Faceswap Forum

p96xl

Is there a way for the eyes to look in the same direction as the model? I dont know how to make it seem more real if the eyes are always looking at the camera, which in B in the picture is doing. Is there an extra setting, or is my training data bad?

p96xl

Whats the difference between all the loss functions? and which one will give me the best results? Also, does Penalize Mask Loss improve results?

p96xl

bryanlyon wrote: ↑Wed Oct 02, 2019 3:17 am
Yeah, problems like that DEFINITELY sound like power issues. Might be a good idea to try some underclock just to even those out.

I slaped in a second PSU that I had, its 650W. Works like a charm on batch size of 32. I guess my PC is some special case.

p96xl

Hmmm, changing the batchsize is unlikely to cause shutdown issues. Lowering the mem clock speed is more likely. However, the important thing is it works now :) My PC is kind of weird, if it doesn't like something then it shuts down. e.g. batch size too high? shutdown. batch size too small? shut dow...

p96xl

Vega FE can easily use 450w and your i5 6600K can easily use 200. Beyond that, many PSUs struggle with peak load and sudden voltage demands which can cause crashes. Unfortunately, we can't do anything to prevent or control hardware issues like this. If you want to try troubleshooting the first step...

p96xl

torzdf wrote: ↑Sun Sep 29, 2019 11:31 pm
If your PC shutsdown it's basically a hardware issue in some way.

Yea, but its random, like it happens at random times

p96xl

torzdf wrote: ↑Sun Sep 29, 2019 11:15 pm
This is a hardware issue. Most likely insufficient power.

The TDP of a Vega FE is just 300W, my PSU is 750W. Im pretty sure its not a power issue, but I could be wrong.

p96xl

When I'm training on RealFace, my PC just randomly decides to turn off and stays off until I press the start button. When I take a closer look, I get error code 0x00000000 and no reason. What can be the cause? Running AMD Vega FE with Intel i5 6600k and 4x2 HyperX ram

p96xl

The following is a simplified explanation. There is more to it than this (the structure of the model etc), but to keep it in simple terms, I will talk purely in terms of resolution. H128 has double the resolution. of the original model. You would think that this would mean it would take twice as lo...

p96xl

To get 100 iterations, it would take 6:45 on my Vega FE. To get 100000 iterations with batch size of 32, it would take almost 5 days. None of the other training methods do this like the original. For original, 100000 iters it would take 16.666 hours with the same setup of batch size of 32. Is H128 s...

p96xl

torzdf wrote: ↑Tue Sep 24, 2019 8:15 am
Not really. Any Overclocks, etc. are detrimental to Machine Learning and can lead to model collapse

What about just the workflow? any way to optimize that? Like batch size in specific things or should I just experiment?

p96xl

Just wondering if there are any ways to optimize my Vega FE in training to either give me better results or faster results.

Faceswap Forum

Search found 12 matches

Eye tracking?

Loss Function

Re: PC crash

Re: PC crash

Re: PC crash

Re: PC crash

Re: PC crash

PC crash

Re: dfl-H128 taking forver to train

dfl-H128 taking forver to train

Re: Optimizations for Vega FE?

Optimizations for Vega FE?