neuralchen / SimSwap

An arbitrary face-swapping framework on images and videos with one single trained model!
Other
4.41k stars 872 forks source link

Can't get Simswap to work after upgrading PC #452

Open FrisbeeDog1 opened 10 months ago

FrisbeeDog1 commented 10 months ago

Hey,

I setup Simswap about a year ago following this guide: https://youtu.be/1NWJz_rFMq0?si=SkbmKXk9gZLrabpt It worked perfectly up until I upgraded my PC.

I've seen online that many people had issues after upgrading their GPU, I went from a 2070 Super to a 4090.

My original issue was that Simswap would get stuck at "END" when trying to process a video, but in trying to fix it I've faced many other issues.

I've tried starting from scratch many times following the instructions here: https://github.com/neuralchen/SimSwap/blob/main/docs/guidance/preparation.md And the original YouTube guide listed above, but I always run into errors now and have no idea what I'm doing.

Is there an updated list of instructions I could use to get this working with my upgraded PC?

Thanks.

instant-high commented 9 months ago

Have you found a solution? I've upgraded today from gtx1050 to rtx3060 but have not tested. RVC working "out of the box" Tpsmm needs to update cuda I think

FrisbeeDog1 commented 9 months ago

Unfortunately not, posting here was my last resort but no one has got back to me on this :/

instant-high commented 9 months ago

Next weekend I will setup the new pc.

instant-high commented 9 months ago

@FrisbeeDog1

Ok, I just got another project working on RTX3060 (Thin plate spline motion model modded for face swap) Simswap will be next...

  1. You can try creating anaconda environment version 3.8 instead of 3.6
  2. pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu117

How it worked for the other project: I have created an environment 3.8 then copied the whole existing old environment into this new folder but kept existing files! At last pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu117

I have not tried onnxruntime-gpu (in my other project I use same detection/alignment as simswap) onnxruntime is fast enough so far

instant-high commented 9 months ago

@FrisbeeDog1

I've got it working on new RTX3060, maybe this will help you:

conda create -n cuda python==3.8 pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu117 pip install imutils==0.5.4 pip install numpy==1.21.6

I also had to update ./modules folder and arcface_checkpoint.tar

I'm not sure which version of cuda you need for 4090

Thomas

FrisbeeDog1 commented 9 months ago

Thanks for the updates!

I'll be sure to try this soon :)

neuralchen commented 9 months ago

What is your problem?