Faceswap Forum

I have prepared the CPU-only docker environment (python3.6, tensorflow1.12, and other libraries as in the requirements.txt) (https://github.com/deepfakes/faceswap/). I prepared the image data (around 600 and 850 PNG files) and tried to start training a new model with the command below.

python3.6 faceswap.py train -A "mydir/faces/b_4" -B "mydir/faces/a" -m "mydir/model "

But the training seemed to stop without any error messages or crash-report. Is there any clue for what should I do next ?

Below is the message in the terminal after running the command.

Setting Faceswap backend to CPU                                                                                                            

02/14/2020 22:10:17 INFO     Log level set to: INFO                                                                                        

Using TensorFlow backend.                                                                                                                  

02/14/2020 22:10:22 INFO     Model A Directory: /srv/mydir/faces/b_4                                                                       

02/14/2020 22:10:22 INFO     Model B Directory: /srv/mydir/faces/a                                                                         

02/14/2020 22:10:22 INFO     Training data directory: /srv/mydir/model                                                                     

02/14/2020 22:10:22 INFO     ===================================================                                                           

02/14/2020 22:10:22 INFO       Starting                                                                                                    

02/14/2020 22:10:22 INFO       Press 'ENTER' to save and quit                                                                              

02/14/2020 22:10:22 INFO       Press 'S' to save model weights immediately                                                                 

02/14/2020 22:10:22 INFO     ===================================================                                                           

02/14/2020 22:10:23 INFO     Loading data, this may take a while...                                                                        

02/14/2020 22:10:23 INFO     Loading Model from Original plugin...                                                                         

02/14/2020 22:10:23 INFO     No existing state file found. Generating.                                                                     

02/14/2020 22:10:25 INFO     Creating new 'original' model in folder: '/srv/mydir/model'                                                   

02/14/2020 22:10:25 INFO     Loading Trainer from Original plugin... 
02/14/2020 22:10:29 INFO     Enabled TensorBoard Logging                                                                                   

Killed                        

Docker isn't officially supported. You're welcome to try Windows, Linux or macOS,

The "Killed" message implies something else killed the operation. You can see if anything caused it to be killed by checking the docker system logs. Unfortunately, I don't know what that is and I doubt you'll find anything in Faceswap's logs. You may TRY setting Faceswap to a high logging level while training and see if it gives you any information in the faceswap.log file.

That said, CPU training (especially in a docker) is going to be an absolute nightmare. I'd fully expect it to take you around a month or two to get even a basic result. We HIGHLY recommend an Nvidia graphics card for training.

deephomage wrote: ↑Fri Feb 14, 2020 11:07 pm

Docker isn't officially supported. You're welcome to try Windows, Linux or macOS,

Thank you for your reply ! The docker environment that I am using is based on Ubuntu 16.04 running from macOS. I will try other environment someday ^^

bryanlyon wrote: ↑Sat Feb 15, 2020 3:26 am

The "Killed" message implies something else killed the operation. You can see if anything caused it to be killed by checking the docker system logs. Unfortunately, I don't know what that is and I doubt you'll find anything in Faceswap's logs. You may TRY setting Faceswap to a high logging level while training and see if it gives you any information in the faceswap.log file.

That said, CPU training (especially in a docker) is going to be an absolute nightmare. I'd fully expect it to take you around a month or two to get even a basic result. We HIGHLY recommend an Nvidia graphics card for training.

Thank you for your reply ! As you said, The faceswap.log does not seem to have information more than the message on terminal. I will take a look on the docker system logs and try setting the logging level.

I understand that training needs much computation cost and time to get some good results. Rightnow I have no good environment and want to start from self-preparing data as practice.

I would recommend setting up in Anaconda on Mac