Academia.eduAcademia.edu

Figure 1 – uploaded by Joaquim Jorge

See full PDFdownloadDownload figure
A variety of models from the Stanford 3D Scanning Repository were used for the evaluation. The methods implemented in this work were done in C++ using the STL and Boost. Assembly code or intrinsics were not used. All timings were done on a computer with a single two-core 3GHz Intel Core 2 Duo CPU with 4GB of memory. Only a single thread was employed. All images were ren- dered at a resolution of 1024 x 1024 with one ray per pixel and diffuse shading.

Table 1 A variety of models from the Stanford 3D Scanning Repository were used for the evaluation. The methods implemented in this work were done in C++ using the STL and Boost. Assembly code or intrinsics were not used. All timings were done on a computer with a single two-core 3GHz Intel Core 2 Duo CPU with 4GB of memory. Only a single thread was employed. All images were ren- dered at a resolution of 1024 x 1024 with one ray per pixel and diffuse shading.

Related Figures (4)

Figure 1: In the middle, is the 618 x 342 x 414 level 1 grid for the Asian Dragon scene at the left. At right, the 103 x 57 x 69 level 2 grid. Vasco Costa*, Joao Pereira, Joaquim Jorge? INESC-ID
Figure 1: In the middle, is the 618 x 342 x 414 level 1 grid for the Asian Dragon scene at the left. At right, the 103 x 57 x 69 level 2 grid. Vasco Costa*, Joao Pereira, Joaquim Jorge? INESC-ID
We empirically chose the grid density parameter by analyzing the behavior for the Buddha scene as can be seen in Figure 2. We se- lected a grid density of p of 32 since it features adequate render time without having a severe impact on time to image.
We empirically chose the grid density parameter by analyzing the behavior for the Buddha scene as can be seen in Figure 2. We se- lected a grid density of p of 32 since it features adequate render time without having a severe impact on time to image.
The build time for hashed grids is short and roughly linear with the number of triangles as can be seen in Figure 3. This gives a short time to image useful for dynamic scenes. Classic multi-level array grids behave poorly regarding build time, getting even worse for the more complex scenes. Figure 3: Jn the top chart, acceleration structure build time statis- tics can be seen. At bottom, the chart has render time statistics for the tested scenes. Timings are the average of several test runs.
The build time for hashed grids is short and roughly linear with the number of triangles as can be seen in Figure 3. This gives a short time to image useful for dynamic scenes. Classic multi-level array grids behave poorly regarding build time, getting even worse for the more complex scenes. Figure 3: Jn the top chart, acceleration structure build time statis- tics can be seen. At bottom, the chart has render time statistics for the tested scenes. Timings are the average of several test runs.
Figure 4: Acceleration structure memory usage statistics for the tested scenes.
Figure 4: Acceleration structure memory usage statistics for the tested scenes.
Related topics:
Computer ScienceComputer GraphicsAnimationRay Tracing
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved