ubuntu 18 tensorflow object detection API

1. 下载

地址	下载文件
https://github.com/protocolbuffers/protobuf/releases	protoc-3.10.0-linux-x86_64.zip
https://github.com/tensorflow/models	models-master.zip
https://github.com/cocodataset/cocoapi	cocoapi-master.zip

安装jupyter

$ pip3 install ipython[notebook]
$ pip3 install jupyter
$ sudo apt install jupyter-core
$ sudo apt install jupyter-notebook

2. 解压

$ unzip models-master.zip
$ unzip cocoapi-master.zip
$ unzip protoc-3.10.0-linux-x86_64.zip -d protoc-3.10.0
$ sudo cp protoc-3.10.0/bin/protoc /usr/bin/protoc

3. 编译cocoapi

$ pip3 install --user Cython
$ cd cocoapi-master/PythonAPI
$ vi Makefile

将pip改为pip3

$ make
$ sudo make install
$ cp -r pycocotools/ ../../models/research/

4. 编译

$ cd ../../models/research/
$ protoc object_detection/protos/*.proto --python_out=.

$ cd slim
$ python3 setup.py build
$ sudo python3 setup.py install

5. 添加环境变量

$ vi ~/.bashrc

添加：

export PYTHONPATH=$PYTHONPATH:'pwd':'pwd'/slim

$ source ~/.bashrc

6. 测试

$ cd ../
$ python3 object_detection/builders/model_builder_test.py

models下打开jupyter-notebook，并运行research/object_detection/object_detection_tutorial.ipynb

如果不显示图片，修改如下:

# plt.imshow(image_np)
# plt.show()
img = Image.fromarray(image_np, 'RGB')
img.show()