softmaxfocalloss

the focal loss in Aritcal Focal Loss for Dense Object Detection

Code is test on MxNet 0.11.0

Installation

1. Clone the softmaxfocalloss repository, and we'll call the directory that you cloned softmaxfocalloss as ${FOCAL_LOSS_ROOT}.

git clone https://github.com/yuantangliang/softmaxfocalloss.git

2. Install MXNet:

   3.1 Clone MXNet and checkout to MXNet by

   git clone --recursive https://github.com/dmlc/mxnet.git
   git submodule update
   

   3.2 Copy operators in $(FOCAL_LOSS_ROOT)/source/softmaxfocal_output.xxx by

   cp -r $(FOCAL_LOSS_ROOT)/source/* $(MXNET_ROOT)/src/operator/contrib/
   

   3.3 Compile MXNet

   cd ${MXNET_ROOT}
   make -j $(nproc) USE_OPENCV=1 USE_BLAS=openblas USE_CUDA=1 USE_CUDA_PATH=/usr/local/cuda USE_CUDNN=1
   

   3.4 Install the MXNet Python binding by

   cd python
   sudo python setup.py install
   

Test

1. run test function to make sure everything is ok by

cd  $(FOCAL_LOSS_ROOT)
python softmaxfocaltest.py