Computer Vision Projects

A vision task tied to a real decision, cameras and lighting surveyed on site, labelling rules and audits, leak-proof splits, honest field testing, and the routines that keep an image or video model working where it is installed.

Free · Mac (Apple silicon) and Windows · v1.0.5 · Other platforms

Then open the Marketplace in the app, search for Computer Vision Projects, and click Install. It goes into your own files, and works offline from then on.

What it installs.

1

Pillar

1

Area

50

Projects

213

Starter tasks

📊 Data & Analytics

Computer Vision Projects

Building image and video models for tasks like detection, classification and inspection, including labelling data and testing in real conditions.

  1. Vision task definition and acceptance bar
  2. Camera, lens and lighting survey on site
  3. Image source and usage rights register
  4. Pilot capture of five hundred real images
  5. Labelling guideline with an edge-case gallery
  6. Choosing an annotation tool
  7. Leak-proof splits by site, camera and scene
  8. Pretrained model baseline before custom training
  9. GPU budget and experiment tracking set-up
  10. Weekly labelling quality audit
  11. Active learning queue for the next labelling batch
  12. Monthly failure gallery review
  13. Versioned dataset releases with release notes
  14. Production frame sampling and human review
  15. Camera health and image quality checks
  16. Evaluation report and model card per release
  17. Labelling supplier management
  18. Face and number plate blurring for stored footage
  19. Detection metrics beyond a single mAP score
  20. Augmentation matched to real-world conditions
  21. Fine-tuning recipes for small labelled sets
  22. Segmentation masks and boundary quality
  23. Tracking and temporal smoothing for video
  24. Per-class confidence thresholds and calibration
  25. Reproducing one published vision result
  26. Long-tail and small-object classes
  27. Architecture shortlist for accuracy against latency
  28. Build, buy or vision API decision
  29. Synthetic data trial with a measured gain
  30. Edge or cloud inference decision
  31. Quantisation and compilation for the target device
  32. Label error hunt in the training set
  33. Low-confidence review queue for operators
  34. Adding a new class without losing the old ones
  35. Shadow-mode pilot at one site
  36. Go-live acceptance test with the site owner
  37. Stakeholder demo built around real failures
  38. Benchmark or challenge submission
  39. Paper submission with dataset and code release
  40. Rollout to a second site or camera model
  41. Solo engineer vision project on a small budget
  42. Factory visual inspection with the quality team
  43. Research image collection with ethics approval
  44. Drone and aerial imagery with tiled inference
  45. Safety-critical imaging under expert oversight
  46. Handover of a vision system to a client team
  47. Open-set handling for unknown objects
  48. Stress testing against occlusion, glare and patches
  49. Self-supervised pretraining on your own images
  50. Multi-camera tracking across overlapping views

Where it starts.

Each project arrives with its purpose, milestones and first tasks written out. Here is how the first three read.

Vision task definition and acceptance bar

Most vision projects that stall were never clear on what the model must output or what happens next with that output. Classification, detection, segmentation and counting need different labels, different metrics and different budgets, and the cost of a missed defect is rarely the cost of a false alarm.

First tasks

  • Ask the person who acts on the output what they do with each prediction
  • Write the cost of a miss and the cost of a false alarm side by side
  • Agree one numeric acceptance bar for the most important class
  • Record the latency and hardware limits of the target location
  • Revisit the acceptance bar with the action owner @recurring(yearly)

Camera, lens and lighting survey on site

Images from a phone at a desk look nothing like frames from a fixed camera at the far end of a loading bay at dusk. Resolution on the object, motion blur, glare, night infrared and the angle of view decide what any model can learn, and fixing them with a lens or a lamp is often cheaper than months of modelling. A half-day survey at the real location tells you whether the task is even visible to the camera.

First tasks

  • Book a half day at the site with someone who knows the cameras
  • Photograph each camera position and note its model and settings
  • Measure the target object's size in pixels at near and far range
  • Capture sample frames in morning, midday and night light
  • List the cheapest hardware fixes for blur, glare and low light

Image source and usage rights register

Teams often train on scraped images, a client's footage or an open dataset with a non-commercial licence, and only find out at launch that they cannot ship the model. Images of people add consent and privacy duties on top. A register of where every image set came from, under what terms and with whose permission protects the work you are about to invest in.

First tasks

  • List every folder and bucket of images the team has collected
  • Write the origin, owner and licence beside each source
  • Flag sources showing people or number plates for privacy review
  • Send the doubtful sources to your legal or privacy adviser
  • Recheck the register against new sources and licence changes @recurring(yearly)

Make it yours.

Install Computer Vision Projects, then tell Aurum the date you're aiming for. It checks whether the plan fits the weeks you actually have.

Free · Mac (Apple silicon) and Windows · v1.0.5 · Other platforms

Blueprints are general information to help you organise your own work. They are not medical, financial, legal or other professional advice. See the Terms.