Papers › MARLIN: Masked Autoencoder for facial video Representation LearnINg

MARLIN: Masked Autoencoder for facial video Representation LearnINg

12 Nov 2022CVPR 2023 1arXiv:2211.06627archive 2025-07-28

Zhixi Cai, Shreya Ghosh, Kalin Stefanov, Abhinav Dhall, Jianfei Cai, Hamid Rezatofighi, Reza Haffari, Munawar Hayat

This paper proposes a self-supervised approach to learn universal facial representations from videos, that can transfer across a variety of facial analysis tasks such as Facial Attribute Recognition (FAR), Facial Expression Recognition (FER), DeepFake Detection (DFD), and Lip Synchronization (LS). Our proposed framework, named MARLIN, is a facial video masked autoencoder, that learns highly robust and generic facial embeddings from abundantly available non-annotated web crawled facial videos. As a challenging auxiliary task, MARLIN reconstructs the spatio-temporal details of the face from the densely masked facial regions which mainly include eyes, nose, mouth, lips, and skin to capture local and global aspects that in turn help in encoding generic and transferable features. Through a variety of experiments on diverse downstream tasks, we demonstrate MARLIN to be an excellent facial video encoder as well as feature extractor, that performs consistently well across a variety of downstream tasks including FAR (1.13% gain over supervised benchmark), FER (2.64% gain over unsupervised benchmark), DFD (1.86% gain over unsupervised benchmark), LS (29.36% gain for Frechet Inception Distance), and even in low data regime. Our code and models are available at https://github.com/ControlNet/MARLIN .

PaperPDFConference PDFCode

Code

ControlNet/MARLIN officialmentioned in papermentioned on GitHubpytorchNOASSERTION report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Action ClassificationAttributeDeepFake DetectionEmotion ClassificationFace SwappingFacial Attribute ClassificationFacial Expression RecognitionFacial Expression Recognition (FER)Multimodal Sentiment AnalysisRepresentation LearningSentiment AnalysisUnconstrained Lip-synchronization

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Action Classification CelebV-HQ MARLIN AUC 0.9406 #1 of 1 Archive leaderboard report
Action Classification CelebV-HQ MARLIN Accuracy 95.48 #1 of 1 Archive leaderboard report
DeepFake Detection FaceForensics++ MARLIN (ViT-L) AUC 0.9377 #3 of 6 Archive leaderboard report
DeepFake Detection FaceForensics++ MARLIN (ViT-B) AUC 0.9305 #4 of 6 Archive leaderboard report
DeepFake Detection FaceForensics++ MARLIN (ViT-S) AUC 0.8863 #5 of 6 Archive leaderboard report
Emotion Classification CMU-MOSEI MARLIN (ViT-L) Accuracy 80.63 #1 of 4 Archive leaderboard report
Emotion Classification CMU-MOSEI MARLIN (ViT-B) Accuracy 80.6 #2 of 4 Archive leaderboard report
Emotion Classification CMU-MOSEI MARLIN (ViT-S) Accuracy 80.38 #3 of 4 Archive leaderboard report
Facial Attribute Classification CelebV-HQ MARLIN AUC 0.9561 #1 of 1 Archive leaderboard report
Facial Attribute Classification CelebV-HQ MARLIN Accuracy 93.9 #1 of 1 Archive leaderboard report
Multimodal Sentiment Analysis CMU-MOSEI MARLIN (ViT-L) Accuracy 74.83 #12 of 15 Archive leaderboard report
Multimodal Sentiment Analysis CMU-MOSEI MARLIN (ViT-B) Accuracy 73.7 #13 of 15 Archive leaderboard report
Multimodal Sentiment Analysis CMU-MOSEI MARLIN (ViT-S) Accuracy 72.69 #14 of 15 Archive leaderboard report
Unconstrained Lip-synchronization LRS2 Wav2Lip + ViT + MARLIN FID 3.452 #1 of 3 Archive leaderboard report
Unconstrained Lip-synchronization LRS2 Wav2Lip + ViT + MARLIN LSE-C 5.528 #1 of 3 Archive leaderboard report
Unconstrained Lip-synchronization LRS2 Wav2Lip + ViT + MARLIN LSE-D 7.127 #1 of 3 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

Introduced by this paper: MARLIN

MAEMARLIN

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections