Skip to content

Determine clear names for each of the FTW 'components' #27

@cholmes

Description

@cholmes

At the last in person meeting we agreed that FTW should be the overall name of the whole initiative, not just for the benchmark dataset. But we need to get clarity on the names of each of the components. Some are fairly obvious for what to do, but some are a little less clear:

  • FTW Benchmark Dataset - this is currently called 'Fields of The World', just append 'benchmark dataset', to be clear it's talking about the data.
  • FTW Baseline Models - they're basically already called this, so that's easy. It's the 4 or so models that we publish.
  • FTW Tools - it does feel like there's 2-3 sets of tools, so not sure if we want to differentiate some between them:
    • command line tools & python libraries for downloading the source benchmark data and creating your own models (currently in ftw-baselines).
    • command line tools and python libraries for downloading scenes to run inference on generate fiboa output (currently in ftw-baselines, but we created ftw-tools with an intent to move them out to there).
    • Tools to format FTW benchmark data. These mostly don't exist or were just internal scripts the ASU team used, but it'd be nice to make the process to add a new dataset to the FTW benchmark dataset be easy. Like to take a source fiboa data and then subset it into test / train, generate the s2 masks, etc.
  • FTW QGIS Plugin - could group under 'tools', name is pretty straightforward.
  • FTW Web App(s) - we have 'inference-app' and 'web-app' (which is just scene selection). Should get to one experience, could get a cool name, or just call 'web app'.
  • FTW 'source data'? This is one of the tricky ones. It's roughly what's on https://fiboa.org/map/ - the data that is public field boundaries, that could potentially go into FTW. I lean towards this not including ML generated output, and calling that something else. But we could also just have a name for the universe of field boundary data that's in fiboa.
  • FTW 'output data'? 'data product'? When we publish like a country-wide map from one of the models we should have a name for that. It's what many people will just care about, ideally we version these, etc.

I'm probably missing some, but just wanted to get the conversation started.

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

Status

In Progress

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions