Fasterrcnn训练自己的数据集

安装cpython&＃xff0c;python-opencv easydict

pip install cpython
pip install easydict
apt-get install python-opencv


进如 py-faster-rcnn/lib 进行make操作

make

进入 py-faster-rcnn/caffe-fast-rcnn

cp Makefile.config.example Makefile.config

对Makefile.config文件进行配置&＃xff0c;我的配置文件如下&＃xff1a;

## 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 :&＃61; 1# CPU-only switch (uncomment to build without GPU support).
# CPU_ONLY :&＃61; 1# uncomment to disable IO dependencies and corresponding data layers
# USE_OPENCV :&＃61; 0
# USE_LEVELDB :&＃61; 0
# USE_LMDB :&＃61; 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 :&＃61; 1# Uncomment if you&＃39;re using OpenCV 3
# OPENCV_VERSION :&＃61; 3# To customize your choice of compiler, uncomment and set the following.
# N.B. the default for Linux is g&＃43;&＃43; and the default for OSX is clang&＃43;&＃43;
# CUSTOM_CXX :&＃61; g&＃43;&＃43;# CUDA directory contains bin/ and lib/ directories that we need.
CUDA_DIR :&＃61; /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 :&＃61; /usr# CUDA architecture setting: going with all of them.
# For CUDA <6.0, comment the *_50 lines for compatibility.
CUDA_ARCH :&＃61; -gencode arch&＃61;compute_20,code&＃61;sm_20 \-gencode arch&＃61;compute_20,code&＃61;sm_21 \-gencode arch&＃61;compute_30,code&＃61;sm_30 \-gencode arch&＃61;compute_35,code&＃61;sm_35 \-gencode arch&＃61;compute_50,code&＃61;sm_50 \-gencode arch&＃61;compute_50,code&＃61;compute_50# BLAS choice:
# atlas for ATLAS (default)
# mkl for MKL
# open for OpenBlas
BLAS :&＃61; 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 :&＃61; /path/to/your/blas
# BLAS_LIB :&＃61; /path/to/your/blas# Homebrew puts openblas in a directory that is not on the standard search path
# BLAS_INCLUDE :&＃61; $(shell brew --prefix openblas)/include
# BLAS_LIB :&＃61; $(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 :&＃61; /usr/local
# MATLAB_DIR :&＃61; /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 :&＃61; /usr/include/python2.7 \/usr/lib/python2.7/dist-packages/numpy/core/include
# Anaconda Python distribution is quite popular. Include path:
# Verify anaconda location, sometimes it&＃39;s in root.
# ANACONDA_HOME :&＃61; $(HOME)/anaconda
# PYTHON_INCLUDE :&＃61; $(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 :&＃61; boost_python3 python3.5m
# PYTHON_INCLUDE :&＃61; /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 :&＃61; /usr/lib
# PYTHON_LIB :&＃61; $(ANACONDA_HOME)/lib# Homebrew installs numpy in a non standard path (keg only)
# PYTHON_INCLUDE &＃43;&＃61; $(dir $(shell python -c &＃39;import numpy.core; print(numpy.core.__file__)&＃39;))/include
# PYTHON_LIB &＃43;&＃61; $(shell brew --prefix numpy)/lib# Uncomment to support layers written in Python (will link against Python libs)
WITH_PYTHON_LAYER :&＃61; 1# Whatever else you find you need goes here.
INCLUDE_DIRS :&＃61; $(PYTHON_INCLUDE) /usr/local/include /usr/include/hdf5/serial
LIBRARY_DIRS :&＃61; $(PYTHON_LIB) /usr/local/lib /usr/lib# If Homebrew is installed at a non standard location (for example your home directory) and you use it for general dependencies
# INCLUDE_DIRS &＃43;&＃61; $(shell brew --prefix)/include
# LIBRARY_DIRS &＃43;&＃61; $(shell brew --prefix)/lib# Uncomment to use &＃96;pkg-config&＃96; to specify OpenCV library paths.
# (Usually not necessary -- OpenCV libraries are normally installed in one of the above $LIBRARY_DIRS.)
# USE_PKG_CONFIG :&＃61; 1BUILD_DIR :&＃61; build
DISTRIBUTE_DIR :&＃61; distribute# Uncomment for debugging. Does not work on OSX due to https://github.com/BVLC/caffe/issues/171
# DEBUG :&＃61; 1# The ID of the GPU that &＃39;make runtest&＃39; will use to run unit tests.
TEST_GPUID :&＃61; 0# enable pretty build (comment to see full commands)
Q ?&＃61; &＃64;

 配置完Makefile.config文件后&＃xff0c;执行&＃xff1a;

make -j4
make pycaffe

• 数据制作&＃xff1a;

目标检测数据集制作工具 labelImg https://blog.csdn.net/caicaiatnbu/article/details/85125018

 修改 ../py-faster-rcnn/data/VOCdevkit2007/VOC2007/ 目录下的 Annotations 和 JPEGImage 以及ImageSets/Main 三个文件夹&＃xff0c;下面依次讲述这个三个文件夹

JPEGImage 文件夹保存着所有图片&＃xff0c;这里我们所有图片的格式统一都是jpg&＃xff0c;命名规则"000001.jpg"开始。

Annotation 文件夹保存着每张图片对应的XML文件&＃xff0c;这里我们将我们的XML文件格式贴出来&＃xff1a;

VOC2007000001.jpgUnknown49937430


ImageSets/Main 文件夹下包含4个.txt文件&＃xff0c;分别为 train.txt val.txt trainval.txt test.txt &＃xff0c;没给txt中保存都是图片的名称。

