py-faster-rcnn + cpu安装及训练自己的数据集

本文安装Python版本的faster-rcnn项目。 
matlab版本请移步：https://github.com/ShaoqingRen/faster_rcnn 
python版本项目主页：https://github.com/rbgirshick/py-faster-rcnn

下面分两部分来讲解，第一部分为py-faster-rcnn及caffe框架安装。第二部分讲解如何修改相关文件，训练自己的数据集。

第一部分 安装

把项目代码clone下来, 注意是递归下载，里面包含项目自带的caffe，跟原版的caffe略有不同，一定要注意：

git clone --recursive  https://github.com/rbgirshick/py-faster-rcnn.git
1
1

安装python包cython, 使用pip安装：

sudo pip install cython
1
1

编译cython

cd /py-faster-rcnn/lib修改setup.py文件，注释掉GPU相关代码，如下：#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']]),setup.py修改完成后，执行make
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19

编译caffe（项目自带的caffe-fast-rcnn） 
具体安装方法，详见我的另一篇博客，地址：http://blog.csdn.net/zhang_shuai12/article/details/52289825 
注意执行完make， 别忘了make pycaffe

到这里， 我们就完成了所得的安装。

第二部分 训练自己的数据集

首先，我们先运行一下项目自身的demo 
下载论文中提到的pascal voc 2007 数据集，pascal voc 2012数据集类似可下载：

wget http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtrainval_06-Nov-2007.tarwget http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtest_06-Nov-2007.tarwget http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCdevkit_08-Jun-2007.tar
1
2
3
1
2
3

下载完成后，解压缩：

tar xvf VOCtrainval_06-Nov-2007.tartar xvf VOCtest_06-Nov-2007.tartar xvf VOCdevkit_08-Jun-2007.tar
1
2
3
1
2
3

解压缩完成后，会形成如下的目录结构：

$VOCdevkit/$VOCdevkit/VOCcode/$VOCdevkit/VOC2007
1
2
3
1
2
3

我们对pascal voc2007 数据集做一个链接：

cd py-faster-rcnn/dataln -s /path/to/VOCdevkit VOCdevkit2007
1
2
1
2

现在,在data目录下会出现VOCdevkit2007文件夹， 结构与VOCdevkit相同。

训练数据集, 论文提供几种训练方式， 具体参数可看代码做相应设置：

cd py-fatser-rcnn./experiments/scripts/faster_rcnn_alt_opt.sh [NET] [--set ...]or./experiments/scripts/faster_rcnn_end2end.sh [NET] [--set ...]
1
2
3
4
1
2
3
4

或者我们也可以运行下面命令进行训练：

./tools/train_net.py --cpu --solver path/to/solver.prototxt --weights path/to/pretrain_model --imdb voc_2007_trainval --iters 100000 --cfgs experiments/cfgs/faster_rcnn_end2end.yml
1
1

测试可使用如下命令行：

./tools/test_net.py --cpu --def path/to/test.prototxt --net path/to/your/final.model --imdb voc_2007_test --cfgs experiments/cfgs/faster_rcnn_end2end.yml
1
1

论文作者已经预训练好几种模型， 我们下载即可：

cd py-faster-rcnn./data/scripts/fetch_faster_rcnn_models.sh./data/scripts/fetch_imagenet_models.sh./data/scripts/fetch_selective_search_data.sh
1
2
3
4
1
2
3
4

下载完成后， 解压缩模型，然后运行：

./tools/demo.py
1
1

会看到图像识别结果。

ok， 下面我们试着修改文件，换成自己的数据集，进行训练和测试。替换过程中，我们依然保持文件夹名称不变，只置换里面的数据。

我们下载解压后的数据集存储在：

/data/VOCdevkit2007/VOC2007VOC2007下有三个文件夹：VOC2007/AnnotationsVOC2007/ImageSetsVOC2007/JPEGImages
1
2
3
4
5
1
2
3
4
5

