Profile

Cover photo
Rodrigo Benenson
Works at Max-Planck-Institut für Informatik
Attends School of life
Lived in Saarbrücken, Germany
648 followers|381,143 views
AboutPostsReviews

Stream

 
Microsoft accelerates the race towards augmented reality in daily life. I have to say: I am impressed.
 
Hands-on with Microsoft’s unbelievable new holographic goggles http://www.wired.com/2015/01/microsoft-hands-on/ #AR, Project #HoloLens, #Baraboo
The prototype is amazing. It amplifies the special powers that Kinect introduced, using a small fraction of the energy. Project HoloLens’ key achievement—realistic holograms—works by tricking your brain into seeing light as matter.
View original post
3
Add a comment...
 
Tomorrow is going to be an exciting day.  
Success or failure, I am so happy to see my tax money spent on bold adventures that push the boundaries what mankind has achieved.

I wish the best outcome to the ESA mission on Comet 67P.
 
You asked us: can we watch Rosetta and Philae's historic landing event live? Yes you can - via our dedicated Rosetta site: rosetta.esa.int.

Streaming starts at 20:00 CET (19:00 UTC).

Alternative link: http://new.livestream.com/ESA/cometlanding

#cometlanding  
7 comments on original post
3
2
Zeeshan Zia's profile photoSabine Sur's profile photo
Add a comment...
6
2
Kosta Derpanis's profile photoGibran Fuentes Pineda's profile photo
Add a comment...

Rodrigo Benenson

Shared publicly  - 
 
+Dyson  is investing in robotics ( https://www.dyson360eye.com ).
Check out this cool (Dyson sponsored) robot concept developed in +Walterio Mayol 's laboratory.
1
2
Walterio Mayol's profile photodavid moloney's profile photoTim Hutton's profile photo
 
Hi Rodrigo, thanks. Yes +Dyson is indeed very strongly investing into robotics and also as part of that, into computer vision, they even sponsored BMVC for us here in Bristol last year. The 360eye is from their main research labs. Our work is actually funded  separately, by their charity, the Dyson Foundation which is actually a very nice concept to engage more PhD research into "making" things. One can only expect other mayor household names to do similar in the coming years!
Add a comment...
 
Here is small demonstration I created showing the power of http://emscripten.org
 
Do you have a C/C++ computer vision algorithm that runs (single core) in ~1 second or less?  
Then you can nowadays show it directly in the web !
The http://emscripten.org compiler works fantastically well.

The attached link is a demonstration I created showcasing structured forest boundaries estimation.
Multi-scale decision trees doing structured regression pixel-wise, running directly in your browser's javascript engine !
The C++ code uses libraries such as boost::gil and Eigen (both header only). The task may seem simple, but its non-trivial, and the demonstration presents state-of-art results.

In my experience, expect a 2x~5x slowdown from non-SIMD single core C++. This is still plenty good to demonstrate fast algorithms !
Being able to show computer vision algorithms to anyone/anywhere (with a modern web browser) is a great gain compared to "go to github, install dependencies, pray it will compile".


#emscripten   #computervision  
Estimating object boundaries in the browser. Object boundary estimation using structured forests from P. Dollar and CL. Zitnick, ported to C++ by P. Krähenbühl, bundled into javascript magic via emscripten. (Tested on Google Chrome, might misbehave on other browsers.) ...
2 comments on original post
5
Add a comment...

Rodrigo Benenson

Shared publicly  - 
 
Just updated my results collection page with a couple of new results for STL-10, CIFAR-10, and Pascal VOC comp3 (covering CVPR 2014, ECCV 2014, and arXiv papers).

Any important paper that I am missing ?
Any other dataset I should be tracking ?

Suggestions for improvement are always welcome.
8
3
Sergey Ten's profile photoXu Jia's profile photoMohsen Amiri's profile photodavid moloney's profile photo
3 comments
 
Then I'll have more stable results, like make average with many random seeds. Right now I have anecdotal evidence - I run 120 epochs, I get 8.4 error, and after I check another variation of convnnet architecture. Then I'll have final layout, and if it stable I'll put it into github. For now in github 8.8 error layout, with very big convolution layers and l2svm cost, and some arithmetic operations on layers
https://github.com/s271/convnet_rw
Add a comment...
In their circles
1,251 people
Have them in circles
648 people
Adnan Doric's profile photo
Roberto Teixeira's profile photo
Anna Senina's profile photo
Christian Aubree's profile photo
Alex Mansfield's profile photo
Peter Sheldrick's profile photo
Xucong Zhang's profile photo
Pierre Sylvain Coq's profile photo
Christopher Mei's profile photo

Rodrigo Benenson

Shared publicly  - 
 
Every year edge.org asks a "non-trivial question" to a group of ~100 thinkers. The question of 2015 is 

"What do you think about machines that think ?".

You can enjoy the 185 answers here: http://edge.org/contributors/what-do-you-think-about-machines-that-think
5
1
Zeeshan Zia's profile photo
Add a comment...

Rodrigo Benenson

Shared publicly  - 
 
Quite impressive demonstration. It somewhat reminds me of the sound quality of international (land-line) phone calls in the 1980s.
Cannot wait to see how awesome this will be in ~10 years.
Microsoft's research department has been building some incredible things of late, and one of them comes from the Skype team and was demoed on stage at..
1
Add a comment...
 
Quote of the day: "part of being human is that we jiggle when we move".

Very nice work on human shape reconstruction from sparse motion capture points.
 
Our new Siggraph Asia paper addresses a common problem; that mocap results in unnatural animations.

