Core concepts

Core concepts

RasterForge is organized around a small set of project objects. These terms appear throughout the app and docs.

Project

A project is the top-level workspace for a segmentation job. It groups source images, AOIs, render choices, segmentation profiles, layers, review state, edits, and exports.

Sample case

A sample case is an installable prepared project. It can include imagery, render state, segmentation profiles, layers, shared profile imports, and review selections so new users can inspect a real workflow before building their own project.

Image and source

An image is a GeoTIFF, PNG, or JPEG inside a project. GeoTIFFs can carry CRS and transform metadata. PNG and JPEG inputs are useful for image-like sources, but they need manual bounds if map placement or geographic export matters.

Bounds

Bounds describe where an image sits on the earth. RasterForge stores map bounds as WGS84 west, south, east, and north coordinates. Correct bounds make map AOI selection, layer display, and geospatial export more useful.

AOI

An area of interest defines the part of the image RasterForge renders and segments. A focused AOI reduces unrelated pixels, lowers patch use, and keeps review centered on the target region.

Render

A render is the visual image made from the source raster. The segmentation workflow sees this rendered image, not every raw raster band. A render can be a preview for tuning or a full segmentation render.

Render profile

A render profile stores the recipe for turning source bands into a visual image. It can define RGB, grayscale, or NDVI mode, band mapping, stretch, gamma, nodata handling, value scaling, and output size.

Segmentation profile

A segmentation profile stores a reusable target definition: name, class label, optional group, prompt text, saved examples, workflow defaults, thresholds, object limits, and postprocess settings. Shared profiles are published reusable profiles that can be imported from the profile library.

Guidance and examples

Guidance is image-specific input such as positive points, negative points, or boxes. Examples are saved image crops attached to a profile or library. They help describe a target visually when text alone is too broad.

Segment workflows

Find things is for repeated targets across one or many images. Select objects is for point or box guided active-image segmentation. Auto masks generates class-agnostic object candidates when you want broad mask proposals before deciding what to keep.

Layer

A layer is the result of a segmentation run or composite operation. A layer can have status, visibility, opacity, class metadata, Pixel mask artifacts, GeoJSON artifacts, errors, review stars, and object-level edit state.

Pixel mask and GeoJSON

A Pixel mask is raster output aligned to the segmented image. GeoJSON is vector output made from segmented objects. Pixel masks are useful for raster workflows and visual overlay; GeoJSON is useful for GIS review, object editing, measurement, and vector export.

Edit draft

The edit page stages GeoJSON object changes in a draft. You can select, reshape, simplify, smooth, fill holes, remove objects, undo, redo, save, or discard without changing the original layer until you commit the draft.

Review master

A review master is the chosen output for an image and class. RasterForge can select candidates automatically, but a user-starred or user-selected layer is the clearest way to mark what should export.

GIS handoff export

The Export page builds downloads from reviewed master layers. GeoPackage is recommended for project handoff, while GeoJSON and Shapefile support interoperability. Export can also include mask PNGs, a review ZIP, Summary PDF, and Summary CSV for selected image/class scope.

Patch

A segmentation patch is a 256 by 256 pixel accounting unit used for SAM 3 runs. Larger renders and broader AOIs use more patches. Segment estimates patch use before runs and the account page shows daily availability.