Research map / Video / Video segmentation and tracking
MARLIN: Masked Autoencoder for Facial Video Representation LearnINg
Zhixi Cai, Shreya Ghosh, Kalin Stefanov, Abhinav Dhall, Jianfei Cai, Hamid Rezatofighi, Reza Haffari, Munawar Hayat
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 |
|---|---|---|---|
| pre-train with MARLIN rather than VideoMAE | ViT-B facial video encoder · CMU-MOSEI 7-class sentiment | accuracy: 33.78 -> 34.63 | Improved |
| pre-train with MARLIN rather than VideoMAE | ViT-B facial video encoder · FaceForensics++ (LQ) | accuracy: 87.57 -> 89.43 | Improved |
| use MARLIN features rather than the Wav2Lip baseline encoder | Wav2Lip on LRS2 · LRS2 | FID: 4.887 -> 3.452 | Improved |
| add Fasking to VideoMAE pre-training (ablation) | ViT-B facial video encoder · CMU-MOSEI 7-class sentiment | accuracy: 33.78 -> 34.58 | Improved |
| add Fasking to VideoMAE pre-training (ablation) | ViT-B facial video encoder · FaceForensics++ | accuracy: 87.57 -> 87.29 | Hurt |
| add adversarial training to VideoMAE pre-training (ablation) | ViT-B facial video encoder · CMU-MOSEI emotion | accuracy: 80.39 -> 80.58 | Improved |
| add Fasking and adversarial training (ablation) | ViT-B facial video encoder · FaceForensics++ | accuracy: 87.57 -> 89.43 | Improved |
| scale the encoder from ViT-B to ViT-L (ablation) | MARLIN · CMU-MOSEI 7-class sentiment | accuracy: 34.63 -> 35.28 | Improved |
| replace Fasking with random masking (ablation) | MARLIN · FaceForensics++ | AUC: 0.9305 -> 0.8797 | Hurt |
| replace Fasking with frame masking (ablation) | MARLIN · CMU-MOSEI emotion | accuracy: 79.33 vs 80.60 | Hurt |
| replace Fasking with standard tube masking (ablation) | MARLIN · FaceForensics++ | AUC: 0.9096 vs 0.9305 | Hurt |
| use a masking ratio near 90% instead of lower or higher ratios (ablation) | MARLIN · CMU-MOSEI | accuracy | Mixed |
Later papers that compared with this method
1 later accepted paper reports MARLIN as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Ex2Eg-MAE: A Framework for Adaptation of Exocentric Video Masked Autoencoders for Egocentric Social Role Understanding (ECCV 2024 · better on 3 of 3 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- VideoMAE — VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training (NeurIPS 2022) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- PrefAce: Face-Centric Pretraining with Self-Structure Aware Distillation (AAAI 2024 · similarity 0.87)
- Pre-training Strategies and Datasets for Facial Representation Learning (ECCV 2022 · similarity 0.87)
- Self-Supervised Facial Representation Learning with Facial Region Awareness (CVPR 2024 · similarity 0.85)
- AVF-MAE++: Scaling Affective Video Facial Masked Autoencoders via Efficient Audio-Visual Self-Supervised Learning (CVPR 2025 · similarity 0.84)
- Extending Video Masked Autoencoders to 128 frames (NeurIPS 2024 · similarity 0.84)
- EVEREST: Efficient Masked Video Autoencoder by Removing Redundant Spatiotemporal Tokens (ICML 2024 · similarity 0.84)
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.
