Practical Aspects of Applying Modern Computer Vision Technologies
summary
Built a factory Quality Inspector from cameras to deployment that runs 200+ use cases across nine plants, claims savings of €10k–€1M per year, and meets client ceilings of 99.5% accuracy and 20 seconds per car. Uses RetinaNet with ResNet101 backbone and a custom deep metric learning classifier; mobile projects use TensorFlow Lite. Camera‑trap work in Ghana reaches ~89% accuracy after three months single‑engineer effort. Pill and damage‑assessment prototypes support self‑retraining and pre/post comparison with EfficientNet backbones.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 1:42Speaker background
- 6:42Car manufacturing quality inspector
- 15:23Wildlife camera traps Ghana
- 25:47Medication error pill detection
- 32:45Disaster damage assessment
- 39:27Project conclusions and future technologies
related talks
24mImage Recognition for safety on the factory floor
36mAI-Powered Real-Time Quality Control
14mComputer vision on edge devices
27mComputer vision for automated quality inspection
23mComputer Vision Toolbox - Connected Computer Vision & AI Quality Inspection at Scale!
40mMachines are Learning
01h 29mComputer Vision: An (Un?)Expected Journey, with Keras and Tensorflow
40mWhen to go deep in Computer Vision... and how