Windows 实现 Faster R-CNN 编译

系统:win10家庭版
显卡:GTX1660Ti
驱动版本:441.22
cuda版本:10.2.89
cudnn版本:7.6.5.32
tensorflow-gpu版本:1.13.1
opencv-python版本:3.4.2
cython
easydict
matplotlib
scipy
Pillow

1.安装cuda、cudnn

参考:https://blog.csdn.net/u013925378/article/details/91046639

2.安装tensorflow-gpu

参考:https://blog.csdn.net/weixin_43318717/article/details/94433790

3.下载Faster-RCNN-TensorFlow-Python3-master.zip压缩包并且解压

链接:https://github.com/dBeker/Faster-RCNN-TensorFlow-Python3

4.cmd到…/Faster-RCNN-TensorFlow-Python3-master目录下,运行安装python软件包(cython,python-opencv,easydict)

cd C:\Users\SSC\Desktop\Faster-RCNN-TensorFlow-Python3-master
pip install -r requirements.txt

5.修改…Faster-RCNN-TensorFlow-Python3-master/data/coco/PythonAPI/setup.py文件:在第15行加上

,
    Extension( 'lib.utils.cython_bbox',
               sources=['../../../lib/utils/bbox.c','../../../lib/utils/bbox.pyx'],
               include_dirs = [np.get_include(), '/lib/utils'], 
               extra_compile_args=[], )

改完之后的setup.py文件如下:

Windows 实现 Faster R-CNN 编译_第1张图片

6.由于没有bbox.c和blob.py文件,所以要cmd到…Faster-RCNN-TensorFlow-Python3-master/lib/utils目录下,执行

cd C:\Users\SSC\Desktop\Faster-RCNN-TensorFlow-Python3-master\lib\utils
python setup.py build_ext --inplace

生成cython_bbox.ccython_bbox.pyx,然后将这两个改名为bbox.cbbox.pyx

7.cmd到…Faster-RCNN-TensorFlow-Python3-master/data/coco/PythonAPI目录下,运行

cd C:\Users\SSC\Desktop\Faster-RCNN-TensorFlow-Python3-master\data\coco\PythonAPI
python setup.py build_ext --inplace
python setup.py build_ext install

8.下载数据集

链接:https://github.com/rbgirshick/py-faster-rcnn#beyond-the-demo-installation-for-training-and-testing-models

VOCtrainval_06-Nov-2007.tar:
http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtrainval_06-Nov-2007.tar
VOCtest_06-Nov-2007.tar:
http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtest_06-Nov-2007.tar
VOCdevkit_08-Jun-2007.tar:
http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCdevkit_08-Jun-2007.tar

9.在…Faster-RCNN-TensorFlow-Python3-master/data目录下,新建VOCdevkit2007文件夹,将三个数据集文件同时解压到VOCdevkit2007文件夹下,并将其中的VOC2007文件夹复制到VOCdevkit2007文件夹下。

Windows 实现 Faster R-CNN 编译_第2张图片

10.下载经过训练的VGG16 ,之后在…Faster-RCNN-TensorFlow-Python3-master/data目录下新建imagenet_weights文件夹,并将下载解压好的vgg_16.ckpt文件改名为vgg16.ckpt后复制到imagenet_weights文件夹下。

下载地址:http://download.tensorflow.org/models/vgg_16_2016_08_28.tar.gz

11.cmd到…Faster-RCNN-TensorFlow-Python3-master目录下,运行

python train.py

之后是漫长的等待,大约七八个小时之后吧。。。

12.修改…Faster-RCNN-TensorFlow-Python3-master目录下的demo.py文件。

39行:NETS 里面修改为自己训练好的模型文件名
40行:DATASETS 删除 “+ voc_2012_trainval”
在这里插入图片描述
108行:default 修改为 vgg16
110行:default 修改为 pascal_voc
Windows 实现 Faster R-CNN 编译_第3张图片

13.新建…Faster-RCNN-TensorFlow-Python3-master/output/vgg16/voc_2007_trainval/default文件夹,并将…Faster-RCNN-TensorFlow-Python3-master/default/voc_2007_trainval/default文件夹下训练的模型复制到刚才新建的default文件夹下。

Windows 实现 Faster R-CNN 编译_第4张图片

14.cmd到…Faster-RCNN-TensorFlow-Python3-master目录下,运行

python demo.py

结果如下:

...
...
Detection took 3.802s for 300 object proposals
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Demo for data/demo/000457.jpg
Detection took 0.094s for 300 object proposals
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Demo for data/demo/000542.jpg
Detection took 0.094s for 300 object proposals
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Demo for data/demo/001150.jpg
Detection took 0.094s for 300 object proposals
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Demo for data/demo/001763.jpg
Detection took 0.101s for 300 object proposals
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Demo for data/demo/004545.jpg
Detection took 0.109s for 300 object proposals

C:\Users\SSC\Desktop\Faster-RCNN-TensorFlow-Python3-master>

你可能感兴趣的:(目标检测,Faster,R-CNN)