Practical computer vision and edge AI, from someone who runs it in production.
Guides on training detectors, running them on small devices and keeping video analytics reliable. Written by a computer vision engineer, with no hype.
Latest articles
Managing Per-Camera Configuration at Scale
Zones, thresholds, schedules and streams for hundreds of cameras: how to layer defaults and overrides, validate them and keep them versioned.
Main Stream or Substream: Choosing Resolution and Bitrate for Analytics
IP cameras offer a full-quality main stream and a lighter substream. Which to analyse depends on object size, model input and decode budget.
Turning Down a Computer Vision Project (and Offering Something Better)
Some vision projects should not be accepted as asked. How to spot them, say no without losing the client, and offer a version that works.
Single-Class or Multi-Class: When to Merge or Split Detection Classes
Should car, van and truck be one class or three? How to decide a detector's classes, and why labelling fine and training coarse keeps options open.
Watchdogs and Self-Recovery for Unattended Edge Boxes
An edge box nobody can reach must recover on its own. Hardware watchdogs, systemd restarts, application health checks and remote alerts, layered.
Deduplicating Video Frames Before Labelling
Most frames from a fixed camera are near-copies. How to pick frames worth labelling with difference checks, hashes and embeddings, keeping rare cases.
Profiling a Python Vision Pipeline: Finding the Real Bottleneck
Before optimising the model, find where the time actually goes. Stage timers, sampling profilers and the usual hidden costs in Python video pipelines.
How Many Frames per Second Does Your Analytics Need?
Analysing every frame is rarely necessary. How to pick an analysis frame rate from object speed, view size, filters and trackers, and test it.
Why a Detector's Confidence Is Not a Probability
A box scored 0.8 is not right 80% of the time. What confidence scores are, why they shift between models and classes, and what that means for thresholds.
YOLO and Detection Basics
Anchor-Free vs Anchor-Based Object Detectors
Anchor-free vs anchor-based detectors: what anchors are, why newer YOLO versions dropped them, and what the change means for training and deployment.
Nano, Small, Medium: How to Choose a YOLO Model Size
Choosing a YOLO model size and input resolution: a method that starts from your hardware budget and accuracy needs, not a guess at nano, small or medium.
Classification vs Detection vs Segmentation: Pick the Right Task
Classification vs detection vs segmentation: how to choose the right task for your project, with a decision guide, cost trade-offs and examples.
All 17 articles in YOLO and Detection Basics
Datasets and Training
Writing an Annotation Guide Labellers Can Follow
Labels are only as consistent as the rules behind them. What to put in an annotation guide, how to show edge cases and how to keep it current.
Annotation Mistakes That Quietly Ruin Your Model
Common detection annotation mistakes, from loose boxes to missing labels and inconsistent classes, and a practical way to audit and fix your dataset.
Class Imbalance and Small Objects: Practical Fixes
Fixes for class imbalance and small objects in detection: more data, tiling, higher resolution, sampling and the right metrics to judge progress.
All 15 articles in Datasets and Training
Edge Hardware and Deployment
Choosing a Camera for Computer Vision Projects
How to choose a camera for computer vision: USB, CSI, IP/RTSP and depth cameras compared, with advice on shutter, lens, lighting and what really matters.
Dockerize a Computer Vision App for Edge and Cloud
Dockerize a computer vision app: Dockerfile patterns for OpenCV and ONNX Runtime, GPU access, smaller images, pinned versions and reproducible builds.
Export YOLO to ONNX and Run It with ONNX Runtime
Export a YOLO model to ONNX with Ultralytics and run it on CPU with ONNX Runtime, including pre-processing, output decoding and NMS differences.
All 17 articles in Edge Hardware and Deployment
Video Analytics
Alerts at the Wrong Time of Day: Time Zones, DST and Camera Clocks
Alerts that fire correctly but at the wrong time usually come from clocks: drifting cameras, servers in UTC, daylight saving. How to handle time properly.
Day/Night Switching: Why Detections Change When the IR Filter Flips
When a camera switches to infrared night mode, the image changes completely. What that does to the model, why cameras flap at dusk, and what to do.
Designing Alert Rules: Schedules, Cooldowns and Escalation
Design video analytics alert rules people trust: time windows, cooldowns, absence rules, grouping, escalation and reliable delivery.
All 17 articles in Video Analytics
Tools and Workflow
Building a Replay Set of Recorded Footage
A small, curated library of recorded clips with known events lets you test every pipeline change against reality. What to include and how to keep it.
Cloud GPUs for Model Training: How to Choose
How to choose a cloud GPU for training vision models: compare cost models, GPU memory, storage and reliability, and avoid the habits that waste money.
Experiment Tracking: TensorBoard, Weights & Biases and MLflow
Why experiment tracking matters for computer vision, what to log on every training run, and how TensorBoard, Weights & Biases and MLflow compare.
All 13 articles in Tools and Workflow
Career and Freelancing
Agreeing a Measurable Target with a Client Before You Start
Turn 'it should work well' into a written acceptance test: what counts as an event, which footage, who labels it and what result counts as a pass.
Best Books and Courses for Learning Computer Vision
How to choose books and courses for learning computer vision by level, with honest pros and cons of each format and a plan that mixes study with projects.
Computer Vision Engineer Roadmap: What to Learn First
A realistic computer vision roadmap from Python and OpenCV to training detectors, edge deployment and video analytics, with projects for each stage.