RainNet: A Large-Scale Imagery Dataset and Benchmark for Spatial Precipitation Downscaling

Accepted by NeurIPS 2022

Xuanhong Chen*, Kairui Feng*, Naiyuan Liu, Bingbing Ni**, Yifan Lu, Zhengyan Tong , Ziang Liu

* Equal contribution ** Corresponding author

[Project Website] [Paper] [NeurIPS2022 Presentation] [Supplementary Material]

The official repository with Pytorch

Top News <img width=8% src="./docs/img/new.gif"/>

2024-09-08: We update the google drive of RainNet [Google Driver] RainNet_HDF5.zip (13.6G). We thank SocialBook for providing us with enough shared storage space to continue making this dataset available.

2024-01-21: We provide the [Supplementary Material].

2022-11-16: The download links are now avaliable: [Google Driver] RainNet_HDF5.zip (13.6G) [Baidu Driver] RainNet_HDF5.zip (13.6G) [Password: sjtu].

2022-11-16: We are working for metric tools and annotation of events.

Download RainNet

[Download Via Google Drive] RainNet_HDF5.zip (13.6G)

[Download Via Baidu Drive] RainNet_HDF5.zip (13.6G) [password: sjtu]

Resources in Zip:

RainNet_HDF5.zip

├ $year$_07.hdf5

├ $year$_08.hdf5

├ $year$_09.hdf5

├ $year$_10.hdf5

└ $year$_11.hdf5

$year$=2002~2018

85 HDF5 files in total;
322GB of hard disk space is required to extract the dataset.

Dependencies

python3.6+
pytorch1.5+
torchvision
h5py
numpy

Usage

Data preparation. Run the 'dataset_prepare_hdf5.py' to process the dataset into patches. In 'dataset_prepare_hdf5.py', variable 'dataset_path' sets the hdf5 file path of RainNet; 'patch_hdf5_root' sets the target path to save processed dataset:
python dataset_prepare_hdf5.py
We provide a example dataloader (pytorch script) to read the processed dataset:
dataloader_hdf5.py
python scripts are archived in fold 'scripts'

Samples in RainNet

High Resolution Precipitation Map:

<img src="./docs/img/HRGT_201009539_201009571.webp" style="zoom: 10%;" /> ### Low Resolution Precipitation Map: <img src="./docs/img/LRGT_201009539_201009571.webp" style="zoom: 10%;" /> ### High Resolution Precipitation Map: <img src="./docs/img/HRGT_201108607_201108655.webp" style="zoom: 20%;" /> ### Low Resolution Precipitation Map: <img src="./docs/img/LRGT_201108607_201108655.webp" style="zoom: 20%;" /> ### High Resolution Precipitation Map: <img src="./docs/img/HRGT_201109091_201109123.webp" style="zoom: 20%;" /> ### Low Resolution Precipitation Map: <img src="./docs/img/LRGT_201109091_201109123.webp" style="zoom: 20%;" />

Citation

If you find this Dataset useful in your research, please consider citing:

@misc{chen2020rainnet,
  title={RainNet: A Large-Scale Dataset for Spatial Precipitation Downscaling},
  author={Xuanhong Chen and Kairui Feng and Naiyuan Liu and Yifan Lu and Zhengyan Tong and Bingbing Ni and Ziang Liu and Ning Lin},
  year={2020},
  eprint={2012.09700},
  archivePrefix={arXiv},
  primaryClass={cs.CV}
}