CS 551/651: Advanced Computer Graphics

Goal: Experiment with different sampling strategies to improve ray tracing quality

Assigned: Thursday, February 25, 1999

Due: Tuesday, March 9, 1999

Relevant reading/files/software:

• See Watt and Watt, Chapter 5
• The class web page will note any issues that arise.
• Some test scenes will also be posted on the course web page.
• You will need your modified version of RSRT.

Synopsis: You will continue modifying the RSRT source code, this time to implement various sampling strategies. Your RSRT should use command-line arguments to choose:

• Point sampling. Exactly one ray is shot through the center of each pixel.
• 3x3 supersampling, box filter. 9 rays are shot per pixel and the results combined as an unweighted average.
• 3x3 supersampling, Barlett filter. As above, but using a Barlett filter for more accurate reconstruction
• 3x3 jittered supersampling, Barlett filter. 9 rays, jittered from a regular grid, but filtered using a Barlett filter as if they were still on a regular grid.

Possible extra features include adaptive supersampling, more sophisticated filtering, and comparing image quality obtained using actual Poisson and Poisson-disc distributions with that obtained using jittering.

In addition to turning in your modified RSRT code, you will turn in two example scenes that illustrate the relative merits of the various sampling strategies. Extra credit may also be given for particularly interesting scenes, though correctness in the ray tracer will be valued over creativity in scene design.

You should include a README file that documents what sampling strategies you implemented, how well they worked, and anything you learned in the process. Also describe what command-line options are necessary to run your version of RSRT with different sampling modes, and how you created your example scenes.

Grading: The usual 10-point scale. An assignment that does everything required will receive 8 points. An assignment implementing "extras" such as Poisson-disc sampling, or with particularly interesting example scenes, can get up to 9 points. The single best assignment will get 10 points.

Turning in the assignment: Before class starts on Tuesday, March 9, you should:

• Tar up a directory containing your code, your README file, and the .ppm files of your two example scenes. Gzip the tar file.
• Send Dale an e-mail message containing the gzipped tar file as an attachment.
• Dale will print out your examples, compile your ray tracer, and test the result against various sample scenes

• RSRT currently uses an extremely crude antialiasing method, which traces two scanlines' worth of rays and averages them together to get pixels comprised of four adjacent rays in the two scanlines (What are some of the problems with this approach?). You should begin by throwing out this primitive averaging hack, embodied in the function pixel_average_two_scanlines()in main.c. Instead, keep track of only a single scanline or even a single pixel at a time, and shoot enough rays per pixel to fully determine the filtered value of that pixel. Begin by writing the code that will simply shoot one ray per pixel (i.e., point-sampling). Then change the call to trace a single pixel ray to call a function that finds the value for a pixel. Depending on the command line options, this function may shoot multiple rays and use various filtering strategies to calculate the final pixel value.
• Start earlier this time! Too many people underestimated how long the first assignment would take, or the difficulty of working within an existing code base. You do not have quite as long to do this assignment, but it should be easier because (1) you know the RSRT code better now, and (2) fewer underlying changes are necessary. Most or all of the code you will need to change is in main.c, though you may want to put filtering functions in another file for convenience.