部署模型上昇腾200dk

前面已经将准备好虚拟机服务器、开发板USB连接并联网、分别在device和host部署好运行和开发环境、安装Mind studio、跑通猫狗识别样例%EF%BC%88Atlas200DK%EF%BC%89)

接下来我要挑选已经训练好的模型，将kares保存的H5文件转换为tf的pb文件进而转换为device上能跑的om文件，在device简单跑通模型，接着小小改动程序加入我们的业务逻辑。

任务还是挺多的，后天上午上课，大后天11月30号-23:31作业截止。 还有一个问题要确定：后天上课是干嘛，上课还是做pre？

模型输入[244,244,3]

tensorflow、pytorch、keras框架模型保存的格式汇总

Tensorflow h5转pb
https://blog.csdn.net/dou3516/article/details/118441871
https://www.tensorflow.org/api_docs/python/tf/keras/models/save_model
https://www.tensorflow.org/tutorials/keras/save_and_load

TF模型h5转换为frozenGraphDef

How to export a TensorFlow 2.x Keras model to a frozen and optimized graph
https://medium.com/@sebastingarcaacosta/how-to-export-a-tensorflow-2-x-keras-model-to-a-frozen-and-optimized-graph-39740846d9eb

老外的的代码有用

Save, Load and Inference From TensorFlow 2.x Frozen Graph
https://leimao.github.io/blog/Save-Load-Inference-From-TF2-Frozen-Graph/

ModelConverter离线转换

通过上面的代码讲H5保存的模型导入再转换为frozengraphdef的pb文件，传到虚拟机中开始atc转换

2021-12-1314:34:49insertopfilecontent:
aipp_op{
related_input_rank:0
src_image_size_w:224
src_image_size_h:224
crop:false
padding:false
input_format:YUV420SP_U8
aipp_mode:static
csc_switch:true
rbuv_swap_switch:false
matrix_r0c0:298
matrix_r0c1:0
matrix_r0c2:409
matrix_r1c0:298
matrix_r1c1:-100
matrix_r1c2:-208
matrix_r2c0:298
matrix_r2c1:516
matrix_r2c2:0
input_bias_0:16
input_bias_1:128
input_bias_2:128
mean_chn_0:104
mean_chn_1:117
mean_chn_2:123
min_chn_0:0.0
min_chn_1:0.0
min_chn_2:0.0
var_reci_chn_0:1.0
var_reci_chn_1:1.0
var_reci_chn_2:1.0
}
2021-12-13 14:34:49  Start to convert model
2021-12-13 14:34:49  export PATH=$PATH:/home/wuboxin/Ascend/ascend-toolkit/5.0.3.alpha002/atc/ccec_compiler/bin:/home/wuboxin/Ascend/ascend-toolkit/5.0.3.alpha002/atc/bin && export PYTHONPATH=$PYTHONPATH:/home/wuboxin/Ascend/ascend-toolkit/5.0.3.alpha002/atc/python/site-packages:/home/wuboxin/Ascend/ascend-toolkit/5.0.3.alpha002/atc/python/site-packages/auto_tune.egg/auto_tune:/home/wuboxin/Ascend/ascend-toolkit/5.0.3.alpha002/atc/python/site-packages/schedule_search.egg:/home/wuboxin/Ascend/ascend-toolkit/5.0.3.alpha002/opp/op_impl/built-in/ai_core/tbe && export LD_LIBRARY_PATH=/home/wuboxin/Ascend/ascend-toolkit/5.0.3.alpha002/atc/lib64:/usr/local/Ascend/driver/lib64:/usr/local/Ascend/driver/lib64/driver:/home/wuboxin/Ascend/ascend-toolkit/5.0.3.alpha002/add-ons:/home/wuboxin/Ascend/ascend-toolkit/5.0.3.alpha002/acllib/lib64:$LD_LIBRARY_PATH:/home/wuboxin/Ascend/ascend-toolkit/5.0.3.alpha002/atc/lib64/stub && export SLOG_PRINT_TO_STDOUT=1 && export ASCEND_OPP_PATH=/home/wuboxin/Ascend/ascend-toolkit/5.0.3.alpha002/opp && /home/wuboxin/Ascend/ascend-toolkit/5.0.3.alpha002/atc/bin/atc  --input_shape="x:-1,224,224,3" --check_report=/home/wuboxin/modelzoo/sleep_or_not/Ascend310/network_analysis.report --input_format=NHWC --output="/home/wuboxin/modelzoo/sleep_or_not/Ascend310/sleep_or_not" --soc_version=Ascend310 --insert_op_conf=/home/wuboxin/modelzoo/sleep_or_not/Ascend310/insert_op.cfg --framework=3 --model="/home/wuboxin/models/mymodel/FrozenGraphDef/sleep_or_not.pb" --dynamic_batch_size="1,2,4,8" 
2021-12-13 14:34:49  ATC start working now, please wait for a moment.
2021-12-13 14:35:42  ATC run success, welcome to the next use.
2021-12-13 14:35:42  Convert model environment variables: 
2021-12-13 14:35:42  export PATH=$PATH:/home/wuboxin/Ascend/ascend-toolkit/5.0.3.alpha002/atc/ccec_compiler/bin:/home/wuboxin/Ascend/ascend-toolkit/5.0.3.alpha002/atc/bin && export PYTHONPATH=$PYTHONPATH:/home/wuboxin/Ascend/ascend-toolkit/5.0.3.alpha002/atc/python/site-packages:/home
wuboxin@ubuntu3:~/modelzoo/sleep_or_not/Ascend310$ll
总用量80068
drwxr-x---2wuboxinwuboxin4096Dec1314:35./
drwxr-x---3wuboxinwuboxin4096Dec1314:34../
-rw-------1wuboxinwuboxin577Dec1314:34insert_op.cfg
-rw-r-----1wuboxinwuboxin3787Dec1314:35ModelConvert.txt
-rw-------1wuboxinwuboxin13462Dec1314:34network_analysis.report
-rw-------1wuboxinwuboxin2196Dec1314:35sleep_or_not_config.json
-rw-------1wuboxinwuboxin81945693Dec1314:35sleep_or_not.om

试运行sleep or not 的demo

复制前天跑通的demo工程目录文件，抽取赖床数据集和将上一部转换的om替换上去，执行，遇到一个问题

Execute model failed for acl.mdl.execute error 145000

参考
https://bbs.huaweicloud.com/forum/thread-132486-1-1.html
将模型重新转换了一次，主要是修改下图红框


再跑一次就能跑通了，不过要测试数据要减少。

现在模型跑通了，下一步小小的加入我们的业务逻辑应该怎么分为几步呢。

• 接通摄像头
  • 跑通王嘉伟同学给的视频demo https://gitee.com/ascend/samples/tree/master/python/level2_simple_inference/2_object_detection/YOLOV3_mask_detection_video#/ascend/samples/blob/master/python/environment

给安装用户赋权
https://support.huaweicloud.com/environment-deployment-Atlas200DK1011/atlased_04_0023.html

跑视频demo

根据王嘉伟同学的依据，跑通了一个解析视频的demo
在此期间，由于我跑上一个demo修改了默认镜像中的环境，所以导致python环境变量冲突，所以不得不重新刷一遍镜像。
另一个重要的原因是，在重启虚拟机时没有及时关闭和开发板的连接，导致重启后不能及时连接到device。