Mocap systems output skeletons, throwing away all the soft tissue motions of real humans. The result is a lifeless animation.

In contrast, MoSh estimates body shape, pose and soft tissue deformations directly from sparse markers. MoSh turns a mocap system into a body scanner. Animation is driven directly by the markers, preserving subtle nuance.

MoSh captures soft tissue motions without large marker sets. Soft tissue motions make animations more realistic. These can be amplified or attenuated and even retargeted to new characters.

MoSh is automatic, does not need a body scan and can work with any marker set.
2 comments on original post
2
1
Pierre Moulon's profile photo
Add a comment...
 
Here is small demonstration I created showing the power of http://emscripten.org
 
Do you have a C/C++ computer vision algorithm that runs (single core) in ~1 second or less?  
Then you can nowadays show it directly in the web !
The http://emscripten.org compiler works fantastically well.

The attached link is a demonstration I created showcasing structured forest boundaries estimation.
Multi-scale decision trees doing structured regression pixel-wise, running directly in your browser's javascript engine !
The C++ code uses libraries such as boost::gil and Eigen (both header only). The task may seem simple, but its non-trivial, and the demonstration presents state-of-art results.

In my experience, expect a 2x~5x slowdown from non-SIMD single core C++. This is still plenty good to demonstrate fast algorithms !
Being able to show computer vision algorithms to anyone/anywhere (with a modern web browser) is a great gain compared to "go to github, install dependencies, pray it will compile".


#emscripten   #computervision  
Estimating object boundaries in the browser. Object boundary estimation using structured forests from P. Dollar and CL. Zitnick, ported to C++ by P. Krähenbühl, bundled into javascript magic via emscripten. (Tested on Google Chrome, might misbehave on other browsers.) ...
2 comments on original post
5
3
Zeeshan Zia's profile photoJeffrey cheung's profile photo
Add a comment...
 
Do you have a C/C++ computer vision algorithm that runs (single core) in ~1 second or less?  
Then you can nowadays show it directly in the web !
The http://emscripten.org compiler works fantastically well.

The attached link is a demonstration I created showcasing structured forest boundaries estimation.
Multi-scale decision trees doing structured regression pixel-wise, running directly in your browser's javascript engine !
The C++ code uses libraries such as boost::gil and Eigen (both header only). The task may seem simple, but its non-trivial, and the demonstration presents state-of-art results.

In my experience, expect a 2x~5x slowdown from non-SIMD single core C++. This is still plenty good to demonstrate fast algorithms !
Being able to show computer vision algorithms to anyone/anywhere (with a modern web browser) is a great gain compared to "go to github, install dependencies, pray it will compile".


#emscripten   #computervision  
Estimating object boundaries in the browser. Object boundary estimation using structured forests from P. Dollar and CL. Zitnick, ported to C++ by P. Krähenbühl, bundled into javascript magic via emscripten. (Tested on Google Chrome, might misbehave on other browsers.) ...
16
8
david moloney's profile photoOliver Hamilton's profile photoZeeshan Zia's profile photoJeffrey cheung's profile photo
2 comments
 
~1 second in C++ so the javacript runs in ~5 seconds. The logic here is more than ~5 seconds is  "way too slow" for persons doing web browsing. 
Add a comment...

Rodrigo Benenson

Shared publicly  - 
 
Just discovered Jibo (created by +Cynthia Breazeal). Not sure it will deliver on the promise, but sure looks like a move in the right direction. More natural "real world" interaction with technology makes our life easier.
Visibly the future is technology is invisible technology.

http://spectrum.ieee.org/automaton/robotics/home-robots/cynthia-breazeal-unveils-jibo-a-social-robot-for-the-home

http://igg.me/at/jiborobot/x

Its tangible aspect makes me think of https://www.sifteo.com
1
Add a comment...
People
In their circles
1,251 people
Have them in circles
648 people
Adnan Doric's profile photo
Roberto Teixeira's profile photo
Anna Senina's profile photo
Christian Aubree's profile photo
Alex Mansfield's profile photo
Peter Sheldrick's profile photo
Xucong Zhang's profile photo
Pierre Sylvain Coq's profile photo
Christopher Mei's profile photo
Education
  • School of life
    present
  • École publique rue de l'Ouest, Paris 14ème
  • Alliance Française de Valparaiso
  • UTFSM
  • Mines ParisTech
Story
Introduction
Dreamer, thinker, doer; 
tries to smile as much as he can, 
tries to do as much good as he can.
Work
Occupation
Engineer & Researcher
Employment
  • Max-Planck-Institut für Informatik
    Post-doc researcher, 2013 - present
  • K.U. Leuven
    Post-doc researcher, 2013
  • INRIA
    PhD student
  • Mines ParisTech
    PhD student
Places
Map of the places this user has livedMap of the places this user has livedMap of the places this user has lived
Previously
Saarbrücken, Germany - Leuven, Belgium - Versailles, France - France and Chile - Valparaiso, Chile - Paris, France
Public - 3 years ago
reviewed 3 years ago
Very good service, clean rooms. And they give green apples for free !
Public - 4 years ago
reviewed 4 years ago
Great value for price. One of the "must eat at" places in Leuven.
Public - 4 years ago
reviewed 4 years ago
7 reviews
Map
Map
Map
The mainstream cinema of Leuven. The only place where you can see Holywood productions in 3d.
Public - 4 years ago
reviewed 4 years ago
Quite good food.
Public - 4 years ago
reviewed 4 years ago
Nice atmosphere, ok food, good service.
Public - 4 years ago
reviewed 4 years ago