ModuleNotFoundError: No module named 'mxnet'

[ilyastoletov]

I have got this error after installing all dependencies and downloading required weights. I can't install mxnet library itself because breaks dependency chain: mxnet requires numpy less than 1.17.0 version, but much of other libraries requires numpy of greater versions (greater than 1.17.0). This what i got after trying to install mxnet library:

Using Windows 10

[AlexanderGroshev]

Hi, @nagibator-hero!

The first thing you need to do is to determine the version of the CUDA with the command nvcc --version.
Then install the correct version of mxnet with source 1 or source 2.
requirements.txt corresponds to CUDA 10.1.
To find the correct version of mxnet you can look at #56, #60 or try solution with docker #51.

[battlesaint]

For Windows

1. install anaconda, visual studio tools 2019, 2022 with c++ development checked

   Cuda v10.1 here

   Cudnn v8.0.5 here

   Extract cudnn, copy all and paste to C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1

2. open up anaconda prompt, conda create -n sber python=3.6

3. conda activate sber

4. cd F:\AI\ghost (where you want to download)

5. git clone https://github.com/sberbank-ai/sber-swap.git
   cd sber-swap
   git submodule init
   git submodule update

   (one by one)

6. pip install -r requirements.txt

   if error, change requests==2.18.4

7. download models and move to the right directory
   here

8. Open model_zoo.py from C:\Users(username)\anaconda3\envs\sber\Lib\site-packages\insightface\model_zoo and edit line 23

   def get_model(self):
   session = onnxruntime.InferenceSession(self.onnx_file, providers=none)

   to

   def get_model(self):
   session = onnxruntime.InferenceSession(self.onnx_file, providers=['TensorrtExecutionProvider', 'CUDAExecutionProvider'])

9. test and enjoy
   python inference.py --source_paths {PATH_TO_IMAGE} --target_faces_paths {PATH_TO_IMAGE} --target_video {PATH_TO_VIDEO}