今天是2019年11月7日
把 https://github.com/tensorflow/models.git 下载下来之后,发现 https://github.com/tensorflow/models/blob/master/research/object_detection/object_detection_tutorial.ipynb 的步骤有些更新,比之前更详细一些,大致几条shell命令为
sudo apt install protobuf-compiler
pip install Cython
pip install contextlib2
pip install pillow
pip install lxml
pip install matplotlib
pip install pycocotools
pip install -U --pre tensorflow=="2.*"
cd models/research/
protoc object_detection/protos/*.proto --python_out=.
cd models/research
pip install .
然后安装就完成了
如果要测试的话,就是前向预测,可以根据 https://github.com/tensorflow/models/blob/master/research/object_detection/object_detection_tutorial.ipynb 来跑一下,看看结果
————————————————————————————————————————
如果训练的话,现在(2019年11月7日)还是要用TensorFlow1.x版本训练,我安装了TensorFlow1.14.0,先处理数据集,参考 https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/preparing_inputs.md, 先下载解压缩(下面创建的路径为models/research/data/)
mkdir data
cd data
wget -c http://www.robots.ox.ac.uk/~vgg/data/pets/data/images.tar.gz
wget -c http://www.robots.ox.ac.uk/~vgg/data/pets/data/annotations.tar.gz
tar -xvf annotations.tar.gz
tar -xvf images.tar.gz
然后应该创建record文件了,然后执行下面的脚本(下面的路径为models/research/):
python object_detection/dataset_tools/create_pet_tf_record.py \
--label_map_path=object_detection/data/pet_label_map.pbtxt \
--da