Create a Python 3.9 Conda environment and install the repo dependencies:
conda create -n demetra python=3.9 -y
conda activate demetra
python -m pip install --upgrade pip
python -m pip install -r requirements.txtNote: the file currently pins torch==1.12.1+cu113, torchvision==0.13.1+cu113
and torchaudio==0.12.1+cu113. If your machine does not use CUDA 11.3, adjust
those lines before installing.
Launch the following script to download image data from Eumetsat (using your account keys) and track with DeMeTra
export EUMETSAT_CONSUMER_KEY=<your_consumer_key>
export EUMETSAT_CONSUMER_SECRET=<your_consumer_secret>
conda activate demetra
python scripts/download_and_track_range.py --start 15-03-2026 --end 17-03-2026 \
--firstpass_model_path <firstpass_model_checkpoint> \
--tracking_model_path <videomae_model_checkpoint>the script will automatically download data from EUMETSAT using your account keys
python scripts/predict_firstpass_and_track_from_folder.py \
--input_dir /media/isacDisk2/source_dataset_by_cyc/jolina \
--output_dir /media/isacDisk2/demetra_output/jolina \
--firstpass_model_path /media/isacDisk1/Demetra/trained_models/firstpass_model.ckpt \
--tracking_model_path /media/isacDisk2/demetra_trained_models/checkpoint_new_tracking2.pth \
--firstpass_threshold 0.2 \
--make_video \
--ffmpeg_path /mnt/share/Demetra_files/VideoMAEv2/ffmpeg-7.0.2-amd64-static/ \
--standard_tiling \
--video_coastlines \
--video_tracking_dot_only