linux 训练DeepFake

参考博客:

https://www.deepfaker.xyz/?p=1202

https://github.com/nagadit/DeepFaceLab_Linux

1. Install Anaconda

Anaconda is the preferred method of installing DeepFaceLab on Linux. Just follow the tutorial.

2. Install System Dependencies

You will need FFMpeg, Git, and the most recent NVIDIA driver for your system to use this project.

If you are here, then you already have everything…

3. Install DeepFaceLab

Just run it in the terminal.

Check latest cudnn and cudatoolkit version for your GPU device.

  1. conda create -n deepfacelab -c main python=3.7 cudnn=7.6.5 cudatoolkit=10.1.243
  2. conda activate deepfacelab
  3. git clone --depth 1 https://github.com/nagadit/DeepFaceLab_Linux.git
  4. cd DeepFaceLab_Linux
  5. git clone --depth 1 https://github.com/iperov/DeepFaceLab.git
  6. python -m pip install -r ./DeepFaceLab/requirements-cuda.txt
  7. cd DeepFaceLab_Linux
  8. # 修改文件权限,如果不修改无法执行(绿色就对了
  9. chmod 777 scripts/*
  10. # 安装ffmpeg
  11. apt install ffmpeg
  12. cd scripts
  13. ./2_extract_PNG_from_video_data_src.sh
  1. linux项目没有默认的dstsrc视频
  2. # 获取测试素材
  3. git clone https://gitee.com/dream80/DFLWorkspace.git workspace

4. Download Pretrain (optional)

Use script 4.1 from the scripts directory.

Or download manually

CelebA

FFHQ

Quick96

5. Navigate to the scripts directory and begin using DeepFaceLab_Linux ᗡ:

Run all scripts with BASH shell

  1. bash 1_clear_workspace.sh

etc

ps:问题

  1. 通过SSH连接无法直接使用Xseg(预处理遮罩文件)

建议:

  1. 将准备好的数据上传到远程服务器,之后再进行训练