OpenCV CUDA Binaries Wiki
OpenCV Pre-built CUDA binaries
Brought to you by:
saharshbishnoi,
sigmoidx
Welcome to your wiki!
This is the default page, edit it as you see fit. To add a new page simply reference it within brackets, e.g.: [SamplePage].
The wiki uses Markdown syntax.
Hello?
Could you revise the binary to support the latest GPU like 980 or Titan X?
NVIDIA GPU arch must be supporting sm_52 (52).
Thanks,
Yeah sure ! Thanks for catching that. I'll do that with Cuda Toolkit 7.0 and upload on Monday together with a build with Visual Studio 2015 and CuToolkit 7.5, hopefully it works. Will try on Monday. Thinking of completely removing the 6.5 build, do you think that's okay ?
Thanks for considering that! I will be greatly thankful for this
revision!!!!! You're a lifesavior!
On Sat, Jul 25, 2015 at 11:31 AM, Saharsh Bishnoi saharshbishnoi@users.sf.net wrote:
Started build a few hours ago, now building with 7.0 and 7.5 with 50 and 52. Doing dynamic first then static, fairly long builds but I will upload as soon as they are finished.
Wow, that's amazing. I will be waiting for it!
Thanks,
Best regards,
On Tue, Jul 28, 2015 at 7:27 AM, Saharsh Bishnoi saharshbishnoi@users.sf.net wrote:
SourceForge upload is currently broken. Here is the link to the latest build. Hope that works ! https://www.dropbox.com/s/v4h48cz51nvexsr/cuda7.5_contrib_viz_dynamic_static.7z?dl=0
Just tested after CUDA 7.5 RC1 was installled. It works on 970 GTX! Thanks
so much!
I hope you could upload it as soon as possible.
Best regards,
On Wed, Jul 29, 2015 at 2:09 AM, Saharsh Bishnoi saharshbishnoi@users.sf.net wrote:
Hello?
I have a question regarding a weird problem occurred in early 2013 version
Macbook Pro Bootcamp. It won't compile completely without any error. And
any executable file that used this library won't run in Macbook Pro. But
the executable that used an ordinary opencv_world300.dll runs well in the
Macbook Pro.
This weird error is only happening in Macbook Pro laptop and they run well
in other laptops and desktop PCs. Is there any CPU specific setting in this
library?
Thanks!
On Wed, Jul 29, 2015 at 8:49 AM, Yongwoo 'Oscar' Jeong sigmoidx@users.sf.net wrote:
Apologies for the late reply, the cuda7.5_contrib_viz_avx build has SSE & AVX2 enabled by default. Your CPU may not support AVX2, a quick check would be to try out the cuda7.5_contrib_viz_noopt build or to try and disable the optims. by calling:
http://docs.opencv.org/modules/core/doc/utility_and_system_functions_and_macros.html#setuseoptimized
You might also want to check hardware support with:
http://docs.opencv.org/modules/core/doc/utility_and_system_functions_and_macros.html#checkhardwaresupport
Please let me know if thats it, otherwise I can provide a test project to try,
OK, I will try it out.
Thanks a lot!
On Fri, Oct 16, 2015 at 8:29 AM, Saharsh Bishnoi saharshbishnoi@users.sf.net wrote:
Hi,
How have you been?
Besides all things you have done nicely, I think Caffe + cuDNN (for CUDA
7.5) could be a great addition to it.
I got an email from NVidia saying that Caffe 1.0 is now officially
supporting cuDNN. Could you add them into current openCV 3.0 (or 3.1) +
CUDA 7.5 + contrib library?
And about the Visual Studio, we could use vs2013 as well as vs2015 version.
(Both versions will be appreciated)
Thanks!
On Fri, Oct 16, 2015 at 10:19 AM, Yongwoo 'Oscar' Jeong sigmoidx@users.sf.net wrote:
Hey, yeah that sounds good. I have been wanting to try out some of the newer deep learning techniques as well. Nvidia Toolkit 8.0 is with pascal support is in RC1 so this is a good time to be thinking about that. Btw, I have made you an Admin so feel free to author this project too. Do you think it ought to be added in as a nuget package or as a stand-alone build like I used to do before ? I am also guessing youre only interested in generating builds for Visual Studio and Windows support, I dont really work on linux systems and I dont think people using linux for development need help with their builds lol I am interested in Clang builds with VS2015 as well though !
Last edit: Saharsh Bishnoi 2016-06-14
Thanks for offering the Admin role but I think it would hurt your previous
works if the packages I made run differently or there is something on my
works..So it would be better to keep yourself as a sole Admin, I guess.
Thanks anyway. :)
And about the nuget packages, it is fine by me. Since they can be unzipped
by 7zip, if anyone doesn't like the nuget packages, just go to the nuget
cache folder then unzipping them with 7zip, making file sortings, and
finally adding the path...it will work also with VS. Having nuget packaes
is a good idea. But if managing nuget packages make you have some troubles,
just make a stand-alone like before.
And about the linux build...we don't need it but what about both python
2.7/3.5 build? It would work on almost any system. There is a rumor opencv
doesn't support opencv-python with cuda support but I'm not sure about it.
If it is not true, cuDNN 4(or 5) + Caffe (latest) + CUDA 8 (or 7.5) +
opencv 3.1 + python 3.5(or, with 2.7) would be great!
Thanks!!!!
On Wed, Jun 15, 2016 at 4:56 AM, Saharsh Bishnoi saharshbishnoi@users.sf.net wrote:
Hello,
Thanks for this project! I've been using it for quite some time and found it to be very useful. It is apparently the only pre-built binary that allows using CUDA on OpenCV. However, as time goes I've been asked to create builds that use the MSVC 2015 or 2017 toolset (if possible, with CUDA 8.0). Do you by any chance plan to release an updated version? If not, would it be possible to give me an idea of how difficult it should be to compile the OpenCV library with CUDA on my own?
Thanks!!!