安装python faster-RCNN
https://blog.csdn.net/t5131828/article/details/53302503
下面两句亲测有效，训练命令
./experiments/scripts/faster_rcnn_end2end.sh 0 VGG16 pascal_voc
./experiments/scripts/faster_rcnn_end2end.sh 0 ZF pascal_voc
然后回碰见faster-rcnn系列assert (boxes[:, 2] >= boxes[:, 0]).all()和loss偶尔为nan的问题
解决方法：https://blog.csdn.net/u013010889/article/details/53193027?utm_source=itdadao&utm_medium=referral
https://blog.csdn.net/zcy0xy/article/details/79614862
https://www.jianshu.com/p/a672f702e596
这篇文章靠谱一些
训练模型训练模型
python ./tools/train_net.py --gpu 0
https://blog.csdn.net/Muzythoof/article/details/76060319
关于训练的文章，说得还行
Fast RCNN训练自己的数据集
http://www.cnblogs.com/louyihang-loves-baiyan/p/4903231.html
用自己的数据集训练Faster-RCNN的几种方法
https://blog.csdn.net/hongxingabc/article/details/79039537
