My computer setting is ubuntu14.04+GeForce GTX TiTan X+cuda 8+opencv 2.4.9

sudo apt-get update

1. install cuda 8

install method can bu found online.

install command:

sudo dpkg -i cuda-repo-ubuntu1404-8-0-local-ga2_8.0.61-1_amd64.deb

sudo apt-get install cuda

remember environment setting  :export PATH=/usr/local/cuda-8.0/bin${PATH:+:${PATH}}

sudo apt-get install nvidia-cuda-toolkit

nvcc -V

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2016 NVIDIA Corporation
Built on Tue_Jan_10_13:22:03_CST_2017
Cuda compilation tools, release 8.0, V8.0.61

test :

cd /usr/local/cuda-8.0/bin/

sudo sh cuda-install-samples-8.0.sh /home/xxx/(object file to test samples)

make

/usr/local/cuda-8.0/bin/nvcc -ccbin g++ -I../../common/inc  -m64    -gencode arch=compute_20,code=compute_20 -o cudaModuleMgr.o -c cudaModuleMgr.cpp
nvcc warning : The 'compute_20', 'sm_20', and 'sm_21' architectures are deprecated, and may be removed in a future release (Use -Wno-deprecated-gpu-targets to suppress warning).

notice: limits of authority ,and can be sudo chmod 777 NVIDIA_CUDA-8.0_Samples/

may face :  376 version problems

./bandwidthTest

[CUDA Bandwidth Test] - Starting...
Running on...

 Device 0: GeForce GTX TITAN X
 Quick Mode

 Host to Device Bandwidth, 1 Device(s)
 PINNED Memory Transfers
   Transfer Size (Bytes)    Bandwidth(MB/s)
   33554432            11423.3

 Device to Host Bandwidth, 1 Device(s)
 PINNED Memory Transfers
   Transfer Size (Bytes)    Bandwidth(MB/s)
   33554432            12845.2

 Device to Device Bandwidth, 1 Device(s)
 PINNED Memory Transfers
   Transfer Size (Bytes)    Bandwidth(MB/s)
   33554432            249933.8

Result = PASS

NOTE: The CUDA Samples are not meant for performance measurements. Results may vary when GPU Boost is enabled.

2.  install  opencv

sudo apt-get install libgtk2.0-dev pkg-configsudo apt-get install build-essential
sudo apt-get install cmake

cmake.

sudo make

[ 94%] Built target opencv_perf_superres_pch_dephelp

[ 94%] Built target pch_Generate_opencv_perf_superres
[ 94%] Built target opencv_superres_pch_dephelp
[ 94%] Built target pch_Generate_opencv_superres
Linking CXX shared library ../../lib/libopencv_superres.so
[ 95%] Built target opencv_superres
Linking CXX executable ../../bin/opencv_perf_superres
[ 95%] Built target opencv_perf_superres
[ 96%] Built target opencv_test_superres_pch_dephelp
[ 96%] Built target pch_Generate_opencv_test_superres
Linking CXX executable ../../bin/opencv_test_superres
[ 96%] Built target opencv_test_superres
[ 97%] Built target opencv_videostab_pch_dephelp
[ 98%] Built target pch_Generate_opencv_videostab
[ 99%] Built target opencv_videostab
[100%] Built target opencv_haartraining_engine
[100%] Built target opencv_createsamples
[100%] Built target opencv_haartraining
[100%] Built target opencv_performance
[100%] Built target opencv_traincascade

sudo make insatll

-- Set runtime path of "/usr/local/bin/opencv_haartraining" to "/usr/local/lib:/usr/local/cuda/lib64"
-- Installing: /usr/local/bin/opencv_createsamples
-- Set runtime path of "/usr/local/bin/opencv_createsamples" to "/usr/local/lib:/usr/local/cuda/lib64"
-- Installing: /usr/local/bin/opencv_performance
-- Set runtime path of "/usr/local/bin/opencv_performance" to "/usr/local/lib:/usr/local/cuda/lib64"
-- Installing: /usr/local/bin/opencv_traincascade
-- Set runtime path of "/usr/local/bin/opencv_traincascade" to "/usr/local/lib:/usr/local/cuda/lib64"

test:

Scanning dependencies of target DisplayImage
[100%] Building CXX object CMakeFiles/DisplayImage.dir/DisplayImage.cpp.o
Linking CXX executable DisplayImage
[100%] Built target DisplayImage

