Search found 624 matches

by bryanlyon
Mon Aug 02, 2021 10:33 pm
Forum: Training Discussion
Topic: Improving quality of sources as training progresses
Replies: 2
Views: 23

Re: Improving quality of sources as training progresses

This is an aspect of what I call "fit training". See the guide https://forum.faceswap.dev/viewtopic.php?f=27&t=74 for some more explanation of the process. Basically once fully trained, you train on just the swap video (and in your case, your best data for B) and let it try to master t...
by bryanlyon
Mon Aug 02, 2021 10:30 pm
Forum: Extract Support
Topic: CUDA driver version is insufficient for CUDA runtime version
Replies: 7
Views: 4119

Re: CUDA driver version is insufficient for CUDA runtime version

That will not solve your problem. The problem is that your driver is too old for Cuda. You could try the v1.0 of Faceswap which uses an older Tensorflow (and therefore Cuda) version, but there is no support for that version anymore. You simply need to upgrade your drivers if you want to use the curr...
by bryanlyon
Mon Aug 02, 2021 10:26 pm
Forum: Training Support
Topic: Using one GPU has more VRAM than running two?
Replies: 2
Views: 23

Re: Using one GPU has more VRAM than running two?

Your error is probably not related to an Out Of Memory (even if it says it is). We don't recommend using different cards in distributed as they can cause issues like this. However, it is possible that it is a true OOM. Using 2 cards doesn't mean you get double the available RAM. One card still works...
by bryanlyon
Mon Jul 26, 2021 10:41 pm
Forum: Extract Support
Topic: CUDA driver version is insufficient for CUDA runtime version
Replies: 7
Views: 4119

Re: CUDA driver version is insufficient for CUDA runtime version

gpu_driver: 390.143

That is definitely not the latest GPU driver for your GPU. You need to find a more up-to-date driver. I recommend checking at nvidia.com if your distro doesn't have any newer drivers.

by bryanlyon
Sun Jul 25, 2021 4:24 am
Forum: Swap Showcase
Topic: Phaze-A - 256px Demo
Replies: 10
Views: 10166

Re: Phaze-A - 256px Demo

When using legacy faces, the face is extracted from the full image at that percentage then scaled to for the model. In the case of a 256x face at 100% coverage, then the while 256x256 image would be scaled and used to train. If you're using a 256x256 model with less than 100% coverage you may need h...
by bryanlyon
Wed Jul 21, 2021 3:39 am
Forum: Training Discussion
Topic: Dual GPU: 3090 FE Installed, any use of keeping 2080 Ti as Dual?
Replies: 3
Views: 3163

Re: Dual GPU: 3090 FE Installed, any use of keeping 2080 Ti as Dual?

I would not recommend mixing. You'll be limited by the speed, ram, and featureset of the worst card option. In general you'll be better served by using the extra GPU on a separate system.

by bryanlyon
Wed Jul 21, 2021 3:34 am
Forum: Installation Support
Topic: Can I install and use Faceswap on Nvidia Jetson?
Replies: 4
Views: 2543

Re: Can I install and use Faceswap on Nvidia Jetson?

Jetson Nano wont be feasible. It simply doesn't have much ram at 2 or 4gb depending on version. With a TX1 or TX2 it's possible, but you'll have to either use a much older version or modify it to remove the pynvml checks. Unfortunately it wont be an easy solution to run on any Jetsons without modifi...
by bryanlyon
Mon Jul 12, 2021 11:56 pm
Forum: Swap Showcase
Topic: Phaze-A - 256px Demo
Replies: 10
Views: 10166

Re: Phaze-A - 256px Demo

Quick question @torzdf - Did you use "no warp" at the end of training for this video? Or was it simply a run of 2M iterations with the StoJo preset model? Torzdf has said in the past that he regularly turns off warp for the last part of training. I would guess that he did the same here. Y...
by bryanlyon
Thu Jul 08, 2021 10:02 pm
Forum: Convert Support
Topic: Strange Performance Limitation Observation (Windows)
Replies: 3
Views: 2405

Re: Strange Performance Limitation Observation (Windows)

System ram is not really critical importance. OOMs apply to vram and not to system ram.

by bryanlyon
Wed Jul 07, 2021 4:59 pm
Forum: Installation Support
Topic: [Guide] Using Faceswap on Nvidia RTX 30xx cards
Replies: 109
Views: 66143

Re: CuDNN instal?

rolnor wrote: Wed Jul 07, 2021 3:18 pm

I am following the protocol for installing FS with a 30xx-card, how do I install cuDNN 8.0? I have downloaded the file and extracted but I dont see any .exe file?

