DLAP221基于mxVision 容器内推理

shiner_chen

已于 2023-01-09 15:04:32 修改

阅读量366

点赞数

分类专栏： docker AI Atlas 文章标签：人工智能 docker linux

于 2023-01-09 10:28:45 首次发布

本文链接：https://blog.csdn.net/shiner_chen/article/details/128610084

版权

Atlas 同时被 3 个专栏收录

22 篇文章 5 订阅

订阅专栏

8 篇文章 1 订阅

订阅专栏

docker

5 篇文章 0 订阅

订阅专栏

DLAP221 基于mxVision 容器内推理

前提条件

DLAP221启动镜像制作成功，查看系统信息如下

HwHiAiUser@davinci-mini:~$ npu-smi info
+------------------------------------------------------------------------------------------------+
| npu-smi 22.0.2                           Version: 22.0.2                                       |
+-----------------------+-----------------+------------------------------------------------------+
| NPU     Name          | Health          | Power(W)     Temp(C)           Hugepages-Usage(page) |
| Chip    Device        | Bus-Id          | AICore(%)    Memory-Usage(MB)                        |
+=======================+=================+======================================================+
| 0       310           | OK              | 12.8         55                2    / 970            |
| 0       0             | NA              | 0            5353 / 7764                             |
+=======================+=================+======================================================+

访问官方推理镜像网站，下载对应的mxVision容器镜像

点击获取镜像，注册华为用户，如果有则直接登录。

选择22.0.RC2-mxvision镜像版本，并点击下载，将弹出以下窗口，根据窗口中说明的步骤，下载推理镜像。

在DLAP221的命令行，运行上图中步骤的命令，如果在step2中输入了正确的密码，但仍然无法下载镜像，请根据step3设置下载凭证.

查看镜像，命令为：

HwHiAiUser@davinci-mini:~$ sudo docker images
REPOSITORY                                             TAG                 IMAGE ID       CREATED        SIZE
ascendhub.huawei.com/public-ascendhub/infer-modelzoo   22.0.RC2-mxvision   ed381e9a8f24   3 months ago   6.08GB

安装Ascend-cann-toolkit包

HwHiAiUser@davinci-mini:~$ sudo ./Ascend-cann-toolkit_5.1.RC2_linux-aarch64.run --install

启动容器

下载mindX测试样例

HwHiAiUser@davinci-mini:~$ git clone https://gitee.com/ascend/ascend_community_projects.git
HwHiAiUser@davinci-mini:~$ cd ascend_community_projects/
HwHiAiUser@davinci-mini:~/ascend_community_projects$ ls
2D_LiDAR_Pedestrain_Detection  AlphaPose  			Assisted-driving   						 ChineseOCR         FCOS        gesturedetection  
insulatorDetection  					 Luggage-Checker  MonocularDepthEstimation_V2    OWNERS             README.md   RosSdkInfer        VideoStitch                    A200DK_GPIO_I2C  AluminumDefectDetection        Atlas200DKMPU6050  deepsort    figures         HelmetIdentification_V2        kws              MedicalKeypointsDetecion       MoveitMechanicalArmSimulation  PointPillars  
RefineDet                      UAV_CrowdCounting  Yolact                       A200DK_UART_SPI                AnimeGAN                 Atlas200DKSPILCD               FairMOT          fire_detection                 Individual_V2      LICENSE     MobileStereoNet           MultiChannelVideoInfer         README.en.md     Retinaface                     Video-ReID         YOLOv5Prune

创建容器启动脚本

docker run -it -u root --rm \
--net=host \
--device=/dev/davinci0 \
--device=/dev/davinci_manager \
--device=/dev/ascend_manager \
--device=/dev/svm0 \
--device=/dev/log_drv \
--device=/dev/event_sched \
--device=/dev/upgrade \
--device=/dev/hi_dvpp \
--device=/dev/memory_bandwidth \
--device=/dev/ts_aisle \
-v /usr/local/Ascend/driver/tools:/usr/local/Ascend/driver/tools \
-v /usr/local/Ascend/driver/lib64:/usr/local/Ascend/driver/lib64 \
-v /usr/lib64/aicpu_kernels:/usr/lib64/aicpu_kernels \
-v /usr/local/sbin/npu-smi:/usr/local/sbin/npu-smi \
-v /var/hdc_ppc:/var/hdc_ppc \
-v /etc/hdcBasic.cfg:/etc/hdcBasic.cfg \
-v /etc/rc.local:/etc/rc.local \
-v /sys:/sys \
-v /usr/bin/sudo:/usr/bin/sudo \
-v /usr/lib/sudo/:/usr/lib/sudo/ \
-v /etc/sudoers:/etc/sudoers/ \
-v /etc/sys_version.conf:/etc/sys_version.conf \
-v /home/HwHiAiUser/ascend_community_projects:/data/ascend_community_projects \
-v /home/HwHiAiUser/ascend_ddk/:/home/HwHiAiUser/ascend_ddk \
ascendhub.huawei.com/public-ascendhub/infer-modelzoo:22.0.RC2-mxvision \
/bin/bash -c "/usr/local/Ascend/driver/tools/minirc_container_prepare.sh;/bin/bash"