./DisplayImage cartoon.jpg

download

1. git clone https://github.com/pjreddie/darknet  
2. cd darknet  
3. vim Makefile  
4. 修改，GPU=1  OPENCV=1  
5. wget http://pjreddie.com/media/files/yolo.weights  
6. make
   
7.  /bin/sh: 1: nvcc: not found make: *** [obj/convolutional_kernels.o] Error 127 ,This is  path problem;addexport PATH=/usr/local/cuda-8.0/bin:$PATHexport LD_LIBRARY_PATH=/usr/local/cuda-8.0/lib64:$LD_LIBRARY_PATH
   
8. ./darknet detect cfg/yolo.cfg yolo.weights data/horses.jpg
   

layer     filters    size              input                output
    0 conv     32  3 x 3 / 1   608 x 608 x   3   ->   608 x 608 x  32
    1 max          2 x 2 / 2   608 x 608 x  32   ->   304 x 304 x  32
    2 conv     64  3 x 3 / 1   304 x 304 x  32   ->   304 x 304 x  64
    3 max          2 x 2 / 2   304 x 304 x  64   ->   152 x 152 x  64
    4 conv    128  3 x 3 / 1   152 x 152 x  64   ->   152 x 152 x 128
    5 conv     64  1 x 1 / 1   152 x 152 x 128   ->   152 x 152 x  64
    6 conv    128  3 x 3 / 1   152 x 152 x  64   ->   152 x 152 x 128
    7 max          2 x 2 / 2   152 x 152 x 128   ->    76 x  76 x 128
    8 conv    256  3 x 3 / 1    76 x  76 x 128   ->    76 x  76 x 256
    9 conv    128  1 x 1 / 1    76 x  76 x 256   ->    76 x  76 x 128
   10 conv    256  3 x 3 / 1    76 x  76 x 128   ->    76 x  76 x 256
   11 max          2 x 2 / 2    76 x  76 x 256   ->    38 x  38 x 256
   12 conv    512  3 x 3 / 1    38 x  38 x 256   ->    38 x  38 x 512
   13 conv    256  1 x 1 / 1    38 x  38 x 512   ->    38 x  38 x 256
   14 conv    512  3 x 3 / 1    38 x  38 x 256   ->    38 x  38 x 512
   15 conv    256  1 x 1 / 1    38 x  38 x 512   ->    38 x  38 x 256
   16 conv    512  3 x 3 / 1    38 x  38 x 256   ->    38 x  38 x 512
   17 max          2 x 2 / 2    38 x  38 x 512   ->    19 x  19 x 512
   18 conv   1024  3 x 3 / 1    19 x  19 x 512   ->    19 x  19 x1024
   19 conv    512  1 x 1 / 1    19 x  19 x1024   ->    19 x  19 x 512
   20 conv   1024  3 x 3 / 1    19 x  19 x 512   ->    19 x  19 x1024
   21 conv    512  1 x 1 / 1    19 x  19 x1024   ->    19 x  19 x 512
   22 conv   1024  3 x 3 / 1    19 x  19 x 512   ->    19 x  19 x1024
   23 conv   1024  3 x 3 / 1    19 x  19 x1024   ->    19 x  19 x1024
   24 conv   1024  3 x 3 / 1    19 x  19 x1024   ->    19 x  19 x1024
   25 route  16
   26 conv     64  1 x 1 / 1    38 x  38 x 512   ->    38 x  38 x  64
   27 reorg              / 2    38 x  38 x  64   ->    19 x  19 x 256
   28 route  27 24
   29 conv   1024  3 x 3 / 1    19 x  19 x1280   ->    19 x  19 x1024
   30 conv    425  1 x 1 / 1    19 x  19 x1024   ->    19 x  19 x 425
   31 detection
mask_scale: Using default '1.000000'
Loading weights from yolo.weights...Done!
data/horses.jpg: Predicted in 0.697368 seconds.
horse: 46%
horse: 59%
cow: 26%
horse: 91%

0.697/s,time is too long ,next will make use of cudnn6

Loading weights from weights/yolo.weights...Done!
data/horses.jpg: Predicted in 0.028641 seconds.
horse: 46%
horse: 59%
cow: 26%
horse: 91%

improve 45 times. 

picture:  ./darknet detect cfg/yolo.cfg weights/yolo.weights data/horses.jpg

video: ./darknet detector demo cfg/coco.data cfg/yolo.cfg weights/yolo.weights videos/PICT6595.mp4

usb-camera: ./darknet detector demo cfg/coco.data cfg/yolo.cfg weights/yolo.weights

anyway, feel recognition not good well. 

next, will be select suitable device to run this algorithm , and optimize in order for our own apply.