A fast labeling canvas
Draw boxes, trace polygons or paint masks with a brush. Zoom, layers, undo, copy-from-previous and keyboard shortcuts keep you moving through thousands of frames.
Vision data platform
From raw footage to a trained detector
Label images and video with your team in real time, let a model pre-label the rest, and export a YOLO dataset with a Colab notebook that trains it — then watch the result run over your own footage.
- Box · Polygon · Brush
- YOLO detect & segment
- ONNX pre-labeling
- Live multi-user
detect · live preview
LIVE
objects 0
mean conf 0.00
frame 0000
3annotation tools
5export formats
1‑clickColab training notebook
WSlive presence & cursors
Everything in one loop
Built for the whole labeling lifecycle
Live collaboration
See teammates' cursors and which frame they are on. Roles keep viewers, editors and owners apart.
Model-assisted pre-labeling
Hand an ONNX detector the unlabeled images. Its output arrives as suggestions you confirm or discard — never as silent truth.
Dataset health
Find what would hurt training before you spend GPU hours: empty classes, thin splits, degenerate boxes.
Charts and training runs
Understand what is in the dataset — sizes, box shapes, class balance, where objects sit — then record runs from results.csv, args.yaml and the exported model to compare them.
Predict on video
Run a model over a project video and scrub through its detections frame by frame.
Export anywhere
YOLO, COCO JSON or Pascal VOC, for detection or instance segmentation, with train, validation and test splits. YOLO archives come with a ready Colab notebook: train on a free T4 with no setup, or copy the command and train on your own machine.
Access control and audit trail
Per-project roles for viewers, editors and owners, API tokens for scripts, and an activity log of who changed what.
Workflow
Five steps, one continuous loop
- 01
Import
Upload images and video, pull frames out of clips, or bring an existing YOLO dataset.
- 02
Label
Annotate with your team, with a model proposing the first pass.
- 03
Export
Pick a format, set the validation and test splits, build the ZIP.
- 04
Train
Open the notebook in Colab or run the command on your own machine.
- 05
Detect
Bring the model back, review its runs, and test it on video — then pre-label the next batch.
API first
Everything the UI does, you can script
The interface is a client of a plain JSON API. Authenticate with a token and drive projects, labels, exports and predictions from a notebook or a pipeline.
bash
# sign in, then ask what a dataset looks like
$ curl -X POST /api/auth/login/ -d username=… -d password=…
$ curl /api/projects/street/health/
$ curl /api/projects/street/charts/
# build a YOLO export and poll the job
$ curl -X POST /api/projects/street/export/
→ 202 {"job": {"status": "queued", "percent": 0, …}}
# see what a model finds in a video
$ curl -X POST /api/projects/street/predict/ \
-F model=@best.onnx -F video=clip.mp4
→ 202 {"job": {"status": "queued", …}}
Start labeling in minutes
Sign in to open your projects, or read the docs to wire it into your own tooling.