faster rcnn训练

一. 基本的训练教程见：

1. github地址：https://github.com/smallcorgi/Faster-RCNN_TF.git

2.教程地址： 

   https://blog.csdn.net/weixin_37203756/article/details/79926543

   https://blog.csdn.net/hitzijiyingcai/article/details/81914200?depth_1-utm_source=distribute.pc_relevant.none-task&utm_source=distribute.pc_relevant.none-task

二. 在训练过程过遇到的问题：

1. 创建环境：
 source activate base
 conda env create --name faster_rcnn python=2.7
 pip install 
2. 安装tensorflow
 2.1 查询本机的cuda和cudnn
 cat /usr/local/cuda/version.txt
 cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A 2
 2.2 选择匹配的tensorflow-gpu的版本：https://www.jianshu.com/p/1d6a508a8d04
 2.3 安装对应的tensorflow-gpu命令：https://blog.csdn.net/z349177893/article/details/101535203

3. 数据上传服务器：rz

4. 生成train.txt、val.txt、trainval.txt、test.txt文件，（参考自https://blog.csdn.net/hitzijiyingcai/article/details/81636455，感谢大佬）

import os
import random
 
trainval_percent = 0.5
train_percent = 0.5
xmlfilepath = '/home/hqd/桌面/VOC2010/Annotations'
txtsavepath = '/home/hqd/桌面/VOC2010/ImageSets/Main'
total_xml = os.listdir(xmlfilepath)
 
num=len(total_xml)
list=range(num)
tv=int(num*trainval_percent)
tr=int(tv*train_percent)
trainval= random.sample(list,tv)
train=random.sample(trainval,tr)
 
ftrainval = open(txtsavepath+'/trainval.txt', 'w')
ftest = open(txtsavepath+'/test.txt', 'w')
ftrain = open(txtsavepath+'/train.txt', 'w')
fval = open(txtsavepath+'/val.txt', 'w')
 
for i  in list:
    name=total_xml[i][:-4]+'\n'
    if i in trainval:
        ftrainval.write(name)
        if i in train:
            ftrain.write(name)
        else:
            fval.write(name)
    else:
        ftest.write(name)
 
ftrainval.close()
ftrain.close()
fval.close()
ftest .close()

4. _ZN10tensorflow7strings6StrCatB5cxx11ERKNS0_8AlphaNumE错误
在lib中的make.sh中加入-D_GLIBCXX_USE_CXX11_ABI=0：
g++ -std=c++11 -shared -o roi_pooling.so roi_pooling_op.cc \
    roi_pooling_op.cu.o -I $TF_INC  -D GOOGLE_CUDA=1 -fPIC $CXXFLAGS  -D_GLIBCXX_USE_CXX11_ABI=0 \
    -lcudart -L $CUDA_PATH/lib64

5. _tkinter.TclError: no display name and no $DISPLAY environment variable错误
    5.1 将import matplotlib.pyplot as plt
    import matplotlib
    matplotlib.use('Agg')
    修改成：
    import matplotlib
    matplotlib.use('Agg')
    import matplotlib.pyplot as plt

   5.2安装xming：https://blog.csdn.net/qq_24036403/article/details/86535401

6. 命令输入：./experiments/scripts/faster_rcnn_end2end.sh $DEVICE $DEVICE_ID VGG16 pascal_voc
应该为： ./experiments/scripts/faster_rcnn_end2end.sh --device gpu --device_id 0 VGG16 pascal_voc

6. raise ValueError("Object arrays cannot be loaded when "
numpy版本太高，切换成低版本的
pip install numpy==1.16.2

7. 在后台运行的命令：nohup ./experiments/scripts/faster_rcnn_end2end.sh gpu 0 VGG16 pascal_voc > test.log 2>&1 &

   实时查看的命令：tail -f test.log/cat test.log

demo运行命令：python ./tools/demo.py --model /share2/home/qingdong/sharedir/Faster-RCNN_TF/tools//model/VGGnet_fast_rcnn_iter_70000.ckpt

8. make时报错：

python setup.py build_ext --inplace
  File "setup.py", line 84
    print extra_postargs
                       ^
SyntaxError: Missing parentheses in call to 'print'. Did you mean print(extra_postargs)?
Makefile:2: recipe for target 'all' failed

可能是你的没有进入python的环境，需要先切换成有tensorflow的python，再运行make，就可以成功。

9. 用自己的数据集报错：

   问题1：https://blog.csdn.net/a8039974/article/details/77592385/
   问题2：

Preparing training data...
Traceback (most recent call last):
  File "./tools/train_net.py", line 83, in <module>
    roidb = get_training_roidb(imdb)
  File "/home/user/work/faster_rcnn/Faster-RCNN_TF/tools/../lib/fast_rcnn/train.py", line 212, in get_training_roidb
    rdl_roidb.prepare_roidb(imdb)
  File "/home/user/work/faster_rcnn/Faster-RCNN_TF/tools/../lib/roi_data_layer/roidb.py", line 27, in prepare_roidb
    roidb[i]['image'] = imdb.image_path_at(i)
IndexError: list index out of rang

     解决方案：删除fast-rcnn-master/data/cache/ 文件夹下的.pkl文件，或者改名备份，重新训练即可。

问题3：https://www.jianshu.com/p/08c1faa38358

更多问题，请参考博客：https://blog.csdn.net/qq_34116292/article/details/90715649