In the first stage we focused on prototyping segmentation and classification methods and in the second part we will improve and help integrate the classification into their existing pipeline.

It came in a decorated box.

Well protected.

All shiny and street with it’s stickers

It’s good looking and definitely will be much easier to integrate in an installation than its predecessor. It’s a regular box with two cameras one hidden and one visible, so the front needs to be unobstructed for the device to work. Underneath the unit the microphone array is placed which makes it much cleaner looking. There is no motorized foot on the device which is compensated with a greater field of view.

Our initial tests with the device look very promising and we will write a future post about our findings. If you have an application that you would like to prototype or test on the device do not hesitate to contact us.

After constructing the graph dynamic markings can be added and modified when needed. These marking is in the example meant to represent runners or bikers and are updated based on time and speed of the runner.

It is possible to register for events triggered when pressing dynamic markings or segments.

You can find an interactive demo here.

The visualization is crafted using THREE.JS and works in both canvas and webGL mode.

The tool was created for Sportstiming and funded by Forsknings og Innovationsstyrelsen.

The National Gallery of Denmark, Redia and the Alexandra Institute developed an app that could bring a Hammershøi painting to life.

We have been involved right from the start and we take part in the latest research. Therefore you will be 3-5 years ahead of your competitors if you collaborate with us.

Invitation in pdf

Venue

19 December 2012 • 13:00 – 16:00

”Lille Auditorium”, INCUBA Science Park, Åbogade 15, 8200 Aarhus N

Registration to jesper.mosegaard@alexandra.dk

Plan

13:00 – 13:15 Introduction by Jesper Mosegaard, CG Lab, the Alexandra Institute
13:15 – 14:00 There and back again – a journey into VFX, Michael Bang, Aarhus University
14:00 – 15:00 OpenVDB, A Free Volumetric Toolkit For Visual Effects, Ken Museth, R&D FX Supervisor and Principle Engineer, Dreamworks Animation
15:00 – 15:15 Coffee and cake
15:15 – 16:00 Mystery guest, TBA

MICHAEL BANG

After completing a postdoc at Aarhus University in 2010-11, I was a research contractor in the research group at Weta Digital – the main visual effects vendor behind movies such as the Lord of the Rings trilogy and Avatar. I will describe some of the research challenges we were faced with during production on the movie Adventures of Tintin – Secret of the Unicorn. Solutions to the challenges were later published at some of the leading conferences and journals in computer graphics and include cutting edge results in rendering and simulation. In this talk I will provide an overview but focus on the results achieved in fluid simulation. I will also share some of my personal experience from working in the VFX industry and learning from the VFX wizards of middle earth. I have now returned to an assistant professor position at Aarhus University, and please note that since I am not a representative of Weta Digital, this talk will only cover material already in the public domain.

KEN MUSETH

Following the recent open sourcing of DreamWorks Animation’s sparse volume toolkit, OpenVDB and SideFX’s subsequent announcement to integrate it in its next major release of Houdini, this presentation is targeting early adopters that wish to experiment with the new technology. Fans of OpenVDB use the phrase “batteries included” to describe the fact that the library ships with a rich toolset of high-level volumetric processing and conversion tools that can be applied directly in VFX pipelines. The presentation will focus on three aspects of Ken Museth’s OpenVDB: 1) a description of its underlying data structures and its accompanying toolset, 2) its integration into Houdini as a first-class citizen, and 3) a tour of the existing Houdini nodes that are included in the open
source distribution.

Our method uses a gradient image $$G$$ which dictates the overall “growth” of the steam. A variable $$t$$ is animated over time from 0 to 2 and the amount of steam $$s$$ for a given position $$mathbf{x}$$ is then defined by a clamped linear interpolation over some range $$r$$:

$$!s = mathrm{clamp}left(frac{t – G(mathbf{x})}{r}, 0, 1right)$$

To make the growth more interesting, we modify $$G$$ with a noise texture. The color of the steam $$c_s$$ is determined from a lookup into a texture-rendered version of the scene which has been blurred by a Gaussian filter. To this value, we add some white noise to give the steam some texture. Finally, we add a value which can be tweaked to give the steam the desired amount of lightness.
The pixel color is then given by the unmodified color $$c$$ interpolated with the steam color:

$$!mathrm{lerp}(c, c_s, s)$$

The final touch to our effect, is the ability to interact with the steam by dragging the mouse around like a finger on the steamed up window. This is implemented by adding a mask texture and modifying the amount of steam by taking the minimum value of the mask and $$s$$. The mask texture is updated by blending black lines into it where the mouse has been when it is being dragged around. Furthermore, a dark grey color is constantly blended into the entire mask in order to make the older strokes steam up again.

To make this effect work we have applied an interesting texture advection shader technique used by Valve for water flow in Left for Dead 2. A presentation of the technique and its improvements for portal 2 was given at Siggraph 2010.

The method combines a few neat shader tricks. Lets say we have a surface texture T and velocities V for the surface. If we advect the texture by the velocities in V, we can make the texture flow according to the velocity field. But rather quickly the displacement of the texture makes it look stretched and wrong. We can remove the distortion by reverting to the original texture after every S seconds, but at the cost of a massive sudden change. We can hide the sudden revert back to the original texture by linearly interpolating between a second texture advection of the same texture. If we offset the advection by half the advection interval we can always show a not too distorted texture advection while reverting the advection of the hidden texture.

What we have added to the method is a layer on top of the advection method. With the Valve texture advection method we create a directional flow of a distortion texture (we use a velocity field with the same speed in the entire domain). The distortion is then used when doing texture lookup in the texture-rendered scene. In this way we can add time-dependent distortion with a directional flow to any surface. A video of the effect on a static background image is shown below.

It would be fairly easy to add depth dependent distortion to the method by using a depth buffer, but for better directability we chose to add a scaling mask instead.

The app is based on intern tests of shader capabilities and UI on the IPad, but now we were at it we made the tests into a small free app.

For instance tomorrow there is a two part course with the title “Advances in Real-Time Rendering in 3D Graphics and games” that covers the entire day. Then there are interesting paper presentations in Boundaries, Edges and Gradients from 9:00-10:30 as well as the Mesh Colors paper in the Textures session and Video tapestries with continuous temporal zoom paper in the Video session both at 10:45-12:15.

There’s also a few interesting talks by NVidia tomorrow which you can find more info on here. It could be interesting to go to the NSight talk 11:00-12:00 to hear a little about what it brings to the GPU programming scene.

At 17:15 there is OpenGL Birds of a feather by Khronos at the Wilshire Grand Hotel you can register here.

Since the paper presentations can be streamed from the ACM digital library after the conference I will probably stick with the course or maybe I will head for a day at the Geek Bar to be able to browse a little of it all.

Look forward to yet another great day at Siggraph