Expansion window local alignment weighted network for fine-grained sketch-based image retrieval

Abstract

Fine-Grained Sketch-Based Image Retrieval (FG-SBIR) is a worthwhile task, which can be useful in many scenarios like recommendation systems, receiving a great deal of attention. In this study, we analyze challenges faced in FG-SBIR and propose a novel Expansion Window Local Alignment Weighted Network (EWLAW-Net). Specifically, it contains two main components: the Expansion Window Local Alignment module (EWLA) and the Local Weighted Fusion module (LWF). The EWLA module adopts an expansion window mechanism to align local features extracted from the backbone with the same semantic meaning between photos and sketches. The LWF module assigns weights to each local feature of the sketch after evaluating their importance and fuses them to calculate the similarity between the sketch and photos for retrieval. Experiments are conducted on five datasets and the results demonstrate the effectiveness of the proposed method.