AllClear
is a comprehensive dataset/benchmark for cloud detection and removal.
Notice: We are actively cleaning up the codebase and uploading our dataset for public access. Stay tuned!
Please navigate to the root directory of this project and run the following commands:
# Clone the repository
git clone https://github.com/Zhou-Hangyu/allclear.git
# Obtain the submodules
cd allclear
git submodule update --init --recursive
# Download the test dataset.zip and metadata json file.
. preprocess.sh
This section provides instructions on how to use the benchmark with the UnCRtainTS
model as an example.
First, set up the environment for UnCRtainTS
. Visit the UnCRtainTS GitHub page and follow the instructions there to create their conda environment.
After setting up the UnCRtainTS
environment, navigate to the root directory of this project and install our package using pip:
pip install -e .
To run the benchmark and see some results, execute the run_benchmark.sh
script located in the demos
directory:
# Run the Least Cloudy baseline
bash demos/run_benchmark_leastcloud.sh
# Run the pretrained UnCRtainTS
bash demos/run_uncrtaints_pretrained.sh
# Run the UnCRtainTS pretrained on our full allclear dataset
bash demos/run_uncrtaints_allclear100pc.sh
This project is licensed under the MIT License.
allclear
. Should only contain reusable code directly related to the use of the dataset and benchmark./baselines
folder.
allclear/baselines.py
with uniform input/output format for easy comparison.demo
folder contains minimal code to demonstrate the use of the dataset and benchmark./experimental_scripts
folder for now.