By Toradex 胡珊逢

機(jī)器學(xué)習(xí)算法對(duì)算力要求較高，通常會(huì)采用 GPU ，或者專用的處理器如 NPU 進(jìn)行加速運(yùn)算。NXP 先后推出的兩款處理器iMX8QuadMax  和 iMX8M Plus 分別可以采用 GPU 和 NPU 對(duì)常用的機(jī)器學(xué)習(xí)算法例如 TensorFlow Lite 等進(jìn)行加速。文章將使用 NXP eIQ 框架在兩個(gè)處理器上測(cè)試不同算法的性能。

這里我們將使用 Toradex 的 Apalis iMX8QM 4GB WB IT V1.1C 和 Verdin iMX8M Plus Quad 4GB WB IT V1.0B 兩個(gè)模塊。BSP 為 Linux BSP V5.3 。eIQ 采用 zeus-5.4.70-2.3.3 版本。Toradex 默認(rèn) Yocto Project 編譯環(huán)境并沒(méi)有直接集成  eIQ 軟件，可以參考這里添加 meta-ml layer 并進(jìn)行編譯。然后修改  meta-ml/recipes-devtools/python/python3-pybind11_2.5.0.bb 中的Python 版本為 3.8 。最后可以生成  multimedia image。

-------------------------------------

EXTRA_OECMAKE = "-DPYBIND11_TEST=OFF \
-DPYTHON_EXECUTABLE=${RECIPE_SYSROOT_NATIVE}/usr/bin/python3-native/python3.8 \ "

-------------------------------------

使用 Toradex Easy Installer 將生成的鏡像安裝到  Apalis iMX8QM 4GB WB IT V1.1C 和 Verdin iMX8M Plus Quad 4GB WB IT V1.0B 兩個(gè)模塊上。

測(cè)試的內(nèi)容參考 NXP 的 i.MX_Machine_Learning_User's_Guide 文檔進(jìn)行，包括 TensorFlow Lite、Arm NN、ONNX、PyTorch。由于目前  OpenCV 還只能運(yùn)行在 iMX8QuadMax  和 iMX8M Plus 的 CPU 上，無(wú)法使用 GPU 或者 NPU 加速，所以本次不做測(cè)試。另外，在使用 Arm NN 測(cè)試 Caffe 模型時(shí)有兩個(gè)限制。第一，batch size 必須為 1。例如  deploy.prototxt 文件修改為

-------------------------------------

name: "AlexNet"

layer {

  name: "data"

  type: "Input"

  top: "data"

  input_param { shape: { dim: 1 dim: 3 dim: 227 dim: 227 } }

}

-------------------------------------

第二， Arm NN 不支持所有的 Caffe 語(yǔ)法，一些老的神經(jīng)網(wǎng)絡(luò)模型文件需要更新到最新的  Caffe 語(yǔ)法。下面是 PC 上用于轉(zhuǎn)換的 Python3 腳本。

-------------------------------------

import caffe

net = caffe.Net('lenet.prototxt', 'lenet_iter_9000-orignal.caffemodel', caffe.TEST)

net.save('lenet_iter_9000.caffemodel')

-------------------------------------

在兩個(gè)模塊上測(cè)試結(jié)果如下。

TensorFlow Lite

l Apalis iMX8QM

label_image

-------------------------------------

root@apalis-imx8:/usr/bin/tensorflow-lite-2.4.0/examples# USE_GPU_INFERENCE=1 ./label_image -m mobilenet_v1_1.0_224_quant.tflite -i grace_hopper.bmp -l labels.txt -a 1

INFO: Loaded model mobilenet_v1_1.0_224_quant.tflite
INFO: resolved reporter
INFO: Created TensorFlow Lite delegate for NNAPI.
INFO: Applied NNAPI delegate.
INFO: invoked
INFO: average time: 12.407 ms
INFO: 0.784314: 653 military uniform
INFO: 0.105882: 907 Windsor tie
INFO: 0.0156863: 458 bow tie
INFO: 0.0117647: 466 bulletproof vest
INFO: 0.00784314: 668 mortarboard

-------------------------------------

benchmark_model

-------------------------------------

root@apalis-imx8:/usr/bin/tensorflow-lite-2.4.0/examples# ./benchmark_model --graph=mobilenet_v1_1.0_224_quant.tflite --use_nnapi=true

STARTING!

