search

arashwan / matrixnet

PyTorch implementation for MatrixNet object detection architecture.
MIT License
171 stars 32 forks source link

some questions about centers in corner-xnet #18

Open liushuai839 opened 4 years ago

liushuai839 commented 4 years ago

Hello, guys, I am confused about why regressing the centers directly in corner-xnet can replace the embeddings in cornerNet to do the corner matching ?

rra94 commented 4 years ago

I had misunderstood the question before. Center regression works as we have multiple outputs so the location of the center is bound by the size of the feature map of a particular layer. We match centers allowing for a 30% error in matching. Please refer to the arxiv paper for more details. https://arxiv.org/abs/2001.03194