CERN Accelerating science

Article
Report number	arXiv:2205.07690 ; FERMILAB-PUB-22-435-PPD
Title	Real-time semantic segmentation on FPGAs for autonomous vehicles with hls4ml
Author(s)	Ghielmetti, Nicolò (CERN ; Milan, Polytech.) ; Loncar, Vladimir (CERN ; Belgrade, Inst. Phys.) ; Pierini, Maurizio (CERN) ; Roed, Marcel (CERN ; Oxford U.) ; Summers, Sioni (CERN) ; Aarrestad, Thea (Zurich, ETH) ; Petersson, Christoffer (Zurich, ETH) ; Linander, Hampus (U. Gothenburg (main)) ; Ngadiuba, Jennifer (Fermilab) ; Lin, Kelvin (Washington U., Seattle) ; Harris, Philip (MIT)
Publication	2022-11-04
Imprint	2022-05-16
Number of pages	11
In:	Mach. Learn. Sci. Tech. 3 (2022) 045011
DOI	10.1088/2632-2153/ac9cb5
Subject category	Computing and Computers ; Detectors and Experimental Techniques
Abstract	In this paper, we investigate how field programmable gate arrays can serve as hardware accelerators for real-time semantic segmentation tasks relevant for autonomous driving. Considering compressed versions of the ENet convolutional neural network architecture, we demonstrate a fully-on-chip deployment with a latency of 4.9 ms per image, using less than 30% of the available resources on a Xilinx ZCU102 evaluation board. The latency is reduced to 3 ms per image when increasing the batch size to ten, corresponding to the use case where the autonomous vehicle receives inputs from multiple cameras simultaneously. We show, through aggressive filter reduction and heterogeneous quantization-aware training, and an optimized implementation of convolutional layers, that the power consumption and resource utilization can be significantly reduced while maintaining accuracy on the Cityscapes dataset.
Copyright/License	preprint: (License: arXiv nonexclusive-distrib 1.0) publication: © 2022-2025 The Author(s) (License: CC-BY-4.0)

Corresponding record in: Inspire