I don't know if I am just dumb.

[Njenjo]

The installation for python is unclear. Do I follow the model preparation section in the C++ readme file or the python readme file?

[nguyenphucnamgiang]

yeah it was vague, how the quick start create engine for python and in the build too. Have you figure out anything?

[nguyenphucnamgiang]

this is how I set up python environment for this :

1. Create a conda virtual environment with python 3.10.12, then follow the step in https://github.com/spacewalk01/depth-anything-tensorrt?tab=readme-ov-file#depth-anything-v2 .
2. The github should have mentioned installing cuDNN, you can follow this blog to install cuDNN https://blog.csdn.net/qq_42042528/article/details/140591685 .
3. Next, I install tensorrt using zip file like in here https://docs.nvidia.com/deeplearning/tensorrt/install-guide/index.html#installing-zip. It should be noted that in the 5 step you are using pip install tensorrt-8.6.0-cp310-none-win_amd64.whl from the github page .
4. The author use cuda-python, and I found on internet about replacing pycuda with python. However the github code still use pycuda so install pycuda with pip .
5. And that's most of it, you can run most of it, just be aware of the path you are in.

Attention

• There is a video on youtube that might be helpful, although it is kind of out of date, but it's worth refering to https://www.youtube.com/watch?v=cWFOKWIDFJ4

• While trying to make this work, I have made some modifications which may or may not have effects,

  • Copy all dll files from lib to bin

  • Add C:/TensorRT-x.x.x.x/ and C:/TensorRT-x.x.x.x/lib to the path

  • Run pip install onnx_graphsurgeon-0.3.12-py2.py3-none-any.whl