Slic Superpixels

I was bored so i implemented a pure python of slic superpixels.
http://ivrg.epfl.ch/supplementary_material/RK_SLICSuperpixels/  for the
paper.
The python code needs vigranumpy just to open the image and to compute
the gradient magnitude of the image. One can easy replace this to remove
the vigra numpy dependency.
The rest is just default stuff like numpy and matplotlib to visualize stuff.

https://github.com/DerThorsten/vision/

A part of my code is based on a great post of the blogger Peekaboo:
peekaboo-vision.blogspot.de region-connectivity-graphs-in-python


update:
 
 i also implemented it in C++ using the vigra-library .
It is soooo fast. Code will be added to github in a few days.
arrsoooo

Sublime: The text editor you fall in love wih....

http://www.sublimetext.com/ a text editor you have to love.

try it and love it!

Optimization Seminar - Fast correlation clustering for image segmentation

I did a seminar where I had to summarize the paper:
 Fast correlation clustering for image segmentation [Yarkony et. al 2012].

Since i think my talk was successful i decided to share my slides an my
short summary of the paper.

I decided to put (all pdfs) on www.pdfcast.org

The slides can be found at: Slides
The short summary at : Summary

C++ Benchmark: std::vector vs. std::list—Baptiste Wicht

A very nice benchmark comparing std::vector vs std::list can be found here.

C++ Benchmark: std::vector vs. std::list—Baptiste Wicht

Seems that i use std::vector where i should =)

OpenGM Python

An (improved) Python-Wrapper for OpenGm 2 OpenGM can be found
at my github https://github.com/DerThorsten/opengm .
This is still a beta version but seems to work without trouble (in most cases).

A major update in the inference system will be added soon.

Generate a FotoMosaik with python

This is just the baseline implemenation.

The implementation can be found at github:
mosaik.py-github
 This implementation is a first running version and has still a lot of bad stuff in the code.

The next version will be added soon.

• it should try to use more different patches
• it should avoid to put the same patches next to each other 

The picture below shows the output of the baseline implementation.
As one can see some pachtes are used to often.
But so far my database has only 500 images.
10k should be a reasonable size.
The result image has been downscales to reduce the size.