Create reusable segmentation profiles
A segmentation profile stores how RasterForge should find or select a target. Profiles are the main way to move from one successful run to repeatable segmentation across a project.
What a profile contains
- A clear profile name, such as
Wetland water edgesorIndustrial roofs. - A class label and optional class group for review and export.
- Prompt text, saved examples, and default workflow settings.
- Run defaults such as detection threshold, mask threshold, max instances, box inference mode, and postprocess settings.
- Optional image-specific guidance, such as points and boxes, stored separately from the shared profile.
Create a profile quickly
The Profiles page and Segment can create a profile from a simple Object or class name. RasterForge infers practical defaults for the profile name, prompt text, class label, and group where it can. Treat those defaults as a starting point: review the class metadata, prompt, and run defaults before running a project batch.
My profiles and Shared
The profile library has two main views:
- My profiles: workspace profiles you can edit, use in Segment, and attach examples to.
- Shared: published reusable profiles that can be imported into your workspace.
Import a shared profile when it describes a simple, repeatable target close to your project. After import, inspect the prompt, class metadata, examples, and run defaults before using it on your imagery.
Sample cases can also install shared profiles. Use those imported profiles as examples of a prepared workflow, then adapt the prompt, examples, thresholds, and class metadata for your own imagery.
Build from a good image
Start from an image where the target is visible and representative. Choose a render that separates the target, create a narrow prompt, add boxes or examples if text is too broad, run on the active image, then inspect the layer before saving or widening scope.
Find versus select
Use Find things when the target repeats and you may run the profile on selected images or a project. Use Select objects when you need targeted object selection on the active image with points or boxes. Auto masks is separate from segmentation profiles because it produces class-agnostic object candidates.
In the current app, SAM 3 powers these workflows with prompt-and-click modes such as text, boxes, saved visual examples, and point or box selection. The profile should describe the target in RasterForge terms first; the model-specific settings can be tuned after the visible workflow is clear.
Image examples
Image examples are reusable crops that can be attached to profiles. You can save examples from useful boxes in Segment, manage them in the image examples library, and attach them to profiles by action or drag and drop. A profile can use up to five examples, so prefer representative examples over near-duplicates.
Use profiles in bulk
Once a profile works on one representative image, run it on selected images before running the whole project. Review failures and update the profile instead of creating many near-duplicate layers.
Keep metadata clean
Class labels and class groups flow into Review and export. Use consistent names early, especially when a project has many images or layers. A clean profile library makes the review matrix easier to scan and keeps exported GIS files easier to interpret.