Research map / Video / Video segmentation and tracking
VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| increase tube mask ratio 75% -> 90% (ablation) | VideoMAE ViT-B · Something-Something V2 | top-1 accuracy: 68.0 -> 69.6 | Helped |
| use tube masking instead of random masking (ablation) | VideoMAE ViT-B · Something-Something V2 | top-1 accuracy: 68.3 -> 69.6 | Helped |
| use tube masking instead of frame masking (ablation) | VideoMAE ViT-B · Something-Something V2 | top-1 accuracy: 61.5 -> 69.6 | Helped |
| increase decoder depth 1 -> 4 blocks (ablation) | VideoMAE ViT-B · Something-Something V2 | top-1 accuracy: 68.5 -> 69.6 | Helped |
| increase decoder depth 4 -> 8 blocks (ablation) | VideoMAE ViT-B · Something-Something V2 | top-1 accuracy: 69.6 -> 69.3 | No clear effect |
| reconstruct full clip instead of center frame (ablation) | VideoMAE ViT-B · Something-Something V2 | top-1 accuracy: 63.0 -> 69.6 | Helped |
| VideoMAE self-supervised pretraining instead of ImageNet-21k supervised (ablation) | ViT-B · Something-Something V2 | top-1 accuracy: 61.8 -> 69.6 | Helped |
| pretrain on target dataset instead of transferring from K400 (ablation) | VideoMAE ViT-B · Something-Something V2 | top-1 accuracy: 68.5 -> 69.6 | Helped |
| MSE loss instead of L1 loss (ablation) | VideoMAE ViT-B · Something-Something V2 | top-1 accuracy: 69.1 -> 69.6 | Helped |
| VideoMAE instead of MoCo v3 pretraining | ViT-B · HMDB51 | accuracy: 39.2 -> 62.6 | Helped |
| VideoMAE instead of MoCo v3 pretraining | ViT-B · Something-Something V2 | pre-training hours: 61.7 -> 19.5 | Helped |
| additionally fine-tune on K400 labels before transfer | VideoMAE ViT-B · AVA v2.2 | mAP: 26.7 -> 31.8 | Helped |
Closest papers by meaning
- Extending Video Masked Autoencoders to 128 frames (NeurIPS 2024 · similarity 0.91)
- EVEREST: Efficient Masked Video Autoencoder by Removing Redundant Spatiotemporal Tokens (ICML 2024 · similarity 0.89)
- VideoMAC: Video Masked Autoencoders Meet ConvNets (CVPR 2024 · similarity 0.89)
- Asymmetric Masked Distillation for Pre-Training Small Foundation Models (CVPR 2024 · similarity 0.88)
- TTVidT: Decoupling the Temporal Axis for Efficient Motion-Centric Video Pretraining (NeurIPS 2026 · similarity 0.87)
- Understanding and Enhancing Mask-Based Pretraining towards Universal Representations (NeurIPS 2025 · similarity 0.87)
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.