Skip to content

[Roadmap] Generating Crop Masks

Jump to bottom
Ivan Zvonkov edited this page Jun 23, 2021 · 1 revision

This roadmap provides the basic requirements for versions of the crop map generation project.

Internal user V1

Primary need: Quickly generate a map in country X for year Y

Release: Ongoing

Example user: internal group member

The system should:

  • Have a clear procedure for a user to provide input dates and region
  • Generate crop map from input dates and region coordinates with a minimal amount of manual steps
  • Support ability to retrain the model with additional training labels

Partner V2

Primary need: Apply the model to generate a map for partner’s country for year Y

Release: April 2021

Example user: RCMRD, Kenya Ministry of Ag

The system should:

  • Generate crop map from input date and region coordinates with no manual steps
  • Method for validating generated crop map (e.g., evaluating on test data or comparing to national statistics)
  • Method of sharing the generated crop map with other team members (e.g., for reviewing in QGIS or Google Earth Engine App)

Partner V3

Primary need: Apply the model to generate a map for their country for year Y and deeply understand mechanisms for independent development of future projects

Release: April 2022

Example user: Rwanda Ag Statistics Bureau The system should:

  • Provide entry points to understand & manipulate mechanics

General Public V4

Primary need: easily evaluate model performance in a [small] ROI/TOI

Example user: someone who is thinking about using our model and wants to test how it does in their ROI

Inspiration from Microsoft Land Cover Mapping

The system should:

  • Allow user to draw ROI and select TOI in browser
  • Allow user to draw new training labels on map & re-train
  • Encapsulate all model mechanics (black box)

The system could:

  • Provide an easy way of sharing the checkpoint/results
Clone this wiki locally