Comparison · Create ML
Compare models against the same reference data and evaluation protocol.
Before you start
At least two compatible completed Create ML results and an available reference dataset.
Quick start
- Open Comparison, choose Create ML and select the candidate runs/models.
- Choose common reference data and compare. Inspect per-class behavior and regression examples, not only overall metrics.
Functions and choices
Choose candidates
Choose one reference and one to five candidates on the same validation, test or guard images.
Check class labels
Model labels may cover a subset of active project classes; unknown or empty labels are rejected.
Use fixed thresholds
Confidence ≥ 25 percent and match IoU ≥ 50 percent are fixed in this view.
Result
Comparable results and documented model differences.
Next step
For Create ML, register a suitable candidate. For Segmentation, select the compatible checkpoint for inference.