Caffe Study

Repository for studying Caffe deep learning framework.

Prerequisite

• Visual studio 2017
• Cuda version 9.2
• cuDNN version 7.6.1
• python 3.5.x
• CMake 3.4 or higher
• Boost 1.65.1 msvc 14.0 x64
• MSBuild 2015

How to build

• First, you have to download or clone this archive and unzip these
• I will call root folder of this repository as $CAFFE$

1. Install Visual studio 2017 (Or higher)

2. And then, install Cuda version 9.2 and cuDNN 7.6.1 (Which compatiable with cuda version)

3. You have to modify ./cmake/Cuda.cmake to proper SM version of your GPU device. (line 7 and 44)

   7   set(Caffe_known_gpu_archs "30 35 50 60 61")
   44  set(__nvcc_out "6.1")
   

4. Install python with version 3.5.x (Must!) and numpy

   pip install numpy
   

5. Must add or check Path Variables of Cuda and python

6. Install Boost 1.65.1 for msvc 14.0

7. Install MSBuild 2015

8. At CMD (Command line) change folder to $CAFFE$ and type like this to build Caffe for window

   ./scripts/build_win.cmd
   

9. Then, in $CAFFE$ there are new folder "build"

10. Open Caffe.sln in "build" folder

11. Build all projects of solutions by "Release" mode

References

• Caffe tutorials
• Caffe in a Day Tutorial
• Intel / TAU Summer Workshop on Deep Learning and Caffe
• API Documentations