MobileNet+SSD+Focal loss by caffe
This is the MobileNet ,SSD and Focal loss.
#######################################################
I have run this deploy on the pi3,ARM A53 1G RAM.
the 600*600 image use time about 4s
this is the compile command on my pi3:
g++ -o mobilenetssd main.cpp pkg-config --cflags --libs opencv
-D CPU_ONLY -I /home/pi/MobileNet-SSD-Focal-loss/include -D CPU_ONLY -I/home/pi/MobileNet-SSD-Focal-loss/build/src -D CPU_ONLY -L /home/pi/MobileNet-SSD-Focal-loss/build/lib -D CPU_ONLY -lcaffe -lglog -lboost_system -lcblas
#######################################################
I refer to many programs from other people.
The proto,src and include files have already changed.
-
make sure you can compile the caffe.
-
git clone https://github.com/canteen-man/MobileNet-SSD-Focal-loss.git and make -j7
-
cd MobileNet-SSD-Focal-loss
-
./jobs/mobilenet+ssd+focal-loss-300*300.sh
A pre training model is provided here,the voc 07 +12 _iter_20000 caffemodel.
If you want to use openblas, you can git clone https://github.com/xianyi/OpenBLAS.git
cd OpenBLAS
make -j7
make install
Of course, you need to change some files,such as Makefile.config。