Joint deep feature learning and unsupervised visual domain adaptation for cross-domain 3D object retrieval

Highlights：

• An unsupervised domain adaption method for 3D object retrieval by jointly learning the deep 3D representation and domain alignment in the end-to-end manner is proposed.

• The proposed method can reduce statistical domain divergence and improve the feature representation from different domains.

• The proposed method focuses on the cross-domain retrieval task, which can effectively use the exist 3D object dataset to explore the generation ability of methods for another dataset.

• The experimental results on CAD Object to CAD Object and RGB-D Object to CAD Object retrieval show the superiority of the proposed method.