【发布时间】：2019-06-30 10:30:43
【问题描述】：
我正在对 caffe 进行培训。网络为：VGG16 + 8 个卷积层 + softmax loss。因为我的机器，我选择只在 CPU 上运行。我的内存是8G，我的数据集是VOC2007 segmentation images(train+val+test不超过1G内存)。
但是，在输入命令后，程序会长时间（~10小时）停留在消息“数据层预取队列为空”的点上。在此期间，系统监视器显示我的 CPU 使用率为 100%，内存为 69%。由于我的火车批次只有 10 个，我认为我的机器可以承受的数量，很难理解为什么会发生这种情况？
以下是网络初始化完成和 Solver 脚手架完成后的输出消息。
I0404 13:08:15.401798 28175 net.cpp:274] Network initialization done.
I0404 13:08:15.401984 28175 solver.cpp:60] Solver scaffolding done.
I0404 13:08:15.402128 28175 caffe.cpp:129] Finetuning from /home/huanghe/caffe-dilation/pretrained/frontend_model.caffemodel
[libprotobuf WARNING google/protobuf/io/coded_stream.cc:537] Reading dangerously large protocol message. If the message turns out to be larger than 2147483647 bytes, parsing will be halted for security reasons. To increase the limit (or to disable these warnings), see CodedInputStream::SetTotalBytesLimit() in google/protobuf/io/coded_stream.h.
[libprotobuf WARNING google/protobuf/io/coded_stream.cc:78] The total number of bytes read was 537388947
I0404 13:08:22.585640 28175 net.cpp:752] Ignoring source layer prob
[libprotobuf WARNING google/protobuf/io/coded_stream.cc:537] Reading dangerously large protocol message. If the message turns out to be larger than 2147483647 bytes, parsing will be halted for security reasons. To increase the limit (or to disable these warnings), see CodedInputStream::SetTotalBytesLimit() in google/protobuf/io/coded_stream.h.
[libprotobuf WARNING google/protobuf/io/coded_stream.cc:78] The total number of bytes read was 537388947
I0404 13:08:23.729429 28175 net.cpp:752] Ignoring source layer prob
I0404 13:08:23.732425 28175 caffe.cpp:219] Starting Optimization
I0404 13:08:23.732533 28175 solver.cpp:279] Solving
I0404 13:08:23.732580 28175 solver.cpp:280] Learning Rate Policy: step
I0404 13:08:23.833487 28175 solver.cpp:337] Iteration 0, Testing net (#0)
I0404 13:08:23.891206 28175 blocking_queue.cpp:50] Data layer prefetch queue empty
我是新手。互联网上的搜索从来没有帮助过我。谁能为我解释一下？非常感谢！
【问题讨论】：
标签： deep-learning caffe