Python for Computer Vision

Motivation

• Improved productivity
• Concise code
• Good libraries for many purposes
• Popular with other scientific organisations
  • Willow Garage and ROS
  • Google
  • NASA
  • and more
• Taught to students
  • More and more students being taught as introduction to programming
• Transition from Matlab relatively painless
  • Numpy for matlab users
  • Mathesaurus

Alternatives

• C/C++
  • Time consuming for common mathematical operations and structures
  • No interactive shell
  • Edit, compile run cycle slows prototyping development
• Matlab
  • Licensing
  • Difficulty in deploying on autonomous systems
  • Closed source
  • Matrix legacy
• Octave
  • No hashtable or set support
  • Buggy 3D visualisation (octaviz)
• Sage - Open source mathematics software
  • Bundle of programs with python interface
  • Viable alternative to Magma, Maple, Mathematica and Matlab

MROL mapping case study

• 3D mapping algorithm
  • https://launchpad.net/mrol
  • Implemented in both Java and python
• Lines of code (Excluding unit tests)
  • Java 1000
  • Python 300
• Comparable performance

Basics

• Getting help - help or following with ?
• Accessing parts of arrays and matrices
• arrays versus lists
• Columns/rows
  • A[:,1]
• Reverse
  • A[::-1,:]

Array versus matrix

• Array - N dimensional array like entity
• Matrix - Similar to Matlab, mathematical entity
• Differences
  • Multiplication etc.
  • M = matrix('[1,2]; [3,4]')
  • A = array([[1,2], [3,4])

Python shells

• Standard, python - Always available
• ipython - Recommended
• bpython - Promising
• Reinteract - Worksheet based
• Sage

ipython -pylab

• tab completion
• help ?
• hist
• run versus import
• pdb
• using ipython
  • %magic for help on ipython functions
• Command history
  • .inputrc file
  • set editing-mode vi
  • hist
  • start type and then press up
  • search through commands history
• ipython qtconsole --pylab
• ipython notebook

Ipython tips

• Shortcuts
  • Automatic parentheses
  • pow 2, 3 -> pow(2, 3)
  • Automatic quoting
• Recursive reload, dreload
• Shell commands
  • ls cd pwd ...
• Can be used as a system shell
  • ipython -p sh
  • x = !ls
• Running code snippets
  • Paste
  • Middle mouse button
  • Run a scratch file
• Namespace
  • who
  • whos
• Support for parallel computing

Typical workflow

• Write a test if possible
• ipython -pylab
  • Similar to Matlab workflow style
  • Work on small sections of code in interpreter
  • When working incorporate into a text file
• Run unit test
• Repeat

Creation and reshaping

• Creation
  • zeros, ones, eye, diag
  • Sequences - arange versus range
• shaping
  • reshape
  • a.shape = 3,2
  • a.T
  • flatten
• Assign all elements
  • X.fill(3)
  • X[:] = 3

Arithmetical functions

• roots and logs
  • sqrt, log, log10, exp
• Rounding
  • round, ceil, floor, fix
• Mathematical constants
  • pi, e
• Complex numbers and trig functions

Concatenation and Selection

• Concatenation
  • hstack, vstack, dstack
  • concatenate((a,a), axis=1)
• Selecting elements
  • ranges, index lists

    a[1:3, :]
    a[[1,2], [0,1]]
    

Gotchas

• index from 0
• slicing refers to intervals rather than array elements
• integer division

  from __future__ import division
  print 2/3, 2//3
  

• Outputs (0.66666666666666663, 0)
  • Add to the path

    import sys 
    sys.path.append('/tmp')
    

• 1D variations
  • Vectors
  • 1 by n
  • n by 1

Linear algebra

a = arange(4).reshape(2,2)
a.T
linalg.det(a)
linalg.inv(a)
norm(a)
lingalg.eig(a)
linalg.svd(a)