A repository containing the implementation of the paper entitled The ROAD to discovery: machine learning-driven anomaly detection in radio astronomy spectrograms
Install conda environment by:
conda create --name road python=3.9.7
Run conda environment by:
conda activate road
Install the appropriate pytorch version:
conda install pytorch torchvision torchaudio pytorch-cuda=<VERSION> -c pytorch -c nvidia
Install dependancies by running:
pip install -r requirements
You will need to download the ROAD dataset and specify the its path using -data_path command line option.
Run the following to replicate the results for the resnet34 used in the paper
./experiments/final_model.sh
or to run for all backbones
./experiments/test.sh
Alternatively the model weights can be downloaded and specified using the -model_name and -model_path flags.
The labelling interface is based on label-studio. To get the label server running for the LOFAR_AD project, run the following:
label-studio start LOFAR_AD --sampling uniform &
and
./webserver /home/mmesarcik/data/LOFAR/compressed/LOFAR_AD/LOFAR_AD_v1/ *.png files 8081
Source code of ROAD is licensed under the MIT License.