C++ program with embedded python running SSD model on tensorflow v2
Command line program developed in MSVC 2019. The program requires the use of tensorflow v2 and opencv as well as numpy. The program is intended to be compiled as a 64 bit release version and the executable will reside in the x64/Release subdirectory. Included in this subdirectory is a sample jpg file for use by the test program. It is necessary to place a downloaded SSD model in the subdirectory as well for the code to be able to find it. The SSD model used during testing was taken from the tensorflow model zoo, ssd_mobilenet_v2_320x320_coco17_tpu-8. The model should be unpacked and placed in the x64\Release subdirectory so the program can find it.
The python portion of the code is also found in the executable subdirectory and may be run independently using the command python model.py to confirm results.
To compile the program, environment variables are used to find include and lib files. $(CONTRIB_PATH) has the location of the opencv files similar to the directory fragment shown below. The PATH environment variable should be appended to include the contrib\bin folder as well for runtime operation.
C:\Users\sr996\contrib
bin
opencv_world451.dll
include
opencv2
lib
opencv_world451.lib
There are additional environment variables required. $(PYTHON_PATH) should point to the python installation directory e.g.
C:\Users\sr996\AppData\Local\Programs\Python\Python36
and $(NUMPY_PATH)
C:\Users\sr996\AppData\Local\Programs\Python\Python36\Lib\site-packages\numpy\core
This program uses code taken from the article published on CodeProject Embedding Python program in a C/C++ code
Python packages used
Python 3.6.8
opencv 4.4.0.42
tensorflow 2.6.0
numpy 1.19.5
To use with a conda installation the following instructions may be helpful
Using the conda prompt, build a conda environment with tensorflow-gpu
conda create --name tf tensorflow-gpu
activate tf
conda install -c conda-forge opencv
set PYTHON_PATH=C:\Users\sr996\anaconda3\envs\tf
set NUMPY_PATH=C:\Users\sr996\anaconda3\Lib\site-packages\numpy\core
set PYTHONHOME=%PYTHON_PATH
Launch the executable from the conda prompt