增加CV原生DNN模块的CUDA加速推理依赖它来了

maoxiaosc

废话不多说，先上效果：
C:\Users\maoxiaosc\Desktop\caffe.png
今日得空，编译了CUDA支持的原生opencv依赖库(非onnx的lite部分)，加强了DNN块的使用效率，部分模型不需要再转换即可享用CUDA加速推理了，如Caffe、Darknet、Tensorflow等。
依赖库的替换注意事项：
1、必须要存在CUDA11.4，CUDNN8.2.2.6的GPU推理环境，否则加载不上库，替换前做好备份；
2、dll文件替换到火山安装目录下 plugins\vprj_win\classlib\sys\cv_ai_lib\libs\opencv\x64\vc16\bin；
3、lib文件替换到火山安装目录下 plugins\vprj_win\classlib\sys\cv_ai_lib\libs\opencv\x64\vc16\lib。
需要的小伙伴到群里下载吧：火山AI机器视觉修真群 792015051

小星星

我也是替换了，也有cuda环境，同样调用不了，楼上哪位朋友有解决了？给小弟指点一下呗这问题怎么这么多人遇到。

a429387333

CUDA11.4，CUDNN8.2.2.26都装好了，无法使用

OpenCV(4.6.0) Error: No CUDA support (OpenCV was not built to work with the selected device. Please check CUDA_ARCH_PTX or CUDA_ARCH_BIN in your build configuration.) in cv::dnn::dnn4_v20220524::Net::Impl::initCUDABackend, file F:\opencv4.6\opencv\sources\modules\dnn\src\op_cuda.cpp, line 30

a429387333

总是报错
[ INFO:0@1.324] global F:\opencv4.6\opencv\sources\modules\videoio\src\videoio_registry.cpp (223) cv::`anonymous-namespace'::VideoBackendRegistry::VideoBackendRegistry VIDEOIO: Enabled backends(8, sorted by priority): FFMPEG(1000); GSTREAMER(990); INTEL_MFX(980); MSMF(970); DSHOW(960); CV_IMAGES(950); CV_MJPEG(940); UEYE(930)
[ INFO:0@1.365] global F:\opencv4.6\opencv\sources\modules\videoio\src\backend_plugin.cpp (383) cv::impl::getPluginCandidates Found 3 plugin(s) for FFMPEG
[ INFO:0@1.377] global F:\opencv4.6\opencv\sources\modules\core\src\utils\plugin_loader.impl.hpp (67) cv::plugin::impl:ynamicLib::libraryLoad load C:\Users\noname\AppData\Roaming\VolDev\??+??\????\???\_sys_sln_tmp\samples\vprj_win\_int\DNN_ObjectDetection\debug\x64\linker\opencv_videoio_ffmpeg460_64d.dll => FAILED
[ INFO:0@1.399] global F:\opencv4.6\opencv\sources\modules\core\src\utils\plugin_loader.impl.hpp (67) cv::plugin::impl:ynamicLib::libraryLoad load opencv_videoio_ffmpeg460_64d.dll => FAILED
[ INFO:0@1.414] global F:\opencv4.6\opencv\sources\modules\core\src\utils\plugin_loader.impl.hpp (67) cv::plugin::impl:ynamicLib::libraryLoad load opencv_videoio_ffmpeg460_64.dll => OK
[ INFO:0@1.423] global F:\opencv4.6\opencv\sources\modules\videoio\src\backend_plugin.cpp (50) cv::impl:luginBackend::initCaptureAPI Found entry: 'opencv_videoio_capture_plugin_init_v1'
[ INFO:0@1.433] global F:\opencv4.6\opencv\sources\modules\videoio\src\backend_plugin.cpp (169) cv::impl:luginBackend::checkCompatibility Video I/O: initialized 'FFmpeg OpenCV Video I/O Capture plugin': built with OpenCV 4.6 (ABI/API = 1/1), current OpenCV version is '4.6.0' (ABI/API = 1/1)
[ INFO:0@1.447] global F:\opencv4.6\opencv\sources\modules\videoio\src\backend_plugin.cpp (69) cv::impl:luginBackend::initCaptureAPI Video I/O: plugin is ready to use 'FFmpeg OpenCV Video I/O Capture plugin'
[ INFO:0@1.457] global F:\opencv4.6\opencv\sources\modules\videoio\src\backend_plugin.cpp (84) cv::impl:luginBackend::initWriterAPI Found entry: 'opencv_videoio_writer_plugin_init_v1'
[ INFO:0@1.467] global F:\opencv4.6\opencv\sources\modules\videoio\src\backend_plugin.cpp (169) cv::impl:luginBackend::checkCompatibility Video I/O: initialized 'FFmpeg OpenCV Video I/O Writer plugin': built with OpenCV 4.6 (ABI/API = 1/1), current OpenCV version is '4.6.0' (ABI/API = 1/1)
[ INFO:0@1.481] global F:\opencv4.6\opencv\sources\modules\videoio\src\backend_plugin.cpp (103) cv::impl:luginBackend::initWriterAPI Video I/O: plugin is ready to use 'FFmpeg OpenCV Video I/O Writer plugin'
O

a429387333

as853749515 发表于 2022-11-15 01:40
为什么我什么都安装好了   也配置好了    运行还是cpu   不是GPU运行呢

我也一样

as853749515

为什么我什么都安装好了   也配置好了    运行还是cpu   不是GPU运行呢

as853749515

为什么我的都配置好了   运行不是用的GPU   而是cpu呢   

as853749515

我的   CUDA11.4，CUDNN8.2.2.6   都安装成功了  dll   lib  也替换了   为啥我运行的时候还是用的CPU呢   设置的英伟达CUDA推理引擎

quary888

膜拜大神

阿丘

CUDA加速真香