|
|
The final project was to take a photograph of a scene and photorealistically render it with our pathtracer. The only caveat was that we could not "cheat" with the colors by tweaking them until they looked right.
My scene was taken in a box that I constructed made out of white cardboard (bottom and left and right sides) and white paperboard (front and back). I took the picture with my digital camera on a overcast day in the parking lot of my apartment complex. I placed my camera on a tripod so I could get a rough estimate of where the CCD (eye point) was located.
The acrylic ball I used I bought from a juggling store, and was advertised as a 3-inch diameter ball (though I did not actually measure it). The cards were from a new pack of playing card, the book was the least mangled book I could find on my bookshelf, and I used two PVC pipe bits (one of which is holding up the ace of spaces and the overturned card).
I marked the locations of the objects on the cardboard before taking the picture so I could recreate the scene for my renderer. The book and cards are textured with images I scanned from the real items. I had to adjust the scanned images in saturation and contrast (since the scanner did a poor job).
I "measured" the index of refraction of the mirror and the acrylic ball by rendering them with different indices of refraction until they looked roughly the same as my photograph. I am not entirely pleased by the results, but I think the problem is that I could not measure the eye position and look at points accurately.
My rendered light is a huge polygon which is about 20 inches above the box and stretches feet in each direction. The emission color was determined the following way: I used a value (0.1 in all spectral color bins) in a scene to test the positioning of the objects. It looked good, except a bit dark so I increased the values to 0.11, which looked too bright, so I settled on a value in the middle of 0.105 for all spectral bins of my light.
The cardboard in the box is rendered with the same lambertian white material (and color) used in the Cornell Box.
Rendering the image with 10648 (223) samples per pixel took roughly 14 hours on sixteen 400 MHz R12000 processors.
Here is a side by side view. Click on the image you think is the
computer generated image:
|
|
|
Last Modified: Sunday, December 16, 2001
Chris Wyman (wyman@cs.utah.edu)