the application can be run on X86 and ARM platform with Neon optimization.
git clone --recursive https://github.com/infinivision/mtcnn_ncnn.git
- ncnn as submodule for the main repo
- to support mtcnn. I update the ncnn code. the repo addr
- for the build and install please refer to the ncnn wiki page
# by default the ncnn was installed in ncnn/build/install folder
mkdir -p lib/ncnn
mkdir -p include/ncnn
cp 3rdparty/ncnn/build/install/lib/libncnn.a lib/ncnn/
cp 3rdparty/ncnn/build/install/include/* include/ncnn/
mkdir -p build
cd build
cmake ..
make
cd bin
./test_picture ../models/ncnn/ ../images/1.jpg
cd bin
./test_video ../models/ncnn 0
we use the WIDER FACE validate dataset as the test data
cd bin
./benchmark ../models/ncnn ../images
Platform | CPU Cores | Memory | total images | detected | min time(ms) | max time(ms) | avg time(ms) |
---|---|---|---|---|---|---|---|
MacOS Sierra | 8 | 16GB | 3226 | 2073 | 8.068 | 287.446 | 47.03 |
firefly 3399 | 6 | 2GB | 3226 | 2084 | 25.541 | 4437.42 | 351.578 |
firefly 3399(Neon) | 6 | 2GB | 3226 | 2084 | 10.578 | 1325.09 | 157.908 |