Skip to content

Latest commit

 

History

History
29 lines (16 loc) · 1.39 KB

README.md

File metadata and controls

29 lines (16 loc) · 1.39 KB

ESA SmartCrop PiSchool autumn 2019

SmartCrop is build upon the Hugin framework version 0.1.x. Hugin is a Python framework designed to help the scientists run Machine Learning experiments on geospatial raster data.

Current extensions include:

  • Z Score standardization performed over entire training set/per channel before training
  • Transfer weights without including last classsification layer
  • Tiling the image without the requirement of having all the input images of the same size
  • Include U-Net model topology
  • Include a proposed implementation for HSN model and W-Net model
  • Include Hugin configuration files for both training and prediction phases for U-Net, HSN and W-Net
  • Add prediction metric computation for multi-class semantic segmentation

This is a proof of concept. The above mentioned extensions are going to be included in the new Hugin release. Please visit the Wiki page for a detailed information about the content of this repository.

Additional documentation for Hugin is available at https://hugin-eo.readthedocs.io/

Acknowledgments

This project was carried out under the supervision of the following stakeholders ESA, E-Geos, UrbyetOrbit, MEEO (Italy) and SISTEMA (Austria) .

Hugin project development is supported by the European Space Agency through the ML4EO project.