no code implementations • 27 Feb 2020 • Z. Gao, K. X Xue, S. H Wan
However, most existing networks do not take into account the impact of multi-view image selection on network training, and the use of contrastive loss alone only forcing the same-class samples to be as close as possible.