動作環境
CUDA10 cuDNN OpenCV3.4.0 darknet YOLOv3
Windows10
VisualStudio2017
install CUDA 10
install cuDNN
install OpenCV(3.4.0)
donwload pre-build 3.4.0
https://sourceforge.net/projects/opencvlibrary/files/opencv-win/
https://docs.opencv.org/3.4.0/d3/d52/tutorial_windows_install.html#tutorial_windows_install_path
darknet
add system variable.
OPENCV_PATH D:\OpenCV\Build\x64\vc15
ここからダウンロードする。
https://github.com/AlexeyAB/darknet
darknet.slnを開き、プロジェクトのプロパティを設定する。
c/c++ > add header folder
D:\opencv\opencv-3.4.0\opencv\build\include
c/c++ > Linker > add library folder
D:\opencv\opencv-3.4.0\opencv\build\x64\vc15\lib
CUDA C/C++
CUDA Toolkit Custom Dir
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0
D:\opencv\opencv-3.4.0\opencv\build\include
copy files.
folder構成
D:\darknet\build\darknet\x64\
darknet.exe
opencv_world340.dll # copy from opencv
opencv_ffmpeg340_64.dll # copy from opencv
cudnn64_7.dll # copy from cuDNN
D:\darknet\build\darknet\x64\card\
card.data
classes.txt
darknet53.conv.74 #Pre-trained file
test.txt
train.txt
yolov3.cfg
D:\darknet\build\darknet\x64\card\img\
IMG1.jpg
IMG2.jpg
IMG3.jpg
IMG1.txt
IMG2.txt
IMG3.txt
# annotation file (labaeImage.pyでバウンディングボックスに囲む）
D:\darknet\build\darknet\x64\card\backup
card-yolov3_last.weights
card.data
classes = 1
train = card/train.txt
valid = card/test.txt
names = card/classes.txt
backup = card/backup
test.txt/train.txt
card/img/IMG_0828.jpg
card/img/IMG_0840.jpg
card/img/IMG_0829.jpg
card/img/IMG_0841.jpg
画像ファイルを20％､80%にわける（spsplitTrainAndTest.py(imgfoler)）
classes.txt(cardという種別一種類のとき)
card
Run
darknet.exe detector train card/card.data card/yolov3.cfg darknet53.conv.74
Annotationfile