python - DeepFaceLab_NVIDIA 在 'data_src faceset extract' - TensorFlow 期间输出错误

Question

在DeepFaceLab从 data_src 和/或 data_dst 中提取面集的第 4 个和/或第 5 个进程输出此错误:

[wf] Face type ( f/wf/head ?:help ) :
wf
[0] Max number of faces from image ( ?:help ) :
0
[512] Image size ( 256-2048 ?:help ) :
512
[90] Jpeg quality ( 1-100 ?:help ) :
90
[n] Write debug images to aligned_debug? ( y/n ) :
n
Extracting faces...

Error while subprocess initialization: Traceback (most recent call last):
  File "C:\DeepFaceLab\DeepFaceLab_NVIDIA\_internal\python-3.6.8\lib\site-packages\tensorflow\python\pywrap_tensorflow.py", line 64, in <module>
    from tensorflow.python._pywrap_tensorflow_internal import *
ImportError: DLL load failed: A dynamic link library (DLL) initialization routine failed.

During handling of the above exception, another exception occurred:

在pywrap_tensorflow.py中的第64行，代码和注释是这样写的:

  # pylint: disable=wildcard-import,g-import-not-at-top,line-too-long,undefined-variable
  try:
    from tensorflow.python._pywrap_tensorflow_internal import *
  # This try catch logic is because there is no bazel equivalent for py_extension.
  # Externally in opensource we must enable exceptions to load the shared object
  # by exposing the PyInit symbols with pybind. This error will only be
  # caught internally or if someone changes the name of the target _pywrap_tensorflow_internal.

类似的问题已经raised in GitHub但找不到具体答案。非常感谢您的帮助。

Answer 1

折腾了好久，终于找到解决办法了。因为我知道训练需要很长时间，所以我决定在没有 GPU 的笔记本电脑上进行训练。出于这个原因，解决方法是先执行第 10 步或名为 10)Make CPU only 的 cmd 文件。这样做会卸载预装的 TensorFlow 并下载旧版本，因此需要 Internet 连接。此外，执行它无法撤消，这意味着 DeepFaceLab 现在将仅通过 CPU 运行 4ever，除非您再次重新安装它。

这应该有望解决问题。