This is a minimal working code to evaluate a pre-trained simple neural compression model on Solar Dynamics Observatory (SDO) images. This code is based on CompressAI, but to keep it uncluttered a tiny clone of the package is provided here.
Model definition can be found in tiny_compressai/models/nasa.py.
A local clone of the CompressAI is provided (tiny_compressai) to make the model integration easier.
In a virtual environment follow the steps below (verified on Ubuntu):
git clone https://github.com/ali-zafari/nasa-sdo-compressor nasa_compressor
cd nasa_compressor
pip install -U pip
pip install -e .For bulk data evaluation of a saved checkpoint, compressai.utils.eval is used. An example to test the rate-distoriton perfomance of a NASACompressor checkpoint:
python -m tiny_compressai.utils.eval_model checkpoint path/to/data/directory -a nasa-compressor --cuda -v -p path/to/a/checkpointTo download a set of data and a checkpoint: Sample Data + Pretrained Network
A jupyter notebook (compression_evaluation.ipynb) is also provided to make clear the procedure of model instantation and single image evaluation. Use the following command to install Jupyter before running the notebook.
pip install jupyterlab