Interview with Henrik Wann Jensen

Interview with Henrik Wann Jensen

Dr. Henrik Wann Jensen is a Research Associate at Stanford University where he is working in the Computer Graphics Group on realistic image synthesis, global illumination and new appearance models. He is the author of "Realistic Image Synthesis using Photon Mapping", AK Peters 2001. Prior to coming to Stanford in 1999, he was working as a postdoctoral research associate at Massachusetts Institute of Technology, and as a research scientist in industry where he added photon maps to a commercial renderer. He received his M.Sc. and Ph.D. in Computer Science from the Technical University of Denmark for developing the photon mapping method.

Interview by Nick Nakadate of MulvannyG2 for CGA.

CGA: Please explain how you came to inventing photon mapping. Who got you started in CG research?

HWJ: I have always been interested in computer graphics, and I have experimented with graphics ever since I got my first computer. My interest quickly turned to realistic image synthesis, and I started writing my own ray tracer and later added radiosity to simulate diffuse color bleeding. However, I wanted more. Radiosity
can only handle diffuse reflection, and it is too costly for complex geometry. As a result I began investigating Monte Carlo ray tracing techniques. Monte Carlo ray tracing techniques are very powerful: they can simulate any type of reflection and they have no problems with complex geometry. However, general Monte Carlo ray tracing is very slow and often results in noisy images. After trying several different algorithms I got the idea that resulted in the photon mapping algorithm. The photon mapping algorithm traces photons from the lights and caches them directly in a data-structure decoupled from the scene geometry. This strategy enables a method with the same general properties as pure Monte Carlo ray tracing, but much faster.

CGA: How do you see photon mapping being beneficial over radiosity based
renderers for architectural renderings?

HWJ: Photon mapping can handle complex scenes, and it works with any type of
material - not just diffuse (i.e. matte) materials.

CGA: Given the nature of photon mapping and its current inability to produce exact energy output (i.e. shoot exactly 200 watts of energy into a scene from a light), do you always see radiosity renderers, such as Lightscape and Radiance, having a place in the CG design community?

HWJ: Photon mapping can easily produce exact energy output. My renderer works with IES lights and is physically based all the way through. Photon mapping is actually more accurate than any radiosity method can be. All radiosity algorithms makes the fundamental assumption that every finite element (patch) in the scene has some simplified (often constant) illumination, and as such it converges to the wrong solution - except in the limit when an infinite number of patches are used. In contrast the construction of the photon map is unbiased which means that the
only error results from too few photons being used. The reconstruction of illumination based on a photon map is biased (i.e. has some error), but this is all local and easy to control. In contrast the error in radiosity is global and hard to analyze.

Radiosity still works well for simple diffuse scenes; however the quest
for realism makes these scenes less common. For all other types of scenes
there is no advantage in using radiosity.

Btw. Radiance does not use a radiosity algorithm (i.e. finite element
method). Radiance uses an irradiance cache which has many advantages
over radiosity - even though it is still limited to diffuse surfaces.

CGA: Is it foreseeable that photon mapping based renderers will attain the
ability to shoot and store accurate and predictable amounts of light energy?

HWJ:Yes, my own renderer does that already.

CGA: Of the current and soon-to-be released commercial GI renderers, which
do you think has the greatest promise?

HWJ: here is a large number of renderers on the market at the moment with
some sort of photon map implementation. The surprising thing is that
the most expensive renderers seem to have the worst implementations.

I cannot say which renderer is the best though.

CGA: Which CG company (ILM, PDI/Dreamworks, Square, etc.) has the most
promising render engine?

HWJ: It depends on what you want to render :) For architecture I think Square has an exciting system. For movie production no one comes close to good old PRman even though this renderer certainly has limitations.

CGA: Your BSSRDF model research is groundbreaking. How do you see this
applied to architecture?

HWJ: Architects are experts in materials and lights. The materiality of things is something architects have to consider. In graphics there has been a tendency to ignore the important interplay between materials and lights. To soften the appearance of a material most graphics programs offer only one option which is to make the light source larger. However, a large number of natural materials look soft because of subsurface scattering. Plants, snow and marble for example look soft even in strong sunlight. To correctly render the appearance of these translucent materials it is necessary to simulate the subsurface light transport - this is the key
idea in the new BSSRDF model.

CGA: Given the complexity of the BSSRDF calculations (8 dimensions I believe) do you see this as a realistic component of architectural renderings in the near future?

HWJ: The BSSRDF computations are very efficient. Most images can be rendered in a few minutes on a standard PC - for example the paper has a marble bust with 1.3 million triangles that was rendered in 1024x1024 with 4 samples per pixel on a dual P3-800 in less than 5 minutes; a standard diffuse reflection model takes 2 minutes.


CGA: Will desktop PC hardware meet the BSSRDF calculation requirements now, or will we be waiting a while? Speed a big concern with the design
community, as well as others, as we must produce under very extreme
deadlines most of the time.

HWJ: Yes. The BSSRDF model is practical on standard PC's

CGA: What do you see on the horizon for Photon Mapping, BSSRDF, and GI? Any new developments in the works?

HWJ: I would like to combine the BSSRDF with photon mapping, and also work on making these algorithms generally available and easy to use for complex models. I am always looking for complex models (I cannot make them myself) to test new algorithms, so I anyone has a new opera house with complex illumination that they would like to render then I am always interested :)

CGA: What have you been most proud of up to this point in your research?

HWJ: That many people use my research results to do real work. I am mostly interested in practical algorithms and I am always happy to hear when someone else has been using photon mapping or the BSSRDF model to achieve cool lighting effects in their images or animations.


CGA: What has been your biggest disappointment in your research? Any amusing stories?

HWJ: My biggest disappointment came when I joined a company after my PhD. I had developed photon mapping as part of my PhD, and they were interested in testing it in their commercial ray tracer. After demonstrating the first prototype for them they got all excited, but instead of saying thanks they threatened me with lawsuits and claimed that they now owned the photon map - even though I published several papers and a PhD thesis before joining them. Naturally, I left quickly.

CGA: Did you have anything to do with the whisky glass render in Final Fantasy?

HWJ: Only indirectly in that the photon map plugin that they used were written
by Jos Stam, and it is based on one of my papers.

CGA: What does the future hold for you? Job plans? Research? Back to Europe?

HWJ: Good question. I am having a lot of fun at Stanford. The graphics group here is great, and after moving to California I have come to the realization that winter is an option :) In some sense I feel like a kid in a candy store; I enjoy practical graphics research so much that I will try to follow the path that will allow me to continue what I do at the moment.

CGA: What is your favorite professional link? What is your favorite personal link?

HWJ:

professional: www.google.com
personal: www.google.com

The Internet is great for finding stuff; though it often is like searching for a needle in a haystack.