
Examples overlapping case. We divide the overlapping region into four sub-regions equally first. CE1 and CE2 are the center positions of example 1 and 2, respectively. CSR1–CSR4 are the center positions of sub-regions. For the sub-region 1 in the upper left part, CSR1, the distance |CSR1−CE1| is smaller than |CSR1−CE2| . Therefore we consider example 1 to override sub-region 1. We use the same strategy to deal with the other sub-regions.
Figures of the Article
-
Pipeline of our method including two major parts: (a) precomputation and (b) real-time shading. The query details in a footprint are shown in (c).
-
Spatial transformations for a tiny example through different transformation matrices. (a) Example. (b) Transformed example through scaling. (c) Transformed example through shearing. (d) Transformed example through symmetry. (e) Transformed example through offsetting. (f) Transformed example through rotation.
-
Examples overlapping case. We divide the overlapping region into four sub-regions equally first. CE1 and CE2 are the center positions of example 1 and 2, respectively. CSR1–CSR4 are the center positions of sub-regions. For the sub-region 1 in the upper left part, CSR1, the distance |CSR1−CE1| is smaller than |CSR1−CE2|. Therefore we consider example 1 to override sub-region 1. We use the same strategy to deal with the other sub-regions.
-
Visualization of the multi-microstructure-layer case.
-
Rendering results of the multi-microstructure-layer case in the Bent quad scene. (a) Single-layer result with an anisotropic noise base (11.2 ms). (b) Multi-layer result with an anisotropic noise base and a scratch clearcoat (14.5 ms). (c) Single-layer result with a scratch base (11.8 ms). (d) Multi-layer result with a scratch base and an anisotropic noise clearcoat (14.5 ms).
-
Comparison between our method and the reference method[1] on the Desktop scene with various types of microstructures. (a) Rendering results of our method. (b) Examples. (c) Result details of ours (38.58 MB, 13.7 ms). (d) Result details of the reference method (
4134.4 MB, 91.8 ms). -
Comparison of our method with other microstructure rendering methods[1-4] on the coffee machine scene containing four types of microstructure. (a) Rendering results of our method. (b) Examples. (c) Result details of ours (113.2 MB, 14.5 ms). (d) Result details of Wang et al.[3] (592.0 MB, 142.7 ms). (e) Result details of Zhu et al.[2] (
1464.5 MB, 142.8 ms). (f) Result details of Tan et al.[4] (149.2 MB, 16.3 ms). (g) Result details of the reference method[1] (17364.4 MB, 142.8 ms). -
Comparison of rendering results using our method and other microstructure rendering methods[1-4] on three different scenes. We compare the glint effect generated by anisotropic noise microstructures using different methods in sphere scene, the continuous highlight of leather type microstructures in the shoes scene, and the scratch effect generated in the bent quad scene. (a) Ours. (b) Wang et al.[3] (c) Zhu et al.[2] (d) Tan et al.[4] (e) Reference method.[1]
-
Comparison between our method and the reference method[1] on the bent quad scene under the same scratch microstructure. (a) Rendering results of our method. (b) Result details of ours (85.25 MB, 11.8 ms). (c) Result details of the reference method (
3307.5 MB, 90.9 ms). (d) Difference visualization between ours and the reference method. -
Comparison between our results and the reference result under the GGX statistical distribution[30] on the deer statue scene. (a) Reference result (9.3 ms). (b) Our result with structured materials (13.3 ms). (c) Our result with scratches (13.9 ms). (d) Our result with isotropic noise (12.0 ms).
-
Glinty results of brushed metal microstructure exhibiting different visual effects in the deer statue scene. (a) Tiling result. (b) Result with a rotation transformation matrix. (c) Result with a scaling transformation matrix.
-
Shiny glints of anisotropic noise microstructure with different spatial distribution noise maps (shown in the upper right corner) in the deer statue scene. (a)–(c) Results with different example spatial distributions determined by the three different noise maps separately.
-
Comparison of different example resolutions on the deer statue scene. (a) Result with an 8×8 tiny example (1.33 MB, 13.8 ms). (b) Result with a 32×32 tiny example (21.31 MB, 13.2 ms). (c) Details of our results.
-
Limitation of our method. (a) Result of the wooden floor. (b) High-resolution normal map.
Others
-
External link to attachment
https://rdcu.be/dUUyW -
PDF format
2024-4-3-4123-Highlights 150KB -
DOCX format
Chinese Information 28KB