Computer vision annotation tool built for CV engineers

A computer vision annotation tool should fit into your training pipeline, not sit next to it. ImageLabeling gives CV engineers an API to upload images and fetch labels, webhooks that fire when a batch is reviewed, frozen dataset versions you can diff, and imports for the labels you already own. Your labelers get a fast editor, and you get data that arrives in the format your code expects.

Label my image See pricing

Robotics engineer next to a camera rig on a workbench, laptop showing labeled images and an API response

From raw images to a training run

City crossing with cars, a cyclist, pedestrians on a zebra crossing and a traffic light

A typical loop looks the same in every team: collect images, label, review, freeze, export, train, find failure cases, label again. ImageLabeling covers each step and keeps the record of who did what.

The image annotation workflow page walks through these steps with screenshots.

Upload by drag and drop, as a zip or from a cloud bucket URL
AI pre-labels for your label schema, with a confidence per object
Labeler then reviewer workflow with comments and an issue queue
Frozen dataset versions with train, validation and test split
Export to COCO JSON, YOLO txt with data.yaml, Pascal VOC XML or CSV

API and webhooks

The REST API is part of Team, Business and Enterprise. It covers the actions you would otherwise click through: upload images to a project, fetch labels for an image or a batch, list dataset versions and trigger an export. API tokens belong to your workspace. A short illustrative request that starts a YOLO export of a frozen version:

Webhooks are part of Business and Enterprise. They send an event to your endpoint when something happens in the workspace when a batch of images is reviewed. A common setup listens for the batch reviewed event, freezes a new version and starts a training job without anyone watching a dashboard. Enterprise adds higher API rate limits for large programs.

curl -X POST "https://api.imagelabeling.com/v1/projects/PROJECT_ID/exports" \
  -H "Authorization: Bearer $IMAGELABELING_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"version": "v4", "format": "yolo"}'

Typical automations

Push images from the line camera or the field app into a project every night
Pull reviewed labels into your feature store as soon as a batch is approved
Export a new version from CI and attach its ID to the training run
Send model failure cases back for labeling with the upload endpoint

Dataset versions you can diff

A version freezes the images, labels, schema and split at one point in time. Compare any two versions to see labels added, changed and removed, which tells you exactly what changed between two model runs. Splits stay fixed inside a version, so test images never leak into training through a later export. Starter keeps 3 versions, Team and above keep as many as you need.

Import the labels you already have

Apples and a few oranges rolling on a stainless steel conveyor belt in a packing plant, a gloved hand sorting

Most teams do not start from zero. Import existing labels as COCO JSON, YOLO txt, Pascal VOC XML or CSV, and they appear in the editor as editable objects ready for review. That lets you audit an old dataset, fix systematic errors and extend it with new classes. Format details are on the COCO annotation tool and YOLO labeling tool pages.

Every annotation type your models need

Boxes, polygons, points and image classes are on every plan. Segmentation masks and attributes start on Team, keypoint skeletons and custom schemas on Business. The data annotation tools page explains when to use each one.

Roles and data control for engineering teams

Roles are owner, admin, reviewer, labeler and viewer, with access set per project. Images are encrypted in transit and at rest, used only to label your dataset and never to train third party models, and deleted on request. Business adds Google and Microsoft SSO, an audit log and a QA score per labeler. Enterprise adds SAML SSO, a DPA, a 99.9 percent uptime SLA and invoicing. More on the data security for image labeling page.

Plans for the computer vision annotation tool

Aerial drone view straight down of farm fields with a red tractor, a white pickup truck, a barn and a dirt road

API access starts on Team at 149 USD per month or 74 USD per month billed yearly, with 10 seats and 3,500 AI pre-labeled images per month. Webhooks start on Business at 499 USD per month or 249 USD per month billed yearly. Compare every limit on the pricing page.

FAQ

Which plans include the API?

Team, Business and Enterprise include the API. Webhooks are part of Business and Enterprise, and Enterprise adds higher rate limits.

Can I upload images through the API?

Yes. The API uploads images to a project, fetches labels, lists versions and triggers exports in COCO, YOLO, Pascal VOC or CSV.

Do API uploads count toward the AI image quota?

Uploads alone do not. An image counts once, when the AI model pre-labels it. Manual labeling and re-exports never count, and deleting images never frees the monthly AI quota.

Can I label video frames?

Yes, as images. Extract frames, upload them through the editor or the API and label them like any other image. There is no native video timeline.

Connect labeling to your pipeline

Create your workspace, upload a first batch and pull the labels back through the API.