Follow the instructions here: https://docs.nvidia.com/deeplearning/cu ... ll-windows

by bryanlyon
Mon Jun 21, 2021 4:21 am
Forum: Training Discussion
Topic: Putting single photo B face on model?
Replies: 2
Views: 3820

Re: Putting single photo B face on model?

One image just isn't how FS works. It will never get unblurry with a single image. I'd suggest if you want to do that go check out First Order Model or "one-shot" solutions.

by bryanlyon
Mon Jun 21, 2021 4:19 am
Forum: Training Discussion
Topic: What is most important in lighting data? Shadows? Resolution?
Replies: 3
Views: 2635

Re: What is most important in lighting data? Shadows? Resolution?

Loss is a poor measure, especially if you control one side of the data. I'd say that the most important thing is Variety. Variety is key because the model can learn all the different types of lighting itself. I'd suggest trying to move a light across the face to get different angles. Getting some fi...
by bryanlyon
Mon Jun 21, 2021 4:01 am
Forum: Training Support
Topic: How much memory for 256 resolution Dfaker?
Replies: 1
Views: 4621

Re: How much memory for 256 resolution Dfaker?

Unfortunately it's not an easy question to answer. It depends on FAR too many variables. You can try it (at a low batch size) and see if it'll work for you. But everything from your OS, to your driver version, to your web browser can effect the amount of vram it takes to train.

by bryanlyon
Mon Jun 21, 2021 3:12 am
Forum: Installation Support
Topic: Problem with new FS version, downgrade
Replies: 3
Views: 2896

Re: Problem with new FS version, downgrade

Model collapses can happen anytime and are COMPLETLEY unrelated to the alignments. In fact, alignments are not ever fed into the model itself, and are only used to store masks and the like. Downgrade is possible, but your data will almost definitely need to be re-extracted and created. To do it you'...
by bryanlyon
Mon Jun 21, 2021 3:07 am
Forum: Extract Discussion
Topic: Masking hats and hair
Replies: 2
Views: 2673

Re: Masking hats and hair

Mask out anything that you don't want to swap in. You might want to check out the new Bisenet masker. It gives you more options and accuracy.

by bryanlyon
Mon Jun 21, 2021 3:05 am
Forum: Training Discussion
Topic: Model collapse while Training Phaze-A
Replies: 2
Views: 2670

Re: Difficulty Training Phaze-A

Unfortunately a model can collapse at any time. I'd suggest trying again, and if you still have problems, try reducing your Learning Rate.

by bryanlyon
Fri Jun 18, 2021 7:42 pm
Forum: Installation Support
Topic: Totally lost on how to install on newest MacBook Air, macOS Big Sur
Replies: 3
Views: 4021

Re: Totally lost on how to install on newest MacBook Air, macOS Big Sur

Unfortunately, we're still at the point where none of the developers have an M1 mac and so cannot develop for it. If anyone who does have an M1 would like to give it a try, you can install https://developer.apple.com/metal/tensorflow-plugin/ the Plugin for tensorflow metal and see if you can get it ...
by bryanlyon
Fri Jun 04, 2021 5:55 pm
Forum: Hardware
Topic: Hardware best practices
Replies: 58
Views: 45767

Re: Hardware best practices

Thank you sir, you are a gentleman and a scholar. Is there a better value card than the 3060 ti? Right now desktop GPUs are all but impossible to get ahold of. Honestly, whichever one you can get is probably going to be the one you want ;). I'd suggest checkout out Cloud options or if those aren't ...
by bryanlyon
Fri Jun 04, 2021 5:35 pm
Forum: Hardware
Topic: Hardware best practices
Replies: 58
Views: 45767

Re: Best CPU, Ram And GPU?

Are more than 4 CPU Cores used effectively? Is an 8 core utilized enough to justify it? What about a 16 core? How much Ram is optimal, 16GB, 32GB, more? Is the amount of VRAM on a GPU the most important aspect? I was thinking about getting a NVIDIA 3060 12GB version. Are there better options for th...
by bryanlyon
Mon May 31, 2021 5:26 pm
Forum: Training Support
Topic: Failed to start training
Replies: 4
Views: 4083

Re: Failed to start training

You can copy/paste the error messages when they show. That said I think your problem is a corrupt model file. Try the restore tool.