Adaptive Sampling: Difference between revisions
More actions
Standingpad (talk | contribs) Add examples with heatmap |
Standingpad (talk | contribs) No edit summary |
||
| Line 4: | Line 4: | ||
<div><ul style="display: table; margin: 0 auto;"> | <div><ul style="display: table; margin: 0 auto;"> | ||
<li style="display: inline-block; vertical-align: top;"> [[File:Adaptive Sampling Example - Noisy Image.jpg|thumb|none|Noisy version of a render using adaptive sampling|alt=Noisy image of a forest scene]] </li> | <li style="display: inline-block; vertical-align: top;"> [[File:Adaptive Sampling Example - Noisy Image.jpg|thumb|none|400px|Noisy version of a render using adaptive sampling|alt=Noisy image of a forest scene]] </li> | ||
<li style="display: inline-block; vertical-align: top;"> [[File:Adaptive Sampling Example - Heatmap Image.jpg|thumb|none|Heatmap demonstrating how many samples were needed to reach the noise threshold, where darker areas represent portions that finished earlier|alt=Heatmap of samples for image of a forest scene]] </li> | <li style="display: inline-block; vertical-align: top;"> [[File:Adaptive Sampling Example - Heatmap Image.jpg|thumb|none|400px|Heatmap demonstrating how many samples were needed to reach the noise threshold, where darker areas represent portions that finished earlier|alt=Heatmap of samples for image of a forest scene]] </li> | ||
</ul></div> | </ul></div> | ||
Adaptive sampling works best with a high amount of fixed samples and a noise threshold that isn't low enough to not cause sampling to stop. While in theory the noise threshold could lead to more noise, when combined with [[Denoiser|denoising algorithms]], the final image quality remains mostly equal.<ref>https://tomclabault.github.io/blog/2024/per-pixel-adaptive-sampling/</ref> | Adaptive sampling works best with a high amount of fixed samples and a noise threshold that isn't low enough to not cause sampling to stop. While in theory the noise threshold could lead to more noise, when combined with [[Denoiser|denoising algorithms]], the final image quality remains mostly equal.<ref>https://tomclabault.github.io/blog/2024/per-pixel-adaptive-sampling/</ref> | ||
Revision as of 07:04, 6 July 2025
Adaptive sampling is an optimization technique for ray tracing in which renderers measure noise in the scene against a pre-defined threshold during the rendering phrase, and when met for a given area of the image, stops sampling in that area. This is based on the observation that in a given scene, certain portions of the image (such as those with flat, well-lit objects) may converge faster than other portions (such as those with more complex geometry, materials, and/or lighting).[1]
-
Noisy version of a render using adaptive sampling -
Heatmap demonstrating how many samples were needed to reach the noise threshold, where darker areas represent portions that finished earlier
Adaptive sampling works best with a high amount of fixed samples and a noise threshold that isn't low enough to not cause sampling to stop. While in theory the noise threshold could lead to more noise, when combined with denoising algorithms, the final image quality remains mostly equal.[2]