fxpodcast #316: Marcos Fajardo

Mike Seymour talks to the founder of Solid Angle’s Arnold, Marcos Fajardo.

Marcos Fajardo won the Sci-tech Scientific and Engineering Award (Academy Plaque) along with Chris Kulla, Alan King, Thiago Ize and Clifford Stein for Arnold. Their renderer was honored for being a highly optimized geometry engine with novel ray-tracing algorithms which unified the rendering of curves, surfaces, volumetrics and subsurface scattering. Arnold started life at Sony Pictures Imageworks and then split into two development streams when Fajardo established Solid Angle SL. Solid Angle was purchased by Autodesk last year.

Arnold’s scalable and memory-efficient single-pass architecture for path tracing, its authors’ publication of the underlying techniques, and its broad industry acceptance were instrumental in leading a widespread adoption of fully ray-traced rendering for motion pictures. We have covered the history of Arnold and spoken to Fajardo on previous occasions but with the major new release of the new version of Arnold coming very soon, Mike Seymour sat down with Marcos Fajardo to discuss the direction the Renderer and the Industry is heading in looking forward.

In particular this interview focuses on research in the area of rendering. A hot topic in rendering is Noise reduction and Solid Angle recently published a really interesting paper on reducing perceived noise at SIGGRAPH 2016.

The paper is available here.

 

Their method uses blue-noise dither masks tiled over the image to correlate the samples between pixels and thus minimize the appearance of noise via a better distribution of the noise or ‘estimation error’.

Without actually reducing the amount of error, this produces images with higher visual fidelity than traditional random pixel ray tracing, especially when using a small number of samples per pixel.

This approach improves the visual appearance of a Monte Carlo rendering without increasing the sampling effort or render time.


Blue noise, was not a Solid Angle invention, it is a concept which uses a different form of noise. Normal noise is called white noise. Blue Noise is a noise ‘color’ with a spectral density (power per hertz) that is proportional to its frequency. This means that the power and energy of the signal increases as frequency increases. It is not blue due to hue it is Blue due to frequency, after all the light spectrum is a range of frequencies from reds to blues..

The idea of BLUE noise is not the only such special form of noise, there are others such as Pink, Gray and each is just a different form of noise and none refer to colour in the visual effects sense.

Below is a random set of dots, but notice that there is natural clumping and that a random distribution is not the same as an evenly distribution.

Source paper

The paper that Marcos refers to in the podcast around 10 mins  (there is an audio drop out) is

Spectrally optimal sampling for distribution ray tracing: (Siggraph 1991) Don P. Mitchell.

(download a copy here: p157-mitchell)

In that 1991 paper he argued that nonuniform sampling of images is a useful technique in computer graphics, because a properly designed pattern of samples can vastly improve an image.

Other Researchers and sources mentioned:

Peter Shirley University of Utah – now Distinguished and Principal Research Scientist, NVIDIA Research

Pat Hanrahan Stanford University

Andrew Glassner, An Introduction to Ray Tracing,

The Pixar Research Papers Library online


 


 

3ds Max to Arnold

In addition to this new podcast, we interviewed Marcos in Sept. 2014 at Siggraph

fxguidetv #194: In-depth with Arnold creator Marcos Fajardo

Leave a Reply