Search found 654 matches

by bryanlyon
Thu Nov 25, 2021 12:30 am
Forum: Training Support
Topic: Manually adjusting thousands of frames individually?
Replies: 3
Views: 113

Re: Manually adjusting thousands of frames individually?

Short scenes let you focus your efforts. Even if you're doing large amounts of scenes, I recommend converting one at a time. It lets you tweak settings and efforts to get the best results for each situation.

by bryanlyon
Wed Nov 24, 2021 11:31 pm
Forum: Training Support
Topic: Manually adjusting thousands of frames individually?
Replies: 3
Views: 113

Re: Manually adjusting thousands of frames individually?

Generally you'll only want to fine-tune data that you're using for actual conversion. It's not worth fine-tuning thousands of files for training. Better to just delete bad images rather than tuning them.

by bryanlyon
Wed Nov 24, 2021 4:50 pm
Forum: Extract Support
Topic: Not enough RAM to perform linkage clustering. Using vector clustering. This will be significantly slower.
Replies: 2
Views: 91

Re: Not enough RAM to perform linkage clustering. Using vector clustering. This will be significantly slower.

Like it says, to use the faster clustering with that many images you'd need more than 700gb of RAM. It uses exponential amounts of RAM to cluster based on how many images you have. You can split it to subses of 50k images to sort it in 4 packs if you want to use the faster method.

by bryanlyon
Fri Nov 12, 2021 4:04 pm
Forum: Artificial Intelligence
Topic: Trying to better understand options for faceswapping.
Replies: 2
Views: 2107

Re: Trying to better understand options for faceswapping.

One-shot models do exist and can run a swap without requiring training. They've come and gone over time, and most of them are nore really built for deployment and are just academic projects. To that end they can be difficult to get running reliably and rarely have any type of support. Even without t...
by bryanlyon
Thu Nov 11, 2021 4:29 pm
Forum: Hardware
Topic: HP Z840 a good system to start traning with?
Replies: 1
Views: 1548

Re: HP Z840 a good system to start traning with?

You lucky duck.

Should work fine. Though P6000 is a bit old, the 24gb of vram keeps it quite usable.

by bryanlyon
Thu Nov 04, 2021 3:50 pm
Forum: Extract Support
Topic: Nirkin_500_softmax_v1 model not available?
Replies: 5
Views: 2274

Re: Nirkin_500_softmax_v1 model not available?

Hmm, perhaps that one is the vgg_obstructed. In that case, you can try again later, manually download and place the file, or use a different mask. It's not anything to do with FS however -- most likely being your internet connection -- and there is nothing that we can do to fix it.

by bryanlyon
Thu Nov 04, 2021 3:46 pm
Forum: Training Support
Topic: Closes EVERY time graph is opened
Replies: 1
Views: 1984

Re: Closes EVERY time graph is opened

We will need the crashlog to see what's going on. But it sounds like your model's graph has gotten corrupted. You can try to restore to an older version of the model. You can also just avoid going to the graph tab. If you want to view the logs for that model, you can try using tensorboard itself dir...
by bryanlyon
Thu Nov 04, 2021 3:43 pm
Forum: Installation Support
Topic: Linux installation without internet
Replies: 2
Views: 2705

Re: Linux installation without internet

The installer cannot be used offline. You can install manually however. That generally does require internet still, but you can get past that by downloading the git and conda libraries on another computer and installing conda libraries from a local source ( https://stackoverflow.com/questions/343169...
by bryanlyon
Thu Nov 04, 2021 3:40 pm
Forum: Extract Support
Topic: Nirkin_500_softmax_v1 model not available?
Replies: 5
Views: 2274

Re: Nirkin_500_softmax_v1 model not available?

It's there, if you're having any problems downloading from Github you'll have to talk to them.

