This is worked example using data sourced from the Land Use and Coverage Area frame Survey (LUCAS) data set, following a strict sampling and acquisition protocol. Landscape objects and elements were fully delineated on the street (eye) level landscape images and can provide critical context for land-use and land-cover assessments and comparison to satellite remote sensing data.
This worked example provides a single class segmentation example framework which should be easily expanded to multiclass problems. The code relies on pytorch and the Lightning modelling framework. Environment and Docker files are provided to ease code deployment.
LUCAS data source:
https://beta.source.coop/repositories/jrc-lucas/jrc-lucas-ml/description/