Templates for Computer Vision

Start with a proven workflow, then make it your own.

24 templates

Dataset intelligence

Run statistical, embedding and prelabeling checks on an imported detection dataset and merge them into one report: findings ranked by severity, and recommended next steps.

  • Build AI index, Your datasetBuild AI indexYour dataset
  • Dataset statsDataset stats
  • Wait for all checksWait for all checks
  • Run flowRun flow
  • Dataset insightsDataset insights
  • +4
  • Data Quality
  • Evaluate
  • Image

Dataset intelligence with zero-shot prelableing check

Run statistical, embedding and prelabeling checks on an imported detection dataset and merge them into one report: findings ranked by severity, and recommended next steps.

  • Dataset statsDataset stats
  • Label qualityLabel quality
  • Prelabeling efficiencyPrelabeling efficiency
  • Build AI index, Your datasetBuild AI indexYour dataset
  • Wait for all checksWait for all checks
  • +7
  • Data Quality
  • Evaluate
  • Image

Zero-shot prelabeling check

Reusable template created from Zero-shot prelabeling check.

  • Prelabeling efficiencyPrelabeling efficiency
  • AI gateAI gate
  • Model accuracyModel accuracy
  • Find missed labels, Auto-label with AIFind missed labelsAuto-label with AI
  • Your datasetYour dataset
  • Data Quality
  • Evaluate
  • Prelabel
  • Image
  • Atomic
  • Grounding Dino
* * * * *

Active learning (object detection)

Reusable template created from Active learning (object detection).

  • Copy to historyCopy to history
  • Sampling results, Rejected, Already sampled, Training Data, Best model predictions, Data Lake, Labeling pool, Build AI index, Annotated data, Training set, Validation setSampling resultsRejectedAlready sampledTraining DataBest model predictionsData LakeLabeling poolBuild AI indexAnnotated dataTraining setValidation set
  • Import from cloud storageImport from cloud storage
  • Skipped, Model registrySkippedModel registry
  • Every 5 hours, Every 15 minutes, Run flowEvery 5 hoursEvery 15 minutesRun flow
  • +28
  • Active Learning
  • Label
  • Train
  • Sample
  • Evaluate
  • Object Detection
*

Live training module

Plug live training into your labeling pipeline: images reach labelers in batches, and once the model is good enough, each batch arrives pre-labeled.

  • Move to labelingMove to labeling
  • Quality gateQuality gate
  • Prelabel batchPrelabel batch
  • Get next batchGet next batch
  • Live trainingLive training
  • Labeling
  • Prelabel
  • Train
  • Image
  • Atomic
  • Grounding Dino
* * *

Active learning with live training

Label a dataset in small batches: a random first batch, then each round the images most similar to the model's validation failures, pre-labeled by a live training session. Evaluate on the test set when you stop.

  • Live trainingLive training
  • Prelabeling with live training, Pre-label and acceptPrelabeling with live trainingPre-label and accept
  • Stop after 10 roundsStop after 10 rounds
  • Learning curveLearning curve
  • Evaluate on test, Evaluate on validationEvaluate on testEvaluate on validation
  • +5
  • Active Learning
  • Label
  • Train
  • Sample
  • Evaluate
  • Object Detection

Labeling with auto quality check

Label a pool of images and let rule-based checks screen every submission before a reviewer sees it; failures return for another pass, and the completed set comes out with a quality report.

  • Labeling, ReviewLabelingReview
  • Auto quality check, Passed QC, Data Quality ControlAuto quality checkPassed QCData Quality Control
  • LabeledLabeled
  • UnlabeledUnlabeled
  • SkippedSkipped
  • +4
  • Labeling
  • Label
  • Review
  • Image
  • Atomic

Detector comparison

Score several ready-made detectors on the same labeled image set with the same eval code and find the best one — per-class mAP/AP50/AP75/F1, no training required.

  • Predict EdgeCrafter, Predict YOLO, Predict DEIMPredict EdgeCrafterPredict YOLOPredict DEIM
  • DatasetDataset
  • Evaluate object detectionEvaluate object detection
  • Compare object detectorsCompare object detectors
  • Wait for allWait for all
  • Object Detection
  • Compare
  • Image
  • Atomic
  • Yolo
  • Edgecrafter
  • Deim

Import images by drag & drop

Drop images from your computer onto the graph and get an OvalBee dataset that any other template can start from. Every drop adds a new version of the dataset.

  • Import drag & dropImport drag & drop
  • DatasetDataset
  • Data Import
  • Import
  • Image
  • Atomic

Run EdgeCrafter Detector on Images

See what a ready-made detector finds on your own images, before you spend anything on labels or training.

  • Predict EdgeCrafterPredict EdgeCrafter
  • DatasetDataset
  • Object Detection
  • Inference
  • Image
  • Atomic
  • Edgecrafter
  • Featured
* * *

Train YOLO detector

Train a YOLO detector on your own labeled images and score it on a test part it never saw: split, train, predict, evaluate in one graph.

  • Source datasetSource dataset
  • Train / val / test splitTrain / val / test split
  • Evaluate on testEvaluate on test
  • Object Detection
  • Train
  • Evaluate
  • Image
  • Atomic
  • Yolo
  • Featured

Live training: quick start

Label images while a model learns from every submit and draws boxes on the images still waiting, so your team moves from drawing to correcting

  • Quality gateQuality gate
  • LabelingLabeling
  • Live trainingLive training
  • Labeled dataLabeled data
  • Input data, Labeling poolInput dataLabeling pool
  • +5
  • Labeling
  • Label
  • Train
  • Image
  • Atomic
  • Grounding Dino
  • Prelabel
  • Featured

Basic labeling pipeline

A labeling loop for your team: images cycle through labeling and review until accepted, so every image in your final dataset has passed review.

