Weights from Precise Forecasting of Sky Images Using Spatial Warping
We have trained the weight from the official respository.
Quick start:
-
Install the dependencies with
pip install -r requirements.txt
-
Download the dataset from here and put them into
SkyNet_Data
folder. -
To run the inference on the pretrained model. The pretrained weight is in
./weights
(SkyImage) user@user:diretory/Precise-Forecasting-of-Sky-Images-Using-Spatial-Warping$ python3 test.py
- To run training for the new model
(SkyImage) user@user:diretory/Precise-Forecasting-of-Sky-Images-Using-Spatial-Warping$ python3 train.py
Our prediction result:
Precise Forecasting of Sky Images Using Spatial Warping
Leron Julian, Aswin Sankaranarayanan
Image Science Lab, Carnegie Mellon University
Paper Dataset
SkyNet imrpoves sky-image prediction to model cloud dynamics with higher spatial and temporal resolution than previous works. Our method handles distorted clouds near the horizon of the hemispherical mirror by patially warping the sky images during training to facilitate longer forecasting of cloud evolution.
# To download dataset for train and test data:
pip install gdown
gdown --folder --id 1BkWx0j6Kt5G8CEMzzREprMeoYfw0v4ge
Installation
# Installation using using anaconda package management
conda env create -f environment.yml
conda activate SkyNet
pip install -r requirements.txt
# How to train the model with default parameters:
python train.py
# For info about command-line flags use
python train.py --help
Thanks
This project makes use of LiteFlowNet for optical-flow estimates:
- LiteFlowNet2 for lightweight optical-flow estimates using a CNN Please refer to their webpage for installation and implementation
Citation
If you use this project in your research please cite:
@INPROCEEDINGS{SkyNet:ICCVW21,
author={Julian, Leron and Sankaranarayanan, Aswin C.},
booktitle={2021 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW)},
title={Precise Forecasting of Sky Images Using Spatial Warping},
year={2021},
}