其中：

/usr/local/Ascend/driver： 为驱动的安装目录，需要映射到容器中
/usr/local/Ascend/ascend-toolkit： CANN toolkit的安装目录需要映射到容器中，如果客户安装的是nnrt运行环境，则替换为nnrt的安装目录
/home/HwHiAiUser/ascend_community_projects： mindx测试样例

修改minirc_container_prepare.sh脚本

#!/bin/bash
[ -d /usr/lib64 ] || mkdir -p /usr/lib64
[ -d /var ] || mkdir -p /var
[ -d /var/driver ] || mkdir -p /var/driver
[ -d /usr/slog ] || mkdir -p /usr/slog
[ -d /run/driver ] || mkdir -p /run/driver
[ -d /var/dmp ] || mkdir -p /var/dmp
[ -d /home/root/hdc_ppc ] || mkdir -p /home/root/hdc_ppc
[ -d /home/HwHiAiUser/hdc_ppc ] || mkdir -p /home/HwHiAiUser/hdc_ppc

cp -arf /usr/local/Ascend/driver/lib64/* /usr/lib64/
cp -arf /usr/local/Ascend/driver/tools/* /var/
###添加以下6行代码
export LD_LIBRARY_PATH=/usr/local/Ascend/driver/lib64:$LD_LIBRARY_PATH
mkdir -p /usr/slog
/var/slogd &
mkdir -p /run/driver
/var/dmp_daemon -I -U 8087 &
chown -R root:root /home/hwMindX/sdk_home
######
chmod 755 /usr/lib64
chown -h HwHiAiUser:HwHiAiUser /usr/slog
chown -h HwHiAiUser:HwHiAiUser /var/dmp
chown -h HwHiAiUser:HwHiAiUser /home/HwHiAiUser/hdc_ppc

if [ ! -f /usr/lib64/libcrypto.so.1.1 ] && [ ! -L /usr/lib64/libcrypto.so.1.1 ]; then
    cp -arf /usr/local/Ascend/driver/lib64/inner/libcrypto.so.1.1 /usr/lib64/
fi

启动容器

HwHiAiUser@davinci-mini:~$ sudo ./docker_run.sh
root@davinci-mini:/home/hwMindX#   <-----进入了容器

运行测试样例

本文运行的测试样例为Luggage-Checker，根据样例的readme，做如下操作

下载om模型

root@davinci-mini:/home/hwMindX# cd /data/ascend_community_projects/Luggage-Checker/
root@davinci-mini:/data/ascend_community_projects/Luggage-Checker# cd python/models/
root@davinci-mini:/data/ascend_community_projects/Luggage-Checker/python/models# wget  -O yolox_pre_post.om https://mindx.sdk.obs.myhuaweicloud.com/ascend_community_projects/laser_detect/yolox_pre_post.om --no-check-certificate

编译测试样例的后处理代码

root@davinci-mini:/home/hwMindX# cd /data/ascend_community_projects/Luggage-Checker/
root@davinci-mini:/data/ascend_community_projects/Luggage-Checker# bash build.sh
-- The C compiler identification is GNU 7.5.0
-- The CXX compiler identification is GNU 7.5.0
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working C compiler: /usr/bin/cc - skipped
-- Detecting C compile features
-- Detecting C compile features - done
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Check for working CXX compiler: /usr/bin/c++ - skipped
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Configuring done
-- Generating done
-- Build files have been written to: /data/ascend_community_projects/Luggage-Checker/postprocess/build
[ 50%] Building CXX object CMakeFiles/YoloxPostProcess.dir/YoloxPostProcess.cpp.o
[100%] Linking CXX shared library libYoloxPostProcess.so
[100%] Built target YoloxPostProcess

