Segment: run segmentation and inspect layers

Segment is where rendered imagery becomes segmentation layers. It combines the map, segmentation controls, layer management, project filmstrip, and run queue in one workspace.

Segment layout

  • The center map displays the active image render and visible layers.
  • The left panel manages segmentation workflows, profiles, prompts, examples, guidance, run settings, and run targets.
  • The right panel manages layers, visibility, opacity, display mode, GeoJSON updates, object state, review stars, composites, and deletes.
  • The bottom filmstrip moves between project images and controls selected images for multi-image runs.
  • The bottom of the right side includes the run queue for active and recent layer runs.
  • Left and right panels can be collapsed or resized so inspection can use more map space.

Segmentation workflows

RasterForge exposes three user workflows:

WorkflowUse whenCurrent SAM 3 flow
Find things (multi-image)You want repeated objects or areas across one or many images.Text prompts, boxes, and optional saved examples.
Select objects (single image)You want targeted segmentation on the active image using points or boxes.Point and box guidance for one image.
Auto masks (object candidates)You want class-agnostic mask proposals before deciding which objects or layers are useful.Automatic mask generation with Balanced, Dense, or Clean presets.

Start with the active image. Once a target works, use the filmstrip to select images and run the profile on selected images before widening scope. Select objects runs only on the active image; Find things and Auto masks can run on selected images or the full project when the run target is available.

Guidance, prompts, and examples

Find things supports prompt text, positive and negative boxes, saved examples, and profile defaults. Select objects supports active-image points and boxes. Auto masks does not use a segmentation profile; it creates class-agnostic candidates with mask generation settings.

Current limits are documented in segmentation settings. The app blocks runs when required inputs are missing, prompt text is too long, image guidance exceeds the limit, example count exceeds the limit, or the daily patch budget is exhausted.

Layers panel

The layers panel is the operational history for segmentation output. Use it to:

  • Filter by all, visible, starred, running, failed, or complete layers.
  • Search layer names, profiles, and prompts.
  • Toggle visibility, opacity, and display mode.
  • Use display mode buttons to switch a layer between GeoJSON and Pixel mask display when both are available.
  • Create a stored Pixel mask from a cache-live model result when the layer supports it.
  • Recalculate GeoJSON from a stored mask bundle with postprocess settings, or export the current layer GeoJSON for the active image.
  • Mark or clear a review master star.
  • Rename layers and edit class or group metadata.
  • Hide, show, remove, focus, or edit individual objects when GeoJSON is available.
  • Combine layers with union, subtract, intersect, or xor composites.

GeoJSON and Pixel mask are different stored results. GeoJSON shows recalculated polygons that can be edited and exported as vectors. Pixel mask shows the underlying raster mask for visual comparison and raster-oriented handoff. A layer can have one, both, or neither depending on run status, saved artifacts, and whether you have generated the missing output.

Segment composites support union, subtract, intersect, and xor operations across selected input layers. Review also has an Add layer action, but that creates a review selection composite for one image/class cell using a union-style mask.

Run queue and status

The run queue sits at the bottom of the right side and groups work by layer run. It shows queued, running, ready, failed, and canceled work, with target information and patch estimates where available. Use it to see whether a batch is still active, which images are waiting, and which items failed.

If a run fails, inspect the visible error, render state, profile inputs, Auto masks settings, and account limits before retrying.

  1. Inspect the render
    Confirm that the target is visible and boundaries are clear enough for segmentation.
  2. Choose a workflow
    Use Find things for repeated targets, Select objects for active-image guidance, or Auto masks for broad object candidates.
  3. Choose or create a profile
    Use an existing segmentation profile when the target repeats. Auto masks does not require a profile.
  4. Add guidance
    Use text, boxes, points, or examples depending on the workflow and target.
  5. Run one image
    Check false positives, missed areas, geometry quality, class metadata, and whether GeoJSON or Pixel mask is the right view.
  6. Widen scope
    Run selected images, then use Review to choose master outputs for export.