eclipse调用caffe后py-faster-rcnn出现的问题几解决方法(cuda8.0+cudnn.V5 )

问题1 : /home/liuyan/caffe-master/include/caffe/util/device_alternate.hpp:34:23: fatal error: cublas_v2.h: 没有那个文件或目录
                #include
                       ^
                 compilation terminated.

                 make: *** [main.o] 错误 1

解决方法: 

             include  此目录/usr/local/cuda-8.0/include


问题2: /home/liuyan/workspace/eclipse_caffe_test/Debug/eclipse_caffe_test: error while loading shared libraries: libcaffe.so.1.0.0: cannot open shared object file: No such file or directory


解决方法: 在eclipse 加载lib库的目录需要和系统中添加的目录一致

                      首先: cd /etc

                                   sudo gedit ld.so.conf

                                    添加 /home/youname/caffe-master/build/lib

                                   sudo ldconfig

                      其次:在工程-》Properties-》c/c++ BUild->setting->GCC C++ Linker->LIbraries   添加 /home/youname/caffe-master/build/lib

问题3: run  eclipse工程出现错误:

               E1120 14:35:01.598381  3072 common.cpp:114] Cannot create Cublas handle. Cublas won't be available.
               F1120 14:35:02.023838  3072 cudnn_conv_layer.cpp:53] Check failed: status == CUDNN_STATUS_SUCCESS (4 vs. 0)  CUDNN_STATUS_INTERNAL_ERROR
             *** Check failure stack trace: ***
             @     0x7f469ecbcdaa  (unknown)
             @     0x7f469ecbcce4  (unknown)
             @     0x7f469ecbc6e6  (unknown)
             @     0x7f469ecbf687  (unknown)
             @     0x7f469f06c999  caffe::CuDNNConvolutionLayer<>::LayerSetUp()
             @     0x7f469f14bfe5  caffe::Net<>::Init()
             @     0x7f469f14e73b  caffe::Net<>::Net()
            @           0x40749b  Classifier::Classifier()
            @           0x403ee1  main
            @     0x7f469c09df45  (unknown)
            @           0x404603  (unknown)
            @              (nil)  (unknown)

解决方法: 将显卡驱动降级,从384版本降到了375版本,又重新安装了cuda,及重新配置caffe。最后解决。

问题4:    eclipse调用py-faster-rcnn生产目标检测用的动态库,出现lib/nms/cpu_nms.o: relocation R_X86_64_32 against ` .....    

解决办法:将nms中的带有cpu的文件去掉:cpu_nms.c,cpu_nms.pyx,cpu_nms.so,py_cpu_nms.py去掉

                                具体问什么会出现上面的错误,参考http://www.cnblogs.com/lightsalt/archive/2011/10/19/2217628.html

问题5: eclipse调用py-faster-rcnn生产目标检测用的动态库,出现/......................./libBOCO_DETECTION_CORE.so:对‘_nms(int*, int*, float const*, int, int, float, int)’未定义的    引用

解决办法:需要将gpu_num.so动态库载入工程中

                   若在Properties-》c/c++ BUild->setting->GCC C++ Linker->LIbrarie  添加gpu_nms,出现 /usr/bin/ld: 找不到 -lgpu_nms:,此方法不正确(若出现类似usr/bin/ld: 找不到 -lxxx:的问题,请参考http://blog.csdn.net/pzw_0612/article/details/52624358)

                    若在Properties-》c/c++ BUild->setting->GCC C++ Linker->Miscellaneous  添加具体路径:/home/liuyan/workspace/BOCO_DETECTION_CORE/lib/nms/gpu_nms.so,成功解决

                    还有一个方法,将gpu_nms.so,重命名为libgpu_nms.so,然后在Properties-》c/c++ BUild->setting->GCC C++ Linker->LIbrarie  添加gpu_nms,记得载入路径,此方法也可以解决

                  


    

你可能感兴趣的:(深度学习)