Assignment 1
Fabian Prada
This assignament explore some features of scene description and rendering options provided by PBRT
1)Resolution
|
|
|
|
|
Resolution |
400x800 |
300x600 |
200x400 |
100x200 |
Rendering Time |
53seg |
29.9seg |
13seg |
3.4seg |
The computational time expended on rendering is linear on the number of pixels of the image. Therefore, duplicating the sides of an image
increases 4x the running time. This is because the number of rays traced (samples) is constant per pixel.
2)Samplers
Sampler |
Result |
Rendering Time |
Random |
|
3.7 seg |
Stratified |
|
3.9 seg |
Halton |
|
4.0 seg |
Low Discrepancy |
|
4.0 seg |
Best Candidate |
|
4.1 seg |
Rendering time was almost independent of the sampler used. This shows that the computational cost of generating samples is just a little portion of the total rendering process. Thereofere, prefering a a higher quality sampling (for instance, LD instead of stratified) can be also justified by the little difference in computational costs. The previous results where generated using 4 samples per pixels and were filtered with Box. Random, Stratified and Halton still have a lot of alias in the back part of the plane, and in the ecuator of the closest sphere. BC do a better work, but is not as good as LD.
When we increase the number of samples per pixel the rendering time increases proportionally. For instance if we take 16 samples per pixel the rendering time is about 15segs. Again, this evidence the linear relation between the computational time and the number of rays traced.
3)Filters
Filter |
Result |
Rendering Time |
Box |
|
2.1 seg |
Triangle |
|
2.3 seg |
Mitchell |
|
2.4 seg |
Gaussian |
|
2.4 seg |
Sinc |
|
3.1 seg |
The previous results were generated using 2 samples per pixel. Rendering time of Box was shorter than Triangle, Mitchell and Gaussian, and these ones were faster than Sinc. Because of the reduced number of samples almost all reconstruction evidence alias in the back part of the plane and in the closest sphere ecuator. Box filtering shows some jagging for the widthest lines, and Triangle filtering is excesivelly blurred.The results of Mitchell, Gaussian and Sinc where slightly better.
4)Camera Coordinate System
Reference |
Change of position |
Change of direction vector |
|
|
|
Change of up vector |
Change of fov |
|
|
5)Object Coordinate System
Reference |
Transation |
Scaling |
Rotation |
|
|
|
|
6)Illumination
Illumination |
Result |
Distant |
|
Point |
|
Infinite |
|
Spot |
|
Area Light: Disk |
|
7)Animation
Reference |
|
Changing Velocities |
|
Changing Start and End Times |
|
Number of samples per pixel |
32 |
512 |
Result |
|
|