Getting started in Computer Vision with Raspberry Pi
Francois Dion
Chief Data Scientist
Dion Research LLC
PyData Triangle
About me
Perceptron
In 1958, Frank Rosenblatt built the Perceptron Mark 1, a machine designed for image recognition. The machine simulated simplified biological neurons. It used an array of 400 photocells, to simulate the retina.
Modern Photocell
Raspberry Pi
Fundamentally, we still need a “machine” to do the work of the neurons, and a camera to do the work of the eye and retina.
Meet the Raspberry Pi family
Getting started
Python Modules
- SimpleCV
- OpenCV
- PyTorch + TorchVision
- Tensorflow 2 with Keras
- MXNet
Limited Power
- no conda, use virtualenv
- pip from piwheels.org/simple
- develop on desktop with GPU
- (retrain) deploy on Raspberry Pi
- consider acceleration:
- Tensorflow Lite
- TPU
Books
- 2011, Computer Vision by Richard Szelisky, Springer
- 2012, Practical Computer Vision with SimpleCV, O’Reilly
- 2012, Mastering OpenCV with Practical Computer Vision Projects, PacktPub
- For PyTorch or Tensorflow, start with the tutorials
- Get a cookbook and a deep learning book covering the package you prefer
Thank you!
fdion@dionresearch.com
https://slides.com/fdion
Right now, available for short to long term engagements, in advisory capacity or full time