Images
- Smart segmentation, label by example and text prompts
- Live Training for boxes and masks
- Pre-labels from your own models
AI-assisted labeling
that learns as you label.
Smart segmentation, labels from a text prompt or one example, pre-labels from your own models and tracking, in every labeling tool. Your team reviews labels instead of drawing them.
The right assistant for every object.
Different objects call for different help. The labeling tools put several AI assistants one click away, and every result stays an ordinary label your team can edit.
Smart segmentation
Box an object and get a precise mask, then refine the edge with a few clicks. It works on objects the model has never seen.
Label by example
Box one object and the assistant finds and segments every similar one in the image or video frame, from products on a shelf to cells under a microscope.
Label from a text prompt
Name a class and foundation models such as Florence-2 and Grounding DINO label every match, with no training at all.
A model that learns while you label.
Live Training fine-tunes an object detection or segmentation model on your project as your team labels it. Each new image arrives pre-labeled, each correction makes the next prediction better, and the project ends with a trained model as well as a dataset.
Label the first images
Label as usual. After two to five images, or five to ten for masks, the model starts to predict, even for classes no public model knows.
Correct instead of draw
Every next image opens with the model’s labels. Fix what it got wrong, and the fix becomes training data for the next update.
Draw less, keep the model
A quality score tracks how closely predictions match your final labels. Checkpoints land on the Experiments page, ready to deploy.
Pre-label whole datasets with your own models.
Run a model over a full project of images or videos before anyone opens it. Annotators start from the model’s labels and only fix what it missed.
Models trained on your data
Train a detector on the labels you already have, or fine-tune Segment Anything on your objects, and it pre-labels the next project for your team.
Videos come back as tracks
On videos, a detector labels every frame and a tracker links its boxes into tracks across the clip.
Models in a chain
A detector can pass its boxes to Segment Anything for masks, to a pose model for keypoints or to a classifier for finer classes.
AI assistance in every labeling toolbox.
Masks on images, tracks in videos, cuboids in LiDAR point clouds and segmentations of medical volumes: each toolbox brings the assistants that fit its data.
Videos
- Auto Tracking follows boxes, masks and keypoints, with a tracker suited to each shape
- Tracks extend as you move forward and end when the object leaves the frame
- Tracking by detection starts tracks for new objects as they appear
3D point clouds
- Circle an object and the 3D AI assistant fits a cuboid around it
- Ground segmentation in one click
- Cuboids tracked across frames, and 3D objects from labels drawn on camera images
DICOM
- Smart segmentation of a structure on any slice, in any plane
- Label key slices, and interpolation fills in the slices between them
- Review the interpolated object in 3D
Frequently asked questions.
What is AI-assisted labeling in Supervisely?
It is the set of AI assistants built into the Supervisely labeling tools: smart segmentation from a box, labeling by example or from a text prompt, pre-labels from your own models, tracking in videos and point clouds, and Live Training, which learns from your labels as you work. People review and correct the results instead of drawing every label.
What is Live Training?
Live Training trains an object detection or segmentation model while your team labels a project. After a few labeled images it pre-labels each new one, learns from every correction, and saves its checkpoints on the Experiments page, so the model can be deployed when the project is done. See the Live Training quickstart.
Can I auto-annotate a whole dataset with my own model?
Yes. Apply a model you trained in Supervisely, a pretrained model or a foundation model to a project or dataset of images or videos, check a preview, and save the predictions to a new project or to the source one. Read more about training and deploying models.
Do the models run on our own hardware?
They can. Models run on GPU machines you connect to Supervisely as agents, in your cloud or on premises. On a self-hosted instance, administrators choose which models the labeling tools offer.
Can I try AI-assisted labeling for free?
Yes. Community teams get smart segmentation that is already running, with 500 requests per day, and can connect a GPU machine of their own to run more models. Compare plans.