An ML project focused on multi-label cloud type identification primarily from the NASA GLOBE CLOUD GAZE dataset. Built with PyTorch, torchvision backbones, multi-label metrics, and MLflow tracking.
Still in progress! More experiments and better models coming soon :)
Check out the current deployment on Hugging Face Spaces.
Take a look at this blog post on my website, which documents part of the process so far.
Special emphasis is placed on object oriented design principles. I took inspiration from Beyond Jupyter. For the data, I also explored using the CCSN dataset on Harvard Dataverse and the plain NASA GLOBE data.