Configuring FASTER-RCNN under Windows

Mark A

http://yun.baidu.com/share/link?shareid=1018944597&uk=1543560377

http://blog.csdn.net/sinat_30071459/article/details/50546891

Configuration environment: windows10x64 matlab2015ra VS2013 Opencv2.4.11 CUDA7.5 gtx950m

2. CUDA7.5 Installation
   Because Cuda7.5 did a lot of optimization improvements, and WIN10 support is better, so here choose to install Cuda7.5, the specific process reference here. However, because the author of the MATLAB version of the faster R-CNN CNN Library is compiled under Cuda6.5, so there will be problems, either need to compile their own Cuda7.5, or can be downloaded here.

4. Download the FASTERR-CNN package
   Download the MATLAB version of FASTER-CNN: link on the author Shaoqingren GitHub. Then unzip to Local.

6. compiling the External/caffe library
   If your PC is installed with CUDA6.5, you can run FETCH_DATA/FETCH_CAFFE_MEX_WINDOWS_VS2013_CUDA65.M directly.
   If you are installing CUDA7.5 like I do, you can download the compiled library here and extract it directly to the root of the author code.

8. generating NMS Mex files
   Run the file root directory: FASTER_RCNN_BUILD.M
   Note: There will be frequent error when running, it is necessary to change the FUNCTIONS/NMS/NVMEX.M under the Cuda_path to your computer Cuda installation path.

10. Set Correlation function Path
    Run the startup.m file, which is the relevant path for the function to run.

12. download models of CNN
    Either run the author code in the package: FETCH_DATA/FETCH_FASTER_RCNN_FINAL_MODEL.M automatically downloaded and saved in the output folder.
    Or you download in the Baidu Cloud, and extract into the output folder.

14. Test Run Demo
    If everything is OK in front of you, copy the EXPERIMENTS/SCRIPT_FASTER_RCNN_DEMO.M to the root directory and run to see the results.
    Note: Because the VGG16 model is too large, so the general computer will appear matlab run, so here will either
    opts.use_gpu = false;Run under the CPU.
    or use ZF models (simpler than VGG16, lower accuracy):

    model_dir= fullfile(pwd, ‘output‘, ‘faster_rcnn_final‘, ‘faster_rcnn_VOC0712_ZF‘)；

This will allow you to see the results and run time of the test picture.

Configuring FASTER-RCNN under Windows