[1]MagicLeap: https://www.magicleap.com/
[2]Daniel DeTone: http://www.danieldetone.com/
[3]paper: https://arxiv.org/abs/1712.07629
[4]slides: https://github.com/MagicLeapResearch/SuperPointPretrainedNetwork/blob/master/assets/DL4VSLAM_talk.pdf
[5]code: https://github.com/MagicLeapResearch/SuperPointPretrainedNetwork
[6]子像素卷积: https://blog.csdn.net/leviopku/article/details/84975282
[7]UCN: https://arxiv.org/abs/1606.03558
[8]SpatialGridSamplerBilinear: https://github.com/pytorch/pytorch/blob/f064c5aa33483061a48994608d890b968ae53fb5/aten/src/THNN/generic/SpatialGridSamplerBilinear.c
[9]Toward Geometric Deep SLAM: https://arxiv.org/abs/1707.07410