0
$\begingroup$

Let's say I designed a housing model and rendered it using a rendering engine. how can I measure the performance of my rendering?

$\endgroup$
  • $\begingroup$ Can you narrow down what you want to know about the rendering performance? There can be many metrics and they too will depend on so many factors in the scene (e.g. complexity of shading models) $\endgroup$ – Simon F Apr 1 '19 at 8:39
0
$\begingroup$

I assume you mean performance of the numerical algorithm responsible for generating the image?

Usually the performance of a rendering algorithm is assessed through a pixel-wise comparison with a ground truth, converged image. This image is typically rendered for a very long time using a standard and (ideally) unbiased technique like unidirectional path tracing. For a fixed time and sampling budget, you then run your custom algorithm and track a meaningful error (e.g. relative MSE) between your intermediate image and the ground truth image. Plotting the error over time on a logarithmic scale should give you a sense of how your algorithm is performing.

When comparing algorithms, you need to make sure that you are actually comparing images that were output by the same rendering engine. This is crucial if you want to report a faithful metric, otherwise the comparison is unfair and inherently depends on the implementation. For instance, if rendering engine $A$ takes twice as long as engine $B$ to fetch and load a texture, your reported error will be higher, even though the main algorithm you are testing is the same (assuming you are treating materials as blackboxes). The main downside is that rendering researchers are typically required to implement other algorithms in their engine to report the true performance of their new algorithm.

| improve this answer | |
$\endgroup$

Not the answer you're looking for? Browse other questions tagged or ask your own question.