Annotations: 存储与图片对应的xml文件，每个xml文件包含object的bbox，name等 
ImageSets: 存储切分好的数据集，包括train,val,test等文件； 
JPEGImages: 存储所有的图片；

我们把自己的训练集替换掉原始的文件，放到相应的上述三个目录下即可。

另外， 在标注数据的过程中， 发现一个非常方便的标注软件labelImg, 在图片上画出框之后，会自动生成训练所需格式的xml文件。 
github项目主页地址：https://github.com/tzutalin/labelImg 
感兴趣的可以自行下载安装使用。

数据集替换问题解决了，接下来，我们一步步的修改文件：

cd /lib/datasets
1
1

首先，修改pascal_voc.py， 
self.CLASS = ( 
‘background‘, #保留这个类别 
。。。 
) 
改成你自己要识别的object 类别，比如我要识别people，只需把people类别加进去即可。

然后修改模型参数， 以训练faster_rcnn_end2end模型为例：

cd /models/pascal_voc/VGG16/faster_rcnn_end2end
1
1

包括三个prototxt文件，分别修改如下： 
对于solver.prototxt, 模型参数初始化，一般不需要修改； 
对于train.prototxt, 修改如下几个地方： 
①num_classes : 设置为你的识别类别数， 比如说我只识别people这一类， 则为2（1 + 背景类） 
②cls_score 层， inner_product_param参数， num_output :也设置为2（1 + 背景类） 
③bbox_pred层， inner_product_param参数， num_output:设置为(1 + 背景类) ×4 
对于test.prototxt文件， 修改如下： 
①cls_score层， inner_product_param参数， num_output: 设置为2（1 + 背景类）

到这里，我们针对自己的数据集，要修改的都已经改完了。训练的命令行再上面已经给出，这里不在写出。接下来就是耐心的等待训练结果啦。

训练完成后， 可以运行demo.py 来查看识别结果。注意，别忘了修改 CLASS的类别。另外，demo.py每次识别一张图片，若包含多个不同的类别，识别结果不能显示在同一张图片上， 可略做修改即可。

参考： 
【1】https://github.com/rbgirshick/py-faster-rcnn 
【2】Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks by Shaoqing Ren, Kaiming He, Ross Girshick, Jian Sun NIPS 2015 
【3】https://github.com/tzutalin/labelImg 
【4】http://caffe.berkeleyvision.org/installation.html#compilation

Faster R-CNN是我科大师弟任少卿在微软研究院实习时完成的，现在用深度学习做图像分割和目标检测最快的算法。

• 下载代码和数据

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

• 下载demo模型数据

[root@localhost py-faster-rcnn]# ./data/scripts/fetch_faster_rcnn_models.sh

Downloading Faster R-CNN demo models (695M)...

。。。

Unzipping...

faster_rcnn_models/

faster_rcnn_models/ZF_faster_rcnn_final.caffemodel

faster_rcnn_models/VGG16_faster_rcnn_final.caffemodel

• 编译cython

进入lib目录，修改setup.py，注释掉GPU相关代码，如下

。。。

#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']]

#    ),

。。。

编译：

[root@localhost lib]# make

• 安装caffe（自带的，不是通用的）

进入caffe-fast-rcnn目录，大部分跟前面caffe安装记录一文一样，修改Makefile.config为

## Refer to http://caffe.berkeleyvision.org/installation.html

# Contributions simplifying and improving our build system are welcome!

# cuDNN acceleration switch (uncomment to build with cuDNN).

# USE_CUDNN := 1

# CPU-only switch (uncomment to build without GPU support).

CPU_ONLY := 1

# uncomment to disable IO dependencies and corresponding data layers

# USE_OPENCV := 0

# USE_LEVELDB := 0

# USE_LMDB := 0

# uncomment to allow MDB_NOLOCK when reading LMDB files (only if necessary)

#    You should not set this flag if you will be reading LMDBs with any

#    possibility of simultaneous read and write

# ALLOW_LMDB_NOLOCK := 1

# Uncomment if you're using OpenCV 3

# OPENCV_VERSION := 3

# To customize your choice of compiler, uncomment and set the following.

# N.B. the default for Linux is g++ and the default for OSX is clang++

# CUSTOM_CXX := g++

# CUDA directory contains bin/ and lib/ directories that we need.

# CUDA_DIR := /usr/local/cuda

# On Ubuntu 14.04, if cuda tools are installed via

# "sudo apt-get install nvidia-cuda-toolkit" then use this instead:

# CUDA_DIR := /usr

# CUDA architecture setting: going with all of them.

# For CUDA < 6.0, comment the *_50 lines for compatibility.

#CUDA_ARCH := -gencode arch=compute_20,code=sm_20 \

#        -gencode arch=compute_20,code=sm_21 \

#        -gencode arch=compute_30,code=sm_30 \

#        -gencode arch=compute_35,code=sm_35 \

#        -gencode arch=compute_50,code=sm_50 \

#        -gencode arch=compute_50,code=compute_50

# BLAS choice:

# atlas for ATLAS (default)

# mkl for MKL

# open for OpenBlas

BLAS := atlas

# Custom (MKL/ATLAS/OpenBLAS) include and lib directories.

# Leave commented to accept the defaults for your choice of BLAS

# (which should work)!

BLAS_INCLUDE := /usr/include/atlas-x86_64-base

BLAS_LIB := /usr/lib64/atlas

# Homebrew puts openblas in a directory that is not on the standard search path

# BLAS_INCLUDE := $(shell brew --prefix openblas)/include

# BLAS_LIB := $(shell brew --prefix openblas)/lib

# This is required only if you will compile the matlab interface.

# MATLAB directory should contain the mex binary in /bin.

# MATLAB_DIR := /usr/local

# MATLAB_DIR := /Applications/MATLAB_R2012b.app

# NOTE: this is required only if you will compile the python interface.

# We need to be able to find Python.h and numpy/arrayobject.h.

PYTHON_INCLUDE := /usr/include/python2.7 \

                  /usr/lib64/python2.7/site-packages/numpy/core/include

# Anaconda Python distribution is quite popular. Include path:

# Verify anaconda location, sometimes it's in root.

# ANACONDA_HOME := $(HOME)/anaconda

# PYTHON_INCLUDE := $(ANACONDA_HOME)/include \

        # $(ANACONDA_HOME)/include/python2.7 \

        # $(ANACONDA_HOME)/lib/python2.7/site-packages/numpy/core/include \

# Uncomment to use Python 3 (default is Python 2)

# PYTHON_LIBRARIES := boost_python3 python3.5m

# PYTHON_INCLUDE := /usr/include/python3.5m \

#                 /usr/lib/python3.5/dist-packages/numpy/core/include

# We need to be able to find libpythonX.X.so or .dylib.

PYTHON_LIB := /usr/lib64

# PYTHON_LIB := $(ANACONDA_HOME)/lib

# Homebrew installs numpy in a non standard path (keg only)

# PYTHON_INCLUDE += $(dir $(shell python -c 'import numpy.core; print(numpy.core.__file__)'))/include

# PYTHON_LIB += $(shell brew --prefix numpy)/lib

# Uncomment to support layers written in Python (will link against Python libs)

WITH_PYTHON_LAYER := 1

# Whatever else you find you need goes here.

INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/include

LIBRARY_DIRS := $(PYTHON_LIB) /usr/lib64

# If Homebrew is installed at a non standard location (for example your home directory) and you use it for general dependencies

# INCLUDE_DIRS += $(shell brew --prefix)/include

# LIBRARY_DIRS += $(shell brew --prefix)/lib

# Uncomment to use `pkg-config` to specify OpenCV library paths.

# (Usually not necessary -- OpenCV libraries are normally installed in one of the above $LIBRARY_DIRS.)

# USE_PKG_CONFIG := 1

BUILD_DIR := build

DISTRIBUTE_DIR := distribute

# Uncomment for debugging. Does not work on OSX due to https://github.com/BVLC/caffe/issues/171

# DEBUG := 1

# The ID of the GPU that 'make runtest' will use to run unit tests.

# TEST_GPUID := 0

# enable pretty build (comment to see full commands)

Q ?= @

修改Makefile

LIBRARIES += satlas tatlas #新版atlas已经不用这两个lib了：cblas atlas

编译caffe和pycaffe

 [root@localhost caffe-fast-rcnn]# make -j8 && make pycaffe

• 跑demo

[root@localhost py-faster-rcnn]# ./tools/demo.py

Traceback (most recent call last):

File "./tools/demo.py", line 17, in

 from fast_rcnn.config import cfg

 File "/root/zhanxiang/work/py-faster-rcnn/tools/../lib/fast_rcnn/config.py", line 23, in

from easydict import EasyDict as edict

ImportError: No module named easydict

缺少Python库easydict，所以安装 pip install easydict

[root@localhost py-faster-rcnn]# ./tools/demo.py

Traceback (most recent call last):

  File "./tools/demo.py", line 18, in

    from fast_rcnn.test import im_detect

  File "/root/zhanxiang/work/py-faster-rcnn/tools/../lib/fast_rcnn/test.py", line 15, in

    import cv2

ImportError: No module named cv2

缺少Python库cv2，这个是openCV里面的。那就来装openCV python库

yum install opencv-python.x86_64

[root@localhost py-faster-rcnn]# python tools/demo.py --cpu

Traceback (most recent call last):

  File "tools/demo.py", line 21, in

    import matplotlib.pyplot as plt

  File "/usr/lib64/python2.7/site-packages/matplotlib/pyplot.py", line 26, in

    from matplotlib.figure import Figure, figaspect

  File "/usr/lib64/python2.7/site-packages/matplotlib/figure.py", line 36, in

    from matplotlib.axes import Axes, SubplotBase, subplot_class_factory

  File "/usr/lib64/python2.7/site-packages/matplotlib/axes/__init__.py", line 4, in

    from ._subplots import *

  File "/usr/lib64/python2.7/site-packages/matplotlib/axes/_subplots.py", line 10, in

    from matplotlib.axes._axes import Axes

  File "/usr/lib64/python2.7/site-packages/matplotlib/axes/_axes.py", line 14, in

    from matplotlib import unpack_labeled_data

ImportError: cannot import name unpack_labeled_data

看起来跟matplotlib库有关，pip install的版本太旧，直接下载源码安装。

按照官网指示，http://matplotlib.org/faq/installing_faq.html#install-from-git

[root@localhost work]# git clone git://github.com/matplotlib/matplotlib.git

[root@localhost work]# cd matplotlib/

安装依赖包

[root@localhost matplotlib]# yum-builddep python-matplotlib

安装

[root@localhost matplotlib]# python setup.py install

[root@localhost py-faster-rcnn]# python tools/demo.py --cpu

Traceback (most recent call last):

  File "tools/demo.py", line 19, in

    from fast_rcnn.nms_wrapper import nms

  File "/root/zhanxiang/work/py-faster-rcnn/tools/../lib/fast_rcnn/nms_wrapper.py", line 9, in

    from nms.gpu_nms import gpu_nms

ImportError: No module named gpu_nms

修改nms_wrapper.py，改force_cpu =True

[root@localhost py-faster-rcnn]# vi lib/fast_rcnn/nms_wrapper.py

def nms (dets, thresh, force_cpu =True):

• 大功告成

[root@localhost py-faster-rcnn]# python tools/demo.py --cpu

就能看到结果了