Log parameter values verbosely: [0]
Graph: [mobilenet_v1_1.0_224_quant.tflite]
Use NNAPI: [1]
NNAPI accelerators available: [vsi-npu]
Loaded model mobilenet_v1_1.0_224_quant.tflite
INFO: Created TensorFlow Lite delegate for NNAPI.
Explicitly applied NNAPI delegate, and the model graph will be completely executed by the
delegate.
The input model file size (MB): 4.27635
Initialized session in 16.746ms.
Running benchmark for at least 1 iterations and at least 0.5 seconds but terminate if exceeding 150
seconds.
count=17 first=305296 curr=12471 min=12299 max=305296 avg=29650 std=68911
Running benchmark for at least 50 iterations and at least 1 seconds but terminate if exceeding 150 seconds.
count=81 first=12417 curr=12430 min=12294 max=12511 avg=12405.6 std=39
Inference timings in us: Init: 16746, First inference: 305296, Warmup (avg): 29650, Inference (avg): 12405.6
Note: as the benchmark tool itself affects memory footprint, the following is only APPROXIMATE to the actual memory footprint of the model at runtime. Take the information at your discretion.
Peak memory footprint (MB): init=1.85938 overall=55.1406

-------------------------------------

l Verdin iMX8M Plus

label_image

-------------------------------------

root@verdin-imx8mp:/usr/bin/tensorflow-lite-2.4.0/examples# USE_GPU_INFERENCE=0 ./label_image -m mobilenet_v1_1.0_224_quant.tflite -i grace_hopper.bmp -l labels.txt -a 1
INFO: Loaded model mobilenet_v1_1.0_224_quant.tflite
INFO: resolved reporter
INFO: Created TensorFlow Lite delegate for NNAPI.
INFO: Applied NNAPI delegate.
INFO: invoked
INFO: average time: 2.835 ms
INFO: 0.768627: 653 military uniform
INFO: 0.105882: 907 Windsor tie
INFO: 0.0196078: 458 bow tie
INFO: 0.0117647: 466 bulletproof vestINFO: 0.00784314: 835 suit

-------------------------------------

benchmark_model

-------------------------------------

root@verdin-imx8mp:/usr/bin/tensorflow-lite-2.4.0/examples# ./benchmark_model --graph=mobilenet_v1_1.0_224_quant.tflite --use_nnapi=true
STARTING!
Log parameter values verbosely: [0]
Graph: [mobilenet_v1_1.0_224_quant.tflite]
Use NNAPI: [1]
NNAPI accelerators available: [vsi-npu]
Loaded model mobilenet_v1_1.0_224_quant.tflite
INFO: Created TensorFlow Lite delegate for NNAPI.
Explicitly applied NNAPI delegate, and the model graph will be completely executed by the delegate.
The input model file size (MB): 4.27635
Initialized session in 16.79ms.
Running benchmark for at least 1 iterations and at least 0.5 seconds but terminate if exceeding 150 seconds.
count=1 curr=6664535
Running benchmark for at least 50 iterations and at least 1 seconds but terminate if exceeding 150 seconds.
count=367 first=2734 curr=2646 min=2624 max=2734 avg=2650.05 std=16
Inference timings in us: Init: 16790, First inference: 6664535, Warmup (avg): 6.66454e+06, Inference (avg): 2650.05
Note: as the benchmark tool itself affects memory footprint, the following is only APPROXIMATE to the actual memory footprint of the model at runtime. Take the information at your discretion.
Peak memory footprint (MB): init=1.79297 overall=28.5117 

-------------------------------------

Arm NN

l Apalis iMX8QM

CaffeAlexNet-Armnn

-------------------------------------

root@apalis-imx8:/usr/bin/armnn-20.08/ArmnnTests# ../CaffeAlexNet-Armnn --data-dir=data --model-dir=models
Info: ArmNN v22.0.0
Info: Initialization time: 0.14 ms
Info: Network parsing time: 1397.76 ms
Info: Optimization time: 195.13 ms
Info: = Prediction values for test #0
Info: Top(1) prediction is 2 with value: 0.706226
Info: Top(2) prediction is 0 with value: 1.26573e-05
Info: Total time for 1 test cases: 0.264 seconds
Info: Average time per test case: 263.701 ms
Info: Overall accuracy: 1.000
Info: Shutdown time: 56.83 ms

