---
license: mit
tags:
- video-classification
- zero-shot-learning
- action-recognition
- videomae
pipeline_tag: video-classification
---

# DeCo-MAE: Decomposing Semantics for Compositional Zero-Shot Action Recognition

This repository contains the **official weights** and code for the paper **"DeCo-MAE"**.

## 🏆 Model Zoo

| Model | Setting | Accuracy | Download |
| :--- | :--- | :--- | :--- |
| **DeCo-MAE (Final SOTA)** | Fully Supervised | **85.80%** | [Download](./final_sota_best.pth) |
| **DeCo-MAE (Zero-Shot)** | Zero-Shot Split | **78.86%** | [Download](./zeroshot_model.pth) |

## 🚀 Usage

Please refer to the [GitHub Repository](https://github.com/SuhangXia/DeCo-MAE) for detailed instructions.