kurteulau / ucb_mids_w207_final_project

UC Berkeley MIDS W207 Applied Machine Learning Final Project

UC Berkeley MIDS W207 Applied Machine Learning Final Project

Spring 2022

Data Sources

Satellite images taken at 4 different bandwidths from the Copernicus Sentinel-2 mission stored on Microsoft Planetary Computer. In total, there are 11,748 chips, each with 5 items: 4 .tif images for each bandwidth and a label mask that details which pixels are cloud pixels and which are not cloud pixels.

Project Organization

├── LICENSE
├── README.md                           <- The top-level README describing the project
├── notebooks                           <- Jupyter notebooks
    ├── archive                         <- Old code from preivous iterations of notebooks
    ├── benchmark_src                   <- Python files returned from running a model
    └── lightning_logs                  <- Logs resulting from PyTorch Lighting
└── references                          
    ├── planetary-computer-containers   <- Pre-loaded containers from Planetary Computer
    └── identify_mis_labeled_chips      <- Code, PNGs, CSVs used to find bad chips

Environment

All files run in Microsoft Planetary Computer when PyTorch - GPU option is selected.