YuShi: Physically-informed volumetric structure modeling for single-radar nowcasting

This repository contains the official implementation of Yushi framework.

🚀 Getting Started

1. Construct Environment

git clone https://github.com/OpenEarthLab/Yushi
cd Yushi
conda create -n yushi python=3.10
conda activate yushi 

2. Build Auxiliary Modules

cd Pytorch-Correlation-extension
pip install -e .
cd Oriented1D/models/dwoconv1d
pip install -e .
cd -
cd Oriented1D/models/dwoconv1d_specialized
pip install -e .

3. Download Checkpoint and Examples

Download the [checkpoint](https://drive.google.com/file/d/1pbiMmdxPWUfuwTClZSr8nTWiaAVx2k3f/view?usp=drive_link) and [examples](https://drive.google.com/drive/folders/1cy2Si3MUma5he9F36n1f8O6IX7jsZwS5?usp=drive_link) from Google Drive.

4. Conduct Inference

CUDA_VISIBLE_DEVICES=0 python inference.py

---

Acknowledgements

We would like to express our gratitude to the following open-source projects, whose high-quality code and research served as a foundation for this work: Pytorch-Correlation-extension, Oriented1D, NowcastNet and Pytorch-UNet.

Contact Information

Please feel free to contact zpq0316@163.com if you have questions.