zxw.sh
/home/tyn/Downloads/center_loss_caffe/build/tools/caffe train -solver face_solver.prototxt --snapshot=center_loss_model_iter_9000.solverstate -gpu=all 2>&1 | tee double_label_centerloss_20575_0.02_train.log
center_loss_model_iter_9000.solverstate #从这个模型开始训练
double_label_centerloss_20575_0.02_train.log #保存的训练日志
感谢陶师姐教会我这么多--虽然在她面前基本上没有叫过师姐
有时候caffe在训练的时候,因某种原因不得不暂停训练,下次怎么接着继续训练呢,下列sh脚本文件
#!/usr/bin/env sh
Path=/home/ccf/CCF/data-me
/home/hjxu/caffe-master/caffe/build/tools/caffe train --solver=/home/hjxu/WSI-metastic/data-me/profile/solver.prototxt \
--snapshot=/home/hjxu/WSI-metastic/data-me/model/caffe_Breast_transfer_train_iter_95000.solverstate 2>&1|tee /home/hjxu/WSI-metastic/data-me/train2.txt
如果已经有了模型,想要第二次加强训练
#!/usr/bin/env sh
Path=/home/ccf/CCF/data-me
/home/hjxu/caffe-master/caffe/build/tools/caffe train --solver=/home/hjxu/WSI-metastic/data-me/profile/solver.prototxt \
--weights=/home/hjxu/WSI-metastic/data-me/model/caffe_Breast_transfer_train_iter_95000.caffemodel 2>&1|tee /home/hjxu/WSI-metastic/data-me/train2.txt
ubuntu下
如果我们在训练途中停电或者有了其他的情况,可以通过之前保存的状态恢复数据,使用的时候直接添加–snapshot参数即可,如:
./build/tools/caffe train --solver=models/bvlc_reference_caffenet/solver.prototxt --snapshot=models/bvlc_reference_caffenet/caffenet_train_iter_10000.solverstate