一、拉取源码
下载 fast-rcnn
因下载解压后 caffe-fast-rcnn是空文件夹，故需要单独下 caffe-fast-rcnn-bcd9b4eadc7d8fbc433aeefd564e82ec63aaf69c.zip
unzip caffe-fast-rcnn-bcd9b4eadc7d8fbc433aeefd564e82ec63aaf69c.zip cp ./caffe-fast-rcnn-bcd9b4eadc7d8fbc433aeefd564e82ec63aaf69c/ /home/cmwang/fast-rcnn-master/caffe-fast-rcnn/二、copy缺失文件夹
因编译中出现缺失layer的问题，故需要copy缺失文件夹layers（caffe-fast-rcnn/include/caffe/ 无 layers文件夹）
可直接从caffe-master中copy或者从
cp /home/cmwang/caffe-master/include/caffe/layers/ /home/cmwang/fast-rcnn-master/caffe-fast-rcnn/include/caffe/layers/ 或者 cp /home/cmwang/py-faster-rcnn/caffe-fast-rcnn/include/caffe/layers/ /home/cmwang/fast-rcnn-master/caffe-fast-rcnn/include/caffe/layers/三、修改Makefile文件
终端输入
cd /home/cmwang/fast-rcnn-master/caffe-fast-rcnn/ cp Makefile.config.example Makefile.config #备份Makefile gedit Makefile.config使用Python层 将# WITH_PYTHON_LAYER := 1修改为 WITH_PYTHON_LAYER := 1
调用matlab 将#MATLAB_DIR := / usr/local/MATLAB/R2015b 中的#去掉。
使用cudnn加速 将# USE_CUDNN := 1修改为USE_CUDNN := 1
保留# CPU_ONLY := 1不变，使用GPU运行faster r-cnn
四、编译Cython模块
终端输入
cd ~/caffe-fast-rcnn/lib/ make五、编译caffe和pycaffe & matcaffe
终端输入
cd ~/caffe-fast-rcnn/caffe-fast-rcnn/ make -j8 && make pycaffe && make matcaffe六、下载模型
终端输入
cd ~/caffe-fast-rcnn/ ./data/scripts/fetch_faste_rcnn_models.sh七、Demo测试
终端输入
cd ~/caffe-fast-rcnn/ ./tools/demo.py出现的问题1
caffe —找不到lhdf5_hl和lhdf5的错误
解决方法：
cd gedit Makefile.config将
INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib修改为
INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include /usr/include/hdf5/serial LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib /usr/lib/x86_64-linux-gnu /usr/lib/x86_64-linux-gnu/hdf5/serial这是因为ubuntu16.04的文件包含位置发生了变化，尤其是需要用到的hdf5的位置，所以需要更改这一路径
建立软连接
cd /usr/lib/x86_64-linux-gnu sudo ln -s libhdf5_serial.so.8 libhdf5.so sudo ln -s libhdf5_serial_hl.so.8.0.2 libhdf5_hl.so出现的问题2
When use fast R-CNN, got error like this:
I0310 08:26:43.672577 144950 net.cpp:340] Input 0 -> data I0310 08:26:43.672601 144950 net.cpp:340] Input 1 -> rois I0310 08:26:43.672621 144950 layer_factory.hpp:74] Creating layer conv1 I0310 08:26:43.672639 144950 net.cpp:84] Creating Layer conv1 I0310 08:26:43.672650 144950 net.cpp:380] conv1 <- data I0310 08:26:43.672664 144950 net.cpp:338] conv1 -> conv1 I0310 08:26:43.672680 144950 net.cpp:113] Setting up conv1 Floating point exception(core dumped).解决方案
gedit lib/fast_rcnn/train.py添加 filter_roidb 范围，示范如下
def filter_roidb(roidb): """Remove roidb entries that have no usable RoIs.""" def is_valid(entry): # Valid images have: # (1) At least one foreground RoI OR # (2) At least one background RoI overlaps = entry['max_overlaps'] # find boxes with sufficient overlap fg_inds = np.where(overlaps >= cfg.TRAIN.FG_THRESH)[0] # Select background RoIs as those within [BG_THRESH_LO, BG_THRESH_HI) bg_inds = np.where((overlaps < cfg.TRAIN.BG_THRESH_HI) & (overlaps >= cfg.TRAIN.BG_THRESH_LO))[0] # image is only valid if such boxes exist valid = len(fg_inds) > 0 or len(bg_inds) > 0 return valid num = len(roidb) filtered_roidb = [entry for entry in roidb if is_valid(entry)] num_after = len(filtered_roidb) print 'Filtered {} roidb entries: {} -> {}'.format(num - num_after, num, num_after) return filtered_roidbIt’s like something about box size. the solution is add filter_roidb function in lib/fast_rcnn/train.py file, like here. Reference: https://github.com/rbgirshick/py-faster-rcnn/issues/159
其他的相关问题可参考 Faster R-CNN的安装及测试中常出现问题部分。
参考文献
caffe —找不到lhdf5_hl和lhdf5的错误
fast-rcnn github
fast-rcnn caffe-fast-rcnn
安装和运行Fast R-CNN的demo
caffe compilation troubleshooting
