Search found 15 matches

by calipheron
Fri Jul 24, 2020 9:19 pm
Forum: Training Discussion
Topic: Comparing model types with examples
Replies: 23
Views: 64936

Re: Comparing models with examples

Fantastic work, it's nice to have a more detailed result to examine. Personally, I have done a MILLION iterations with both DFL-SAE and Realface, and am unable to get results as detailed as my model B source files provide. (5400~ faces) I'm quite disheartened at this point. I'm certain that my sourc...
by calipheron
Fri Jul 24, 2020 9:04 pm
Forum: Extract Discussion
Topic: Manual Landmarks
Replies: 5
Views: 5216

Re: Manual Landmarks

I looked for a way to assist editing / clearing multiple garbage faces from long videos using the manual tool. I ended up making an AutoHotKey script for this purpose. As this post is in a similar ballpark, I thought I would post it here: The manual tool is definitely not the best tool for this. In...
by calipheron
Mon Jul 13, 2020 5:12 pm
Forum: Extract Discussion
Topic: Manual Landmarks
Replies: 5
Views: 5216

Re: Manual Landmarks

I looked for a way to assist editing / clearing multiple garbage faces from long videos using the manual tool. I ended up making an AutoHotKey script for this purpose. As this post is in a similar ballpark, I thought I would post it here: ^j:: Loop 100 { Send {1} Sleep 10 Send {d} Sleep 10 Send {x} ...
by calipheron
Sun Jun 28, 2020 11:49 am
Forum: General Discussion
Topic: Nvidia Studio Driver VS Game Ready Driver
Replies: 1
Views: 3032

Re: Nvidia Studio Driver VS Game Ready Driver

My experience has been the opposite.
With a 2060 Super, I could not get reliable training with the studio driver.
Moved to the latest Game Ready driver and - as long as I run nothing else that needs GPU / VRAM - it's pretty stable.

by calipheron
Sun Jun 28, 2020 11:45 am
Forum: Training Support
Topic: Caught exception in thread: '_training_0' Error: OSError: exception: access violation reading 0x0000000000000010
Replies: 2
Views: 6204

Re: Caught exception in thread: '_training_0' Error: OSError: exception: access violation reading 0x0000000000000010

Also consider sticking with simpler models.
Lightweight, Original or DFaker.
They get a bit too complicated after that, for AMD Polaris GPUs.

by calipheron
Sun Jun 21, 2020 10:20 am
Forum: Training Support
Topic: Failed to get convolution algorithm : CRASH
Replies: 12
Views: 6624

Re: Brand new 2060 Super. Trouble training.

This looks like an upstream issue, to be honest ;( https://github.com/tensorflow/tensorflow/issues/33536 Thank you for replying Torzdf! I take it this means we have to wait for Tensorflow itself to receive a fix? Well, so far I have found that, unlike with my RX 580, I basically have to have Facesw...
by calipheron
Sat Jun 20, 2020 6:44 pm
Forum: Training Support
Topic: Failed to get convolution algorithm : CRASH
Replies: 12
Views: 6624

Re: Brand new 2060 Super. Trouble training.

Scratch that. Another crash. Will try the latest gaming drivers instead. Edit: Fresh install of the gaming ready drivers, v446.14 and after 40 minutes of training, no crashes . So for Faceswapping with 2060 Super at least, I would say that the studio ready drivers are not recommended. Edit: 8 hours ...
by calipheron
Sat Jun 20, 2020 6:31 pm
Forum: Training Support
Topic: Failed to get convolution algorithm : CRASH
Replies: 12
Views: 6624

Re: Brand new 2060 Super. Trouble training.

Cuda errors in that last lot of messages tipped me off as to the possible problem. Faceswap might be configured for Nvidia, but I was still using the old conda environment from the "AMD install." I completely removed Anaconda / miniConda, Python and Faceswap. Made sure no conda files were ...
by calipheron
Sat Jun 20, 2020 6:03 pm
Forum: Training Support
Topic: Failed to get convolution algorithm : CRASH
Replies: 12
Views: 6624

Re: Brand new 2060 Super. Trouble training.

Now I'm getting the following: 020-06-20 19:00:06.112961: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cublas64_100.dll 2020-06-20 19:00:06.300959: E tensorflow/stream_executor/cuda/cuda_blas.cc:238] failed to create cublas handle: CUBLAS_STATUS...
by calipheron
Sat Jun 20, 2020 5:34 pm
Forum: Training Support
Topic: Failed to get convolution algorithm : CRASH
Replies: 12
Views: 6624

Re: Brand new 2060 Super. Trouble training.

Those errors could just mean that they weren't able to load the data that they have on the card, or could be that the drivers are messing up. My advice is to (in this order) install the latest drivers, reboot, and close all applications to try again. This is with a fresh install of the drivers, tod...
by calipheron
Sat Jun 20, 2020 5:33 pm
Forum: Training Support
Topic: Failed to get convolution algorithm : CRASH
Replies: 12
Views: 6624

Re: Brand new 2060 Super. Trouble training.

Trying with allow growth, no other settings changed. Getting a lot of this: 2020-06-20 18:29:20.692407: I tensorflow/stream_executor/cuda/cuda_driver.cc:831] failed to allocate 2.49G (2673629440 bytes) from device: CUDA_ERROR_OUT_OF_MEMORY: out of memory There's more: 2020-06-20 18:29:10.471097: I t...
by calipheron
Sat Jun 20, 2020 5:27 pm
Forum: Training Support
Topic: Failed to get convolution algorithm : CRASH
Replies: 12
Views: 6624

Re: Brand new 2060 Super. Trouble training.

Thanks Bryan. nvidia-smi reports only 939MiB of 8192MiB is used. Well, that is with Firefox open. I will try with "allow growth". I don't mind slower training, but this is disappointing so far: I never had ONE single memory related error on my RX 580. I just had this crash: 2020-06-20 18:2...
by calipheron
Sat Jun 20, 2020 5:17 pm
Forum: Training Support
Topic: Failed to get convolution algorithm : CRASH
Replies: 12
Views: 6624

[Mitigated] Brand new 2060 Super. Trouble training.

Hello, I've just taken delivery of a new 2060 Super. 2060 Super is the current recommended baseline hardware for Faceswap. I assumed I'd be able to move up to slightly more hefty models than DFaker or Original, compared to the RX 580 I was using. I have the latest Studio drivers installed. Trying to...
by calipheron
Fri May 15, 2020 9:56 am
Forum: General Discussion
Topic: Interesting AMD performace using the ROCm stack
Replies: 10
Views: 34829

Re: Interesting AMD performace using the ROCm stack

Thanks for replying, I'll try my 580 instead and see if that behaves any differently.

by calipheron
Thu May 14, 2020 7:44 pm
Forum: General Discussion
Topic: Interesting AMD performace using the ROCm stack
Replies: 10
Views: 34829

Re: Interesting AMD performace using the ROCm stack

So I've just spent the last few hours trying to get this to work on a fresh install of Lubuntu 18.04 i7-4770 system with a pokey RX 460, but this should be enough to get Original training going. I managed to get faceswap to run and it even says ROCM is in use. But it complains about MLOpen not being...