Skip to main content

Datasets Page

Route: /datasets — Source: ui/src/pages/DatasetsPage.tsx

The Datasets page is the home for everything dataset-related: browse, drill into one to see videos and labeling jobs, kick off new auto-label runs, and import existing data.

Datasets page

What you can do here​

  • Create datasets — name them; they're scoped to your active workspace.
  • Search & sort — the toolbar filters live as you type; sort by newest, name, video count, or label count.
  • Drill into a dataset card — see videos, labeled frames, classes, and the latest labeling job state.
  • Import — pull existing YOLO directories from disk or register MinIO objects without re-uploading.
  • Upload videos — drop files or batch-upload via the URL list.
  • Configure prompts — define text or visual prompts at the dataset level so new uploads auto-label automatically.

Anatomy of a dataset card​

FieldMeaning
STATUSpending, running, completed, failed from the most recent labeling job
VIDEOSCount of Video rows in the dataset
LABELSCount of Annotation rows across all jobs
CLASSESDistinct class_name values currently labeled

Notable interactions​

  • Card expansion calls GET /api/v1/jobs/{job_id}/overview for inline stats so the list itself stays fast.
  • The import panel is hidden by default — toggling it sets a ?showImport=1 query param so URLs are shareable.
  • TanStack Query caches results with a 5-minute staleTime; pull-to-refresh forces a refetch.

Export after review​

Open a labeling job's export menu and select its task format. Classification exports padded object crops; pose exports one centroid keypoint per saved polygon. Pending labels are included unless rejected. Review the shapes and classes before export. The returned download link points to an immutable ZIP. Bulk export uses each job's task format.

Native and text-prompt labeling retain observations for this explicit export step. They do not automatically replace a reviewed dataset on retry. If review changes invalidate the training artifact, export again before starting training.