Wednesday was the first full day of BMVC. Most of the talks were out of my areas, so I could follow enough to understand, but not to really engage with stuff much. The hilight of the day (which went on to win the best paper award) was a paper from Oxford Brookes on joint classification and stereo reconstruction for street scenes using graph cuts. It was a nice paper and quite similar in some ways to work a colleague of mine has done on football data.

I finally arrived at Aberystwyth at 1am after a nightmare train journey which ended in a taxi ride from Shrewsbury to Aberystwyth (paid for by First Great Western thankfully)!

Anyway, the pain was worth it as the day started with Simon Price's rather good tutorial on dimensionality reduction. He covered everything from PCA to Multifactor Gaussian Process Latent Variable Models discussing the advantages, drawbacks and interrelations of all the different methods. Everything was being videoed so I assume it will be made available online at some point.

So, not really much computer vision in this, but I love this kind of stuff. I wonder if you could do something like this with a mobile where you can attach a message to a wall using the mobile and then read it by looking at the wall through another mobile...

Seems like someone has put together a half decent stab at proving that P != NP.

What does it mean? Nothing much, people have been assuming that P != NP for a long time. Still, it would be nice to know we were right.

Listen to the words of Gael Varoquaux.

There is much truth in his words. Particularly the following:

"My view of code design and software engineering has progressively evolved to favor extreme simplicity over sophistication. I believe that a good programmer should know design patterns, powerful language features, libraries dark corners, and not use them unless absolutely necessary."

and:

Here is some pretty cool work presented recently at Siggraph on a camera-less laser projector.

Watch the video here:

V&L NET is a new academic community / group that has been set up to help "researchers from the fields of Computer Vision and Language Processing to meet, exchange ideas, expertise and technology, and form new partner- ships".

You can sign up for a quarterly newsletter, and Associate membership is free. It's still early days yet, but worth a look if you are in any of the relevant fields.

I do quite a lot of work on 3D triangle meshes, making them, deforming them, matching them, etc. etc. So I am on the lookout for a good mesh manipulation library that I can use rather than writing all the code myself.

My checklist of points I am looking for is:

An interesting blog post on optical illusions over at brain blogger. I think that computer vision could benefit a lot from studying optical illusions. The brain uses a lot of tricks in order to interpret the visual information we receive from our eyes and optical illusions occur when these tricks go wrong.

Wow, I've only been neglecting this site for a few weeks, and it has really been hit by the spam commenters.

I've tried to beef up the security a bit and cleaned out all the rubbish. Let's see if that slows them down a bit....

Joe