That said, we don't use that model for anything in FS. It's just there for legacy reasons.

by bryanlyon
Thu Oct 28, 2021 4:30 pm
Forum: Hardware
Topic: Hardware best practices
Replies: 60
Views: 74729

Re: RTX 3060

I'm not having any luck with my Raedon (Asrock Radeon RX 6700 XT Challenger Pro 12G OC) So I've decided I'm going to purchase a new video card. Is the RTX 3060 chipset have fully compatiable? Or can you suggest the best (mid range) card that can be purchased new (29/10/2021) Unfortunately, due to c...
by bryanlyon
Wed Oct 27, 2021 3:45 pm
Forum: Convert Support
Topic: No diference between 25k and 500k itteractions
Replies: 1
Views: 2272

Re: No diference between 25k and 500k itteractions

Check your commands and folders to make sure you're selecting everything correctly. If you still have problems you're going to need to provide more information (screenshots are likely the best).

by bryanlyon
Tue Oct 26, 2021 2:34 pm
Forum: Installation Support
Topic: Error module
Replies: 4
Views: 4252

Re: Error module

The app.php/faqpage#f1r1 still applies, though it's written more for windows. Another option is to open the faceswap conda env and run update_deps.py to update all requirements.

by bryanlyon
Tue Oct 19, 2021 6:09 pm
Forum: Training Support
Topic: Gray faces after 200k iterations
Replies: 5
Views: 3315

Re: Gray faces after 200k iterations

Corruption can happen randomly (though overclocks make it more likely). Your only choice is to revert to an older version and continue training from there.

by bryanlyon
Tue Oct 19, 2021 2:21 am
Forum: Training Support
Topic: Faceswap backend set to NVIDIA
Replies: 2
Views: 2909

Re: Faceswap backend set to NVIDIA

"Backend" doesn't mean not being used it means that the software backend is running on Nvidia.

by bryanlyon
Wed Oct 13, 2021 2:20 pm
Forum: Training Support
Topic: Training stops after ~2000 iterations without crash or error
Replies: 2
Views: 3180

Re: Training stops after ~2000 iterations without crash or error

I've never seen this before. Sometimes if there are power problems (Noisy, aging, or insufficient power supply) it can cause issues similar though. I'd see if restarting your Nvidia drivers do anything to the freeze ( Win+Ctrl+Shift+B ). Unfortunately if there are no messages it's very hard to troub...
by bryanlyon
Thu Oct 07, 2021 10:38 pm
Forum: Hardware
Topic: It's it possible to use edge-TPU to train a modle?
Replies: 5
Views: 6100

Re: It's it possible to use edge-TPU to train a modle?

No, we only support GPUs. TPUs, NPUs, the Apple M1, and any other "neural accelerators" are not supported. Most of those devices support inference ONLY and wont work for training at all. The only one we'd consider supporting is the M1 but that's hardware we can't justify purchasing and can...
by bryanlyon
Thu Oct 07, 2021 7:52 pm
Forum: Hardware
Topic: NVidia gtx760m
Replies: 3
Views: 4050

Re: NVidia gtx760m

That is an old card, but the biggest problem is the ram. 2gb of vram isn't really enough to be useful. You may want to check out cloud options instead.

by bryanlyon
Tue Oct 05, 2021 12:16 am
Forum: General Discussion
Topic: Windows 11 is OUT
Replies: 1
Views: 5287

Re: Windows 11 is OUT

We are not aware of any issues with Windows 11, but cannot yet guarantee any support as we haven't had a chance to test.

Assume it'll work and report if it doesn't.

by bryanlyon
Mon Oct 04, 2021 6:30 pm
Forum: Training Support
Topic: Crashing after 70 iterations
Replies: 1
Views: 3884

Re: Crashing after 70 iterations

This is a network/remote storage error. It's saying that your network data is becoming unavailable and causing faceswap to be unable to access the files. Unfortunately this is nothing to do with Faceswap and is entirely due to your setup. Either get your network storage more reliable or move the fil...
by bryanlyon
Mon Oct 04, 2021 6:28 pm
Forum: Extract Support
Topic: old extracted faces with .fsa upgrade?
Replies: 2
Views: 4645

Re: old extracted faces with .fsa upgrade?

Chances are you're asking about legacy vs face extract?

If so, you can keep using the old set by changing to "legacy" faces in the configuration settings.

If it's not that, please explain what your issue is in more detail so we an help understand what you need.