Cornell vs. Blender

Here are two renderings of the famous Cornell Box. The top image was rendered by the software (and possibly some hardware) at the Cornell University Program of Computer Graphics. There seems to be a small error in this image. Take a look at the bottom left corner of the small box and you will see a light leak, which gives that corner the illusion that it is floating.

The bottom image was rendered with Blender 1.80. As you can see, Blender's radiosity solution is very similar. There is a slight shadow leak on the bottom left edge of the tall box, but there is no light leak on the small box. I obtained the original Cornell Box Data and tried to guess the rgb values from the reflectance spectra. Anyone know the right way to do this? If the rgb color values were more accurate, the images would be even more identical.

This image was solved in about 30 minutes and rendered in 3 seconds on a PII 450MHz. Please note that I let the radiosity calculation run until it was finished. In case you want to render your own tests, you can download the cornell-box.blend scene file 21KB. The following settings were used:

PaMax: 250

PaMin: 100

Hemires: 1,000

ElMax: 50

ElMin: 10

MaxEl: 20,000

Radiosity hints for Blender 1.8 users:

  • Quality and rendering time increase as the Patch and Element variables approach their minimum values, and as the Hemires and Maximum Elements variables approach their maximum values.
  • You can divide the 'Radiosity solving time' by 6,000 to get the time in minutes.

This site was last revised on January 05, 2000
Copyright © 2000 Anthony D'Agostino
All rights reserved.