Search found 488 matches

by bryanlyon
Sun Nov 29, 2020 7:06 am
Forum: Hardware
Topic: MULTI GPU - double speed at same batc size or not?
Replies: 6
Views: 30

Re: MULTI GPU - double speed at same batc size or not?

SLI is just for gaming and does NOT make the system treat it as a single card. Faceswap specifically handles multiple GPUs allowing for each card to compute different images at the same time. Many of our users (including myself) have used MultiGPU in Faceswap. It definitely works and increases the s...
by bryanlyon
Sat Nov 28, 2020 8:25 pm
Forum: Hardware
Topic: MULTI GPU - double speed at same batc size or not?
Replies: 6
Views: 30

Re: MULTI GPU - double speed at same batc size or not?

We do not support DFL. Faceswap does get greatly increased speed on a properly setup Multi GPU system. It will not cut your time in half as there are a lot of things that cannot be parallelized but it will reduce total time per iteration.

by bryanlyon
Sat Nov 28, 2020 6:29 am
Forum: Swap Showcase
Topic: Bill Murray as Han Solo
Replies: 4
Views: 1249

Re: Bill Murray as Han Solo

No, as he said in the description

Made with https://faceswap.dev/
Villan model 450K iterations.

by bryanlyon
Sat Nov 28, 2020 6:26 am
Forum: Convert
Topic: DLight model still blurry after 400k iterations
Replies: 4
Views: 496

Re: DLight model still blurry after 400k iterations

Do not do that. It is the exact opposite of what you should do for a quality swap. (Keep A and B separate)

by bryanlyon
Sun Nov 22, 2020 5:47 am
Forum: Training
Topic: Dark or shadow on faces.
Replies: 2
Views: 69

Re: Dark or shadow on faces.

Was it an HDR rip that you were training on? If so you probably gave it different data than the data in the trailer. But I'd first try turning up the brightness and contrast on the trailer to see if you can get a clearer swap, then turn it back in postprocessing.

by bryanlyon
Sun Nov 22, 2020 5:44 am
Forum: Training
Topic: Good settings for keeping Glasses
Replies: 1
Views: 56

Re: Good settings for keeping Glasses

This is down mostly to your choice of masks. Most masks will keep glasses as long as they're in all the data, but there is no guarantee that they'll look right at the end.

by bryanlyon
Thu Nov 19, 2020 6:59 pm
Forum: Training
Topic: crash report
Topic: crash report
Replies: 3
Views: 57

Re: crash report

DO NOT RUN FACESWAP WITH ADMIN.

The error is, as you said, a permissions error. This is usually caused by your antivirus or cloud sync software locking this file which is interfering with moving the logs.

by bryanlyon
Thu Nov 19, 2020 6:14 am
Forum: Hardware
Topic: Replacing graphics cards
Replies: 1
Views: 121

Re: Replacing graphics cards

Except for viewtopic.php?f=14&t=1043 . As long as you stay Nvidia you'll be fine.

by bryanlyon
Tue Nov 17, 2020 4:34 pm
Forum: General Chat
Topic: Can you deep fake two diffrent people on the screen at the same time?
Replies: 3
Views: 94

Re: Can you deep fake two diffrent people on the screen at the same time?

Yes, you can. What you would do is convert once for one face, then use that same video to convert for the second face.

by bryanlyon
Mon Nov 16, 2020 8:20 pm
Forum: Extract
Topic: You do not have enough GPU memory
Replies: 7
Views: 214

Re: You do not have enough GPU memory

Yes, that is exactly your problem. Try the steps included in that message. 11/16/2020 19:54:11 MainProcess MainThread launcher execute_script ERROR You do not have enough GPU memory available to run detection at the selected batch size. You can try a number of things: 11/16/2020 19:54:11 MainProcess...
by bryanlyon
Mon Nov 16, 2020 3:54 pm
Forum: Training
Topic: First attempt: Weird learning graph
Replies: 3
Views: 91

Re: First attempt: Weird learning graph

No, the float has gotten "stuck" in a representation of infinity that's used for error states. Local minimums can recover, infinity cannot.

by bryanlyon
Mon Nov 16, 2020 3:52 pm
Forum: General Chat
Topic: Warning on Ampere (30xx series) cards.
Replies: 1
Views: 2404

Re: Warning on Ampere (30xx series) cards.

Tensorflow has released 2.4 RC1. This is a good first step toward support. However, we're still waiting on the full version to be released and deployed to pip/Conda.

by bryanlyon
Sat Nov 14, 2020 3:06 am
Forum: Hardware
Topic: Hardware best practices
Replies: 36
Views: 16832

Re: Hardware best practices

It's impossible to specify one card vs another to that degree. It depends on a lot. Generally read the guide at the beginning of this thread and if you have SPECIFIC questions feel free to ask. But that's a very vague (and circumstantial) question.

by bryanlyon
Fri Nov 13, 2020 11:18 pm
Forum: Convert
Topic: Converted Clip is about Half Length
Replies: 1
Views: 60

Re: Converted Clip is about Half Length

Is it missing frames or is it just going faster?

by bryanlyon
Thu Nov 12, 2020 7:11 pm
Forum: Training
Topic: Modifying training data for partially trained model
Replies: 3
Views: 139

Re: Modifying training data for partially trained model

Yes, go ahead and add data. That's fine and a normal part of training.

by bryanlyon
Thu Nov 12, 2020 4:07 pm
Forum: Hardware
Topic: AMD Vega 64 issues
Replies: 3
Views: 96

Re: AMD Vega 64 issues

Exccluding a card is only used for Tensorflow not PlaidML. Not sure how PlaidML does it. With reducing the power used by your card (assuming you're on Linux because of your rocm mention) check out https://www.phoronix.com/scan.php?page=news_item&px=Linux-4.20-Increase-AMD-GPU-TDP which unlocks y...
by bryanlyon
Tue Nov 10, 2020 6:10 pm
Forum: Hardware
Topic: What Do you think of this MB
Replies: 4
Views: 223

Re: What Do you think of this MB

The only way for a motherboard to have multiple x16 PCI-E lanes is for the CPU to have a lot of lanes. Some Threadripper and Xeon boards can support that. All consumer chips, however, have too few PCI-E lanes to support 2 x16 PCI-E ports (after some are used for the southbridge/USB ports/SATA connec...
by bryanlyon
Tue Nov 10, 2020 6:04 pm
Forum: Training
Topic: Double Eyebrow still
Replies: 3
Views: 184

Re: Double Eyebrow still

With that high of eyebrows you're going to need to increase coverage. They're just too far up the face for default coverage to capture.