imagetune 0.1.0rc0

Live tuning of image processing algorithms

IMAGETUNE

imagetune is a simple GUI to interactively tune image-processing parameters. It utilizes the hardware accellerated speed of fastplotlib to provide visual feedback that is not slowed down by plotting.

imagetune is simple to use requiring only minimal changes to your code. Simply wrap the functions you wish to tune the parameters of with tune and launch tuneui.

How to use

Here is a simple script to make a binary version of an image:

from skimage.filters import gaussian
from skimage import data


def threshold(im, thres_val):
    return im > thres_val


def preprocessing(im):
    bg = gaussian(im, 10)
    fg = im - bg
    segmented = threshold(fg, 0.1)
    return segmented


im = data.coins()
result = preprocessing(im)

The script depends on parameters sigma=10 for Gaussian filtering and thres_val=0.1 for the thresholding. In order to tune these live, simply wrap those functions in tune and launch tuneui:

from skimage.filters import gaussian
from skimage import data
+from imagetune import tune, tuneui


def threshold(im, thres_val):
    return im > thres_val


def preprocessing(im):
-    bg = gaussian(im, 10)
+    bg = tune(gaussian)(im, 10)
     fg = im - bg
-    segmented = threshold(fg, 0.1)
+    segmented = tune(threshold)(fg, 0.1)
    return segmented


im = data.coins()
+tuneui(preprocessing, im)

This launches a window in which the parameters are tunable interactively:

You can also decorate your functions instead

+@tune
def threshold(im, thres_val):
    return im > thres_val

Note that the function still works as normal:

output = preprocessing(im)

Only when you run tuneui(preprocessing, im) does it behave differently.

Installation

Install directly with pip:

pip install imagetune

Function requirements

All tunable functions are asssumed to take an image as its first input and return an image as its output.

Choosing parameters

Given a function:

def adjust(im, alpha, gamma):
    return alpha * im**gamma

The decorator tune will per standard assume you wish to tune the first parameter (alpha). You can change this by specifying argnames:

tune(adjust, argnames='gamma')

or argnums:

tune(adjust, argnums=2)

The behavior of the above is identical. You can also tune both parameters, by either

tune(adjust, argnames=('alpha', 'gamma'))

or

tune(adjust, argnums=(1, 2))

Misc

To overwrite function names in the UI, simply pass along a name:

bg = tune(gaussian, name='blur')(im, 10)

You can also choose a min, max of the parameter bar:

bg = tune(gaussian, min=0, max=300)(im, 10)