lyuxianqiang / RainyScape

Code for 'RainyScape: Unsupervised Rainy Scene Reconstruction using Decoupled Neural Rendering'

RainyScape

Code for ACM MM'24 paper: 'RainyScape: Unsupervised Rainy Scene Reconstruction using Decoupled Neural Rendering'

Get the RainyScape_dataset

Please download the dataset via the following Onedirve link.

https://portland-my.sharepoint.com/:u:/g/personal/xianqialv2-c_my_cityu_edu_hk/EWLAW9WZ99lPq1j7OPA-1kEBwvq7k4MMwWpgXkTMh5pmnA?e=NnpnvZ

After get the dataset, put the dataset into data folder.

Training and Testing RainyScape

NeRF Rendering

Setting

Install the environment.

cd RainyScape-NeRF
pip install -r requirements.txt

Changing the data path and log path in the configs/Yard_deraining_unsup.txt

Training

To train and test the RainyScape-NeRF model, use the following command:

CUDA_VISIBLE_DEVICES=0 python torch_nerf_rainyscape.py --config configs/Yard_deraining_unsup.txt

3DGS Rendering

Setting

Install the environment. Before running the 3DGS rendering, make sure to configure the basic environment for 3DGS. Refer to the official 3DGS project page for the setup instructions.

Convert the dataset by COLMAP for 3DGS input.

Training

Once the environment is configured, you can run the rendering with:

CUDA_VISIBLE_DEVICES=0 python torch_3dgs_rainyscape.py -s data/Yard/

RainyScape for NeRF rendering

RainyScape for 3DGS rendering

Acknowledgments

If you use the code or the dataset in you own paper, please cite

@misc{lyu2024rainyscapeunsupervisedrainyscene,
   title={RainyScape: Unsupervised Rainy Scene Reconstruction using Decoupled Neural Rendering}, 
   author={Xianqiang Lyu and Hui Liu and Junhui Hou},
   year={2024},
   eprint={2404.11401},
   archivePrefix={arXiv},
   primaryClass={cs.CV},
   url={https://arxiv.org/abs/2404.11401}, 
}

This project is based on the following implementations:

• The pytorch implementation of NeRF: https://github.com/yenchenlin/nerf-pytorch

• The 3DGS official code: https://github.com/graphdeco-inria/gaussian-splatting