Blog

A new song I made yesterday. As usual it is not finished yet, but I'm too impatient.

Post NYE stress by gijzelaerr

In Februari 2012 I will start a new job. I will be working for the Anton Pannekoek institute, the astronomical group of the University of Amsterdam. More specifically i'll work on AARTFAAC.

The AARTFAAC (Amsterdam—ASTRON Radio Transients Facility and Analysis Centre) Project will take advantage of the advent of wide-field radio astronomy to detect extreme objects like black holes. We will use the LOFAR wide-field radio telescope to hunt for these extreme transients and systematically collect enough well-documented examples of the behaviour of each type of transient. Furthermore, we will expand LOFAR with a true 24/7 all-sky monitor to catch and study even the rarest of events.

I'll be working on pattern detection in the signals coming from the numerous radio antennas. Since this is data in extremely large quantities I can gain a lot of experience in the field of High Performance Computing and GPU computing. I can't wait to get started.

I've backported the latest OpenCV 2.3.1 package from Debian Unstable. You can find them in my OpenCV launchpad PPA. This means the package is split up into different packages per lib again. I want to have my package as similar to the original ones as much ass possible, so no future conflict emerge. Since Ubuntu is x86 and i686+ plus only, I can enable some more optimization flags. I don't know if they have a really big impact.

Changes from my previous packages:

• Split up into package per lib
• You should remove libopencv2.3 before upgrading to the new packages. If your upgrade fails, just remove libopencv2.3 and try again

Changes from the original debian packages:

• Enabled SSE3 only for amd64
• Enabled SSE2
• Enabled CUDA
• Enabled TBB
• Removed i486 optimisation patch, Ubuntu is i686 and amd64

I'm organising the Amsterdam Computer Vision Meetup (ACVM). This meetup well take place at the 16th of January 2012 from 20:00 till 23:00 and will be hosted at the Waag. The event is sponsored by Waag Society and Appsterdam - participation is free of costs. ACVM is an independent and open meetup where Computer Vision (CV) experts and amateurs can meet.

CV is a fascinating and fast growing field. The last years there have been major improvements in computer speed and therefore the ability to process images and video. Also more and more people are wearing small supercomputers with integrated camera in their pocket. There is a lot of knowledge and curiosity in Amsterdam related to CV, both from a professional and artistic perspective. But since this field is so cutting edge this knowledge is protected and locked away inside companies and universities. The recent popularly and success of social tech meetings around the world learns us that a lot can be gained by opening up, meeting and sharing your experiences and knowledge.

I think this event is the perfect opportunity for

• The curious mind to find out more about CV and techniques used
• The professional who wants to meet fellow specialists and share their pain and love
• The artist or entrepreneur looking for inspiration
• A place for employers and potential employees to meet.

These people will give a presentation:

• Cees Snoek (euvt.eu) - Video searchengines based on visual categories and events
• Jan-Mark Geusebroek (perceptech.eu) - Visual classification of banknotes for re-usage
• matthijs van Henten (ischen.nl) - Video To Sound
• Mark de Greef (VicarVision.nl) - FaceReader
• Jan van Gemert (UvA) - Correlating Photograpic Style with Object Categorization

Please let me know if you are interested and want to come by leaving a comment below. There is only a limited number of seats.

I've updated the OpenCV packages in the opencv2.3 launchpad PPA. FFMPEG is working again, and I also enabled CUDA support. I made a separate CUDA Toolkit PPA.

Differences compared to the original Ubuntu packages:

Installation:

$ sudo add-apt-repository ppa:gijzelaar/cuda
$ sudo add-apt-repository ppa:gijzelaar/opencv2.3
$ apt-get update
$ apt-get install libopencv-dev

Let me know if there are some problems in the comments, but please don't e-mail me about OpenCV related questions. Use the OpenCV user group for this.

Slowly I start to enjoy making music again. lat week I finally took the time to startup my music software again, I didn't do that since my accident. I wanted to do more old school sampling. RJD2 has become a big inspiration to me. Very productive, and sampling like a mad man. Below is the result. There are some sampling timing errors that are a artifact of the final audio rendering, I don't know where they come from. Ontzettende teringzooi by gijzelaerr

A new hobby of mine is beer brewing. I'm doing this together with some other guys. we're slowly preparing and planning everything for about a year now, and we are getting close to some action. I've prepared a separate website to capture our progress.

