FCHD在window10（pytorch1.0,cuda9.0,cudnn7.1,python3.6）上测试

1.安装相关环境

pytorch安装
https://blog.youkuaiyun.com/loco1223/article/details/91354337

其他小环境

pip install ipdb
pip install opencv-python
pip install scipy
pip install scikit-image
pip install matplotlib
pip install cupy-cuda80//我的cuda版本是8.0

以及torchnet
https://blog.youkuaiyun.com/weixin_43264516/article/details/83187775
下载torchnet：
https://github.com/pytorch/tnt
用Anaconda Prompt激活环境后cd到所在文件目录下
安装torchnet：

python setup.py build
python setup.py install

2.初步检测

在github上下载相关模型


两个都是需要的，VGG模型应该是用于前部分衔接。

将两个模型直接放置于解压后的工程下

修改部分代码

src目录下的config.py文件内caffe_pretrain_path路径
一开始这些路径都是为ubuntu设计的。

在终端输入代码

python head_detection_demo.py --img_path 1.png --model_path head_detector_final

会报错，但会有结果，暂时不知道怎么修改。

3.改正框位置不正确

参考
https://blog.youkuaiyun.com/missyougoon/article/details/90408970

4.some

1、a bytes-like object is required,not ‘str’：
https://blog.youkuaiyun.com/csu_vc/article/details/78372932/

2、安装SQLite3：
https://blog.youkuaiyun.com/say_hi_andhelloworld/article/details/86062102

3、TypeError: le() received an invalid combination of arguments - got (numpy.ndarray), but expected one of:

• (Tensor other)
  didn’t match because some of the arguments have invalid types: (numpy.ndarray)
• (Number other)
  didn’t match because some of the arguments have invalid types: (numpy.ndarray)
  https://github.com/aditya-vora/FCHD-Fully-Convolutional-Head-Detector/issues/28
  add this code ‘min_size=np.array(min_size)’ before ‘keep = np.where((hs >= min_size) & (ws >= min_size))[0]’

4、 the python code for non_maximum_suppression is about 2x slow
It is strongly recommended to build cython code:
cd model/utils/nms/; python3 build.py build_ext --inplacecd model/utils/nms/; python3 build.py build_ext --inplace’’’)
https://github.com/aditya-vora/FCHD-Fully-Convolutional-Head-Detector/issues/27

5、pytorch报错Implicit dimension choice for softmax has been deprecated. Change the call to include dim=X
https://blog.youkuaiyun.com/legalhighhigh/article/details/82819747

注：

目前基本放弃在win10上测试了，需要的python2.7环境等无法win10上很好的创建，比如CUDA9.0 没有支持python2.7的版本，以及遇到的无法联网记录等问题，决定接下来在ubuntu上进行测试，注定是个很漫长的过程。
待续吧。。。