Get startedData · Statistics and quality
Diagnose dataset health
Use coverage, balance, dimensions, spatial plots, and split statistics to find data risk before training.
Source reviewed 2026-09-16. Availability depends on your installation, permissions, and compatible runtimes. A supported path is not a guarantee of model quality or production readiness.
Outcome
Turn dataset statistics into concrete curation, annotation, and split decisions.
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
- — One immutable dataset version
- — Expected domain and deployment distribution
Expected output and limits
- — Readiness findings
- — A prioritized curation or relabeling queue
Core workflow
Choose a version
Open the immutable version whose readiness and distribution you want to understand.
Review coverage
Inspect asset counts, annotation coverage, classes, splits, and label distribution.
Read the spatial plots
Use annotation area, centroid, dimensions, and asset dimensions to identify bias or malformed labels.
Resolve the risk
Route weak samples to review, rebalance splits, or create a corrected version.
What this surface supports
Class distribution
Annotation area and centroid plots
Asset and annotation dimensions
Split and coverage summaries
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.
Quality and operating signals
- — Class entropy and split balance
- — Annotation area and centroid density
- — Resolution, aspect-ratio, and coverage outliers
Common failure modes
- — Treating aggregate balance as representativeness
- — Ignoring rare critical slices
- — Reading plots without comparing production data
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.
Engineering safeguards
- — Statistics identify their dataset version
- — Warnings remain explainable
- — Corrections create a new version