TY - GEN
T1 - Performance Analysis of PointPillars on CPU and GPU Platforms
AU - Choi, Yuho
AU - Kim, Byungguk
AU - Kim, Seon Wook
N1 - Funding Information:
Hyundai Motorgroups supported this work (Grants No. Q2028321).
Publisher Copyright:
© 2021 IEEE.
PY - 2021/6/27
Y1 - 2021/6/27
N2 - Recently, the safety of ADAS (Autonomous Driving Assistance System) has become a critical issue in industrial areas. Modern intelligent automobile control systems adopt various sensors for recognizing environmental obstacles, and LiDAR (Light Detection and Ranging) sensor has been selected due to its consistently accurate sensing abilities regardless of external conditions. Also, many neural network models have been proposed to process LiDAR data. Short inference latency and high accuracy matter for ADAS. From this perspective, the encoder of Pointpillars shortens inference latency by converting the 3D LiDAR point cloud into a 2D pseudo-image for object detection.In this paper, we study the architecture of PointPillars and analyze its performance by VTune [15] on CPU and NVIDIA profiler on GPU [16]. It has been observed that RPN (Region Proposal Network) shows the most dominant execution time to the overall model due to its internal convolution and transposed convolution operation. Therefore, the adoption of PointPillars to real ADAS requires RPN network optimization.
AB - Recently, the safety of ADAS (Autonomous Driving Assistance System) has become a critical issue in industrial areas. Modern intelligent automobile control systems adopt various sensors for recognizing environmental obstacles, and LiDAR (Light Detection and Ranging) sensor has been selected due to its consistently accurate sensing abilities regardless of external conditions. Also, many neural network models have been proposed to process LiDAR data. Short inference latency and high accuracy matter for ADAS. From this perspective, the encoder of Pointpillars shortens inference latency by converting the 3D LiDAR point cloud into a 2D pseudo-image for object detection.In this paper, we study the architecture of PointPillars and analyze its performance by VTune [15] on CPU and NVIDIA profiler on GPU [16]. It has been observed that RPN (Region Proposal Network) shows the most dominant execution time to the overall model due to its internal convolution and transposed convolution operation. Therefore, the adoption of PointPillars to real ADAS requires RPN network optimization.
KW - LiDAR
KW - Performance analysis
KW - PointPillars
KW - convolution
UR - http://www.scopus.com/inward/record.url?scp=85121220152&partnerID=8YFLogxK
U2 - 10.1109/ITC-CSCC52171.2021.9611297
DO - 10.1109/ITC-CSCC52171.2021.9611297
M3 - Conference contribution
AN - SCOPUS:85121220152
T3 - 2021 36th International Technical Conference on Circuits/Systems, Computers and Communications, ITC-CSCC 2021
BT - 2021 36th International Technical Conference on Circuits/Systems, Computers and Communications, ITC-CSCC 2021
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 36th International Technical Conference on Circuits/Systems, Computers and Communications, ITC-CSCC 2021
Y2 - 27 June 2021 through 30 June 2021
ER -