datitran--generate_tfrecord.py使用踩坑记录

版权声明：本文为博主原创文章，未经博主允许不得转载。 https://blog.csdn.net/m0_38088298/article/details/87714005

• 系统环境：
  macOS Mojave
  virtualenv
  python 2.7
• 坑点一：No module named ‘object_detection’
  解决方式：添加环境变量
  a.使用终端打开文件

  touch .bash_profile
  open -e .bash_profile

b.在文件中添加并保存环境变量
添加

PYTHONPATH=$PYTHONPATH:/your/path/to/tensorflow/models/research:/your/path/to/tensorflow/models/research/slim

c.到配置文件中
保存

source .bash_profile

如还不能解决参考：https://github.com/datitran/raccoon_dataset/issues/14

 - https://github.com/tensorflow/models
   Download.
 - Go to folder "research".
 - Open terminal: cd Downloads/models-master/research //!!!进入到tensorflow/models/research目录下
4.(for 2.7python) python setup.py build
(for 3.6python) python3.6 setup.py build
5.(for 2.7python) sudo python setup.py install
(for 3.6python) sudo python3.6 setup.py install
 - You'll see : "Finished processing dependencies for object-detection==0.1"

• 坑点二：tensorflow.python.framework.errors_impl.NotFoundError: image_0068.png; No such file or directory
  解决方法：
  执行时添加一个路径指定图片位置

python generate_tfrecord.py  --csv_input=flower1_train_labels.csv --output_path=flower1_train.record --image_dir=img

• 坑点三：return tf.train.Feature(int64_list=tf.train.Int64List(value=value))
  TypeError: None has type NoneType, but expected one of: int, long
  解决方法：
  修改源码，https://github.com/datitran/raccoon_dataset/issues/14
  
  将红圈处类名修改为你用labelImg保存的类名。