Start here
Custom pack · 2,500 images
from $1,500one-time
2,500 annotated images on your own cameras and classes, hard negatives included.
- Your scenes and classes
- Hard negatives included
- YOLO + COCO, train / val split
- One revision round
Custom data
Your cameras. Your classes. Events you can't film.
A model trained on stock footage fails on your mounting height, your lens and your lighting. We render the incidents you need into your own empty scenes and deliver them labelled — as images, video clips, or both.
01How it works
From five photos to a labelled training set.
- 01
Send your scenes
5–20 photos of your empty camera views. Same mounting height, lens and lighting you run in production.
- 02
Agree the spec
Classes, look-alikes to include, lighting and weather mix, image count and label format. Fixed price, in writing.
- 03
We render and label
Your scenes are populated with the events you need. Every box is checked, and a sample is reviewed by a person.
- 04
Review and revise
You get a preview batch first, then the full set with a dataset card. One revision round is included.
Generated sequenceThreat
CCTV video datasets, as well as images
Video clips of the event in your own fixed camera views: fights, falls, weapon draws, shoplifting, smoke or fire starting, loitering.
Video labels: Per-frame boxes, tracking IDs and event start/end times. Useful for temporal models and for checking how fast an alarm fires.
Quoted per project — tell us what you need and we'll quote. More on what goes into a synthetic CCTV video dataset.
02Weather & lighting
The same scene at night, in rain and in fog.
Most real footage is daytime and dry. Detectors trained on it fail when the light drops or the weather turns — exactly when cameras are hardest to watch.
Every custom dataset can include weather and lighting variations of your scenes: night and low light, rain, fog and storms. The event and its labels stay the same; only the conditions change, so you can measure how your model holds up.
Variations can also be applied to footage you already have. Read more in Synthetic weather variations.
03Packs
Fixed scope, fixed price.
One-time prices that include hard negatives, a dataset card and one revision round. The more images, the lower the price per image. Video is quoted per project.
Quoted per project
Video, or 5,000+ images
Quote
Video clips from your fixed camera views, or larger image sets at a lower price per image.
- Per-frame boxes and tracking IDs
- Event start / end times
- Several scenarios in one order
- Staged deliveries
Ongoing
Enterprise / Data Partner
from $5,000/ month
New classes built for you, validation reports, re-tests and a dedicated account manager.
- New scenarios every month
- Retraining data on demand
- Re-tests after every model release
- Priority delivery
04Questions
Common questions
What do you need from us to start?
5–20 still images of your empty scenes, the classes you detect, and an example of what your model gets wrong today. No real footage of incidents is needed.
Who owns the delivered data?
You do. Custom packs are made for you and are not resold or added to the public catalogue.
Do you make video as well as images?
Yes. Video clips of the event in your own fixed camera views: fights, falls, weapon draws, shoplifting, smoke or fire starting, loitering. Per-frame boxes, tracking IDs and event start/end times. Video is quoted per project — tell us what you need and we'll quote.
Which label formats do you deliver?
YOLO and COCO by default, with a fixed train / val split. Segmentation masks, keypoints and VLM captions are available on request.
What if the first batch isn't right?
Every pack includes a preview batch and one revision round. We adjust the spec and re-render before the full delivery.
Start
Tell us about your cameras.
A few quick questions. We reply within one working day with a quote — images start at $1,500 for 2,500; video is quoted per project.