Supplemental - Glossy Probe Reprojection for Interactive Global Illumination SIMON RODRIGUEZ, Université Côte d’Azur and Inria THOMAS LEIMKÜHLER, Université Côte d’Azur and Inria SIDDHANT PRAKASH, Université Côte d’Azur and Inria CHRIS WYMAN, NVIDIA Corporation PETER SHIRLEY, NVIDIA Corporation GEORGE DRETTAKIS, Université Côte d’Azur and Inria ACM Reference Format: Simon Rodriguez, Thomas Leimkühler, Siddhant Prakash, Chris Wyman, Peter Shirley, and George Drettakis. 2020. Supplemental - Glossy Probe Reprojection for Interactive Global Illumination. ACM Trans. Graph. 39, 6 (December 2020), 6 pages. https://doi.org/10.1145/3414685.3417823 In this supplemental document, we present error maps corresponding to fig. 16, 17 and 18 in the article, along with a qualitative comparison of different probe counts and resolution when rendering the Bathroom scene. 1 ERROR MAPS In Fig. 1,2,3,4, we evaluate the accuracy of our method on the images presented in the main article. We display the per-channel squared difference between the offline path-traced ground truth generated with Mitsuba and our method. Values are scaled by 10 to improve visibility. We provide similar error maps for the real-time ray-tracing with lightmap baseline, as it is the one that had the closest error scores and is available on all our test scenes. Our method is able to more accurately render glossy reflections thanks to the fully precomputed lighting, where the baseline has difficulties sampling the area emitters properly. On the other hand our method exhibit errors on reflector edges, partially caused by specular aliasing. 2 VARYING PROBE COUNT AND RESOLUTION We processed the Bathroom scene for 125, 252 and 504 probes, at resolutions 768x384, 1024x512 and 1440x720, doubling the probe texel count each time (see Fig. 5). Increasing the probe resolution improves details around object edges in reflections, for instance the reflection of the furniture in the wall mirror. At low probe count, coverage of glossy effects can become incomplete, as can be observed on the top of the bin for 125 probes. The corresponding ground truth rendering is visible in Fig. 7 and error maps in Fig. 6. In this case 252 probes is probably sufficient; minor quality degradation with 504 probes (e.g., side panel highlight) is probably because we do not test for intersections of the probes with geometry, which happens more often at higher probe counts. Authors’ addresses: Simon Rodriguez, Université Côte d’Azur and Inria, [email protected]; Thomas Leimkühler, Université Côte d’Azur and Inria, [email protected]; Siddhant Prakash, Université Côte d’Azur and Inria, siddhant. [email protected]; Chris Wyman, NVIDIA Corporation, [email protected]; Peter Shirley, NVIDIA Corporation, pshirley@ nvidia.com; George Drettakis, Université Côte d’Azur and Inria, [email protected]. © 2020 Copyright held by the owner/author(s). Publication rights licensed to ACM. This is the author’s version of the work. It is posted here for your personal use. Not for redistribution. The definitive Version of Record was published in ACM Transactions on Graphics, https://doi.org/10.1145/3414685.3417823. ACM Trans. Graph., Vol. 39, No. 6, Article . Publication date: December 2020.
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Supplemental - Glossy Probe Reprojection for InteractiveGlobal Illumination
SIMON RODRIGUEZ, Université Côte d’Azur and InriaTHOMAS LEIMKÜHLER, Université Côte d’Azur and InriaSIDDHANT PRAKASH, Université Côte d’Azur and InriaCHRIS WYMAN, NVIDIA CorporationPETER SHIRLEY, NVIDIA CorporationGEORGE DRETTAKIS, Université Côte d’Azur and Inria
ACM Reference Format:Simon Rodriguez, Thomas Leimkühler, Siddhant Prakash, Chris Wyman, Peter Shirley, and George Drettakis.2020. Supplemental - Glossy Probe Reprojection for Interactive Global Illumination. ACM Trans. Graph. 39, 6(December 2020), 6 pages. https://doi.org/10.1145/3414685.3417823
In this supplemental document, we present error maps corresponding to fig. 16, 17 and 18 in thearticle, along with a qualitative comparison of different probe counts and resolution when renderingthe Bathroom scene.
