Real-time spatiotemporal division multiplexing electroholography for 1,200,000 object points using multiple-graphics processing unit cluster

Hiromi Sannomiya; Naoki Takada; Kohei Suzuki; Tomoya Sakaguchi; Hirotaka Nakayama; Minoru Oikawa; Yuichiro Mori; Takashi Kakue; Tomoyoshi Shimobaba; Tomoyoshi Ito

doi:doi:10.3788/COL202018.070901

Chinese Optics Letters, 2020, 18 (7): 070901, Published Online: Jun. 15, 2020

Real-time spatiotemporal division multiplexing electroholography for 1,200,000 object points using multiple-graphics processing unit cluster Download： 792次

Hiromi Sannomiya ¹Naoki Takada ^2,*Kohei Suzuki ¹Tomoya Sakaguchi ¹Hirotaka Nakayama ³Minoru Oikawa ²Yuichiro Mori ²Takashi Kakue ⁴Tomoyoshi Shimobaba ⁴Tomoyoshi Ito ⁴

Author Affiliations

¹ Graduate School of Integrated Arts and Sciences, Kochi University, Kochi 780-8520, Japan

² Research and Education Faculty, Kochi University, Kochi 780-8520, Japan

³ National Astronomical Observatory of Japan, Mitaka 181-8588, Japan

⁴ Graduate School of Engineering, Chiba University, Inage-ku 263-8522, Japan

Figures & Tables

Fig. 1. Spatiotemporal division multiplexing approach for suppressing the deterioration of a 3D holographic video reconstructed from a point-cloud model comprising a huge number of object points.

下载图片查看原文

Fig. 2. Spatiotemporal division multiplexing approach using moving image features.

下载图片查看原文

Fig. 3. Reconstructed 3D image from a 3D object “fountain” comprising 1,064,462 object points.

下载图片查看原文

Fig. 4. Multi-GPU cluster system with multiple GPUs connected by a gigabit Ethernet network and a single SLM.

下载图片查看原文

Fig. 5. Pipeline processing for the spatiotemporal electroholography system shown in Fig. 2.

下载图片查看原文

Fig. 6. Read data processing and CGH calculation on each CGH calculation node in the multi-GPU cluster system shown in Fig. 4. (a) Serial computing. (b) Parallel computing.

下载图片查看原文

Fig. 7. Comparison of the total display time for every 12 frames using serial computing shown in Fig. 6(a) with that using parallel computing shown in Fig. 6(b) when the number of object points is 1,200,000.

下载图片查看原文

Fig. 8. Display-time interval $T$ shown in Fig. 5 plotted versus the number of object points when using the spatiotemporal division multiplexing approach using moving image features implemented on the multi-GPU cluster system shown in Fig. 4.

下载图片查看原文

Fig. 9. Snapshot of a reconstructed 3D video (Video 1).

下载图片查看原文

Table1. Specifications of Each Node in the Multi-GPU Cluster System

CPU	Intel Core i7 7800X (clock speed: 3.5 GHz)
Main memory	DDR4-2666 16 GB
OS	Linux (CentOS 7.6 x86_64)
Software	NVIDIA CUDA 10.1 SDK, OpenGL, MPICH 3.2
GPU	NVIDIA GeForce GTX 1080 Ti

查看原文

Table2. Frame Rate of the Reconstructed 3D Video from the Original 3D Video “Fountain” Comprising 1,064,462 Object Points Against the Number of Space Divisions

Number of Space Divisions	Object Points	Frame Rate (fps)
No division	1,064,462	5.43
Two divisions	532,231	10.86
Four divisions	266,116	21.70
Six divisions	177,411	32.70

查看原文

Hiromi Sannomiya, Naoki Takada, Kohei Suzuki, Tomoya Sakaguchi, Hirotaka Nakayama, Minoru Oikawa, Yuichiro Mori, Takashi Kakue, Tomoyoshi Shimobaba, Tomoyoshi Ito. Real-time spatiotemporal division multiplexing electroholography for 1,200,000 object points using multiple-graphics processing unit cluster[J]. Chinese Optics Letters, 2020, 18(7): 070901.

Real-time spatiotemporal division multiplexing electroholography for 1,200,000 object points using multiple-graphics processing unit cluster Download： 792次

Fig. 1. Spatiotemporal division multiplexing approach for suppressing the deterioration of a 3D holographic video reconstructed from a point-cloud model comprising a huge number of object points.

Fig. 2. Spatiotemporal division multiplexing approach using moving image features.

Fig. 3. Reconstructed 3D image from a 3D object “fountain” comprising 1,064,462 object points.

Fig. 4. Multi-GPU cluster system with multiple GPUs connected by a gigabit Ethernet network and a single SLM.

Fig. 5. Pipeline processing for the spatiotemporal electroholography system shown in Fig. 2.

Fig. 6. Read data processing and CGH calculation on each CGH calculation node in the multi-GPU cluster system shown in Fig. 4. (a) Serial computing. (b) Parallel computing.

Fig. 7. Comparison of the total display time for every 12 frames using serial computing shown in Fig. 6(a) with that using parallel computing shown in Fig. 6(b) when the number of object points is 1,200,000.

Fig. 8. Display-time interval $T$ shown in Fig. 5 plotted versus the number of object points when using the spatiotemporal division multiplexing approach using moving image features implemented on the multi-GPU cluster system shown in Fig. 4.

Fig. 9. Snapshot of a reconstructed 3D video (Video 1).

Table1. Specifications of Each Node in the Multi-GPU Cluster System

Table2. Frame Rate of the Reconstructed 3D Video from the Original 3D Video “Fountain” Comprising 1,064,462 Object Points Against the Number of Space Divisions

关于本站 Cookie 的使用提示

全站搜索

Real-time spatiotemporal division multiplexing electroholography for 1,200,000 object points using multiple-graphics processing unit cluster Download： 792次

Fig. 1. Spatiotemporal division multiplexing approach for suppressing the deterioration of a 3D holographic video reconstructed from a point-cloud model comprising a huge number of object points.

Fig. 2. Spatiotemporal division multiplexing approach using moving image features.

Fig. 3. Reconstructed 3D image from a 3D object “fountain” comprising 1,064,462 object points.

Fig. 4. Multi-GPU cluster system with multiple GPUs connected by a gigabit Ethernet network and a single SLM.

Fig. 5. Pipeline processing for the spatiotemporal electroholography system shown in Fig. 2.

Fig. 6. Read data processing and CGH calculation on each CGH calculation node in the multi-GPU cluster system shown in Fig. 4. (a) Serial computing. (b) Parallel computing.

Fig. 7. Comparison of the total display time for every 12 frames using serial computing shown in Fig. 6(a) with that using parallel computing shown in Fig. 6(b) when the number of object points is 1,200,000.

Fig. 8. Display-time interval T shown in Fig. 5 plotted versus the number of object points when using the spatiotemporal division multiplexing approach using moving image features implemented on the multi-GPU cluster system shown in Fig. 4.

Fig. 9. Snapshot of a reconstructed 3D video (Video 1).

Table1. Specifications of Each Node in the Multi-GPU Cluster System

Table2. Frame Rate of the Reconstructed 3D Video from the Original 3D Video “Fountain” Comprising 1,064,462 Object Points Against the Number of Space Divisions

相关论文

相关资讯

关于本站 Cookie 的使用提示

全站搜索

Fig. 8. Display-time interval $T$ shown in Fig. 5 plotted versus the number of object points when using the spatiotemporal division multiplexing approach using moving image features implemented on the multi-GPU cluster system shown in Fig. 4.