Semantic Embedding for Sketch-Based 3D Shape Retrieval

The main challenge for sketch-based 3D shape retrieval lies with the large domain gap between 2D sketch and 3D shape. Most existing works attempt to overcome the domain gap by learning a joint feature embedding space to align the two domains. In this work we argue that the large domain gap cannot be effectively bridged in a shared feature space. Instead, we propose to align them in their common class label space. To this end, a novel deep cross-domain semantic embedding model is proposed. Extensive experiments are carried out on two large benchmarking datasets, SHREC’13 and SHREC’14. The results show that the proposed model drastically improves over the state-of-the-art alternatives.