SegNet - 训练我自己的数据集

Question

我试图在 SegNet 上训练我自己的数据集（使用 caffe），我准备了与 segnet tutorial 相同的数据集。当我尝试运行火车时，它显示了这个错误：

I0914 11:38:04.496928 37274 layer_factory.hpp:74] Creating layer data
I0914 11:38:04.496963 37274 net.cpp:90] Creating Layer data
I0914 11:38:04.496974 37274 net.cpp:368] data -> data
I0914 11:38:04.497005 37274 net.cpp:368] data -> label
I0914 11:38:04.497020 37274 net.cpp:120] Setting up data
I0914 11:38:04.497033 37274 dense_image_data_layer.cpp:41] Opening file /home/ubuntu/full_conv_net/train.txt
I0914 11:38:04.497061 37274 dense_image_data_layer.cpp:51] Shuffling data
I0914 11:38:04.497422 37274 dense_image_data_layer.cpp:56] A total of 0 examples.
*** Aborted at 1473853084 (unix time) try "date -d @1473853084" if you are using GNU date ***
PC: @     0x7fd912325f20 (unknown)
*** SIGSEGV (@0x0) received by PID 37274 (TID 0x7fd9132a8a40) from PID 0; stack trace: ***
    @     0x7fd911cc6cb0 (unknown)
    @     0x7fd912325f20 (unknown)
    @     0x7fd912b84a5c std::operator+<>()
    @     0x7fd912bc0fa0 caffe::DenseImageDataLayer<>::DataLayerSetUp()
    @     0x7fd912ba2e09 caffe::BasePrefetchingDataLayer<>::LayerSetUp()
    @     0x7fd912b16be3 caffe::Net<>::Init()
    @     0x7fd912b18952 caffe::Net<>::Net()
    @     0x7fd912afebf0 caffe::Solver<>::InitTrainNet()
    @     0x7fd912affbc3 caffe::Solver<>::Init()
    @     0x7fd912affd96 caffe::Solver<>::Solver()
    @           0x40c5d0 caffe::GetSolver<>()
    @           0x406611 train()
    @           0x404bb1 main
    @     0x7fd911cb1f45 (unknown)
    @           0x40515d (unknown)
    @                0x0 (unknown)

我试过 make clean / all / test / runtest 但还是不行。感谢您的帮助。

Answer 1

出现以下错误

I0914 11:38:04.497033 37274 dense_image_data_layer.cpp:41] Opening file /home/ubuntu/full_conv_net/train.txt
I0914 11:38:04.497061 37274 dense_image_data_layer.cpp:51] Shuffling data
I0914 11:38:04.497422 37274 dense_image_data_layer.cpp:56] A total of 0 examples.

这表明网络无法找到训练数据。检查 'train.txt' 的格式是否正确。最好在 .txt 文件中包含图像的整个路径。

SegNet - 训练我自己的数据集

SegNet - train my own dataset

deep-learning

caffe

conv-neural-network