nowcastnet-rewritten 0.0.2

A personal reimplementation of NowcastNet inference framework

NowcastNet-Rewritten

1. Introduction

This project is a personal reimplementation of the NowcastNet inference framework. The original research, titled "Skilful nowcasting of extreme precipitation with NowcastNet," by Yuchen Zhang, Mingsheng Long et al., was published in Nature and can be accessed at https://www.nature.com/articles/s41586-023-06184-4. Additionally, the original code by Yuchen Zhang is available at https://doi.org/10.24433/CO.0832447.v1.

Q: Why reimplement? A: Just for learning :)

2. Getting Started

1. Cloning the repository:

   git clone https://github.com/VioletsOleander/nowcastnet-rewritten.git
   

2. Install the package from PyPI:

   pip install -U nowcastnet-rewritten
   

Notes:

• You may need to implement your own code to read the dataset. Sample code for reading the radar dataset is provided in the datasets directory.
• To ensure compatibility with this reimplementation's architecture, weights have been modified and are available for download from Hugging Face.

3. Usage

To start inference, run inference.py with required arguments.

To get an overview of the arguments, start with the basic command:

python inference.py -h

Here is an example shell script do_inference.sh to streamline the process. You can adjust it accordingly:

#!/bin/bash
python inference.py \
    --case_type normal \
    --device cuda:0 \
    "path_to_weights" \
    "path_to_data" \
    "path_to_result" \

4. Example Inference Result

1024 x 1024:

512 x 512: