py-faster-rcnn配置CPU下运行demo.py

1.git数据

git clone --recursive https://github.com/rbgirshick/py-faster-rcnn.git

2.下载model文件,

bash ./data/scripts/fetch_faster_rcnn_models.sh

下载并解压：py-faster-rcnn/data/faster_rcnn_models/VGG16_faster_rcnn_final.caffemodel    py-faster-rcnn/data/faster_rcnn_models/ZF_faster_rcnn_final.caffemodel

3.进入py-faster-rcnn/lib,修改setup.py

#CUDA = locate_cuda()

...

#            self.set_executable('compiler_so', CUDA['nvcc'])

...

#    Extension('nms.gpu_nms',

#        ['nms/nms_kernel.cu', 'nms/gpu_nms.pyx'],

#        library_dirs=[CUDA['lib64']],

#        libraries=['cudart'],

#        language='c++',

#        runtime_library_dirs=[CUDA['lib64']],

#        # this syntax is specific to this build system

#        # we're only going to use certain compiler args with nvcc and not with

#        # gcc the implementation of this trick is in customize_compiler() below

#        extra_compile_args={'gcc': ["-Wno-unused-function"],

#                            'nvcc': ['-arch=sm_35',

#                                     '--ptxas-options=-v',

#                                     '-c',

#                                     '--compiler-options',

#                                     "'-fPIC'"]},

#        include_dirs = [numpy_include, CUDA['include']]

#    ),

运行 make

4.安装caffe,进入目录 py-faster-rcnn/caffe-fast-rcnn,首先 cp Makefile.config.example Makefile.config,并修改Makefile.config

CPU_ONLY := 1

BLAS_INCLUDE := /usr/include/atlas
BLAS_LIB := /usr/lib/atlas-base

WITH_PYTHON_LAYER := 1

INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include  /usr/include/hdf5/serial/

修改Makefile:

LIBRARIES += glog gflags protobuf boost_system boost_filesystem m hdf5_serial_hl hdf5_serial

再运行 make -j8 && make pycaffe

5.回去目录 py-faster-rcnn 运行python tools/demo.py --cpu

如果：ImportError: No module named easydict，运行pip install easydict

如果：ImportError: No module named gpu_nms，修改lib/fast_rcnn/nms_wrapper.py，

#from nms.gpu_nms import gpu_nms
from nms.cpu_nms import cpu_nms
def nms(dets, thresh, force_cpu=True):

至此，运行python tools/demo.py --cpu 成功。