COMPUTER VISION · WHERE NNCODE STARTED

Computer vision that works in the real world.

Computer vision is where NNCODE started, and it's still what we nerd out about most. We build the data, models, video pipelines, and deployment for the place it actually has to run, and the dashboards, APIs, and integrations around it too. One team, the whole system.

WHAT IT LOOKS LIKE LIVE

From pixels to "hey, something happened".

Finding and tracking objects is just the start. Zones, events, and alerts turn every frame into something a person or another system can actually act on.

Fig 1Inspection line
Watch, or press a button
line stopped
01

Detection & segmentation

Custom models that find and outline exactly the things you care about, trained on your data, with a clear look at where they fail.

02

Tracking & video analytics

Follow objects across frames, trigger events when something happens, and show it all on a live dashboard.

03

Datasets that don't lie to you

What to collect, how to label it, and automatic checks so bad data doesn't sneak in and quietly ruin your model.

04

Fast on GPUs and small devices

DeepStream, TensorRT, and OpenCV, tuned so it runs in real time, even on a small edge box.

LESSONS LEARNED THE HARD WAY

Accuracy isn't everything.

A real vision system has to deal with cameras dropping out, the light changing, motion blur, new environments, data drifting over time, tight latency, and the real cost of getting it wrong.

So we start with the decision it needs to help with. Then we figure out the data, pick or train the right model, test the whole pipeline end to end, and add monitoring for the things most likely to go wrong.

What you get is a working application you can measure and trust, not a model file and a "good luck".

HOW IT'S BUILT

Capture, detect, track, reason, deliver.

Every system we build follows the same line: cameras and calibration, optimized models on GPU or edge, identities across frames, zones and rules, then alerts, dashboards, and APIs. Arm the zone rule and watch a person in zone 1 turn into an alert.

Fig 2Vision pipeline
Watch, or press a button
pipeline idle · 5 stations

THINGS WE GET ASKED TO BUILD

Where computer vision pays off.

Industrial inspectionDefect detection, quality control, traceability, and human review.
Video intelligenceDetection, tracking, zones, events, counting, and live analytics.
3D and spatial visionDepth estimation, calibration, point clouds, visual odometry, and measurement.
Dataset infrastructurePrivate annotation, quality validation, versioning, and export workflows.

GUIDES

How vision systems actually work.

Computer visionHow multi-object tracking keeps the same ID on the same person7 min readComputer visionBounding boxes vs segmentation masks: what masks actually add6 min readComputer visionHow OCR reads a shipping label, and how to catch its mistakes7 min readComputer visionFrom pixels to floor positions: making camera zones work in metres6 min read3D visionDepth from two cameras: how stereo vision measures distance6 min readComputer visionNon-maximum suppression: why your detector draws one box, not six6 min readComputer visionChoosing the confidence threshold: misses versus false alarms7 min readComputer visionLighting decides what your inspection camera can see7 min readPerformanceQuantization: making a vision model smaller and faster without breaking it7 min read

Let's talk

Got something you want to build?

Tell us what you're working on. You talk directly with the engineers who build it, from the first call to everything after.

  • Reply within a day
  • Straight to the engineers
  • No obligation