Today set up on the server CTPN training environment, the problem can be described as one after another, the whole I have the whole person is not good, so I write this one is recording it, and second, hope that their experience can help to everyone stepped pit, ado, open the whole!
Let me talk about my configuration, cuda10.0, tensorflow1.13 (As for my use of tensor why this is due to the 1.13 version of my cuda restrictions on correspondence between tensorflow and cuda version you can refer to this article https: / /blog.csdn.net/qq_35451572/article/details/92772749 )
other technical documents are in accordance with ctpn in the versions of the environment in order to facilitate migration of Han, personal recommendations, ctpn the virtual environment to build the best in the conda conduct environment.
According to tensorflow technical documentation requirements in the setup of the corresponding version installed, cython, opencv-python There easydict then execute code in the shell (remember we must execute in the corresponding directory, or else a lot of trouble) if the execution process occurs the error to open the setup.sh file, then all the commands one by one input to the terminal, if you are using a graphics card and lazy, then, that paramenters can skip directly to the demo this step, perform demo will find 
this happen then install pillow (pip install pillow)
implementation of new problems have emerged once again 
installed PyYAML (pip install PyYAML)
After the installation is complete you can successfully run the demo
Next is the training process, or be in accordance with the technical documentation, prepare data, first download the pre-training model put it in the data / pretrain / VGG_imagenet.npy in
and then tune the parameters of good split_label.py to ensure that the training data to find data processing continues in accordance with the document
to the ln -s TEXTVOC VOCdevkit2007 this step, the TEXTVOC copied from prepare_training_data to data, and then data in VOCdevkit2007 deleted, and then execute ln -s TEXTVOC VOCdevkit2007
implementation of training script, and found a problem
then loaded pip install scipy
Then came the highlight of
this problem needs to be changed is /ctpn/lib/fast_rcnn/train.py this file, log_image 74th row = gen_logging_ops._image_summary (log_image_name, tf.expand_dims (log_image_data , 0), max_images = 1 ) into log_image = gen_logging_ops.image_summary (log_image_name, tf.expand_dims ( log_image_data, 0), max_images = 1) ok, just one more pit father _
The next question then continue
execution pip install numpy == 1.16.2 will be happy de trained
ctpn / lib / fast_rcnn 134 plus a back, end = '_'