PyTorch code for CVPR'2020 paper “Pose-guided Visible Part Matching for Occluded Person ReID”
113
stars
27
forks
source link
Why does the output of graph matching have two loss terms, one positive and one negative, so will they be effective if they are added together to make the loss? #15
@lhc11171552
Hi,
The positive term is the binary classification loss for training the PVP module to predict the right visibility score of each part. And the negative term aims to maximize the object function of graph matching. The final loss function has already been the sum of them.
@lhc11171552 Hi, The positive term is the binary classification loss for training the PVP module to predict the right visibility score of each part. And the negative term aims to maximize the object function of graph matching. The final loss function has already been the sum of them.