Fotoshi Labs
Internal demos for map, photo, and model-assisted workflows.
Prioritise photo collection using corpus coverage, held-out category accuracy, and zero-photo targets.
Isolated low-zoom palette study for FOT-579; ordinary map detail remains intact.
Try style-only backings for real @2x POI sprites over satellite imagery.
Synchronised MapTiler, Azure Maps, and Esri imagery at the same places and zoom levels.
MapLibre prototype that converts coarse land imagery into existing Fotoshi basemap colours.
Click covered map objects and inspect aggregate AI knowledge by fid plus per-photo observations by oid.
Android-flow simulator for model-assisted place tagging and missing-object creation.
Crop visible text from attached place photos and use it as a place-name suggestion.