V

Videomae Base Short Ssv2

Developed by MCG-NJU
VideoMAE is a self-supervised pretraining model for videos based on Masked Autoencoder (MAE), pretrained for 800 epochs on the Something-Something-v2 dataset.
Downloads 112
Release Time : 8/2/2022

Model Overview

This model learns internal video representations through self-supervision, primarily for fine-tuning downstream tasks such as video classification.

Model Features

Video Self-supervised Learning
Uses Masked Autoencoder (MAE) method for video self-supervised pretraining, requiring no labeled data
Efficient Pretraining
Pretrained for 800 epochs on Something-Something-v2 dataset to learn internal video representations
Downstream Task Adaptation
Pretrained model can be fine-tuned for various video understanding tasks

Model Capabilities

Video Feature Extraction
Self-supervised Learning
Video Representation Learning

Use Cases

Video Understanding
Video Classification
Fine-tune the pretrained model for video content classification
Action Recognition
Can be used for human action recognition tasks in videos
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
Š 2025AIbase