faster rcnn snapshot断点训练

训练自己的网络：sudo ./experiments/scripts/faster_rcnn_end2end.sh 0 JZ pascal_voc

修改

'/home/jz/py-faster-rcnn/lib/fast_rcnn/config.py' snapshot=6000

不进行断点训练，恢复solver中：snapshot=0

======================================================================================================================

1. solver.prototxt中的snapshot_prefix不能是output.(不知道why)

2. sudo python ./tools/train_net.py --gpu 0 --solver models/subway0/1type/solver.prototxt --snapshot output/faster_rcnn_end2end/voc_2012_train/jz0_faster_rcnn_iter_10000.solverstate --weights output/faster_rcnn_end2end/voc_2012_train/jz0_faster_rcnn_iter_10000.caffemodel --iters 40000 --imdb voc_2012_train --cfg experiments/cfgs/faster_rcnn_end2end.yml