Guided capture
Participants open a private link, read clear on-screen steps and record with their own webcam. No app, no login.
Computer-vision data collection, simplified
Real-world video in.
Training-ready datasets out.
Annotiv guides people through short webcam recordings, labels every frame on their own device, and hands you a clean dataset plus a notebook that trains your model. No installs. No labelling backlog.
Free plan · No card needed · Participants never need an account
Export-ready forYOLO11YOLOv8YOLOv5Faster R-CNNRetinaNetSSDEfficientDetDETRMask R-CNN
What we do
Everything between "we need data" and "the model is training"
On-device auto-labelling
A YOLO11 model runs right in the participant's browser and boxes everyday objects in every frame before upload.
Fast review studio
Move, resize, relabel or draw boxes with keyboard shortcuts. Mark frames reviewed and track progress per participant.
Exports for 9 models
YOLO11, YOLOv8, YOLOv5, Faster R-CNN, RetinaNet, SSD, EfficientDet, DETR and Mask R-CNN, in the layout each expects.
Ready-to-run notebooks
Every zip ships with a Colab/Jupyter notebook pre-filled with your epochs, batch size and learning rate.
Fair train/test splits
With enough people, whole participants go to the test set so your model is judged on faces and hands it has never seen.
How it works
Four steps, one afternoon
- Design the studyWrite short steps, name the objects, and draw guide boxes (optionally from an uploaded or webcam demo, or on a blank canvas). Ask participants to use a plain white or light background.
- Invite participantsEach person gets a private, no-login link. Your guide boxes show them where to hold each object.
- Review in minutesFrames arrive seeded from your boxes. Propagate labels across similar frames, then pass a final accuracy check.
- Export and trainAfter the quality check passes, export a YOLO-ready zip and start training (cloud training stays stubbed until approved).
Watch a walkthrough
From sign-in to a downloaded dataset
A 7½-minute tour of Annotiv: create a project, invite a participant, record with the in-browser labeller, review frames, export YOLO or COCO, and open the training notebook. No account needed to watch.
Built for
Teams that need their own data
- Research labs running studies with real participants
- Robotics and manufacturing teams teaching machines new objects
- Retail and product teams detecting their own items
- Educators and students learning computer vision end to end
Privacy and security by design
- Consent is recorded before any camera use
- Video only, never audio
- Auto-labelling runs on the participant's device
- Every project is private to its owner, enforced on the server
- Encrypted in transit and at rest on Google Cloud
Our mission
Make good training data easy to get, fairly and safely
Most computer-vision projects stall on data. We want any team to collect consented, well-labelled video of the things they care about, and turn it into a working model the same week.
About AnnotivWhat we plan to do
- Team workspaces shared projects and reviewer roles
- More label types outlines, keypoints and tracking
- Smarter open-vocab labels optional YOLO-World style vocabulary
- Hosted training train in one click, no notebook needed
Your first dataset is a link away
Create a project, invite a participant and export a dataset today.