Data for computer vision

Images and footage captured in your conditions, and labelled to the precision your model needs.

What Computer Vision needs

Vision models learn the lighting, angles and clutter in their training set. If production looks different, accuracy drops without warning.

We capture images and video across the conditions you specify, and label them with boxes, masks, tracks and attributes, with per-class quality scores.

Data types

Image
Video
Annotations

Industries

Automotive
Retail and e-commerce
Safety and surveillance
Healthcare
Real estate
AR and VR

The data it takes

Image collection
Photos taken to a brief across devices, lighting and locations.
Video collection
Clips of actions, gestures and scenes at set frame rates.
Bounding boxes and masks
Object and region labels in COCO, VOC or YOLO.
Tracking and events
Persistent object IDs and event boundaries across frames.
Attributes
Pose, expression, occlusion and condition tags on each item.

How we collect it

01
Capture to a brief
Contributors photograph or film to your shot list, in the environments you name.
02
Metadata kept
Device, lighting, angle and location type recorded for each item.
03
Labelled to a taxonomy
Annotators follow written guidelines with worked edge cases.
04
Reviewed per class
IoU and agreement scored per class, with gold tasks in the queue.