3D Reconstruction of a face from a single image

I don’t know if scary, jaw-dropping or simply impressive, but it turns out that there is already an algorithm developed here in Germany by Professors Volker Blanz and Thomas Vetter, that reconstructs a face in 3D from one image. Yes! out of a single image!

Maybe it wouldn’t sound that amazing at first sight, but you have to think about all the possible applications. Besides identifications, you can entirely play around with the information out of peoples’ faces, prediction, tracking, augmented reality, medical procedures, etc.

Is it something new? No way! It was presented in the International Conference on Computer Graphics and Interactive techniques of 1999! That’s right, fella, 14 years ago!

But why nobody used this impressive technique? Well, my guess is, of course, because of the computation times of it. Yes yes, very good, looks great and reconstructs faces pretty good, BUT took 50 minutes to reconstruct a face.

But wait a second! It was 14 years ago, right? With slower computers. Certainly, according to their paper, they used an SGI R10000 processor, which at that time was as powerful as 250 MHz… Well, I think is time to try it again with our new multi-core processors at several GHz.

Is it still on development? Did the creators give up? Are they preparing some new surprises? Well, we don’t know; but so far I can say that this technique has tons of applications with the new devices focusing in computer vision applications.

Just check the demo about this great procedure and the explanation of its development:

Advertisements

Predator Algorithm

The more I go into Machine Vision, the more I get amazed by the Engineering behind the most common thing around me.

Now, this post is not abut something that we can see daily now, but I’m sure it’s gonna be soon among us. Very soon. I’m talking about the Predator Algorithm (originally named TLD), developed by czech scientist Zdenek Kalal. This algorithm is implemented in such a way that the computer learns continuously detecting the features and objects in it.

Yes, it learns and with time detects easily every object. But well, nothing better than a graphical explanation of what I’m trying to explain, by the inventor himself:

The creator of this algorithm had such a success that he already started his own company to start the implementation of this great Engineering tool: TLD Vision.

These are the kind of things that really inspire.