Tag: ai segmentation

One French suburb at six pixel sizes. At 10 m, 0.2% of its buildings cover 16 pixels; at 50 cm, 96% do. Counts of 20,911 buildings and 2,966 pools.

Semantic segmentation labels every pixel with a class: building or not. Instance segmentation draws every building as its own object.

Browser panel > XYZ Tiles > double-click OpenStreetMap. Then satellite tiles by URL, QuickMapServices, a national WMS, and the ground size of each zoom.

Check OpenStreetMap first. Then four ways to map rooftop PV from aerial imagery in QGIS, with the menu paths and where each one fails.

Start with QuickOSM on leisure=swimming_pool. Where the map is thin, use a segmentation model, threshold the blue, or run a water index. Five methods compared.

Start with QuickOSM: in four French towns OSM held 96 to 98% of paved road length. Where it is thin, five more methods and the line-or-surface choice.

For multispectral imagery, threshold NDVI in the raster calculator, then polygonize. Six methods, what each one needs, and what a canopy layer costs.

Draw a zone, name the object, run it: you get a polygon layer in minutes. Open data, AI detection and hand-digitizing compared, with the filtering step.

SAM outlines an object from a click or a word. Seven QGIS plugins run it: what each costs, which needs an NVIDIA card, and how to install one.

Draw a zone, name the object, get every polygon as a clean vector layer. The full run in QGIS, the three-step review in any order, and Semi-Auto mode.