Zero-Shot Sketch-Based Image Retrieval via Graph Convolution Network

• DOI: 10.1609/aaai.v34i07.6993
• Publication Date: 2020-06-19T08:21:49Z
• Authors: Zhaolong Zhang; Yuejie Zhang; Rui Feng; Tao Zhang; Weiguo Fan
• Keywords: Autoencoder; Sketch; Semantic gap; Convolution (computer science)

Zero-Shot Sketch-based Image Retrieval (ZS-SBIR) has been proposed recently, putting the traditional (SBIR) under setting of zero-shot learning. Dealing with both challenges in SBIR and learning makes it become a more difficult task. Previous works mainly focus on utilizing one kind information, i.e., visual information or semantic information. In this paper, we propose SketchGCN model graph convolution network, which simultaneously considers Thus, our can effectively narrow domain gap transfer knowledge. Furthermore, generate from using Conditional Variational Autoencoder rather than only map them back space to space, enhances generalization ability model. Besides, feature loss, classification loss are introduced optimize Our gets good performance challenging Sketchy TU-Berlin datasets.