background

	This article describes how to pack Python depth learning frameworks in Anaconda; use Yolo-Fastestv2
 For the case, record all packaged processes and adapt to CUDA deployment to other computers.

Environment version

win10——x64
CUDA——10.0
Nuitka—— 0.6.16.4
torch ——1.5.1
torchvision—— 0.6.1
tqdm——4.36.1
opencv_python——4.4.0
torchsummary——1.5.1
numpy——1.16.6

Note: Here the packaging process only involves Torch 1.5.1 is possible to run normally.

Project code structure adjustment

• hereYolo-FastestV2The project file directory structure is re-adjusted to the following figure structure, which is easy to pack:

• After the directory structure is adjusted, you need to test the code operation; run Demo.py normally.

Nuitka packaging

• Nuitka common packaging command
• It is best to install MINGW64, the installation process is not described here; no packaging process will be downloaded.
• Ignore this step of installing MingW64- The packaging process will automatically download MINGW64 and other dependence, may not be downloaded, and Baidu link is available here.

Link:
Extraction code: DKDR
…
After downloading, you can decompressed to C: \ Users ***** \ appdata \ local \ nuitka \ nuitka overlays.

• Project package

python -m nuitka --mingw64 --standalone  --show-progress --show-memory  --follow-import-to=need --nofollow-imports   ./demo.py

• Package success will generate 2 folders: demo.dist and demo.build; just leave DEMO.DIST.

Dependence

• Only the BASE environment is currently supported under Anaconda.
• Library and model files required for copy programs

Copying the corresponding library from the E: \ ### \ ### \ Anaconda3 \ lib \ site-packages directory; if you need OpenCV, copy the CV2 folder to the corresponding directory.
- As shown below: The red circle portion is copied from Anaconda to Demo.dist; the green circle is a model related file, etc.

Run test

• The package process is completed, running the Demo.exe program test.
• Other Computers If you use CUDA, you need to call the GPU acceleration.