Open Access

表5

与其他基于FPGA平台CNN加速器对比

加速器 [10] [11] [12] [13] [14] 本文
FPGA Virtex7
VC709
Aria 10
GX 1150
Aria 10
GT 1150
Zynq
XC7Z030
Virtex-7
485T
Ultra96-V2 Ultra96-V2 ZCU102
频率/MHz 200 240 232 150 150 150 150 200
CNNs VGG-16 ResNet-50 VGG-16 VGG-16 VGGNet-D VGG-16 ResNet-50 ResNet-50
检测精度 - - - 67.7% - 71.6% 75.2% 75.2%
DSP 2877(80%) 3036(100%) 3000(99%) 400(100%) 2800(100%) 343(95%) 336(93%) 2315(92%)
BRAMs 1765(60%) 2356(87%) 1668(61%) 203(77%) 7275(89%) 54.5(40%) 204(97%) 743(82%)
计算性能 1713.0 599.6 1171.3 105.2 809.0 216.6 214.0 931.8
计算效率 0.59 0.20 0.39 0.26 0.29 0.63 0.64 0.40

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.