性能数据

可以参考benchmark_tools,推荐一键benchmark

测试环境

  • 测试模型
    • fp32模型
      • mobilenet_v1
      • mobilenet_v2
      • squeezenet_v1.1
      • mnasnet
      • shufflenet_v2
    • int8模型
      • mobilenet_v1
      • mobilenet_v2
  • 测试机器(android ndk ndk-r17c)
    • 骁龙855
      • xiaomi mi9, snapdragon 855
      • 4xA76(1@2.84GHz + 3@2.4GHz) + 4xA55@1.78GHz
    • 骁龙845
      • xiaomi mi8, 845
      • 2.8GHz(大四核),1.7GHz(小四核)
    • 骁龙835
      • xiaomi mix2, snapdragon 835
      • 2.45GHz(大四核),1.9GHz(小四核)
    • 麒麟970
      • HUAWEI Mate10
  • 测试说明
    • branch: release/v2.3.0
    • warmup=10, repeats=30,统计平均时间,单位是ms
    • 当线程数为1时,DeviceInfo::Global().SetRunMode设置LITE_POWER_HIGH,否者设置LITE_POWER_NO_BIND
    • 模型的输入图像的维度是{1, 3, 224, 224},输入图像的每一位数值是1

测试数据

fp32模型测试数据

paddlepaddle model

骁龙855armv7armv7armv7armv8armv8armv8
threads num124124
mobilenet_v133.2719.5211.1431.7218.7610.24
mobilenet_v229.0815.799.2525.8914.178.38
shufflenet_v24.403.092.304.283.022.35
squeezenet_v1.119.9612.618.7618.2511.467.97
mnasnet21.0012.547.2819.6511.656.96
骁龙845armv7armv7armv7armv8armv8armv8
threads num124124
mobilenet_v166.3635.9719.4562.6633.8717.85
mobilenet_v245.8625.5314.641.5823.2413.39
shufflenet_v27.584.893.417.444.913.58
squeezenet_v1.137.1522.7413.5134.6921.2712.74
mnasnet40.0921.7311.9138.1921.0212.11
骁龙835armv7armv7armv7armv8armv8armv8
threads num124124
mobilenet_v196.9853.9232.2489.3148.0227.58
mobilenet_v267.7237.6623.8260.1034.3621.05
shufflenet_v210.726.624.6310.106.444.63
squeezenet_v1.153.8933.2820.7350.8332.3119.51
mnasnet59.5533.5320.3256.2131.5819.06

caffe model

骁龙855armv7armv7armv7armv8armv8armv8
threads num124124
mobilenet_v133.3619.4511.2631.6318.7410.31
mobilenet_v231.6319.2111.6128.3417.1410.16
shufflenet_v24.463.082.324.262.982.35
骁龙845armv7armv7armv7armv8armv8armv8
threads num124124
mobilenet_v166.3235.8319.5662.5233.7917.91
mobilenet_v258.4632.6918.5653.7229.8616.80
shufflenet_v27.654.823.467.554.973.62
骁龙835armv7armv7armv7armv8armv8armv8
threads num124124
mobilenet_v195.3854.0932.0395.0548.3327.54
mobilenet_v288.4648.9830.2379.2844.6427.10
shufflenet_v210.076.514.6110.316.504.66

int8量化模型测试数据

骁龙855armv7armv7armv7armv8armv8armv8
threads num124124
mobilenet_v136.8021.5811.1214.018.134.32
mobilenet_v228.7219.0812.4917.2411.557.82
骁龙835armv7armv7armv7armv8armv8armv8
threads num124124
mobilenet_v160.7632.2516.6656.5729.8415.24
mobilenet_v249.3831.1022.0747.5228.1819.24
麒麟970armv7armv7armv7armv8armv8armv8
threads num124124
mobilenet_v165.9534.3918.6860.8630.9816.31
mobilenet_v268.8739.3924.4365.5737.3120.87