Ordinal Regression

Caffe Loss Layer for Ordinal Regression with Multiple Output CNN for Age Estimation.

How to

You need to install Caffe first. Copy relative files to Caffe's source code tree.

export CAFFE_HOME=/path/to/caffe
cp layers/ordinal_regression_loss_layer.hpp $CAFFE_HOME/include/caffe/layers/ordinal_regression_loss_layer.hpp
cp layers/ordinal_regression_loss_layer.cpp $CAFFE_HOME/src/caffe/layers/ordinal_regression_loss_layer.cpp
cp layers/ordinal_regression_loss_layer.cu $CAFFE_HOME/src/caffe/layers/ordinal_regression_loss_layer.cu
cp layers/test_ordinal_regression_loss_layer.cpp $CAFFE_HOME/src/caffe/test/test_ordinal_regression_loss_layer.cpp

Modify $CAFFE_HOME/src/caffe/proto/caffe.proto according to layers/caffe.proto

After all, follow Caffe's documents and compile it.

Test the layer

make runtest GTEST_FILTER='OrdinalRegressionLossLayerTest/*'

References

• Caffe
• Ordinal Regression with Multiple Output CNN for Age Estimation