After a few days of fighting finally configured the configuration of title, now the configuration of the general process to write down for everyone to configure the reference (due to the computer hardware and systems vary widely, it is not suitable for writing detailed)
(All configuration tutorials that do not declare configuration environment are bullying)
Environment: Inter set + gtx1070 single display
Ubuntu14.04lts (ubuntu System, if the two video card driver at the same time there will be conflict, seemingly shut down what LIGHTDM can be resolved, I will not toss, installed N card driver I will be in the BIOS switch off the set display only with a single display )
Cuda_8.0.61_375.26_linux.run cudnn-8.0-linux-x64-v5.1 opencv-3.0.0-rc1 nvidia-linux-x86_64-367.44.run (note!) Be sure to choose the software version that is compatible with your hardware environment, and the compatibility of Nvidia software is not flattering. For example, here because I am the 10 series of graphics card, you must use cuda8.0, can not use cuda7.5, do not ask me how to know, CUDNN is also the best use V5, did not try V4)
Reference blog:http://blog.csdn.net/baidu_32173921/article/details/53510764(I basically configure it according to this blog)
http://blog.csdn.net/ai_smith/article/details/53000973
http://blog.csdn.net/samylee/article/details/50922601
First, configure the Caffe
And then the general configuration process: 1. Install opencv3.0 (because the first to install Cuda Reload OpenCV will conflict , so first installed it), reference: http://www.cnblogs.com/zf-blog/p/6649612.html;
2. Install the Cuda and then install NVIDIA driver ( first install driver and then installed Cuda Easy conflict ), Be sure to install at the command line interface , refer to this blog: http://blog.csdn.net/baidu_ 32173921/article/details/53510764 (very detailed), where the driver is installed as follows:
Otherwise it is possible to fall into the circular login interface, before the set opened in the environment installed ubuntu14.04, now installed n card driver, and then login screen resolution will be a problem (Ubutntu does not seem to support the dual graphics driver), so I turned off the set display only with the unique display.
After installation Be sure to test the driver and Cuda has not been installed successfully, there are many online tutorials, nvidia-smi command test drive, there is a test cuda do not remember.
3. Install CUDNN, reference: http://blog.csdn.net/baidu_32173921/article/details/53510764
http://blog.csdn.net/ai_smith/article/details/53000973
http://blog.csdn.net/samylee/article/details/50922601
4. Now is the beginning to configure the Caffe, first install a variety of dependencies, and so on, note that Python is best to use the system comes with , anaconda although the integration of a lot, but still insufficient, and easily with the system comes from the Python conflict, I suggest that the system comes with the python,ide can download jupyter notebook on the line, the following steps on the internet I do not say, but to note that is make runtest best to pass, it can ensure that your caffe is intact In the process of compiling everyone should have a variety of errors, I do not say here, because each person's hardware and software environment of the different problems are not the same, the problem is likely to solve the same method is not the same, suggest Baidu, can fq as far as possible Google, the quality of the online solution is really very high! Configuration good Caffe after run a bit imagenet network, speed really fast ...
Second, the configuration py-faster-rcnn
Then configure PY-FASTER-RCNN, refer to:http://blog.csdn.net/samylee/article/details/51086153
One thing to note here is that since Caffe on GitHub can already support CUDNNV5, PY-FASTER-RCNN cannot support CUDNNV5, only V4, so there will be an error when making all. The workaround is to use git merge to merge the Caffe on GitHub into the Caffe in Py-faster-rcnn, as follows:http://blog.csdn.net/10km/article/ details/62418583 , the following problem may occur when you use the git merge command here:
Then execute it in the main directory:
Name can be changed, other forms do not change, and then merge on the line, followed by the above blog http://blog.csdn.net/10km/article/details/62418583 in the line. Then recompile (make-clean) step-by-step, and at the end of making runtest, you may get an error, saying you can't find Vision_ CAFFE.HPP file, because it is in the test file, probably because the merge Caffe version when a little bit of a problem, but I later tried not to affect, and finally run the demo, as follows (no pictures):
ubuntu14.04 + cuda8.0 + cudnnv5 + Caffe + PY-FASTER-RCNN configuration