从零搭建Mask R-CNN环境

1.下载安装Anaconda

https://www.anaconda.com/products/individual

2.配置tensorflow-keras环境

conda create –n tensorflow-gpu python=3.6
activate tensorflow-gpu 
pip install tensorflow-gpu==1.13.2
pip install keras==2.1.5

3.配置mask rcnn依赖环境

3.1安装skimage

pip install scikit-image==0.15.0

3.2安装matplotlib

pip install matplotlib

3.3安装opencv

conda install opencv==3.4.2

3.4安装imgaug

pip install imgaug

3.5安装pycocotools

3.5.1下载

https://github.com/philferriere/cocoapi下载源码，下载位置任意，并进行解压。打开 conda prompt 终端，并切换到cocoapi\PythonAPI目录下

3.5.2运行指令

# install pycocotools locally （先这一条，结束后再下一条）
python setup.py build_ext --inplace
		
# install pycocotools to the Python site-packages
python setup.py build_ext install

3.5.3可能出现的问题

（1）No module named 'Cython'

pip install Cython --install-option="--no-cython-compile"

（2）No module named 'cv2.cv2'

因为装的opencv是4.0.2版本的

卸载重装为3.4.2版本

 https://stackoverflow.com/questions/55360459/no-module-named-cv2-cv2

（3）No module named 'pycocotools._mask'

https://www.twblogs.net/a/5ca71e63bd9eee59d33308d5/zh-cn

完成上述安装与配置，将pycocotools复制到项目路径下

 （4）有效解决出现AttributeError: module 'labelme.utils' has no attribute 'draw_label'错误

https://blog.csdn.net/qq_34997364/article/details/105152927?utm_medium=distribute.pc_relevant.none-task-blog-BlogCommendFromBaidu-5&depth_1-utm_source=distribute.pc_relevant.none-task-blog-BlogCommendFromBaidu-5

换labelme版本

（5）AttributeError: 'Model' object has no attribute 'metrics_tensors'

https://blog.csdn.net/DeMeng33/article/details/103953984