1 ERROR MAPSIn Fig. 1,2,3,4, we evaluate the accuracy of our method on the images presented in the mainarticle. We display the per-channel squared difference between the offline path-traced groundtruth generated with Mitsuba and our method. Values are scaled by 10 to improve visibility. Weprovide similar error maps for the real-time ray-tracing with lightmap baseline, as it is the onethat had the closest error scores and is available on all our test scenes. Our method is able to moreaccurately render glossy reflections thanks to the fully precomputed lighting, where the baselinehas difficulties sampling the area emitters properly. On the other hand our method exhibit errorson reflector edges, partially caused by specular aliasing.
2 VARYING PROBE COUNT AND RESOLUTIONWe processed the Bathroom scene for 125, 252 and 504 probes, at resolutions 768x384, 1024x512and 1440x720, doubling the probe texel count each time (see Fig. 5). Increasing the probe resolutionimproves details around object edges in reflections, for instance the reflection of the furniture inthe wall mirror. At low probe count, coverage of glossy effects can become incomplete, as canbe observed on the top of the bin for 125 probes. The corresponding ground truth rendering isvisible in Fig. 7 and error maps in Fig. 6. In this case 252 probes is probably sufficient; minor qualitydegradation with 504 probes (e.g., side panel highlight) is probably because we do not test forintersections of the probes with geometry, which happens more often at higher probe counts.
Authors’ addresses: Simon Rodriguez, Université Côte d’Azur and Inria, [email protected]; Thomas Leimkühler,Université Côte d’Azur and Inria, [email protected]; Siddhant Prakash, Université Côte d’Azur and Inria, [email protected]; ChrisWyman, NVIDIA Corporation, [email protected]; Peter Shirley, NVIDIA Corporation, [email protected]; George Drettakis, Université Côte d’Azur and Inria, [email protected].
© 2020 Copyright held by the owner/author(s). Publication rights licensed to ACM.This is the author’s version of the work. It is posted here for your personal use. Not for redistribution. The definitive Versionof Record was published in ACM Transactions on Graphics, https://doi.org/10.1145/3414685.3417823.
ACM Trans. Graph., Vol. 39, No. 6, Article . Publication date: December 2020.
2 Simon Rodriguez, Thomas Leimkühler, Siddhant Prakash, Chris Wyman, Peter Shirley, and George Drettakis
Ground truth image RTRT+lightmap error Our method error
Fig. 1. Per-pixel square error on Bathroom, scaled 10x, for the RTRT + lightmap baseline and our method.
Ground truth image RTRT+lightmap error Our method error
Fig. 2. Per-pixel square error on Kitchen, scaled 10x, for the RTRT + lightmap baseline and our method.
ACM Trans. Graph., Vol. 39, No. 6, Article . Publication date: December 2020.
Supplemental - Glossy Probe Reprojection for Interactive Global Illumination 3
Ground truth image RTRT+lightmap error Our method error
Fig. 3. Per-pixel square error on Livingroom, scaled 10x, for the RTRT + lightmap baseline and our method.
Ground truth image RTRT+lightmap error Our method error
Fig. 4. Per-pixel square error on Staircase, scaled 10x, for the RTRT + lightmap baseline and our method.
ACM Trans. Graph., Vol. 39, No. 6, Article . Publication date: December 2020.
4 Simon Rodriguez, Thomas Leimkühler, Siddhant Prakash, Chris Wyman, Peter Shirley, and George Drettakis
768x384px 1024x512px 1440x720px
125prob
es252prob
es504prob
es
Fig. 5. Rendering Bathroom using different probe count and resolution.
ACM Trans. Graph., Vol. 39, No. 6, Article . Publication date: December 2020.
Supplemental - Glossy Probe Reprojection for Interactive Global Illumination 5
768x384px 1024x512px 1440x720px
125prob
es252prob
es504prob
es
Fig. 6. Corresponding error maps (per-channel squared error, scaled 10x for visibility).
ACM Trans. Graph., Vol. 39, No. 6, Article . Publication date: December 2020.
6 Simon Rodriguez, Thomas Leimkühler, Siddhant Prakash, Chris Wyman, Peter Shirley, and George Drettakis
Fig. 7. Ground truth rendering of the same viewpoint.
ACM Trans. Graph., Vol. 39, No. 6, Article . Publication date: December 2020.
Related Documents
