This is the code for SkyCloud: A Neural Network-based Approach for Sky and Cloud Segmentation from Natural Images published at 8th International Conference on Image, Vision and Computing (ICIVC).

The repository currently only contains the code for the evaluation. The training code is currently optimized for multi-GPU training and will be published at the time of publication. The code was tested on Ubuntu 22.04 LTS.

• Python: >3.2

• CUDA: 10.2

To install the required python packages run:

pip install -r requirements.txt

The SkyCloud data set and the pretrained weights are available here

Download the SkyCloud data set. Open the config file config/config.yaml and add the path to the data set root folder at root_dataset

Download the pretrained weights and save them to the weights folder. Alternatively change the DIR folder to the location of the weights.

Test the network:

python3 eval.py --cfg config/config.yaml