MrDeepFakes Forums

Some content may not be available to Guests. Consider registering an account to enjoy unrestricted access to guides, support and tools

  • We are looking for community members who are intested in helping out. See our HELP WANTED post.

Are model files across platforms compatible?

mondomonger

DF Admirer
Verified Video Creator
Trying to upgrade from Fakeapp 2.2 today.   I've been able to get both OpenFaceSwap and MyFakeApp working properly, but my Fakeapp 2.2 model files (all done at Fakepp 2.2 defaults:  Nodes 512, Layers 4) are not recognized.


Are there particular settings I need to use to get older Fakeapp models to work with OpenFaceSwap/MyFakeApp?
 

wifepics

DF Vagrant
Repeat from another thread. but just to make sure you get my answer....

I believe they may be fundamentally different. I know that MyFakeApp specifically is a Python script. I don't know if FakeApp2.2 is Python. If they are both Python, I'm not sure if it's worth your time to try to learn the syntax nuances between the two. I would just start training a new model in MyFakeApp. Just use the same face sets. You won't need to re-extract faces. Expect the same amount of training time as FakeApp2.2. But you should see generally more accurate replication that FakeApp2.2.
I have had some people help me with personal fakes using different apps. so far, the MyFakeApp produces the most accurate model I have seen. OpenFaceSwap almost "cartoonizes" the face. FakeApp makes almost too robotic face expressions. Of course, knowing my subject personally, i'm able to recognize if an app is replicating facial expressions like the real person.
So far, MyFakeApp is the only one that has made me go "wow, it really nailed that."
 

mondomonger

DF Admirer
Verified Video Creator
Thanks for the advice. I'm about 8 hours into training my "base model" that I use to start all new models with MyFakeApp with a GTX1080. I assume I am 24-48 hours away from having a useful model file and then maybe a month away (!) from converting old model files over to MyFakeApp.

Question: I see that MyFakeApp has a github page with updates. I'm good with computers, but no programmer. Will updates to MyFakeApp's Feb 18, 2018 release only be available through Github type hacking or will a new release eventually come out?

Thanks for the reply by the way. Most forums these days are dead.
 

dpfks

DF Enthusiast
Staff member
Administrator
Verified Video Creator
mondomonger said:
Thanks for the advice.    I'm about 8 hours into training my "base model" that I use to start all new models with MyFakeApp with a GTX1080.    I assume I am 24-48 hours away from having a useful model file and then maybe a month away (!) from converting old model files over to MyFakeApp.

Question:   I see that MyFakeApp has a github page with updates.    I'm good with computers, but no programmer.    Will updates to MyFakeApp's Feb 18, 2018 release only be available through Github type hacking or will a new release eventually come out?

Thanks for the reply by the way.   Most forums these days are dead.

A simple way to update the app is to just download the .zip file from github, and replace your files that you've installed. I think faceapp and faceswap models are only interchangeable. I am not too sure about the rest but I am looking to change to a new app too (currently using faceswap).
 

LingCHN19

DF Vagrant
dpfks said:
A simple way to update the app is to just download the .zip file from github, and replace your files that you've installed. I think faceapp and faceswap models are only interchangeable. I am not too sure about the rest but I am looking to change to a new app too (currently using faceswap).

Do you mind if I ask a question about Faceswap, do you know how to compile dlib with CUDA support?

I followed tutorials and tried several times, but even after it's built successfully, dlib-cnn is still very slow for me. I don't know how to fix it, so I can't use dlib-cnn in Faceswap.
 

Pocketspeed

DF Admirer
Verified Video Creator
LingCHN19 said:
dpfks said:
A simple way to update the app is to just download the .zip file from github, and replace your files that you've installed. I think faceapp and faceswap models are only interchangeable. I am not too sure about the rest but I am looking to change to a new app too (currently using faceswap).

Do you mind if I ask a question about Faceswap, do you know how to compile dlib with CUDA support?

I followed tutorials and tried several times, but even after it's built successfully, dlib-cnn is still very slow for me. I don't know how to fix it, so I can't use dlib-cnn in Faceswap.

dlib is probably causing tensorflow to use CPU and not GPU, hence the slow speed.  So here's what worked for me (everything gets installed in the default directories):


Note: always add to PATH if asked during installations, even Anaconda (conda will warn you when you check the box. Ignore it.) All commands should be run through your virtual environment (see below). If you get messages about upgrading pip, don't. You want pip version 10. Just ignore the warnings.

I use Anaconda 3 with a Python 3.6.5 build on Win10 machine. Install CUDA 9.0. Install cudNN 7.05 (for CUDA 9.0). Install Visual Studio 2015. Install Cmake (latest stable build, I think it's 3.12). Install OpenCV. Make sure all your environment variables are set to PATH correctly. Not technically necessary in Anaconda, but for utility and safety purposes, create and activate a virtual environment through the conda prompt. In your virtual env, at the conda cmd line, enter "chcp 866", then after that enter "chcp 65001" (No quotes, of course.) If chcp is not a recognized internal or external command by conda, in your environment variables add to PATH [font=Arial,]"C:\Windows\System32\". [/font]Now get the Faceswap files. I prefer to use the git for Faceswap. (If you don't know how, Youtube is your friend. It's easy. Search github tutorial.) I do not use the docker file. Still in your virtual env, now type "python setup.py" and let it do its thing. Watch for errors. If you have an error or errors, you can usually find good solutions by simply Googling the error message. When it's done, if you have no errors (or you've fixed them), still in your virtual env, type "pip uninstall dlib". Let it uninstall. Then get the latest dlib version from github (it's the davisking\dlib one). This time I downloaded the .zip file and extracted to my virtual env folder in Anaconda. Now type "python setup.py install --yes DLIB_USE_CUDA". Let it do it's thing. Once that is done, if you had no errors, you will need to change your tensorflow version, as it will not be compatible with cudNN 7.05. Type "pip install --upgrade --force-reinstall tensorflow-gpu==1.9.0  --user". Now you will find that face_recognition needs to be updated. Simply type "pip install face-recognition --upgrade". Now you should be good to go.

If this still doesn't work, then it is possible that you have another version or versions of dlib somewhere on your PC. You will need to find those, and remove them.

I hope this was helpful.
 
Top