注意&＃xff1a;trainval.txt 为 train.txt 和 val.txt的总和。train.txt 和 val.txt 里面保存的样本数一定要为偶数&＃xff0c;否则会报错

•  代码部分修改

修改 py-faster-rcnn/models/pascal_voc/ZF/faster_rcnn_alt_opt/stage1_rcnn_train.pt   和  stage2_fast_rcnn_train.pt 两个文件

name: "ZF"
layer {name: &＃39;data&＃39;type: &＃39;Python&＃39;top: &＃39;data&＃39;top: &＃39;rois&＃39;top: &＃39;labels&＃39;top: &＃39;bbox_targets&＃39;top: &＃39;bbox_inside_weights&＃39;top: &＃39;bbox_outside_weights&＃39;python_param {module: &＃39;roi_data_layer.layer&＃39;layer: &＃39;RoIDataLayer&＃39;param_str: "&＃39;num_classes&＃39;: 6" #按训练集类更改&＃xff0c;该值为类别数&＃43;1}
}layer {name: "cls_score"type: "InnerProduct"bottom: "fc7"top: "cls_score"param { lr_mult: 1.0 }param { lr_mult: 2.0 }inner_product_param {num_output: 6 #按训练集类更改&＃xff0c;该值为类别数&＃43;1weight_filler {type: "gaussian"std: 0.01}bias_filler {type: "constant"value: 0}}
}layer {name: "bbox_pred"type: "InnerProduct"bottom: "fc7"top: "bbox_pred"param { lr_mult: 1.0 }param { lr_mult: 2.0 }inner_product_param {num_output: 24 #按训练集类更改&＃xff0c;该值为&＃xff08;类别数&＃43;1&＃xff09;* 4weight_filler {type: "gaussian"std: 0.001}bias_filler {type: "constant"value: 0}}
}

修改 py-faster-rcnn/models/pascal_voc/ZF/faster_rcnn_alt_opt/stage1_rpn_train.pt  和 stage2_rpn_train.pt

name: "ZF"
layer {name: &＃39;input-data&＃39;type: &＃39;Python&＃39;top: &＃39;data&＃39;top: &＃39;im_info&＃39;top: &＃39;gt_boxes&＃39;python_param {module: &＃39;roi_data_layer.layer&＃39;layer: &＃39;RoIDataLayer&＃39;param_str: "&＃39;num_classes&＃39;: 6" #按训练集类更改&＃xff0c;该值为类别数&＃43;1}
}


修改py-faster-rcnn/models/pascal_voc/ZF/faster_rcnn_alt_opt/faster_rcnn_test.pt

layer {name: "cls_score"type: "InnerProduct"bottom: "fc7"top: "cls_score"inner_product_param {num_output: 6 #按训练集类更改&＃xff0c;该值为类别数&＃43;1}
}
layer {name: "bbox_pred"type: "InnerProduct"bottom: "fc7"top: "bbox_pred"inner_product_param {num_output: 24 #按训练集类更改&＃xff0c;该值为&＃xff08;类别数&＃43;1&＃xff09;* 4}
}

修改 py-faster-rcnn/lib/datasets/pascal_voc.py

注意 标签名尽可能为英文小写。否则会报bug

self._classes &＃61; (&＃39;__background__&＃39;, # always index 0&＃39;label_a&＃39;,&＃39;label_b&＃39;,&＃39;label_c&＃39;,&＃39;label_d&＃39;,&＃39;label_e&＃39;)# &＃39;aeroplane&＃39;, &＃39;bicycle&＃39;, &＃39;bird&＃39;, &＃39;boat&＃39;,# &＃39;bottle&＃39;, &＃39;bus&＃39;, &＃39;car&＃39;, &＃39;cat&＃39;, &＃39;chair&＃39;,# &＃39;cow&＃39;, &＃39;diningtable&＃39;, &＃39;dog&＃39;, &＃39;horse&＃39;,# &＃39;motorbike&＃39;, &＃39;person&＃39;, &＃39;pottedplant&＃39;,# &＃39;sheep&＃39;, &＃39;sofa&＃39;, &＃39;train&＃39;, &＃39;tvmonitor&＃39;)

修改 py-faster-rcnn/lib/datasets/imdb.py

def append_flipped_images(self):num_images &＃61; self.num_images#widths &＃61; self._get_widths()widths &＃61; [PIL.Image.open(self.image_path_at(i)).size[0]for i in xrange(self.num_images)]for i in xrange(num_images):boxes &＃61; self.roidb[i][&＃39;boxes&＃39;].copy()oldx1 &＃61; boxes[:, 0].copy()oldx2 &＃61; boxes[:, 2].copy()boxes[:, 0] &＃61; widths[i] - oldx2 - 1boxes[:, 2] &＃61; widths[i] - oldx1 - 1for b in range(len(boxes)):if boxes[b][2] &＃61; boxes[:, 0]).all()entry &＃61; {&＃39;boxes&＃39; : boxes,&＃39;gt_overlaps&＃39; : self.roidb[i][&＃39;gt_overlaps&＃39;],&＃39;gt_classes&＃39; : self.roidb[i][&＃39;gt_classes&＃39;],&＃39;flipped&＃39; : True}self.roidb.append(entry)self._image_index &＃61; self._image_index * 2


删除缓存文件

删除 py-faster-rcnn文件夹下所有的.pyc 文件 以及 data文件件下的cache文件夹&＃xff0c;

删除 data/VOCdevkit2007/annotations_cache 文件夹

删除 py-faster-rcnn/lib/datasets/ 下的imdb.pyc 和 pascal_voc.pyc

开始训练&＃xff1a;

./experiments/scripts/faster_rcnn_alt_opt.sh 0 ZF pascal_voc