搭建及训练py-R-FCN遇到的问题

搭建py-R-FCN遇到的问题记录。

make pycaffe 报错

src/caffe/layers/box_annotator_ohem_layer.cu(49): error: a template argument may not reference a local type
          detected during instantiation of "void caffe::BoxAnnotatorOHEMLayer<Dtype>::Forward_gpu(const std::vector<caffe::Blob<Dtype> *, std::allocator<caffe::Blob<Dtype> *>> &, const std::vector<caffe::Blob<Dtype> *, std::allocator<caffe::Blob<Dtype> *>> &) [with Dtype=float]" 
(80): here


src/caffe/layers/box_annotator_ohem_layer.cu(49): error: a template argument may not reference a local type
          detected during instantiation of "void caffe::BoxAnnotatorOHEMLayer<Dtype>::Forward_gpu(const std::vector<caffe::Blob<Dtype> *, std::allocator<caffe::Blob<Dtype> *>> &, const std::vector<caffe::Blob<Dtype> *, std::allocator<caffe::Blob<Dtype> *>> &) [with Dtype=double]" 
(80): here


2 errors detected in the compilation of "/tmp/tmpxft_0000339c_00000000-19_box_annotator_ohem_layer.compute_61.cpp1.ii".
Makefile:595: recipe for target '.build_release/cuda/src/caffe/layers/box_annotator_ohem_layer.o' failed
make: *** [.build_release/cuda/src/caffe/layers/box_annotator_ohem_layer.o] Error 1

解决方法：

box_annotator_ohem_layer uses C++11 features, you need to make changes to the makefile 

修改makefile

319行部分改成：

ifeq ( $(DEBUG), 1)

    

    COMMON_FLAGS += -DDEBUG -g -O0 -std=c++11

    NVCCFLAGS += -G

else

    

    COMMON_FLAGS += -DNDEBUG -O2 -std=c++11

endif

训练的时候遇到的问题：AttributeError: 'module' object has no attribute 'text_format'

之前遇到过，没记录，现在记录下来。解决方法：

在出错的文件里加上：import google.protobuf.text_format

我这里错误提示在train.py里，所以找到train.py加上import google.protobuf.text_format