Use cases | OvalBee

Each page here describes one job: what the flow looks like, which nodes do the work, how to set it up for your data and how teams usually extend it. Every one of them starts from a template in the library, and most have a live example you can open and a case study with real numbers.

Bring data into OvalBee

Upload files, pull a public dataset from Dataset Ninja, or import a Supervisely project.

Organize labeling

Labeling with review, expert escalation, automated quality checks and rework loops.

Check and clean a dataset

Profile a dataset, find label errors, duplicates and outliers, and fix them with rules.

Try a pretrained detector

See what an off-the-shelf model finds on your images before you label anything.

Train a detector

Split a labeled dataset and fine-tune YOLO, DEIM or EdgeCrafter on it.

Evaluate and compare models

Score a model against ground truth and pick the best of several.

Templates at a glance

Open the library with Start from template on the Home page, or Templates in the flow toolbar.

Template	Use case
Import images by drag & drop	Bring data in
Import from Dataset Ninja	Bring data in
Import from Supervisely	Bring data in
Labeling with review loop	Organize labeling
Labeling with expert	Organize labeling
Labeling with auto quality check	Organize labeling
Supervisely queue labeling	Organize labeling
Labeling jobs in Supervisely	Organize labeling
Dataset quality control	Check and clean a dataset
Filter-based data cleanup	Check and clean a dataset
Run EdgeCrafter Detector on Images	Try a pretrained detector
Train EdgeCrafter Detector on Images	Train a detector
Detector evaluation	Evaluate and compare models
Detector comparison	Evaluate and compare models

Combining them

The use cases are building blocks of one larger process. Imported data feeds a labeling flow, its accepted dataset feeds training, the trained model is evaluated and compared, and its predictions become the starting point for the next round of labeling. Because datasets are versioned and references can follow the latest version, you can connect these pieces in one flow and let each part pick up the newest results of the one before it.

Not sure where to start? Describe your goal to the AI agent — it searches the Node Library and assembles a first version of the flow for you.