Package Details: opencv-cuda 4.5.3-2

Git Clone URL: https://aur.archlinux.org/opencv-cuda.git (read-only, click to copy)
Package Base: opencv-cuda
Description: Open Source Computer Vision Library with CUDA support
Upstream URL: http://opencv.org/
Licenses: BSD
Conflicts: opencv, opencv-samples
Provides: opencv, opencv-samples
Submitter: billypilgrim
Maintainer: billypilgrim
Last Packager: billypilgrim
Votes: 9
Popularity: 0.122420
First Submitted: 2018-08-11 08:14
Last Updated: 2021-09-13 15:36

Dependencies (26)

Required by (307)

Sources (3)

Latest Comments

1 2 3 4 5 6 ... Next › Last »

billypilgrim commented on 2021-10-22 10:36

Hi all! For those of you who are interested in having this package in community, I've opened a feature request to add this by extending the existing opencv PKGBUILD in community: https://bugs.archlinux.org/task/72512

Please go and make your voices heard if this is something you'd like too :-)

DAC324 commented on 2021-09-14 14:45

@billypilgrim: This is just for experimenting :) We have gcc-git in the AUR, that's where I get gcc-12 from :)

billypilgrim commented on 2021-09-14 14:00

How are you building it with gcc v12? It's not in the main repos yet...

DAC324 commented on 2021-09-14 12:57

@billypilgrim and all: Indeed, I have tried building from scratch. But it looks like I have found the cause: OpenCV does not seem to like gcc-12 yet. Compiling with gcc-11 (tested on another system) seems to work.

Researched a bit more. According to https://forum.opencv.org/t/building-opencv-from-source-gcc-and-cuda-versions/3278 , the GCC version suitable for building opencv depends on the CUDA version used.

As, due to my old graphics card, I only have CUDA 11.4, I did the following:

$ cd /opt/cuda/bin
$ sudo ln -sf /usr/bin/gcc-10 gcc
$ sudo ln -sf /usr/bin/g++-10 g++

After that, the compile process continues.

billypilgrim commented on 2021-09-14 12:13

@DAC324: Have you tried doing a clean build?

DAC324 commented on 2021-09-14 11:55

This package does not compile:

[136/2401] Building NVCC (Device) object mo...da/cuda_compile_1_generated_gpu_mat_nd.cu.o
FAILED: modules/core/CMakeFiles/cuda_compile_1.dir/src/cuda/cuda_compile_1_generated_gpu_mat_nd.cu.o /opt/.cache/yay/opencv-cuda/src/build/modules/core/CMakeFiles/cuda_compile_1.dir/src/cuda/cuda_compile_1_generated_gpu_mat_nd.cu.o 

and

CMake Error at cuda_compile_1_generated_gpu_mat_nd.cu.o.Release.cmake:280 (message):
  Error generating file
  /opt/.cache/yay/opencv-cuda/src/build/modules/core/CMakeFiles/cuda_compile_1.dir/src/cuda/./cuda_compile_1_generated_gpu_mat_nd.cu.o


[137/2401] Building NVCC (Device) object mo.../cuda/cuda_compile_1_generated_gpu_mat.cu.o
FAILED: modules/core/CMakeFiles/cuda_compile_1.dir/src/cuda/cuda_compile_1_generated_gpu_mat.cu.o /opt/.cache/yay/opencv-cuda/src/build/modules/core/CMakeFiles/cuda_compile_1.dir/src/cuda/cuda_compile_1_generated_gpu_mat.cu.o 

Win8Error commented on 2021-08-28 13:20

Please add nvidia, cuda and cudnn as make dependencies.

soslowboy commented on 2021-08-18 14:47

Dear @billypilgrim, Opencv builds fine for me, but it doesn't build the nvcuvid support. The output of cmake for CUDA is the following. NVIDIA CUDA: YES (ver 11.4, CUFFT CUBLAS FAST_MATH) Any ideas how i can fix this? I haven't changed anything in the PKGBUILD file just to be sure. Also, this is on a newly installed system. Thanks a lot,

ThiasTux commented on 2021-07-01 16:10

@billypilgrim Thank you for your quick response and fix of #2. I noticed that bug, but since there was a fix for lapack-3.9.1, I thought it might just needed to be adjusted for 3.10. Do you know if loosing lapack support for the time being affect performance?

For #1, you were right, I followed literally tens of tutorials, just to discover that for some reason, the initialisation of the classifier was wrong in all of them. Reading the official documentation fixed it the error in my code. Sorry about that.

billypilgrim commented on 2021-07-01 08:03

@ThiasTux I've fixed #2 by not building against lapack at all for now. Unfortunately there's an upstream issue (https://github.com/opencv/opencv/issues/19846) and I don't have time to fix it myself. Once it's fixed I'll re-enable lapack support.

Re #1: Those file paths are purely debug information. *.cpp files aren't used by the program once it's compiled. So my suspicion is that your calling code is doing something wrong, hence why OpenCV is giving you an admittedly not-very-helpful error message. Try delving into the OpenCV documentation or the source code to see why you might be getting this error.