Extract process hangs on pass 2, stuck at 0%
Code: Select all
Loading...
Setting Faceswap backend to AMD
04/20/2022 21:30:57 INFO Log level set to: INFO
04/20/2022 21:30:57 INFO Setting up for PlaidML
04/20/2022 21:30:57 INFO Using GPU(s): ['Advanced Micro Devices, Inc. - gfx1032 (experimental)']
04/20/2022 21:30:57 INFO Successfully set up for PlaidML
04/20/2022 21:31:00 INFO Output Directory: C:\Users\brend\Desktop\Data Nemesis Clips\Data - Nemesis Extract
04/20/2022 21:31:00 INFO Loading Detect from S3Fd plugin...
Using plaidml.keras.backend backend.
04/20/2022 21:31:00 INFO Loading Align from Fan plugin...
04/20/2022 21:31:00 INFO Loading Mask from Components plugin...
04/20/2022 21:31:00 INFO Loading Mask from Extended plugin...
04/20/2022 21:31:00 INFO Starting, this may take a while...
04/20/2022 21:31:00 INFO Initializing S3FD (Detect)...
04/20/2022 21:31:00 INFO Opening device "opencl_amd_gfx1032.0"
04/20/2022 21:31:06 INFO Initialized S3FD (Detect) with batchsize of 12
Running pass 1 of 4: Detect: 11111it [53:28, 3.68it/s]
Running pass 1 of 4: Detect: 11124it [53:31, 3.78it/s]
Running pass 1 of 4: Detect: 11136it [53:34, 3.80it/s]
Running pass 1 of 4: Detect: 11148it [53:37, 3.79it/s]
Running pass 1 of 4: Detect: 11160it [53:41, 3.77it/s]
Running pass 1 of 4: Detect: 11172it [53:44, 3.68it/s]
Running pass 1 of 4: Detect: 11184it [53:47, 3.67it/s]
Running pass 1 of 4: Detect: 11196it [53:51, 3.69it/s]
Running pass 1 of 4: Detect: 11208it [53:54, 3.66it/s]
Running pass 1 of 4: Detect: 11220it [53:57, 3.68it/s]
Running pass 1 of 4: Detect: 11232it [54:00, 3.70it/s]
Running pass 1 of 4: Detect: 11244it [54:04, 3.68it/s]
Running pass 1 of 4: Detect: 11256it [54:21, 1.59it/s]
04/20/2022 22:25:27 INFO Initializing FAN (Align)...
04/20/2022 22:25:35 INFO Analyzing Ops: 2162 of 2707 operations complete
04/20/2022 22:26:20 INFO Initialized FAN (Align) with batchsize of 12
New user, hi, I've been searching for an answer on the forums, and tried a few different things as suggested in other posts already, so hopefully I'm not out of line for making a new post.
Here's more details: I've tried several times to get this working, but it seems to bog down on pass 2 of extraction. Currently, I'm trying to extract from an almost 15 minute video (just over 11,000 frames), but I've tried shorter clips that also hung up at this point. I tried changing the batch size in the Align "fan" settings, and in the Detect settings. I've tried using S3Fd and Mtcnn.
My system has a AMD Ryzen 5 5600G CPU (3901 Mhz. 6 cores), 16 GB of system memory (8 GB of virtual memory available), and a Radeon RX 6600 XT with 8 GB of ram. I tried uninstalling and deleting any conda files on my system and reinstalling tFaceswap (choosing the AMD GPU install option).
I also tried to use images instead of a video as a source, which is what is running right now, and after about an hour, it just finally got to 1% on pass 2 of 4, and lists over 100 hours remaining (currently says 32.21 s/it as well).
I did succeed in getting two other videos to fully extract before, but there were pretty small, just running as a test. I don't remember how long each of them took to complete, as I left my PC to run without watching it, but they did finish. I realize my current attempt is using more frames/images, so I expected it to take longer, but I don't think it's supposed to go this slow.
I had tried the standard batch settings without success. Currently I'm using settings of 12 for both Align and Detect,as it seemed to speed up pass 1 over the lower settings.