Segmentation · Contours and masks
Train an experimental instance segmentation model from reviewed contours.
Before you start
Choose a published contour dataset with Train and Val. Set up the Segmentation environment and prepare masks.
Quick start
- Choose the dataset and verify its annotations. Prepare the contour/mask packet.
- Check Segmentation settings, step budget and output destination. Start the run.
- Inspect the saved checkpoint and run evidence; evaluate using the Segmentation backend.
Functions and choices
Use a checkpoint
This path saves a checkpoint rather than a Core ML detector. Compatible saved results can be selected for Segmentation inference in Capture Lab and Annotator/Failure Bank.
Model name
Give the model a descriptive name, or keep the suggested name. Every run has a stable id.
Prepare masks
Masks use saved hand-drawn contours. Correct missing or invalid contours before preparation.
Result
A Segmentation checkpoint with settings and evaluation evidence.
Next step
Open Evaluation or Comparison and choose Segmentation.