Abstract
Most existing 3D object recognition algorithms focus on leveraging the strongdiscriminative power of deep learning models with softmax loss for theclassification of 3D data, while learning discriminative features with deepmetric learning for 3D object retrieval is more or less neglected. In thepaper, we study variants of deep metric learning losses for 3D objectretrieval, which did not receive enough attention from this area. First , twokinds of representative losses, triplet loss and center loss, are introducedwhich could learn more discriminative features than traditional classificationloss. Then, we propose a novel loss named triplet-center loss, which canfurther enhance the discriminative power of the features. The proposedtriplet-center loss learns a center for each class and requires that thedistances between samples and centers from the same class are closer than thosefrom different classes. Extensive experimental results on two popular 3D objectretrieval benchmarks and two widely-adopted sketch-based 3D shape retrievalbenchmarks consistently demonstrate the effectiveness of our proposed loss, andsignificant improvements have been achieved compared with thestate-of-the-arts.