Get startedSegmentation learns object shapes. Training your own model is separate from using the built-in segmentation service.
Source reviewed 2026-09-23. Availability depends on your installation, permissions, and compatible runtimes. A supported path is not a guarantee of model quality or production readiness.
Start here
What you need to know first
Check the inputs below before starting. If you are new, begin with the first-project guide. A dataset holds media and labels; a model produces results; a deployment makes a selected model version callable. Creating one does not create the others.
Bring these inputs
- Confirmed polygon annotations
- Training and validation splits
- Local training runtime
Expected output and limits
- A checkpoint and mask ONNX artifact if successful
- Training report and dataset lineage
Core workflow
- 01
Select reviewed outlines
Choose a segmentation dataset version with confirmed simple polygon instances. Boxes alone, holes, disconnected shapes, and self-crossing polygons are not supported training annotations.
- 02
Check the split
Prepare training split if training or validation is missing. The new version retains test images and review status. Generated data must retain its complete brief, unique images, scenario coverage, and confirmed annotations before training.
- 03
Train on This machine
Use the local supervised path with one, two, or three configurations. The total epoch ceiling is configurations × epochs. Score keeps each validation score and both artifacts, then selects the strongest mask score. Remote segmentation and architecture search are not enabled.
- 04
Test the saved model
Inspect actual masks on separate reviewed images. An exported model and a small local smoke test do not establish customer accuracy.
What this surface supports
- Local supervised segmentation and bounded configuration search
- Reviewed polygon validation
- Checkpoint and ONNX export
- Mask-based evaluation
Expert section
Contracts, signals, and failure modes
Use this section when you are defining acceptance criteria, automating the surface, or reviewing whether its output is safe to promote downstream.
Expert release checklist
- Inputs and dependencies are pinned to immutable versions.
- Acceptance metrics include critical classes and operating slices.
- Failure, retry, cost, and rollback behavior are understood.
- The resulting artifact has an owner and a downstream review path.