-------------------------------------

CaffeMnist-Armnn

-------------------------------------

root@apalis-imx8:/usr/bin/armnn-20.08/ArmnnTests# ../CaffeMnist-Armnn --data-dir=data --model-dir=models

Info: ArmNN v22.0.0
Info: Initialization time: 0.09 ms
Info: Network parsing time: 8.70 ms
Info: Optimization time: 2.67 ms
Info: = Prediction values for test #0
Info: Top(1) prediction is 7 with value: 1
Info: Top(2) prediction is 0 with value: 0
Info: = Prediction values for test #1
Info: Top(1) prediction is 2 with value: 1
Info: Top(2) prediction is 0 with value: 0
Info: = Prediction values for test #5
Info: Top(1) prediction is 1 with value: 1
Info: Top(2) prediction is 0 with value: 0
Info: = Prediction values for test #8
Info: Top(1) prediction is 5 with value: 1
Info: Top(2) prediction is 0 with value: 0
Info: = Prediction values for test #9
Info: Top(1) prediction is 9 with value: 1
Info: Top(2) prediction is 0 with value: 0
Info: Total time for 5 test cases: 0.015 seconds
Info: Average time per test case: 2.927 ms
Info: Overall accuracy: 1.000
Info: Shutdown time: 1.56 ms 

-------------------------------------

CaffeVGG-Armnn

-------------------------------------

root@apalis-imx8:/usr/bin/armnn-20.08/ArmnnTests# ../CaffeVGG-Armnn --data-dir=data --model-dir=models

Info: ArmNN v22.0.0
Info: Initialization time: 0.08 ms
Info: Network parsing time: 1452.35 ms
Info: Optimization time: 491.98 ms
Info: = Prediction values for test #0
Info: Top(1) prediction is 2 with value: 0.692014
Info: Top(2) prediction is 0 with value: 9.80887e-07
Info: Total time for 1 test cases: 2.723 seconds
Info: Average time per test case: 2722.846 ms
Info: Overall accuracy: 1.000
Info: Shutdown time: 115.74 ms

-------------------------------------

l Verdin iMX8M Plus

CaffeAlexNet-Armnn

-------------------------------------

root@verdin-imx8mp:/usr/bin/armnn-20.08/ArmnnTests# ../CaffeAlexNet-Armnn --data-dir=data --model-dir=models

Info: ArmNN v22.0.0
Info: Initialization time: 0.12 ms
Info: Network parsing time: 1250.55 ms
Info: Optimization time: 141.40 ms
Info: = Prediction values for test #0
Info: Top(1) prediction is 2 with value: 0.706225
Info: Top(2) prediction is 0 with value: 1.26573e-05
Info: Total time for 1 test cases: 0.110 seconds
Info: Average time per test case: 110.124 ms
Info: Overall accuracy: 1.000
Info: Shutdown time: 15.04 ms

-------------------------------------

CaffeMnist-Armnn

-------------------------------------

root@verdin-imx8mp:/usr/bin/armnn-20.08/ArmnnTests# ../CaffeMnist-Armnn --data-dir=data --model-dir=models

Info: ArmNN v22.0.0
Info: Initialization time: 0.11 ms
Info: Network parsing time: 8.96 ms
Info: Optimization time: 3.01 ms
Info: = Prediction values for test #0
Info: Top(1) prediction is 7 with value: 1
Info: Top(2) prediction is 0 with value: 0
Info: = Prediction values for test #1
Info: Top(1) prediction is 2 with value: 1
Info: Top(2) prediction is 0 with value: 0
Info: = Prediction values for test #5
Info: Top(1) prediction is 1 with value: 1
Info: Top(2) prediction is 0 with value: 0
Info: = Prediction values for test #8
Info: Top(1) prediction is 5 with value: 1
Info: Top(2) prediction is 0 with value: 0
Info: = Prediction values for test #9
Info: Top(1) prediction is 9 with value: 1
Info: Top(2) prediction is 0 with value: 0
Info: Total time for 5 test cases: 0.008 seconds
Info: Average time per test case: 1.608 ms
Info: Overall accuracy: 1.000
Info: Shutdown time: 1.69 ms 

