Python for Computer Vision 2

Operators

Arithmetic
- power, modulus
- 2**2
Inplace operations
- +=
- methods on array
Relational - > < = !=
Logical
- any(a) and all(a)
- logical_and
- A and B for lists
where

Data structures

hashtable, dict
lists, queues, deques
array <-> list
import collections

Sets and Statistics

Sets
```
a = set([1, 2, 4, 5, 5])
b = set([3, 5, 5])
```
- Operations - union, intersect, unque1d, intersect1d
Statistics
- mean, median, std, var

Debugging example

Standard python debugger
- pdb
- Similar to gdb
Post mortem debugging
- pdb -m pdb buggy.py
Ipython debugger
- auto break on error with %pdb
- Can be used standalone or within ipython
- Advantages over pdb
- Tab completion
- Syntax colouring
- Plot functions to help see problems
Place breakpoints in code
- import pdb; pdb.set_trace()
Set break point by line number
```
pdb buggy.py
b 23
c 23
```

Fancy indexing

Ndimage

Ndvision

Refactoring

Generally easier
- Duck typing
- Changing method signatures
- add default argument
- Named/labelled arguments makes order less important
Tools
- Rope
- Bicycle Repair man
- Search and replace easier

Publication quality plots

Latex within plots
Combining with pdflatex
Vector graphics ideal
2D plots
3D plots
- Vector graphics possible but often not suitable
Fonts for publication
- pdffonts utility
- Have to be embedded and subsetted for publication

Publication quality plots code

import matplotlib.pyplot as plt
from matplotlib import rc

fig = plt.gcf()
rc('text', usetex=True)
rc('font', family='serif')
rc('legend', fontsize='medium')

plt.plot(np.arange(10)**2)
plt.xlabel('$x$')
plt.ylabel('$x^2$')
plt.show()
plt.savefig('graph.pdf')

Visualisation

2D is a subset of 3D
- Use a 3D visualiser
matplotlib
- import matplotlib.pyplot as plt
- Capable of 3D as well (not GPU rendered)
3D plotting with mayavi2
- Running mlab scripts
Alternatives
- Vpython/python visual
- python-vtk - Visualization toolkit bindings for python

Example Visualisation

Video cuboid

Example Visualisation Code

#! /usr/bin/ipython -wthread
import pylab
import enthought.mayavi.mlab as mlab
I = pylab.imread('dichroic.png')
mlab.surf(100*I[...,0])
mlab.savefig('dichroicsurf.png')

Simple animations

from pylab import *
import time

ion()

tstart = time.time()               # for profiling
x = arange(0,2*pi,0.01)            # x-array
line, = plot(x,sin(x))
for i in arange(1,200):
    line.set_ydata(sin(x+i/10.0))  # update the data
    draw()                         # redraw the canvas

print 'FPS:' , 200/(time.time()-tstart)

Examining Video

Load video as 4D array
- image height, image width, RGB, frame number
- Large arrays
- memmap
- hdf5
Display frame
- imshow(V[:, :, :, 0])
Plot pixel colour variation over time
- plot(V[10, 10, :, :])
Example video_array.py
Exercise for reader
- Generate video thumbnails/comic strip

Computer Vision

python opencv
Grab frame from camera
Frame differencing demo
Camscan demo
Surf to look at images
- Hardware accelerated
Dichroic filter eample
- dichroicsurf.py

Frame differencing demo

import sys
import time
import cv

def get_image():
    return cv.QueryFrame(camera)

#camera = cv.CreateCameraCapture(0)
camera = cv.CaptureFromFile('/jcr/coding/opencv/tree.avi')

cv.NamedWindow('Frame Difference')
cv.NamedWindow('Original')
cv.StartWindowThread()
lastim = get_image()
D = cv.CloneImage(lastim)

while True:
    im = get_image()
    if im == None: break
    cv.AbsDiff(im, lastim , D)
    cv.ShowImage('Frame Difference', D) 
    cv.ShowImage('Original', im) 
    lastim = cv.CloneImage(im)
    keypress = cv.WaitKey(200)

Data analysis

Data
- Loading
- Plotting
- Analysing
- Processing
plotfile example

Example Parsing of a data file

Side by side of matlab code
Equivalent python code
Load whole file into memory as an array
load and parse line by line
load chunks of file as arrays and parse
Make a string appear like a file
- known as memory files
- cStringIO
- StringIO

Performance

Use built in numpy functions
Apply along axis
Calling C
Inline C with weave
Google working on fast python
- Unladen swallow

Misc

openrave with python
RANSAC demo
- ransac.py
Optimizing
- Profiling
Python C integration
- Calling C from python
- Scripting C
- Calling python from C
- Calling Python

Debugging C called from python

Possible when using cpython implementation of python
get the PID
- ps -ef | grep python
gdb PID
More info
- Debugging with GDB

Relevant Software Libraries and Tools

OpenCV - Large C++/Python/C computer vision library
PCL - Point cloud processing library
Ndimage - python N-dimensional array library
Ndvision - video processing based on N-dimensional arrays
MROL - Multi-resolution occupied voxel lists for efficient environment representation and mapping

Further information

Scipy website
python help files
Ipython
- Ipython manual
Mathesaurus
- Mathesaurus
Interesting links

Scientific programming with Python

Scientific programming with python and svn
- http://openwetware.org/wiki/Open_writing_projects/Scientific_Programming_with_Python_and_Subversion/Outline
- http://openwetware.org/wiki/Open_writing_projects
Why python?
- Argument for python in scientific programming http://openopt.org/Python_For_Scientific_Engineering
- Software carpentry
- How to think like a computer scientist Good introduction to python and software engineering