Same-style clothes retrieval is a task to search images which contain exactly the same designing style clothes. For such a task, too limited training data makes the problem of how to gain suitable same-style feature representations challenging but significant. In this paper, we adopt a memory-augmented deep neural network, also called as a few-shot learning model, to collect possibly same-style images. Besides, we present an object-aware clothes retrieval framework to further enhance the same-style feature representations, in which object focusing regions through object detection are first obtained, and a multi-task Siamese network is designed for ranking feature learning provided with some same-style or non-same-style image pairs. Experiments results show that our proposed solution is effective to discover more same-style images precisely, and further achieve the satisfied performance on same-style clothes retrieval.
CITATION STYLE
Shen, Z., Fang, Z., & Liu, J. (2018). Same-style products mining for clothes retrieval. In Communications in Computer and Information Science (Vol. 819, pp. 449–458). Springer Verlag. https://doi.org/10.1007/978-981-10-8530-7_44
Mendeley helps you to discover research relevant for your work.