Working hard to know your neighbor's margins:Local descriptor learning loss
arXiv e-Print archive - 2017 via Local arXiv
First published: 2017/05/30 (6 years ago) Abstract: We introduce a novel loss for learning local feature descriptors that is
inspired by the SIFT matching scheme. We show that the proposed loss that
relies on the maximization of the distance between the closest positive and
closest negative patches can replace more complex regularization methods which
have been used in local descriptor learning; it works well for both shallow and
deep convolution network architectures. The resulting descriptor is compact --
it has the same dimensionality as SIFT (128), it shows state-of-art performance
on matching, patch verification and retrieval benchmarks and it is fast to
compute on a GPU.
This paper learns deep local patch descriptor (for replacing SIFT) by hard negative mining using current mini-batch. It outperforms SIFT and deep competitors on Oxford5K and Paris6K retrieval datasets.