khdlr / SoundingEarth

Self-supervised Audiovisual Representation Learning for Remote Sensing Data

SoundingEarth

Code for Self-supervised Audiovisual Representation Learning for Remote Sensing Data

Data

Download the dataset at https://zenodo.org/record/5600379 and configure DataRoot in config.yml to point to the extracted archive.

Code

• train.py is the main engine for the project.
• Configuration is done in config.py (defaults) and config.yml (experiments)
• Data loading in data_loading.py

All other interesting stuff goes in lib/:

• Models to lib/models/
• Loss functions in lib/loss_functions.py

Pre-trained Weights

Find model weights in the Releases tab