Deep Live cam Notes

Project address: https://github.com/hacksider/Deep-Live-Cam

This note will not cover cuda, Python, Conda, Git, and other related computer knowledge. There are only records of steps and problems encountered when using the deployment.

The following operations are all performed under the Windows operating system. My system version number is:

Microsoft Windows 11 Pro
10.0.22631 Build 22631

image-20240906221442070.png

Prerequisites

  1. Desktop or laptop computer with Nvidia graphics card

  2. Stable network environment

  3. Software or compilation tools required by DeepLiveCam

    • Python 3.10.0
    • CUDA
    • cuDNN
    • Git
    • ffmpeg
    • Conda-Py 3.10
    • Microsoft C++ Build Tools

  4. DeepLiveCam project ontology

  5. Model download

Environment installation

Python 3.10.0

Download address: https://www.python.org/ftp/python/3.10.0/python-3.10.0-amd64.exe

Just install it like a normal application software, remember to check Add to System Variables

CUDA

Download address: https://developer.nvidia.com/cuda-toolkit-archive

After the installation is complete, run nvidia-smi on the command line to check the maximum CUDA version supported by the graphics card.

nvidia-smi

image-20240906212930253.png

cuDNN

You need to download a version supported by CUDA

Download address: http://developer.nvidia.com/rdp/cudnn-archive

After the installation is complete, run nvcc -V cuDNN version number on the command line.

If the prompt command does not exist, please check the system’s environment variables and user variables

nvcc -V

image-20240906213133117.png

Git

Download address: https://git-scm.com/downloads

After the installation is complete, run git --version on the command line to check the Git version number.

git --version

If the prompt command does not exist, please check the system’s environment variables and user variables

image-20240906213419363.png

Conda

Download address: https://docs.anaconda.com/miniconda/

Select Miniconda3-py310_24.7.1-0-Windows-x86_64.exe

After the installation is complete, run conda --version on the command line to check the conda version number.

conda --version

image-20240906214102387.png

Note: If the prompt command does not exist, please check the environment variables and user variables

The path is as follows: C:\ProgramData\miniconda3\Scripts

ffmpeg

Download address: https://www.ffmpeg.org/download.html

After the installation is complete, run ffmpeg --version on the command line to check the conda version number.

ffmpeg -version

image-20240906214520225.png

Note: If the prompt command does not exist, please check the environment variables and user variables

The path is: yourpath\ffmpeg-6.1.1-essentials_build\bin

Microsoft C++ Build Tools

Download address: https://visualstudio.microsoft.com/visual-cpp-build-tools/

During the installation process, make sure “C++ build tools” and “Windows 10 SDK” (or equivalent) are selected

image-20240906214754523.png

Modify the execution policy of Windows PowerShell

  1. View current execution strategy

    Get-ExecutionPolicy -List

  2. ** Set execution policy **: 可以将执行策略设置为 Unrestricted 或 RemoteSigned。Unrestricted 允许所有脚本运行，而 RemoteSigned 仅要求从互联网下载的脚本必须经过签名。

    • 设置为 Unrestricted：

      Set-ExecutionPolicy Unrestricted -Scope LocalMachine

    • 或者设置为 RemoteSigned：

      Set-ExecutionPolicy RemoteSigned -Scope LocalMachine

  3. ** Confirm change **: 在提示时输入 Y 确认更改。

  4. ** Verification execution policy **: 再次运行以下命令以确认执行策略已更改：

    Get-ExecutionPolicy -List

Project deployment

The administrator runs PowerShell to create an English directory at the location you want to specify on the command line.

cd to the directory you created to clone the project

Note: The entire environment should not have Chinese directories or names

  1. Clone project

    git clone https://github.com/hacksider/Deep-Live-Cam.git

  2. CD to project directory

    cd /yourpath/Deep-Live-Cam

  3. Use conda to create an env named DeepLiveCam

    #初始化conda 环境
conda init 

#创建DeepLiveCam的 env
conda create -n deeplivecam python=3.10.0

#查看env 列表
conda env list

#根据提示按Y
#等待创建完成之后激活env
conda activate deeplivecam  #激活之后，命令行前会显示(deeplivecam)

  4. Install DeepLiveCam project requriements.txt

    确认在项目目录内 yourpath/deeplivecam 执行以下命令

    #安装环境包 （需要完全稳定的互联网）
pip install -r requirements.txt

#等待安装完成之后 重新安装一下以下软件包
pip uninstall onnxruntime onnxruntime-gpu

#指定安装onnxruntime-gpu版本
pip install onnxruntime-gpu==1.16.3

  5. Copy model file

    如果以上步骤和命令都已经完成安装且没有任何报错，将之前下载好的模型文件拷贝到项目的models 目录下, 文件树如下

     Directory: D:\Deep-Live-Cam\models


Mode                 LastWriteTime         Length Name
---- ------------- ------ ----
-a---- 9/2/2024   9:09 AM      348632874 GFPGANv1.4.pth
-a---- 9/2/2024  10:27 PM             34 instructions.txt
-a---- 9/2/2024   9:11 AM      277288649 inswapper_128_fp16.onnx

Start the DeepLiveCam project

The administrator runs Windows Power Shell and executes the following command in Windows Power Shell

#进入DeepLiveCam项目目录
cd D:\Deep-Live-Cam\  #根据自己的实际路径操作
conda activate deeplivecam # 激活deeplivecam 的env
python run.py --execution-provider cuda #运行deeplivecam 项目

Finally, select a photo in the project interface for Live verification and the effect is achieved

image-20240906221442070.png

Note: Other functions in the function options need to download other resources when running for the first time. Please check the stability of the network.