window10+tensorflow+Faster-RCNN

（1）基本的配置，可以去百度，包括tensorflow numpy scipy Image什么的，注意安装顺序，和包的安装方式，最好都pip install

         从这里下载相应的wel，http://www.lfd.uci.edu/~gohlke/pythonlibs/

，然后安装

（2）在https://github.com/dBeker/Faster-RCNN-TensorFlow-Python3.5 下载了源码。然后按照他的read me 一步一步。

（3）- Install tensorflow, preferably GPU version. Follow [instructions]( https://www.tensorflow.org/install/install_windows). If you do not install GPU version, you need to comment out all the GPU calls inside code and replace them with relavent CPU ones.


（4）Install python packages (cython, python-opencv, easydict)pip install 就好


（5）Checkout this branch


（6）编译，有坑，见上一篇文章Go to  ./data/coco/PythonAPI


      Run `python setup.py build_ext --inplace`


      Run `python setup.py build_ext install`


（7） Follow this instruction to download PyCoco database. [Link]( https://github.com/rbgirshick/py-faster-rcnn#beyond-the-demo-installation-for-training-and-testing-models)


I will be glad if you can contribute with a batch script to automatically download and fetch. The final structure has to look like


        "data/VOCDevkit2007/annotations_cache"
  
        "data/VOCDevkit2007/VOC2007"
  
 （8） Download pre-trained VGG16 from [here](http://download.tensorflow.org/models/vgg_16_2016_08_28.tar.gz) and place it as "data\imagenet_weights\vgg16.ckpt"
 
 For rest of the models, please check [here](https://github.com/tensorflow/models/tree/master/research/slim#pre-trained-models)
 
 （9）- Run train.py
  
  Notify me if there is any issue found. Please note that, I have compiled cython modules with sm61 architecture (GTX 1060, 1070 etc.). Compile support for other architectures will be added. 
 

在开始编译setup.py的时候，一直出现各种错误，比如找不到cant open vcvarsall.bat，改了vs的版本，还是有问题。在E:\Program Files\Anaconda3\Lib\distutils\_msvccompiler.py里面修改了if version >= 14 and version > best_version:将他改成更低版本，又出现了新的错误。最后重装vs2015解决了问题。

三个python库。

然后在编译运行setup.py就没有错了。运行train 可以用python 直接运行，也可以用 cmd   python train.py --run 运行。