Practical Intro to OpenCV & TensorFlow Lite
summary
Walks through building a smart shopping cart used with Caper Cart / Instacart hardware: seven cameras, integrated scale, LED feedback, and a working prototype that pairs camera events with weight changes. Shows using background subtractor in OpenCV to isolate items, fine‑tuning mobile net V2 in a Jupyter notebook to ~0.97 test accuracy, converting to TensorFlow Lite with metadata, and using Perfetto traces to find a ~120 ms per frame inference hotspot and pixel‑copy bottlenecks.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 6:50Smart shopping cart overview
- 14:38OpenCV background removal
- 17:24CameraX setup
- 19:55Image classifiers
- 23:05Training with Jupyter
- 27:00Fine-tuning MobileNetV2
- 34:50Model training metrics
- 38:47Converting TensorFlow models to TensorFlow Lite
- 41:12Perfetto optimization
- 46:27Q&A
related talks
47mUnlocking Data-Driven Apps with TensorFlow
22mOpenCV and TensorFlow in practice
01h 29mComputer Vision: An (Un?)Expected Journey, with Keras and Tensorflow
31mMake your IoT even Smarter with Tensorflow Lite to Design the Future of Vertical Farming
14mComputer vision on edge devices
27mMachine Learning on the edge using TensorFlow Lite
41mPractical Aspects of Applying Modern Computer Vision Technologies
38mReal-world Machine Learning with TensorFlow and Cloud AI