  • ReviewReview
  • RejectedRejected
  • Return to labelingReturn to labeling
  • SkippedSkipped
  • AcceptedAccepted
  • +3
  • Labeling
  • Label
  • Review
  • Image
  • Atomic
  • Featured

Detector evaluation

Score one detector against a labeled reference set — mAP/AP50/AP75/F1 with per-class breakdown and diagnostic examples, the acceptance check before shipping a model.

  • Predict EdgeCrafterPredict EdgeCrafter
  • Evaluate object detectionEvaluate object detection
  • DatasetDataset
  • Object Detection
  • Evaluate
  • Image
  • Atomic
  • Edgecrafter

Case study: PCB inspection with DEIM

From the public PCB Component Detection dataset to an exported detector: import, split, train DEIM, evaluate on a held-out test part, export ONNX and TensorRT.

  • Import PCB Component DetectionImport PCB Component Detection
  • Train / val / test splitTrain / val / test split
  • Train DEIMv2-STrain DEIMv2-S
  • Predict DEIM on testPredict DEIM on test
  • Evaluate on testEvaluate on test
  • +1
  • Object Detection
  • Import
  • Train
  • Evaluate
  • Export
  • Image
  • Deim
  • Pcb Component Detection
*

Train EdgeCrafter detector

Fine-tune an EdgeCrafter detector on your own labeled images and score it on a held-out test part: split, train, predict, evaluate.

  • Source datasetSource dataset
  • Train / val / test splitTrain / val / test split
  • Train EdgeCrafterTrain EdgeCrafter
  • Predict EdgeCrafter on testPredict EdgeCrafter on test
  • Evaluate on testEvaluate on test
  • Object Detection
  • Train
  • Evaluate
  • Image
  • Atomic
  • Edgecrafter
*

Train DEIM detector

Train a DEIM detector on your own labeled images and score it on a test part it never saw: split, train, predict, evaluate in one graph.

  • Source datasetSource dataset
  • Train / val / test splitTrain / val / test split
  • Train DEIMTrain DEIM
  • Predict DEIM on testPredict DEIM on test
  • Evaluate on testEvaluate on test
  • Object Detection
  • Train
  • Evaluate
  • Image
  • Atomic
  • Deim

Labeling with expert

Route difficult annotations to a specialist while routine work goes straight to final review; anything the reviewer rejects returns to the labeling pool for another pass.

  • Review poolReview pool
  • RejectedRejected
  • Return to labelingReturn to labeling
  • Labeling, ExpertLabelingExpert
  • Labeling poolLabeling pool
  • +4
  • Labeling
  • Label
  • Review
  • Image
  • Atomic

Filter-based data cleanup

Profile a dataset, then fix what the findings point at with rules — remove the images that should carry labels but don't, rename a misspelled class — and see the cleaned copy re-profiled next to the original. No labeling, no training, no model.

  • Dataset, Filtered DatasetDatasetFiltered Dataset
  • Filter imagesFilter images
  • Transform AnnotationsTransform Annotations
  • Data Quality ControlData Quality Control
  • Data Quality
  • Evaluate
  • Clean
  • Image
  • Atomic
  • Pcb Component Detection

Dataset quality control

Point a dataset at a quality-control step and get a report of what is inside — asset and annotation statistics. No labeling, no training, no model.

  • DatasetDataset
  • Data Quality ControlData Quality Control
  • Data Quality
  • Evaluate
  • Image
  • Atomic
*

Supervisely queue labeling

Keep the team labeling in Supervisely and let the queue run itself: annotators pull the next image from one shared pool instead of waiting on a batch, so nobody sits idle, and a reviewer sends a reject straight back to its author. Stage views show where every image stands right now.

  • ReviewReview
  • Rework, Waiting for labeling, Waiting for review, AcceptedReworkWaiting for labelingWaiting for reviewAccepted
  • LabelingLabeling
  • Labeling queue managerLabeling queue manager
  • Source datasetSource dataset
  • Labeling
  • Label
  • Review
  • Image
  • Atomic
*

Labeling jobs in Supervisely

Hand out the images in a dataset as named batches — one Supervisely labeling job per annotator, one reviewer per job — and watch every image move through its stages without leaving OvalBee. An annotator labels their whole batch before submitting it; rejected images are reissued as a fresh job, and accepted ones collect in a final dataset. Runs on a schedule, so images added later join the same pipeline on their own. No model, no metrics.

  • LabelingLabeling
  • Rejected, Waiting for review, Accepted, Waiting for labelingRejectedWaiting for reviewAcceptedWaiting for labeling
  • ReviewReview
  • Source datasetSource dataset
  • Labeling jobs managerLabeling jobs manager
  • Labeling
  • Label
  • Review
  • Image
  • Atomic

Import from Supervisely

Copy images and their annotations from a Supervisely project, dataset or collection into an OvalBee dataset that any other template can start from.

  • Import imagesImport images
  • DatasetDataset
  • Data Import
  • Import
  • Image
  • Atomic

Import from Dataset Ninja

Bring a public dataset from Dataset Ninja into OvalBee by its link, images and annotations included, ready for any other template to start from.

  • Import from Dataset NinjaImport from Dataset Ninja
  • DatasetDataset
  • Data Import
  • Import
  • Image
  • Atomic