The problem of caffe Multi-task

[chenjoya]

薛大牛，您好！在看了您的文章《一箭N雕：多任务深度学习实战》之后，深受启发，在自己进行测试的时候，发现在数据读入的时候，在这一步骤就卡住了：

I0119 20:50:00.314520 10200 layer_factory.hpp:77] Creating layer data
I0119 20:50:00.314520 10200 net.cpp:100] Creating Layer data
I0119 20:50:00.315522 10200 net.cpp:418] data -> data

我是finetune的AlexNet，在fc7层分流，两个fc8层和两个loss层。前面一切都没有问题，看起来网络似乎已经初始化了：
......

I0119 20:50:00.269047 10200 layer_factory.hpp:77] Creating layer loss_type
I0119 20:50:00.269047 10200 net.cpp:150] Setting up loss_type
I0119 20:50:00.270496 10200 net.cpp:157] Top shape: (1)
I0119 20:50:00.271216 10200 net.cpp:160] with loss weight 0.5
I0119 20:50:00.271759 10200 net.cpp:165] Memory required for data: 8348048
I0119 20:50:00.271759 10200 layer_factory.hpp:77] Creating layer loss_surface
I0119 20:50:00.272760 10200 net.cpp:100] Creating Layer loss_surface
I0119 20:50:00.272760 10200 net.cpp:444] loss_surface <- fc8_surface
I0119 20:50:00.275764 10200 net.cpp:444] loss_surface <- surface
I0119 20:50:00.276763 10200 net.cpp:418] loss_surface -> loss_surface
I0119 20:50:00.277765 10200 layer_factory.hpp:77] Creating layer loss_surface
I0119 20:50:00.277765 10200 net.cpp:150] Setting up loss_surface
I0119 20:50:00.278766 10200 net.cpp:157] Top shape: (1)
I0119 20:50:00.279767 10200 net.cpp:160] with loss weight 0.5
I0119 20:50:00.279767 10200 net.cpp:165] Memory required for data: 8348052
I0119 20:50:00.280766 10200 net.cpp:226] loss_surface needs backward computation.
I0119 20:50:00.281766 10200 net.cpp:226] loss_type needs backward computation.
I0119 20:50:00.281766 10200 net.cpp:226] fc8_surface needs backward computation.
I0119 20:50:00.282783 10200 net.cpp:226] fc8_type needs backward computation.
I0119 20:50:00.283769 10200 net.cpp:226] fc7_drop7_0_split needs backward computation.
I0119 20:50:00.283769 10200 net.cpp:226] drop7 needs backward computation.
......
可是在读入数据时就是不能动。十分渴望得到您的解答！
万分抱歉打扰您的。新年就要到了，提前预祝您万事如意！

[xuguozhi]

问题解决了没？

[chenjoya]

(｡･∀･)ﾉﾞ这个问题很简单的... 检查一下你的solver文件，没有测试数据就不用加test相关的东西