Closed segalinc closed 5 years ago
@cristinasegalin can you please make sure that you have at least cuda 9.2 and the latest pytorch version (1.0)? Some people were reporting issues with lower versions of cuda. Regarding a potential out of memory, it is possible but unlikely. To rull this out make sure that your input image is relatively small since the image detectors may go out of memory at large resolutions.
I have cuda 10. And using Yolov3 as face detector on a separate gpu. I was trying to use the same image used in the example shared with the repo
Cristina
Sent from my OnePlus
On Fri, Jan 4, 2019, 17:23 Adrian Bulat <notifications@github.com wrote:
@cristinasegalin https://github.com/cristinasegalin can you please make sure that you have at least cuda 9.2 and the latest pytorch version (1.0)? Some people were reporting issues with lower versions of cuda. Regarding a potential out of memory, it is possible but unlikely. To rull this out make sure that your input image is relatively small since the image detectors may go out of memory at large resolutions.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/1adrianb/face-alignment/issues/123#issuecomment-451615180, or mute the thread https://github.com/notifications/unsubscribe-auth/AI63kom4ix-e6N9OvYQ04Ofyf7JaSHlzks5u__50gaJpZM4ZubBd .
solved compiling pytorch from source
Great to know that you found and fixed the issue!
Compiling pytorch from source ? @cristinasegalin How to do this???
if using np.copy before cuda(). the error may disappear! pre = pre.transpose((2, 0, 1))[np.newaxis,].copy() x = torch.from_numpy(pre).cuda()
I tried to run the code only changing the face detector with cvlib, and then I get this error. RuntimeError: cuda runtime error (11) : invalid argument at /pytorch/aten/src/THC/THCGeneral.cpp:
if I change the dace detector to the one I implemented using as well cuda I get the error even before the landmark detection. I am not sure where it comes from..as the variables and input are all correct, could it be something realted to out of memory? I have 4 RTX 2080