Julien's fuzzy blog

I wish you all a Merry Christmas !

By the way my paper has been accepted to the Body Sensor Network Workshop that will be held at the MIT in April

Back to work after a deserved lazy week-end. No paper to focus on now. I can concentrate on long-term objectives again but I guess it's also time to re-evaluate my way of working and interacting with people. I'll start probably by reading again Richard Hamming's talk: You and Your Research. A short quotation to start with:

"You find this happening again and again; good scientists will fight the system rather than learn to work with the system and take advantage of all the system has to offer. It has a lot, if you learn how to use it. It takes patience, but you can learn how to use the system pretty well, and you can learn how to get around it. After all, if you want a decision `No', you just go to your boss and get a `No' easy. If you want to do something, don't ask, do it. Present him with an accomplished fact. Don't give him a chance to tell you `No'. But if you want a `No', it's easy to get a `No'."

Writing a paper at the moment, an e-mail was suddenly sent to the \LaTeX users:

We MUST move your writing to MS Word for all your future articles.
I won't waffle about the reasons (real or official) (*), but I'm \textbf{really} unhappy about it \includegraphics{sad.png}. First on a technical point of view, I'm going to have to care about the style and so on. Versions won't be compatible on all computers, EndNote never works, I can't work from home as I don't have MS Windows, etc ...
Secondly, this seriously hurts my ethics. Lately I've been more and more reluctant to proprietary software. Should I really compromise and use this shit ? Is that a lab of computer scientists or of businessmen ?

\begin{Huge}\textbf{ ARRRGGGGG }\end{Huge}

(*) Actually, I can't resist to give you the first reason I was told: \textit{MS Word is more compatible} - troll inside !

After a new session of data acquisition with Dan and a long morning of calibration, I got some new images to show you, starring Benny .
As the colors of one camera were not very good (over-saturated), I've tried to use this camera only for 3D modelling, but not for shading. Some parts were not textured at all, so I decided to use this bad camera view only if there was no other texture available. In which case the upper body shading is much better, but the legs are quite problematic.

After correction of a bunch of bugs and hole filling.

The back is still strange, because this part of the fish is viewed rather tangentially by only one camera. Therefore the resolution is not sufficient for a proper reconstruction. Its right eye is probably blurred do to some inaccuracy in the camera calibration: images don't match perfectly.

My first results with Image-Based Visual Hulls. Not perfect as you can see, but I'll try to improve them. This is mostly due to an inaccurate camera cross-calibration. Also results using images of myself and my red T-shirt are not beautiful enough to been demonstrated


blended textures from various cameras (according to their positions)


texture of the nearest camera

Yesterday I acquired some videos of me moving around, aiming to reconstruct a 3D model. In order to get the camera parameters, I also acquired a couple of sequences with a grid for further calibration.



By the way, you can also see a bit of my "fashionable" red tight T-shirt. Theoretically used as a hiking underwear, I found it suitable to contrast with the background... but also to get lots of comments !

Today I had once more in my life to use MS Windows to write a couple of slides. I don't really understand how people can use that s***. After waiting 30 minutes for all the updates, I got loads of error messages all the time, crash notifications, etc. The CPU was always busy for some reason. I wanted to change to wallpaper, but it only takes BMP files and 'Paint' was not able to open my JPEG to translate it (convert wp.jpeg wp.bmp rules). I couldn't even launch IE once, it crashed before. Using Mozilla, I couldn't listen to the radio online although the plugin was installed. Then PowerPoint drove me crazy with the spacing and alignment in bullet boxes that were always everything but what I wanted.

Small game: can you reproduce this bug of a MS software ?

... and I received this letter:

Your registration as an Internal postgraduate student of the University of London has now been approved as shown below.

Beside my litterature review, I've started to test a bit some silhouette-based modeling algorithms. Nothing really new, but I can show you some preliminary results on background subtraction.

   

Original image (part of), after statistical background removal, after removing noise.

Next stages: shadows subtraction, MPEG-2 artifacts correction (not visible here) and generation of a view-dependent 3D model.

I tried a method relying on the Space Carving output specificities to generate in a low time a hull of the result. To date, I can say I failed... The hull doesn't cover all the fish at all. But rendering is incredibly faster.

Thanks to Mommas, a nice bug was fixed today !

Ok, sorry for the bad joke ... After 2 weeks of bug tracking, I eventually achieved to get a fairly good result. As you can see it's not perfect, in particular in the uncolored parts (white, grey or black). I'll now work on the improvement of these technics.



This was generated using only 3 low-resolution images of the real fish (the fish size is approx. 100-150 pixels). It was carved in a bloc of approx. 2Mvoxel (voxel size: 1mm, rendered 1.5mm to fill holes) in 5 minutes.

Many thanks to Dan for the camera calibration.

Yesterday I spent the day at the Nextwave Interface Annual Seminar. It was nice - I had too many coffees I was "wearing" a wireless electro-cardiogram (ECG) sensor the whole day: 3 electodes and a small box stuck on my chest. To show on a PDA how my heart was going to people wandering around ...

I've started to implement a "shape by space carving" algorithm. Nothing new for the moment, I'm just "playing" to see roughly what can be done. This algorithm takes a couple of camera images of a 3D object as input. The basic principle is to start with a cube and to carve every voxel that is not consistent with the set of images. This should converge to the "best" 3D modeling of the object regarding to the amount of information given by the set of images.
The application is still fully bugged, but I can't resist to show a first screen shot ! This modeling has been done with only 4 synthetic images as input, carving in a bloc of approx. 3,500,000 voxels