Spectroscopic Transformer

This repository contains code for the Spectroscopic Transformer, a pixel-wise deep learning model for spectral tasks.

Subdirectories of this repository contain research code for specific publications and/or applications.

Note

Active development on a more formalized package for easier installation and deployment is ongoing in the dev branch.

Dependencies

Note

Conda environment definitions will be provided in the future.

• Pytorch
• ISOFIT
• Spectral Python
• Rasterio
• Schedulefree
• sklearn
• pyyaml
• numpy
• (for training logging) WandB

Publications

SpecTf: Transformers Enable Data-Driven Imaging Spectroscopy Cloud Detection

Jake H. Lee, Michael Kiper, David R. Thompson, Philip G. Brodrick. In Review

Preprint: https://arxiv.org/abs/2501.04916

For details see the cloud directory