My personal Paris-Nice in 3 days cycling challenge took place end of June. It's been a phantastic three days on the bike and I look forward to repeating it in 2011. A few differences to previous years: A) I stayed south of the Loire between Nevers and Digoin during my first night, this turned out to be much flatter and quieter than the departmental road a few kilometers to the north. B) Once in the Rhone valley at Andance, I stayed west of the river until almost 40km before Orange, and thus avoided having to go through Montelimar and Valence. The road is quieter and the scenic value a lot greater than on the N7. C) Once at the Mediterranean in Frejus I took the beautiful mountain road through the Foret d'Esterel instead of following the gruelling ups and downs along the coast. It's one 25km climb up and then downhill all the way to Cannes. Awesome! All in all, these changes made it quite a smooth ride. I didn't expect to beat my 2002 time and, frankly, I am happy that I managed to equal it.

It was the first time I rode without my shimano flightdeck or any other speedometer, and so only found out afterwards how much I had cycled each day: a whopping 350km on the first day, just over 250km on the second day, another 310km the third day, and 50km to the finish line in the early morning hours of the fourth day. Some pictures of the Paris-Nice trip.

Empora.com is currently looking for a lead developer to join our engineering team.

A video from our Spring trip to Morocco, produced by one of the other pilots.

and a few paragliding pictures from the morocco 2010 trip.

This conference paper is a simple extension of the 1-bit quantisation of wavelet coefficients proposed by Jacobs, Finkelstein and Salesin in the mid-90s. We show that a quantiser with four partition regions improves retrieval results and prove a few theorems that help to find optimal partition sizes.

We developed an unsupervised version of Grabcut in which the initialisation step is automated, and propose a new stopping criterion based on the KL divergence between background and foreground models.