MTCNN-Tensorflow
1、得到一个人脸图片,使用MTCNN算法进行图片中的人脸检测,返回人脸边框坐标数据、人脸特征5个关键点(左眼、右眼、鼻尖、左嘴角、右嘴角)
https://github.com/LeslieZhoa/tensorflow-MTCNN
https://github.com/AITTSMD/MTCNN-Tensorflow
Deepgaze 头部姿态估计
Deepgaze基于OpenCV和Tensorflow,是一个利用卷积神经网络(Convolutional Neural Networks, CNNs)进行人脸检测、头部姿态估计和分类的人机交互、人物检测和跟踪库。
2、头部姿态识别,筛选正脸的图片进行使用
参考文章:https://www.jianshu.com/p/9844be7e33fd
GitHub:https://github.com/mpatacchiola/deepgaze
MXNet+insightface人脸分析
提取特征 欧氏距离计算相似度
https://blog.csdn.net/weixin_44005915/article/details/96484330
https://github.com/deepinsight/insightface
MXNet-GPU 运行时 记录
win10 显卡RTX2080i CUDA 9.0 运行MXNetError错误:Check failed: e == CUBLAS_STATUS_SUCCESS (13 vs. 0) : cuBLAS: CUBLAS_STATUS_EXECUTION_FAILED
需要安装CUDA9.0的补丁2 地址:https://developer.nvidia.com/cuda-90-download-archive?target_os=Windows&target_arch=x86_64&target_version=10&target_type=exenetwork

这个方法对我不管用,于是重新安装CUDA等
tensorflow-gpu 环境搭建(win10)
安装CUDA
CUDA 10.0 下载地址:https://developer.nvidia.com/cuda-toolkit-archive
安装完成之后,添加环境变量。


Path中添加
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0\bin
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0\libnvvp
查询版本验证安装是否成功
nvcc -V

安装cuDNN
cuDNN和CUDA版本有对应关系。
cuDNN 7.6.5 for CUDA 10.0] 下载地址:https://developer.nvidia.com/rdp/cudnn-archive
下载之后,将对应的文件复制到cuda目录下,具体如下:
1. cuda\bin\cudnn64_7.dll to C:\Program Files
\NVIDIA GPU Computing Toolkit\CUDA\v10.0\bin.
2. cuda\ include\cudnn.h to C:\Program Files
\NVIDIA GPU Computing Toolkit\CUDA\v10.0\include.
3. cuda\lib\x64\cudnn.lib to C:\Program Files
\NVIDIA GPU Computing Toolkit\CUDA\v10.0\lib\x64
安装tensorflow-gpu
tensorflow和CUDA之间有版本关系限制,可自行百度,这里我安装的是gpu1.14版本。
pip install tensorflow-gpu==1.14 -i https://mirrors.aliyun.com/pypi/simple/
安装MXNet
mxnet-cu100 对应cuda10
pip install mxnet-cu100 -i https://mirrors.aliyun.com/pypi/simple/
遇到的问题
Tensorflow报错
WARNING:tensorflow:Entity <bound method SeparableConv2D.call of <tensorflow.python.layers.convolutional.SeparableConv2D object at 0x0000011F446E9668>> could not be transformed and will be executed as-is. Please report this to the AutgoGraph team. When filing the bug, set the verbosity to 10 (on Linux, export AUTOGRAPH_VERBOSITY=10) and attach the full output. Cause: converting <bound method SeparableConv2D.call of <tensorflow.python.layers.convolutional.SeparableConv2D object at 0x0000011F446E9668>>: AssertionError: Bad argument number for Name: 3, expecting 4
可以卸载掉高版本的gast 再安装 pip install gast==0.2.2