代码地址:https://github.com/smallcorgi/Faster-RCNN_TF
软件要求:tensorflow 和python包:cpython、python-opencv、easydict
python包安装命令行:sudo apt-get install cython python-opencv
sudo pip install easydict
硬件要求:For training the end-to-end version of Faster R-CNN with VGG16, 3G of GPU memory is sufficient (using CUDNN)
demo运行步骤
1、获取代码
git clone --recursive https://github.com/smallcorgi/Faster-RCNN_TF.git
2、建立cyphon模块
cd $FRCN_ROOT/lib
make (在下载的faster rcnn根目录下的lib文件下执行make)
3、 下载在PASCAL VOC 2007上训练的模型去测试
模型地址:https://drive.google.com/open?id=0ByuDEGFYmWsbZ0EzeUlHcGFIVWM或
https://www.dropbox.com/s/cfz3blmtmwj6bdh/VGGnet_fast_rcnn_iter_70000.ckpt?dl=0
4、运行demo
cd $FRCN_ROOT (在faster rcnn根目录下运行)
python ./tools/demo.py --model model_path
运行demo过程中遇到的错误
1、g++: error: roi_pooling_op.cu.o: No such file or directory
解决:Setting the following should make it work (it worked at least for me)
(1) export PATH=$PATH:/usr/local/cuda-8.0/bin/ (or your corresponding cuda path) for your terminal
(2) In make.sh file inside lib directory, do CXXFLAGS='-D_MWAITXINTRIN_H_INCLUDED'
2、importerror: no module named gpu_nms