SAM 3 satellite segmentation workflow in RasterForge

SAM 3 satellite segmentation workflow in RasterForge

Use this workflow when you want promptable segmentation for satellite or aerial imagery and need the result to stay tied to project images, render choices, review state, and export decisions.

RasterForge treats SAM 3 as one step in a larger workflow. The model sees the rendered image, so import quality, AOI choice, render profile, examples, Auto masks settings, and review process all affect the final output.

Inputs

  • GeoTIFF, PNG, or JPEG imagery imported into a RasterForge project.
  • A render where the target can be seen clearly by a human reviewer.
  • Prompt text, boxes, points, saved examples, or Auto masks settings that describe the segmentation task.
  • A review plan for false positives, missed areas, noisy boundaries, and export format.

Workflow

  1. Import imagery and confirm bounds if the image has geographic meaning.
  2. Prepare the render. Use a profile that exposes the target and avoids unnecessary clipping or low-contrast boundaries.
  3. Open Segment and choose Find things, Select objects, or Auto masks.
  4. Create, select, or import a segmentation profile when using a profile-based workflow. Keep the target name and class metadata specific enough for review and export.
  5. Add guidance. Use prompt text for repeated objects, boxes for target areas, points for targeted selection, and examples when a profile should transfer across images.
  6. Run one image first. Inspect the output layer, object count, geometry quality, Pixel mask, GeoJSON, and confusing background classes.
  7. Run selected images or the full project only after the workflow behaves well on representative cases.
  8. Use Review and Export to turn selected masters into GIS files or handoff packages.

Quality checks

Look for plausible but wrong masks. Roads, shadows, bright bare ground, clouds, and vegetation edges can look convincing until compared with the source imagery. Review failure cases before saving a profile as reusable.

For project-scale runs, use the filmstrip, layers panel, run queue, and Review image overview to compare images with different lighting, seasons, and target sizes. A profile that works on one scene may need separate variants for another sensor or land-cover context.

Export

Use Review to choose master outputs, then use Export for GeoPackage, GeoJSON, Shapefile, mask PNGs, Summary PDF, Summary CSV, or Review ZIP. For geometry-sensitive outputs, edit the vector layer before handoff.

Read more in SAM 3 satellite segmentation, Segmentation settings, and Segment: run segmentation and inspect layers.

Related guides