运行测试样例

root@davinci-mini:/data/ascend_community_projects/Luggage-Checker/python/Main# python3 pre_post.py
Begin to initialize Log.
WARNING: Logging before InitGoogleLogging() is written to STDERR
W20230109 02:15:26.178339    65 ConfigUtil.h:103] [1016][Object, file or other resource doesn't exist] Fail to read (log_prefix) from config, default is: 1
W20230109 02:15:26.178485    65 ConfigUtil.h:103] [1016][Object, file or other resource doesn't exist] Fail to read (colorlogtostderr) from config, default is: 1
W20230109 02:15:26.178542    65 ConfigUtil.h:103] [1016][Object, file or other resource doesn't exist] Fail to read (stop_logging_if_full_disk) from config, default is: 1
The output directory of logs file doesn't exist.
Create directory to save logs information.
Save logs information to specified directory.
.....
.....
I20230109 02:15:28.072665    65 MxpiImageDecoder.cpp:120] Begin to deinitialize MxpiImageDecoder(mxpi_imagedecoder0).
I20230109 02:15:28.074342    65 MxpiImageDecoder.cpp:139] End to deinitialize MxpiImageDecoder(mxpi_imagedecoder0).
I20230109 02:15:28.074736    65 MxGstBase.cpp:822] element(mxpi_dataserialize0) gst_change_state READY_TO_NULL.
I20230109 02:15:28.074843    65 MxGstBase.cpp:822] element(mxpi_objectpostprocessor0) gst_change_state READY_TO_NULL.
I20230109 02:15:28.074934    65 MxGstBase.cpp:822] element(mxpi_tensorinfer0) gst_change_state READY_TO_NULL.
I20230109 02:15:28.075084    65 MxGstBase.cpp:822] element(mxpi_imageresize0) gst_change_state READY_TO_NULL.
I20230109 02:15:28.075177    65 MxGstBase.cpp:822] element(mxpi_imagedecoder0) gst_change_state READY_TO_NULL.
I20230109 02:15:28.076000    65 MxsmStream.cpp:917] Destroys the stream(detection) successfully.
I20230109 02:15:28.076083    65 MxStreamManager.cpp:72] Begin to destroy management threads.
I20230109 02:15:28.077147    65 DeviceManager.cpp:120] DestroyDevices begin
I20230109 02:15:28.082032    65 DeviceManager.cpp:138] DestroyDevices successfully
I20230109 02:15:28.082115    65 MxStreamManager.cpp:263] Destroy streams successfully.

遇到一下错误

Traceback (most recent call last):
  File "/data/ascend_community_projects/Luggage-Checker/python/Main/pre_post.py", line 25, in <module>
    from visualize import plot_one_box
  File "/data/ascend_community_projects/Luggage-Checker/python/Main/visualize.py", line 20, in <module>
    import webcolors
ModuleNotFoundError: No module named 'webcolors'
WARNING: Logging before InitGoogleLogging() is written to STDERR
I20230109 02:13:37.540711    52 MxStreamManager.cpp:54] MxStreamManager destroy.
I20230109 02:13:37.540863    52 MxStreamManager.cpp:61] log rotate thread exits successful.

解决方法：

root@davinci-mini:/data/ascend_community_projects/Luggage-Checker/python/Main# pip3 install webcolors
Looking in indexes: http://pypi.doubanio.com/simple/
Collecting webcolors
  Downloading http://pypi.doubanio.com/packages/af/a9/71beaebcacd7dd848af470740dbe7f39aa6bd23a60b8162f8c050090eed4/webcolors-1.12-py3-none-any.whl (9.9 kB)
Installing collected packages: webcolors
Successfully installed webcolors-1.12
WARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv

[notice] A new release of pip available: 22.2.2 -> 22.3.1
[notice] To update, run: pip install --upgrade pip

测试结果

测试结果图片为pre_post_bgr.jpg

root@davinci-mini:/data/ascend_community_projects/Luggage-Checker/python/Main# cd ../test_img/
root@davinci-mini:/data/ascend_community_projects/Luggage-Checker/python/test_img# ls
pre_post_bgr.jpg  test.jpg

[外链图片转存失败,源站可能有防盗链机制,建议将图片保存下来直接上传(img-SDT3nuQ5-1673231104070)(null)]