OpenCV 2.3.1 has been released so I updated the launchpad OpenCV repository.

Here they are, Ubuntu Natty packages of the OpenCV 2.2 branch, revision 5206. This is, at least at the moment of writing, the most recent, stable and bugless version of OpenCV available. You can see this as the probably/maybe/hopefully upcoming 2.2.1 release.

Differences to original Ubuntu packages:

Installation:

$ sudo add-apt-repository ppa:gijzelaar/opencv2
$ apt-get update
$ apt-get install libopencv-dev

For more information have a look at the PPA. Let me know if there are some problems in the comments, but please don't e-mail me about OpenCV related questions. Use the OpenCV user group for this.

To show the damage to my skull, a CT-scan of my head was made. Luckily, I could get a copy of the CT scan. After playing around with VTK I managed to make a nice 3D rendering of the dataset.

This is only one side of my skull, the complete dataset is too big to fit in the memory of my computer. I still need to find out how to resize the dataset, if I will ever do this.

I recompiled the 5.6.1 VTK Debian packages for Ubuntu Natty Narwhal and have put them in a public launchpad PPA. The only change is the enabling of the Ogg Theora video codec. To install run:

 $ sudo add-apt-repository ppa:gijzelaar/vtk
 $ sudo apt-get update
 $ sudo apt-get install libvtk5.6

Bad luck, I had an accident 2 months ago. Broke many bones, but the biggest problem is that I have quite a bad concussion. I'm at home, can't do much at the moment. I also have a scull fracture, which caused permanent hearing loss. I made a simulation of the experience:

Hearing loss simulation by gijzelaerr

Sonic Gesture is finished and I am graduated! Sonic gesture can be downloaded from the project site. Prebuild Ubuntu binaries are available from a launchpad repository. If you have problems getting the release to work feel free to contact me.

Here you can see an example of how to use sonic gesture:

After I graduated a company called Third Sight Vision offered me a job and I accepted. I'm going to focus on Computer Vision, Augmented reality and mobile devices.

Slowly I'm making more progress with my Sonic Gesture project. I've trained the program with 12 hand signs now, and i've been working on getting the output more stable. Check out the movie below for the results:

Yesterday OpenCV 2.1 was released. There are not many new features, this is more a stability update. Last week I did an update of my OpenCV 2.0 Ubuntu packages so they include the new Python API. I'm happy I did this, since now the packaging is OpenCV 2.1 ready. I've made packages for 2.1 and put them in my PPA.

I've made some small changes:

• libcv4, libhighgui4 and libcvaux4 are replaced by one opencv library package libopencv0.
• This is the same for libcv-dev, libhighgui-dev and libcvaux-dev, they are replaced by libopencv-dev.
• Python API can use Numpy, this is enabled. This means the Python package depends on python-numpy.
• There is a new package named opencv which contains the training binaries. This package depends on the libraries, so it also acts as a sort of meta package.
• All documentation is in the opencv-doc package and is installed in /usr/share/doc/opencv.

I've received a lot of e-mails asking about my modified Ubuntu packages and how to use the new Python API. The old packages I made didn't contain the new Python API. I had some problems getting this to work, but I finally managed to do this.

I've made new packages, based upon the latest Debian OpenCV packages. I've modified them to include the new python API, and the OpenCV cmake configure script is included now also.

Run the following command to enable my repository:

sudo add-apt-repository ppa:gijzelaar/opencv2

Below is a screen capture of my first prototype for translating gestures into sound. Left hand controls pitch, right hand controls Frequency Modulation. Note that this is real time, and without prior knowledge about skin color. Skin color is extracted from my frace. Want to know more? Check out the source code on the project site.

This month I'm starting my graduation project for my Master. I will research the possibilities of translating sign language into sound, and implement this. I made a Google Code project for this, where you can keep track of my progress.

here you can see a preview of the first step of the project: