ImportError: No module named pathlib

[nerdprogrammergm]

I have already installed pathlib in python3.6:Requirement already satisfied: pathlib in /usr/local/lib/python3.6/dist-packages

Command executed: python3 faceswap.py extract -i ~/faceswap/photo/trump -o ~/faceswap/data/trump

Traceback (most recent call last):
File "faceswap.py", line 3, in
from scripts.extract import ExtractTrainingData
File "/home/ubuntu/data/faceswap/scripts/extract.py", line 2, in
from lib.cli import DirectoryProcessor
File "/home/ubuntu/data/faceswap/lib/cli.py", line 6, in
from lib.utils import get_image_paths, get_folder, load_images, stack_images
File "/home/ubuntu/data/faceswap/lib/utils.py", line 4, in
from pathlib import Path
ImportError: No module named pathlib

Can anyone help me out with this issue?

[yangchen8710]

use python3.6 instead of python3?

[Ganonmaster]

Make sure you are running python 3.6 or otherwise try adding pathlib through pip:

pip install --upgrade pathlib

[Ganonmaster]

Otherwise, try running the script in a virtualenv as described in the guide.

[nerdprogrammergm]

Thanks for quick reply.
I figured out the issue and now working on further part.
Presently I am getting this error:
Using TensorFlow backend.
usage: faceswap.py [-h] {extract,train,convert} ...
faceswap.py: error: invalid choice: 'extract/trump' (choose from 'extract', 'train', 'convert')

I have extracted the faces in extract/trump and extract/cage,
using this command to train the model: python3.6 faceswap.py -A extract/trump -B extract/cage -m models -p

What wrong am I doing?

[lukaville]

Most likely you are trying to start the script with invalid arguments. You have to specify 'extract', 'train' or 'convert' right after faceswap.py

[nerdprogrammergm]

I am following the instructions mentioned on this lnk:https://github.com/deepfakes/faceswap/blob/master/USAGE.md

which says to execute the following command to train the model!
python3.6 faceswap.py -A ~/faceswap/data/trump -B ~/faceswap/data/cage -m ~/faceswap/models/

[lukaville]

Looks like it's a typo in the doc. Use python faceswap.py train -A ~/faceswap/data/trump -B ~/faceswap/data/cage -m ~/faceswap/models/ instead.

[nerdprogrammergm]

Got the issue, its the "train" missed out from the command!
This command should be used to train the model: python3.6 faceswap.py train -A extract/trump -B extract/cage -m models -p
@lukaville @yangchen8710 @Ganonmaster

[nerdprogrammergm]

Yes, thanks man. Have you tried to setup this on aws ec2 p2 instance?
Overthere, I am dealing with the issue for cuda9, got to wait until tensorflow support cuda9.
For cuda8, I tried but somewhere dlib through error about lib related to 9.so . Try to troubleshoot it but found that its tensorflow which is not compatible with cuda9 yet :/
You got any solution for this?

[nerdprogrammergm]

There is also a typo in the doc for converting:
Using TensorFlow backend.
usage: faceswap.py [-h] {extract,train,convert} ...
faceswap.py: error: invalid choice: 'data/trump' (choose from 'extract', 'train', 'convert')

Proper command to convert is: python3.6 faceswap.py convert -i data/trump -o output/trump -m models

@Ganonmaster @yangchen8710

Sorry for tagging ya all, m noob at using github too!

[lukaville]

@nerdprogrammergm I successfully used floydhub, I made a post about it https://www.reddit.com/r/deepfakes/comments/7mgpki/floydhub_cloud_training/

[nerdprogrammergm]

Cool @lukaville , great job.
Will give it a try ;)
In the mean time can you have a look on this link and help me out?
deepfakes/faceswap-playground#10
Thanks once again.

[Ganonmaster]

The doc changes have been merged. I'll look at your other issue shortly. 👍