The goal of this extension is to compare the different samplers. We proceed as follows:
-
We choose a scene and render this in high quality. This becomes our reference image, our "holy grail" that we try to approximate as well as possible.
-
Next, we render the same scene using the different samplers using different parameters. We store both the result and the time it took to render.
-
Next, we determine the "quality" of each image by measuring the "distance" between a sampler-image and the reference image.
-
Finally, we compare the quality with the rendering time necessary to achieve it. For example, the single sampler could be fast, but low quality, whereas nrooks could be slow but good.
1. Choosing a Scene
First, we need a scene to render. Choose whatever scene you want.
|
Create a scene that will be used to compare the rendering results of different samplers. |
2. Reference Image
How do we create our reference image? Let’s simply render the scene using a sampler with many samples. This will take a long time rendering, but that does not matter.
|
Render the scene using a sampler parameterized to use many samples (at least 25 per pixel). Save the result as a WIF file. |
3. Samplers
Next, we render approximations of this "ideal image".
|
Render the same scene using each sampler, using multiple different parameters. For example,
Store all results. |
4. Quality Metric
Now we need to be able to compute how much each of the approximations differs from the reference image. There is no single best way of doing this.
-
You could count how many pixels differ.
-
Probably better would be to measure the distance between corresponding pixels. You can see the R, G, B values as (x, y, z) coordinates and use Euclidian distance
\[\sqrt{(r_1-r_2)^2+(g_1-g_2)^2+(b_1-b_2)^2}\]
|
Write a script that, given two WIF files, can compute the "distance" between them. Use any language you want.
|
5. Comparing
Now compare every approximating image with the reference image. The smaller the distance, the better an approximation it is.
|
Use your script to measure the distance between each sampler produced image and the reference image. |
6. Drawing Conclusions
In a last step, we want to compare rendering times to quality.
|
Example
|
Say you have the following data:
In this case, |
|
Make a graph (e.g., using Excel) that visualizes the relation between rendering times and quality for each sampler. See if you can identify a sampler that clearly performs better than others (lower rendering times for same level of quality). |