Mask_RCNN安装(ROS环境下)

github-code
因为一开始MASKRCNN使用COCO训练的，所以其识别分割种类有80种

教程步骤:
1.conda create -n maskrcnn python=3.4
2.pip3 install -r requirements.txt
报错如下，但是不做处理

ERROR: scikit-image 0.15.0 has requirement pillow>=4.3.0, but you'll have pillow 3.1.2 which is incompatible.
ERROR: markdown 3.1.1 has requirement setuptools>=36, but you'll have setuptools 20.7.0 which is incompatible.
ERROR: tensorboard 2.1.0 has requirement requests<3,>=2.21.0, but you'll have requests 2.9.1 which is incompatible.
ERROR: tensorboard 2.1.0 has requirement setuptools>=41.0.0, but you'll have setuptools 20.7.0 which is incompatible.
ERROR: google-auth 1.11.0 has requirement setuptools>=40.3.0, but you'll have setuptools 20.7.0 which is incompatible.

后面在maskrcnn的环境下再次运行，(可能上一次在其他conda环境)重新下载，后中途报错

ERROR: grpcio 1.26.0 requires enum34>=1.0.4, which is not installed.
ERROR: grpcio 1.26.0 requires futures>=2.2.0, which is not installed.
ERROR: tensorboard 1.14.0 has requirement setuptools>=41.0.0, but you'll have setuptools 27.2.0 which is incompatible

使用pip3 install setuptools==41.0 不全setuptools的版本

3.python3 setup.py install
Finished processing dependencies for mask-rcnn==2.1
4.Download mask_rcnn_coco.h5(我是用手机挂tizi然后传到电脑)存在root director下
5.(最好安装)下载pycocotools来train or test on MS COCO
github上也有关于MS COCO Requirements的说明
Linux中直接使用：
pip3 install pycocotools

How To Get Started
各种jupyter notebook的作用，这里不做翻译了

    demo.ipynb Is the easiest way to start. It shows an example of using a model pre-trained on MS COCO to segment objects in your own images. It includes code to run object detection and instance segmentation on arbitrary images.

    train_shapes.ipynb shows how to train Mask R-CNN on your own dataset. This notebook introduces a toy dataset (Shapes) to demonstrate training on a new dataset.

    (model.py, utils.py, config.py): These files contain the main Mask RCNN implementation.

    inspect_data.ipynb. This notebook visualizes the different pre-processing steps to prepare the training data.

    inspect_model.ipynb This notebook goes in depth into the steps performed to detect and segment objects. It provides visualizations of every step of the pipeline.

    inspect_weights.ipynb This notebooks inspects the weights of a trained model and looks for anomalies and odd patterns.

在maskrcnn的目录cmd下输入jupyter notebook
运行samples/demo
如果安装了ros，则会报错

ImportError: /opt/ros/kinetic/lib/python2.7/dist-packages/cv2.so: undefined symbol: PyCObject_Type
参考:https://blog.csdn.net/qq_34544129/article/details/81946494
分析:不一定是ROS的缘故，还有可能是多版本python的缘故

解决方案:
在bashrc加上

export PYTHONPATH="/home/asber/Document/anaconda2/lib/python2.7/site-packages:$PYTHONPATH"

之后进入maskrcnn执行jupyter notebook报错

importerror: this package should not be accessible on python 3. either you are trying to run from the python-future src folder or your install…

这个问题在shell执行unset PYTHONPATH可以解决

我在base环境下jupyter notebook报错
UnicodeDecodeError: ‘ascii’ codec can’t decode byte 0xe5 in position 4: ordinal not in range(128)
在gedit /home/asber/Document/anaconda2/lib/python2.7/site-packages/jupyter_core/application.py
import sys
reload(sys)
sys.setdefaultencoding(‘utf-8’)
之后还是编码报错
[C 11:28:39.943 NotebookApp] Bad config encountered during initialization:
[C 11:28:39.943 NotebookApp] Could not decode ‘\xe6\x9c\xaa\xe5\x91\xbd\xe5\x90\x8d’ for unicode trait ‘untitled_notebook’ of a LargeFileManager instance.
最后gedit /home/asber/Document/anaconda2/lib/python2.7/site-packages/traitlets/traitlets.py
change line 2050, from return value.decode(‘ascii’, ‘strict’) to return value.decode(‘UTF-8’, ‘strict’)
解决错误，参考

运行结果

PS:最后github上还介绍了一个3Ddataset for 点云分割
参考:
1.此处还有使用cv2读取摄像头进行分割的代码
2.Mask RCNN 算法笔记
3.Mask-RCNN模型的实现自定义对象(无人机)检测
4.『深度应用』一小时教你上手训练MaskRCNN·Keras开源实战（Windows&Linux）