extraction error for ampere architecture maybe

The Extraction process failing on you, and you aren't getting an error back with clear instructions? Tell us about it here


Forum rules

Read the FAQs and search the forum before posting a new topic.

This forum is for reporting errors with the Extraction process. If you want to get tips, or better understand the Extract process, then you should look in the Extract Discussion forum.

Please mark any answers that fixed your problems so others can find the solutions.

Locked
User avatar
Linhaohoward
Posts: 23
Joined: Sat Dec 21, 2019 1:23 pm
Has thanked: 3 times

extraction error for ampere architecture maybe

Post by Linhaohoward »

Code: Select all

Loading...
Setting Faceswap backend to NVIDIA
09/27/2020 15:43:14 INFO     Log level set to: INFO
09/27/2020 15:43:15 INFO     Output Directory: D:\Faceswap Images\jalimika src
09/27/2020 15:43:15 INFO     Loading Detect from S3Fd plugin...
09/27/2020 15:43:15 INFO     Downloading model: 's3fd_keras' from: https://github.com/deepfakes-models/faceswap-models/releases/download/v1.11.1/s3fd_keras_v1.zip

09/27/2020 15:43:20 INFO     Extracting: 's3fd_keras'

09/27/2020 15:43:20 INFO     Loading Align from Fan plugin...
09/27/2020 15:43:20 INFO     Downloading model: 'face-alignment-network_2d4_keras' from: https://github.com/deepfakes-models/faceswap-models/releases/download/v1.13.2/face-alignment-network_2d4_keras_v2.zip

09/27/2020 15:43:49 INFO     Extracting: 'face-alignment-network_2d4_keras'

09/27/2020 15:43:50 INFO     Loading Mask from Components plugin...
09/27/2020 15:43:50 INFO     Loading Mask from Extended plugin...
09/27/2020 15:43:50 INFO     Loading Mask from Vgg_Obstructed plugin...
09/27/2020 15:43:50 INFO     Downloading model: 'Nirkin_500_softmax' from: https://github.com/deepfakes-models/faceswap-models/releases/download/v1.5.1/Nirkin_500_softmax_v1.zip

09/27/2020 15:44:42 INFO     Extracting: 'Nirkin_500_softmax'

09/27/2020 15:44:44 INFO     Reset batch sizes due to available VRAM: Detect: 20, Align: 19, Mask: 2
09/27/2020 15:44:44 INFO     Starting, this may take a while...
09/27/2020 15:44:44 INFO     Initializing S3FD (Detect)...
09/27/2020 15:48:47 INFO     Initialized S3FD (Detect) with batchsize of 20
09/27/2020 15:48:47 INFO     Initializing FAN (Align)...
C:\Users\Howard\faceswap\plugins\extract\detect\s3fd.py:496: RuntimeWarning: invalid value encountered in greater
poss = zip(*np.where(ocls[:, 1, :, :] > 0.05))
09/27/2020 16:03:22 INFO     Initialized FAN (Align) with batchsize of 19
09/27/2020 16:03:22 INFO     Initializing Components (Mask)...
09/27/2020 16:03:22 INFO     Initialized Components (Mask) with batchsize of 1
09/27/2020 16:03:22 INFO     Initializing Extended (Mask)...
09/27/2020 16:03:22 INFO     Initialized Extended (Mask) with batchsize of 1
09/27/2020 16:03:22 INFO     Initializing VGG Obstructed (Mask)...
09/27/2020 16:03:25 INFO     Initialized VGG Obstructed (Mask) with batchsize of 2