-------------------------------------

CaffeVGG-Armnn

-------------------------------------

root@verdin-imx8mp:/usr/bin/armnn-20.08/ArmnnTests# ../CaffeVGG-Armnn --data-dir=data --model-dir=modelsInfo: ArmNN v22.0.0

Info: Initialization time: 0.15 ms
Info: Network parsing time: 2842.95 ms
Info: Optimization time: 316.74 ms
Info: = Prediction values for test #0
Info: Top(1) prediction is 2 with value: 0.692015
Info: Top(2) prediction is 0 with value: 9.8088e-07
Info: Total time for 1 test cases: 1.098 seconds
Info: Average time per test case: 1097.593 ms
Info: Overall accuracy: 1.000
Info: Shutdown time: 130.65 ms 

-------------------------------------

ONNX

l Apalis iMX8QM

onnx_test_runner

-------------------------------------

root@apalis-imx8:~# time onnx_test_runner -j 1 -c 1 -r 1 -e vsi_npu ./mobilenetv2-7/

result:  
Models: 1
Total test cases: 3
 Succeeded: 3
 Not implemented: 0
 Failed: 0
Stats by Operator type:
 Not implemented(0):  
 Failed:
Failed Test Cases:
 
real 0m0.643s
user 0m1.513s
sys 0m0.111s

-------------------------------------

l Verdin iMX8M Plus

onnx_test_runner

-------------------------------------

root@verdin-imx8mp:~# time onnx_test_runner -j 1 -c 1 -r 1 -e vsi_npu ./mobilenetv2-7/

result:  
Models: 1
Total test cases: 3
 Succeeded: 3
 Not implemented: 0
 Failed: 0
Stats by Operator type:
 Not implemented(0):  
 Failed:
Failed Test Cases:
 
real 0m0.663s
user 0m1.195s
sys 0m0.073s 

-------------------------------------

PyTorch

l Apalis iMX8QM

pytorch_mobilenetv2.py

-------------------------------------

root@apalis-imx8:/usr/bin/pytorch/examples# time python3 pytorch_mobilenetv2.py

('tabby, tabby cat', 46.348018646240234)
('tiger cat', 35.17843246459961)
('Egyptian cat', 15.802857398986816)
('lynx, catamount', 1.161122441291809)
('tiger, Panthera tigris', 0.20774582028388977)
 
real 0m8.806s
user 0m7.440s
sys 0m0.593s 

-------------------------------------

l Verdin iMX8M Plus

pytorch_mobilenetv2.py

-------------------------------------

root@verdin-imx8mp:/usr/bin/pytorch/examples# time python3 pytorch_mobilenetv2.py

('tabby, tabby cat', 46.348018646240234)
('tiger cat', 35.17843246459961)
('Egyptian cat', 15.802857398986816)
('lynx, catamount', 1.161122441291809)
('tiger, Panthera tigris', 0.20774582028388977)
 
real 0m6.313s
user 0m5.933s
sys 0m0.295s 

-------------------------------------

匯總對(duì)比

根據(jù)具體測(cè)試應(yīng)用不同，兩者之間的性能差距大小不一?？傮w來(lái)看常用機(jī)器學(xué)習(xí)算法在 Verdin iMX8M Plus 的 NPU 上的表現(xiàn)會(huì)優(yōu)于 Apalis iMX8QM 的 GPU。

總結(jié)

機(jī)器學(xué)習(xí)是較為復(fù)雜的應(yīng)用，除了硬件處理器外，影響算法性能表現(xiàn)的還包括對(duì)模型本身的優(yōu)化。尤其是對(duì)嵌入式系統(tǒng)有限的處理能力來(lái)講，直接將 PC 上現(xiàn)成的模型拿過(guò)來(lái)用通常會(huì)表現(xiàn)不佳。同時(shí)根據(jù)項(xiàng)目需求選擇合適計(jì)算機(jī)模塊，畢竟 Verdin iMX8M Plus 和 Apalis iMX8QM 的用途側(cè)重點(diǎn)不同。