This program is a C++ toolbox for extracting CNN feature map from image. For achieving high-performance computation, it support CUDA acceleration. The average computation of 3 layers feature maps of a 640x480 image is 20 ms. You can easly integrated the code to you embedded program. The core functions are extracted from matconvnet (http://www.vlfeat.org/matconvnet/)

• OpenCV 2.4.9 (or above)
• OpenBLAS (include in the package at ./Thirdparty/OpenBLAS)
• gfortran (sudo apt-get install gfortran)
• CUDA 5.0 (or above)
• PIL (included in the code at ./Thirdparty/PIL)

1. build OpenBLAS

 cd ./Thirdparty/OpenBLAS
 tar xzf OpenBLAS-0.2.14.tar.gz
 make 
 sudo make install

2. build PIL

 cd ./Thirdparty/PIL
 make

3. build pi-cnn

 cd cnn_models/
 wget http://www.adv-ci.com/download/pi-cnn/imagenet-vgg-f.cm 
 cd ..
 make

 # GPU calculation
 ./test_CNN useGPU=1

 # CPU calculation
 ./test_CNN useGPU=0

 # show feature maps
 ./test_CNN act=showFeatureMap

 # match feature points
 ./test_CNN act=matchWholeImage

Only test on LinuxMint 17.1 64-bit, may be other distributions are also support.

- -

http://www.adv-ci.com/blog/source/pi-cnn/