Object detection and tracking
Detecting, counting and following people, vehicles, products or defects across frames, with YOLO-family and transformer-based models.
Roles · AI & Automation
Remote computer vision engineers for detection, tracking and video analytics
Models that work on your cameras, your lighting and your hardware, measured on your own footage. Your engineer takes the work from labeled data to a model running in production.
From a promising notebook to a model on real cameras
Pretrained models look impressive on benchmark images and then miss half the objects in your warehouse at night. Getting vision to work in the field means collecting the right footage, labeling it consistently, training and evaluating on your conditions, and fitting the model onto the hardware you actually have. A computer vision engineer owns that whole loop and keeps measuring after launch.
What they do for you
- Define what the model must detect, count or track, and how accuracy will be measured
- Set up data collection and labeling, with clear rules for annotators
- Train and fine-tune detection, segmentation and classification models
- Build tracking and counting logic on top of detections across video frames
- Optimize models for edge devices or GPUs with ONNX, TensorRT or OpenVINO
- Build video pipelines that read camera streams and send events to your systems
- Monitor accuracy in production and retrain when conditions change
How strong computer vision engineers think about accuracy
Tools they work in
PythonPyTorchOpenCVUltralytics YOLODetectron2Hugging FaceONNXTensorRTOpenVINONVIDIA JetsonDeepStreamGStreamerFFmpegCVATLabel StudioRoboflowAWSDockerMany more
A good fit if:
- You have camera or image data and a clear question you want answered from it
- A model works in testing but falls short on real footage
- You need vision running on edge hardware with limited power or bandwidth
- Labeling is slow, inconsistent or done in spreadsheets
Not the right fit if:
- Your AI work is mainly text, chat or document processing; see AI Engineer
- You need someone to connect tools and automate business workflows; see AI Workflow Specialist
- The main job is the app or dashboard around the model; see Full-Stack Developer
Interview questions
01
"Our detector misses people at night. Walk me through how you'd fix it."
Strong answer: check the data for night coverage, look at failure cases by condition, add and label targeted footage, consider camera settings, and measure on a held-out night set.
02
"How do you decide between precision and recall for this use case?"
Strong answer: they ask what a false alarm costs against a missed event, then pick thresholds per class with you.
03
"The model must run at 15 frames per second on a Jetson. What are your options?"
Strong answer: a smaller architecture, lower input resolution, quantization, TensorRT, frame skipping and batching, each measured for accuracy loss.
04
"How do you keep labeling consistent across several annotators?"
Strong answer: a written guide with edge-case examples, a review step, agreement checks on overlap samples, and fixing the guide when disagreements show up.
Signs of a strong computer vision engineer:
- Builds a test set from your real conditions before training anything
- Looks at failure images, not only summary metrics
- Knows the hardware limits before choosing a model
- Versions data and models so results can be reproduced
Signs of a weak computer vision engineer:
- Reports accuracy on the training data
- Picks the biggest model and worries about speed later
- Treats labeling as someone else's problem
- Can't say how the model will be monitored once it ships
Image and video models
Classification, segmentation, pose estimation, OCR and video analytics, including fine-tuning open models on your own footage.
Edge deployment
Models compiled and tuned for NVIDIA Jetson, Intel hardware, mobile devices or on-site servers, with streaming pipelines built on DeepStream or GStreamer.
Data labeling pipelines
Annotation tooling in CVAT, Label Studio or Roboflow, labeling guides, quality review and active learning to label the images that matter most.
What happens once you brief us on a vision role
- 1
Talk it through
We agree the work, the tools, the hours you share and who they report to. For this role: what the model has to recognize, your cameras or image sources, the hardware it will run on, and the data you already have.
- 2
Meet a shortlist and choose
We screen a wide pool for every role and introduce only the few who fit your brief. We source, screen and test candidates against your brief, and run background and reference checks. You interview and decide.
- 3
Start with support in place
They start on the work in the brief, and we stay in contact once they've begun.
Computer vision is a specialist field. Plan for this search to run longer than most, especially when you need edge deployment experience as well.
How pricing works
A computer vision engineer can be hired under any of the options shown. It's your call which one, and we can go through the trade-offs with you. Compare the arrangements on ways to hire.
Who does what
Swipe to see more →
| Task | You | JiveTalent |
|---|---|---|
| What the model must detect, and the accuracy you'll accept | Own | – |
| Data pipelines, training, deployment and monitoring | Review | Built by your engineer; we follow up on progress against the goals you set |
Questions about computer vision engineers
Do we need lots of data before we start?
Not always. Many projects start with a pretrained model and a few hundred labeled images from your cameras, then grow the dataset where the model struggles.
Will they work during our business hours?
The brief sets the hours and the overlap with your team before the search starts, so every candidate is hired for them.
Who owns the models and the training data?
You do. Datasets, labels, model weights and pipeline code stay in your accounts and repositories.
Can they work with our existing cameras?
Usually. They work with RTSP streams, video files and most IP cameras. Share the camera models and resolution in the brief.
What if it isn't working out?
Talk to your account lead. We go through it with you and agree a way forward.
What about privacy when footage shows people?
They can build in blurring, on-device processing and short retention. Which rules apply to your footage is a decision for you and your legal adviser.
How do we see what our computer vision engineer is working on?
Their hours are tracked in the client app and reported there for you to see.
Put your camera data to work
Book a call. We'll agree the brief for your computer vision engineer and talk you through the ways to hire.