Error with operands cannot be broadcasted about deepfacelab HOT 19 CLOSED

cant help

from deepfacelab.

1 hour of extracting?
your dlib compiled without CUDA

from deepfacelab.

try MT extractor

from deepfacelab.

Hey, thanks for the reply. As I mentioned above I already tried with MT and got the same error. I will try again by making dlib use Cuda and will report back here the results.

from deepfacelab.

Ok I just tried dlib with cuda and the extraction was quick (under 3-4 mins for 550 images). But I still see the same error. I think its related to Numpy but I am not sure. I tried MT as well again and I get the same error on pass 3. Please help!

from deepfacelab.

@Rohit-Shah you can simply: git reset --hard COMMIT, to more old commit, eg: git reset --hard a74b57d. Maybe this will help. I think the problem is one of the last commits.

from deepfacelab.

I'm going to try moving from Windows 10 to Ubuntu 17.10 to free up some more VRAM, and I'll report if there's a similar error during extraction.

from deepfacelab.

@mrsmith0x00 Going to try that and will report back. Thanks for the tips.
@gessyoo were you able to reproduce this?

from deepfacelab.

rohit-shah, I was not able to reproduce the error on Ubuntu 17.10 or 18.04 using the latest commit. Under 18.04, I'm running a GTX 1070, Anaconda python 3.65, CUDA 9.0 and CUDNN 7.05. The extract command worked as expected with detector dlib (compiled to use CUDA) and mt. The nvidia driver is 396. When you ran sudo pip3 install -r requirements-gpu-cuda9-cudnn7.txt from the terminal, were there any error messages?

from deepfacelab.

@gessyoo Hmm, I can reproduce ValueError on latest commit. How do you have installed Python packages? Via anaconda install (by following the instructions) or simply pip install (install all packages from requirements file)?

from deepfacelab.

I installed Anaconda first, then ran sudo pip3 install -r requirements-gpu-cuda9-cudnn7.txt.

from deepfacelab.

So the problem in some Anaconda package. I until a bit busy, but after, all be sure to check and will create PR with fixes in the instruction. Big thanks!

from deepfacelab.

@gessyoo
Not related to the error, but you shouldn't run pip as root as it poses a security risk. Adding --user to the command will only install it for the current user. No elevated privileges needed.
For software like DeepFaceLab or faceswap I'd even recommend using a virtual environment. Install virtualenv via pip, create the virtual environment via virtualenv env_dir/ and activate it with source env_dir/bin/activate. Once activated you can install all the other requirements. No need for root access or the --user option. Also if something goes wrong, you can just delete env_dir/ and start over with a clean slate.

from deepfacelab.

Thanks for the tips. I resorted to using sudo just to get Ubuntu up and running ASAP, without permission errors. I wanted more VRAM for training the models, because, for reasons known only to Microsoft, Windows 10 reserves a big chunk of VRAM from CUDA. I've been using python environments under Windows 10.

from deepfacelab.

Same error for me on step 4. I thought there were zero dependencies ?

from deepfacelab.

I thought there were zero dependencies ?

read better, its only for windows prebuilt package

from deepfacelab.

read better, its only for windows prebuilt package

and i'm having the error on windows

from deepfacelab.

windows binaries 100% tested on clean windows install with latest updates.

from deepfacelab.

windows binaries 100% tested on clean windows install with latest updates.

wow, this has been so thoroughly tested that nobody is having this issue whatsoever
developers with tremendous ego...

https://steamuserimages-a.akamaihd.net/ugc/861739932490930245/E7B7A3E9A483958914068E01DCF92261E63F3FDA/

from deepfacelab.