Learning on Differentiated Reconstruction for Face Forgery Detection
AAAI 2025 · Table 1
OOD and anomaly detection
499.22SFDG (Ours)
EfficientNet-b4
Dynamic Graph Learning With Content-Guided Spatial-Frequency Relation Reasoning for Deepfake Detection
CVPR 2023 · Table 1
CNN features
598.98M2F2-Det
CLIP/ViT-L-patch14-336
Rethinking Vision-Language Model in Face Forensics: Multi-Modal Interpretable Forged Face Detector
CVPR 2025 · Table 2
Human and embodied modeling
698.59RECCE
Xception
End-to-End Reconstruction-Classification Learning for Face Forgery Detection
CVPR 2022 · Table 1
CNN features
798.48Ours (SIA)An Information Theoretic Approach for Attention-Driven Face Forgery Detection
ECCV 2022 · Table 3
CNN features
895.11UNITE (FF++ + GTA-V training)
SigLIP-So400M
Towards a Universal Synthetic Video Detector: From Face or Background Manipulations to Fully AI-Generated Content
CVPR 2025 · Table 1
Video generation
991.5FakeCatcherFakeCatcher: Detection of Synthetic Portrait Videos using Biological Signals
TPAMI 2020 · Table 9
Video segmentation and tracking
1090.47FInferFInfer: Frame Inference-Based Deepfake Detection for High-Visual-Quality Videos
AAAI 2022 · Table 3
Video segmentation and tracking

Not affiliated with any conference, journal or the benchmark's authors. Corrections: contact@latenticlabs.com.