如何在caffe Forward_gpu期间打印变量

Question

我试图在 Forward_gpu 函数中打印底部形状和底部数据。但是会造成段错误。有人可以告诉我这样做的正确方法以及出现段错误的原因吗？

我做到了：

void Forward_gpu(const vector<Blob<Dtype> *> &bottom, const vector<Blob<Dtype> *> &top)
{
  ...
  puts("bottom shape: ");
  std::cout << bottom[2*i]->gpu_shape()[0] << std::endl;
  ...
}

我得到的：

bottom shape:
*** Aborted at 1500362213 (unix time) try "date -d @1500362213" if you are using GNU date ***
PC: @     0x7fe4be27cbfd caffe::DeformableConvolutionLayer<>::Forward_gpu()
*** SIGSEGV (@0x1020d602c00) received by PID 17153 (TID 0x7fe4bea969c0) from PID 224406528; stack trace: ***
    @     0x7fe4bc615cb0 (unknown)
    @     0x7fe4be27cbfd caffe::DeformableConvolutionLayer<>::Forward_gpu()
    @     0x7fe4be204471 caffe::Net<>::ForwardFromTo()
    @     0x7fe4be2047e7 caffe::Net<>::ForwardPrefilled()
    @     0x7fe4be224455 caffe::Solver<>::Step()
    @     0x7fe4be224d59 caffe::Solver<>::Solve()
    @           0x4084be train()
    @           0x405cac main
    @     0x7fe4bc600f45 (unknown)
    @           0x40647d (unknown)
    @                0x0 (unknown)
Segmentation fault (core dumped)

Answer 1

你必须像这样访问它们

bottom[2*i]->shape(0)

并确保 2*i 在可用底部范围内，首先尝试

bottom[0]->shape(0)

确保至少可以打印出来

如何在caffe Forward_gpu期间打印变量

How to print variables during Forward_gpu in caffe

c++

machine-learning

deep-learning

caffe