Template - Optimisation
There is a high time cost increase (essentially 4x) when rendering 4 view images in comparison to a standard a 2D image. Aside from general best practices (such as material properties and reducing bounce or indirect lighting) a specific shortcut can be taken to help mitigate the time cost related to rendering 4 view images.
Since the final image size of each view is 1/4 that of the total image, we can make concessions with the quality of each rendered view (ideally we would render the images at 1/4 the resolution, but are unable with the current system).
There is significant time cost reduction when decreasing the pixel render samples on each camera view.
Any increase noise from lowered samples can be removed by using the Intel AI de-noiser node on each view.
These changes coupled with the already existing scaling compression yield great results.
Samples per pixel
128
32
16
Render Time
252
95
56
Improvement
-
2.5x
4.5x
16 Samples
128 Samples
On first impression the two (16 and 128 sample) images are identical, but on closer inspection there is a slight loss of fidelity with finer details.
NOTE: Pay close attention to the edges of objects.
The final output was rendered at 32 samples per pixel. This seems like a reasonable balance to preserve details, and the result is indistinguishable from a higher pixel sample (128) rate.
Last modified 1yr ago
Copy link