C:\Users\Howard\faceswap\plugins\extract\align\fan.py:150: RuntimeWarning: invalid value encountered in subtract
x_subpixel_shift = batch["prediction"][offsets[0]] - batch["prediction"][offsets[1]]
C:\Users\Howard\faceswap\plugins\extract\align\fan.py:153: RuntimeWarning: invalid value encountered in sign
subpixel_landmarks[:, :, 0] = indices[1] + np.sign(x_subpixel_shift) * 0.25 + 0.5
C:\Users\Howard\faceswap\plugins\extract\align\fan.py:151: RuntimeWarning: invalid value encountered in subtract
y_subpixel_shift = batch["prediction"][offsets[2]] - batch["prediction"][offsets[3]]
C:\Users\Howard\faceswap\plugins\extract\align\fan.py:154: RuntimeWarning: invalid value encountered in sign
subpixel_landmarks[:, :, 1] = indices[0] + np.sign(y_subpixel_shift) * 0.25 + 0.5
C:\Users\Howard\MiniConda3\envs\faceswap\lib\site-packages\numpy\linalg\linalg.py:2116: RuntimeWarning: invalid value encountered in det
r = _umath_linalg.det(a, signature=signature)
C:\Users\Howard\MiniConda3\envs\faceswap\lib\site-packages\numpy\linalg\linalg.py:1872: RuntimeWarning: invalid value encountered in greater
return count_nonzero(S > tol, axis=-1)
C:\Users\Howard\faceswap\plugins\extract\detect\s3fd.py:512: RuntimeWarning: invalid value encountered in subtract
return np.exp(inp - logsumexp(inp, axis=axis, keepdims=True))
C:\Users\Howard\faceswap\plugins\extract\mask\extended.py:49: RuntimeWarning: invalid value encountered in floor_divide
ml_pnt = (landmarks[36] + landmarks[0]) // 2
C:\Users\Howard\faceswap\plugins\extract\mask\extended.py:50: RuntimeWarning: invalid value encountered in floor_divide
mr_pnt = (landmarks[16] + landmarks[45]) // 2
C:\Users\Howard\faceswap\plugins\extract\mask\extended.py:53: RuntimeWarning: invalid value encountered in floor_divide
ql_pnt = (landmarks[36] + ml_pnt) // 2
C:\Users\Howard\faceswap\plugins\extract\mask\extended.py:54: RuntimeWarning: invalid value encountered in floor_divide
qr_pnt = (landmarks[45] + mr_pnt) // 2
C:\Users\Howard\faceswap\plugins\extract\mask\extended.py:65: RuntimeWarning: invalid value encountered in floor_divide
landmarks[17:22] = top_l + ((top_l - bot_l) // 2)
C:\Users\Howard\faceswap\plugins\extract\mask\extended.py:66: RuntimeWarning: invalid value encountered in floor_divide
landmarks[22:27] = top_r + ((top_r - bot_r) // 2)
C:\Users\Howard\faceswap\lib\umeyama.py:117: RuntimeWarning: divide by zero encountered in true_divide
scale = 1.0 / src_demean.var(axis=0).sum() * np.dot(S, d)
C:\Users\Howard\faceswap\lib\aligner.py:110: RuntimeWarning: invalid value encountered in double_scalars
y_scale = (mat[0, 0] * mat[1, 1] - mat[0, 1] * mat[1, 0]) / x_scale

have anyone had any problems with rtx 3000 series? just setup my 3090 yesterday so its the first run. seems like detector isnt working.

Attachments
Untitled2.png
Untitled2.png (96.07 KiB) Viewed 5347 times
Untitled.png
Untitled.png (81.79 KiB) Viewed 5349 times
User avatar
torzdf
Posts: 2649
Joined: Fri Jul 12, 2019 12:53 am
Answers: 159
Has thanked: 128 times
Been thanked: 623 times

Re: extraction error for ampere architecture maybe

Post by torzdf »

30xx cards require Cuda 11. Faceswap installs with Cuda 10.1 which is incompatible.

As soon as Conda have a Cuda 11 release of Tensorflow, then we will upgrade, until then you won't be able to use your new card for Face swapping, or you will need to compile a custom Tensorflow, which is outside of the scope of support we will give.

My word is final

User avatar
Linhaohoward
Posts: 23
Joined: Sat Dec 21, 2019 1:23 pm
Has thanked: 3 times

Re: extraction error for ampere architecture maybe

Post by Linhaohoward »

torzdf wrote: Sun Sep 27, 2020 11:32 am

30xx cards require Cuda 11. Faceswap installs with Cuda 10.1 which is incompatible.

As soon as Conda have a Cuda 11 release of Tensorflow, then we will upgrade, until then you won't be able to use your new card for Face swapping, or you will need to compile a custom Tensorflow, which is outside of the scope of support we will give.

i heard a special +nv tensorflow version might solve the problem with A100 which by right is also on ampere architecture. did a search but it seems that this tensorflow version is limited to ubuntu. do you think this might work?

Attachments
94213159-c0f00c80-ff08-11ea-8fc1-7d8634459dc5.png
94213159-c0f00c80-ff08-11ea-8fc1-7d8634459dc5.png (8.7 KiB) Viewed 5314 times
User avatar
bryanlyon
Site Admin
Posts: 793
Joined: Fri Jul 12, 2019 12:49 am
Answers: 44
Location: San Francisco
Has thanked: 4 times
Been thanked: 218 times
Contact:

Re: extraction error for ampere architecture maybe

Post by bryanlyon »

No, it wont work. Ampere REQUIRES Cuda 11 to operate. Further, we require Tensorflow 2.2.

At this time it is NOT POSSIBLE to use 30xx cards (or A100) in Faceswap.

We will fix this as soon as possible, but are waiting on other, upstream, fixes/updates before we can.

Locked