wednesday 01 february 2012
We (N. Thome, M. Cord, A. Rakotomamonjy and me) received the notification of acceptance for a poster presentation at ESANN 2012. This work is on learning product combinations of kernels.
The sketch is as follows: suppose you have several types of features and signatures leading to a variety of kernels (typically Gaussian kernels). This is quiet a common scheme in Computer Vision. You might want to combine them, and usually people use MKL approaches (i.e. a weighted sum of kernel). However, in most cases these kernels are redundant, and you would better do a product combination of these (think of the different scales in Spatial Pyramid or different scales of the same descriptors). The product is like an 'AND' gate while the sum more like an 'OR' gate, thus if your features are redundant, the product is more likely to denoise than the sum.
The bad thing about this product counterpart of MKL is that it is non-convex (we have a nice proof about this). So we proposed and algorithm finding a local optimum. While this might not be the best combination possible, it is sufficiently robust in practice to remove non-informative kernels. The good thing is that it also performs the kernels parametrization without any need for cross-validation.
Once I've cleaned the paper of all remaining typos and corrections as suggested by the reviewers, I'll put the code online.
monday 23 january 2012
I'll be in Vienna this week. This is part of the annual trip to a German speaking country with the second year students. This year, we'll have the excellent opportunity to visit the University of Technology.
Most important of all, it's also the occasion to go to the opera and see die Zauberflöte!
monday 09 january 2012
Tomorrow, Peter (University of Surrey) will be showing us the latest advances of his work in computer vision. The seminar is open to anyone willing to attend, but the entrance of the lab can be difficult if you're not working here. Please send me an email if you're coming so I can make things easier.
This presentation focuses on a number of steps in Bags-Of-Words including: i) segmentation-based descriptor design, ii) descriptor-to-visual-vocabulary coding step including Soft Assignment and its connection to Linear Coordinate Coding methods, and iii) Spatial Coordinate Coding to reduce histogram representations with Dominant Angle and Colour Pyramid Match to exploit non-spatial bias in images.
Regarding i), segmentation-based image descriptors for object category recognition were investigated. In contrast to commonly used interest points the proposed descriptors are extracted from pairs of adjacent regions given by a segmentation method. In this way we exploit semi-local structural information from the image.
Regarding ii), we show that one can take two views on Soft Assignment: an approach derived from Gaussian Mixture Model or special case of Linear Coordinate Coding. The latter view helped us propose how to optimise smoothing factor of Soft Assignment in a way that minimises descriptor reconstruction error and maximises classification performance.
Regarding iii), alternative ways of introducing spatial information during formation of histograms were investigated. Specifically, we proposed to apply spatial location information at a descriptor level (Spatial Coordinate Coding). Lastly, we demonstrated that Pyramid Match can be applied robustly to other measurements: Dominant Angle and Colour.
Info: Tuesday 10th jan, 15h, room 384, ENSEA
tuesday 03 january 2012
That's how I like new year beginning: a project I've been working on hard has been approved and will begin shortly. It's called TerraRush
The main topic is the indexing of video rushes (up to 50h+ of video per hour of movie) to facilitate their commercial exploitation. More on this soon...
friday 25 november 2011
The fact is I had to buy a new phone, so I went shopping for it this morning. While there, I tried the Kinect from Microsoft on a golfing game.
This is certainly the ultimate game. I mean, the Kinect is a great piece of technology with top algorithms, but there is certainly no way it can detect your arms movement so as to precisely estimate how you would hit the ball and aim at the hole.
On easy mode, this is just flagrant. You do a random swing of your arm and the computer displays a random animation of the ball flying in the air and then falling at a random location nearby the hole. You don't even have to concentrate and aim, every thing is randomly generated to make you feel you achieve a pretty good result. Everything is faked.
That's really a new kind of game, because the results don't rely on your performance. You have minimal input, but still it is entertaining. It is a kind technological version of flipping a coin. The difference being that flipping a coin for 2 hours is not entertaining. You do some random gestures, and the game creates a totally virtual response, giving you a virtual sensation of you achieving something. Totally disturbing.
What you feel, is not real. Virtual Emotions...