Paddlelite compilation and code running through the disk

I am a tm Big enemy species of , notice Paddle lite Just be afraid

Sometimes run Paddle Lite If the code is wrong , stay spyder Give it directly to me inside restart kernel, I tm…

The basic operation is based on this document ：
https://paddle-lite.readthedocs.io/zh/develop/source_compile/windows_compile_windows.html

（PaddleLite The update speed of is faster , Pay attention to timeliness ）

---

1. git Basic operation problems of

I met all kinds of git Basic operation problems of ：

1. The success of ping github.com,ping Change if it doesn't work host file ( Don't dwell on )
2. git Throughout clone Not coming down , Use an acceleration channel https://github.91chi.fun/, A kind of , Previous hub.fastgit.org Hang up
3. Then is third-party The directory should include third-party github package clone Come down , Recursive modification .gitmodule and .git/config file , Until you execute the command git submodule update --init --recursive No problem

Add here , Current version Paddle Lite Windows Compiled scripts
lite\tools\build_windows.bat
Yes bug, His own document says , hold third-party Delete the table of contents , The script will be downloaded automatically third-party, However, there are actually problems

Look at him batch Script , Download directly from this link ：
https://paddlelite-data.bj.bcebos.com/third_party_libs/third-party-91a9ab3.tar.gz

2. Compile start ：

I met VS2019 There is a problem with the compilation , So I downloaded VS2015

About VS2015 Installation , You can refer to here ：
https://blog.csdn.net/HaoZiHuang/article/details/126020139

Then directly compile ：

lite\tools\build_windows.bat with_extra with_profile with_precision_profile

10000+ Multiple warning…

---

There are some episodes in the middle ：

CMake Error at lite/kernels/CMakeLists.txt:118 (message):
  Traceback (most recent call last):

    File "xxxxx/Paddle-Lite/lite/tools/cmake_tools/create_fake_kernel_registry.py", line 243, in <module>
      parse_fake_kernels_from_path(faked_kernels_list_path)
    File "xxxxx/Paddle-Lite/lite/tools/cmake_tools/create_fake_kernel_registry.py", line 101, in parse_fake_kernels_from_path
      paths = set([path for path in f])
    File "xxxxx/Paddle-Lite/lite/tools/cmake_tools/create_fake_kernel_registry.py", line 101, in <listcomp>
      paths = set([path for path in f])

  UnicodeDecodeError: 'gbk' codec can't decode byte 0x95 in position 37:
  illegal multibyte sequence



-- Configuring incomplete, errors occurred!
See also "xxxxx/Paddle-Lite/build.lite.x86/CMakeFiles/CMakeOutput.log".
See also "xxxxx/Paddle-Lite/build.lite.x86/CMakeFiles/CMakeError.log".

Need to give the upper open Add encoding='utf-8', To the authorities PR, I don't know whether it will be merged ：
https://github.com/PaddlePaddle/Paddle-Lite/pull/9301

---

If you use VS2019 perhaps VS2017 compile , Then use use_vs2017 and use_vs2019 Specify the parameter , Do not add these two parameters , It's the default VS2015 compile , If your installation location changes or there is no installation , He will let you fill in the path manually , If you fill in VS2019 perhaps VS2017 The path of , This command line exits directly … There is no hint

For specific parameters of compilation, please refer to ：
https://paddle-lite.readthedocs.io/zh/develop/source_compile/windows_compile_windows.html

3. Precompiled translation package

You can download the precompiled package from here , But there is no such thing as Py3.8 And the versions above ：
https://paddle-lite.readthedocs.io/zh/latest/quick_start/release_lib.html#windows

About the meaning of each compiled file , The following references are from here ：
https://paddle-lite.readthedocs.io/zh/develop/source_compile/windows_compile_windows.html#id14

The compilation result is located in build.lite.x86\inference_lite_lib

The details are as follows ：

1. cxx Folder ： contain c++ Library file and corresponding header file
   include ： The header file
   lib ： The library files
   Static library files ：
   libpaddle_api_full_bundled.lib ：full_api Static library
   libpaddle_api_light_bundled.lib ：light_api Static library

2. third_party Folder ： Dependent third-party prediction Library mklml
   mklml ： Paddle Lite The prediction library depends on mklml Math library

3. demo\cxx Folder ：C++ Example demo
   mobilenetv1_full ： Use full_api perform mobilenet_v1 Predicted C++ demo
   mobilenetv1_light ： Use light_api perform mobilenet_v1 Predicted C++ demo

4. demo\python： Python Example demo
   mobilenetv1_full_api.py： Use full_api perform mobilenet_v1 Predicted Python demo
   mobilenetv1_light_api.py： Use full_api perform mobilenet_v1 Predicted Python demo

5. python Folder ： contain Python Library file and corresponding .whl package
   install Folder ： Compiled successfully .whl Package is located in the install\dist\*.whl
lib Folder ：.whl Library files that the package depends on

actually , The result of compilation , The content is the same as that of this precompiled package

however , My compiled package cannot be used , stay github It was mentioned that issue:
https://github.com/PaddlePaddle/Paddle-Lite/issues/9298

4. pip Direct installation ( Suggest this …)

Due to the present py3.8 None of the above precompiled packages , So create a new one conda Environmental Science Python=3.7

then ：

python -m pip install paddlelite==2.11

5. verification ( Conversion part )

PaddleLite You need to convert the static graph model into nb Format , stay linux The environment has opt Tools , stay windows Environmental Science , Script only ：

import paddlelite.lite as lite

a = lite.Opt()

#  Not  combined  form 
# a.set_model_dir("mobilenet_v1")

# combined  form , Specific model and parameter name , Please modify according to the actual situation 
a.set_model_file("model.pdmodel")
a.set_param_file("model.pdiparams")

a.set_optimize_out("pplite")
a.set_valid_places("x86")

a.run()

It should be noted that

• model.pdmodel File with the __model__ File corresponding to the
• model.pdiparams File with the __param__ File corresponding to the

in addition , There are two static graph models ： Not combined form And combined form

The latter is the two above , The former is dark brown ：

__model__ The file is the topology of the model , The other files are parameter files

The script above , Will generate a .nb file

6. verification ( The reasoning part )

Excerpt from PaddleLite file ：
https://paddle-lite.readthedocs.io/zh/latest/api_reference/python_api_doc.html

from paddlelite.lite import *
import numpy as np
from PIL import Image

# (1)  Setting configuration information 
config = MobileConfig()
config.set_model_from_file("./mobilenet_v1_opt.nb")

# (2)  Create predictors 
predictor = create_paddle_predictor(config)

# (3)  Read data from pictures 
image = Image.open('./example.jpg')
resized_image = image.resize((224, 224), Image.BILINEAR)
image_data = np.array(resized_image).transpose(2, 0, 1).reshape(1, 3, 224, 224)

# (4)  Set input data 
input_tensor = predictor.get_input(0)
input_tensor.from_numpy(image_data)

# (5)  Execution forecast 
predictor.run()

# (6)  Get the output data 
output_tensor = predictor.get_output(0)
print(output_tensor.shape())
print(output_tensor.numpy())

Just run