I have four ultra-high-resolution aerial photos that need precise AI-based extraction of 4 key object classes: roads/pathways, buildings, cleared land/construction sites, and water bodies (reservoir, ponds, lakes). Your job is to run the segmentation or detection workflow you trust most, then convert the resulting polygons into a clean GeoJSON file that opens flawlessly in any GIS viewer as GIS LINE or GIS filled Polygon shape. The geotiffs provided have all the geo metadata included. And will need to be used for most standard GIS platforms (Esri, QGis). Please make sure every structure and road edge is tightly traced—no gaps, over-merging, or dangling nodes. The geometry should be topologically valid, use WGS-84, and include an attribute field that tags each feature as either “building” or “road”. The output should be a smoothed centerline for roads (dirt roads in one color, paved roads in another), the buildings, cleared areas, and waterbodies are clear to see and should be GIS Shapes that are filled and no jagged edges. Deliverables • A GeoJSON file containing all classes outlines as separate layers, ready for import and use • Typical outputs of prj shp etc for each class • a python based github stored Model that processes and delivers the above output using similar Geotiff files I will verify your work by overlaying your output in QGIS and measuring IoU against spot-checked ground truth. Speed to developing this output and model is crucial. As in start today, could complete today as well. If you routinely work with tools such as Mask R-CNN, Yolo, SAM2/3, AI Vision, U-Net, Detectron2, or other segmentation detection—and you’re comfortable post-processing in GDAL/OGR—this task should be straightforward. If you know this world of image data analysis and segmenting, should be straightforward easy money. Let me know your preferred model pipeline and a realistic turnaround time; I’m aiming to approve the file as soon as possible. We have a few models internally if needed for adapting/use or insight.