MCG-NJU/VideoMAE
Official PyTorch implementation of VideoMAE, a masked autoencoder for self-supervised video representation learning.
Not currently ranked — collecting fresh signals.
star history
VideoMAE is a NeurIPS 2022 Spotlight paper implementing masked autoencoders for data-efficient self-supervised video pre-training. The repository provides PyTorch training code and pretrained models for video understanding tasks including action recognition on Kinetics-400, Something-Something, and UCF-101 datasets. It uses vision transformers (ViT) adapted for video data with masked token reconstruction as the pretraining objective.
Frequently asked
- What is MCG-NJU/VideoMAE?
- Official PyTorch implementation of VideoMAE, a masked autoencoder for self-supervised video representation learning.
- Is VideoMAE open source?
- Yes — MCG-NJU/VideoMAE is an open-source project tracked on heatdrop.
- What language is VideoMAE written in?
- MCG-NJU/VideoMAE is primarily written in Python.
- How popular is VideoMAE?
- MCG-NJU/VideoMAE has 1.8k stars on GitHub.
- Where can I find VideoMAE?
- MCG-NJU/VideoMAE is on GitHub at https://github.com/MCG-NJU/VideoMAE.