Large Scale Long-tailed Product Recognition System at Alibaba
A practical large scale product recognition system suffers from the phenomenon of long-tailed imbalanced training data under the E-commercial circumstance at Alibaba. Besides product images at Alibaba, plenty of image related side information (e.g. title, tags) reveal rich semantic information about images. Prior works mainly focus on addressing the long tail problem in visual perspective only, but lack of consideration of leveraging the side information. In this paper, we present a novel side information based large scale visual recognition co-training~(SICoT) system to deal with the long tail problem by leveraging the image related side information. In the proposed co-training system, we firstly introduce a bilinear word attention module aiming to construct a semantic embedding over the noisy side information. A visual feature and semantic embedding co-training scheme is then designed to transfer knowledge from classes with abundant training data (head classes) to classes with few training data (tail classes) in an end-to-end fashion. Extensive experiments on four challenging large scale datasets, whose numbers of classes range from one thousand to one million, demonstrate the scalable effectiveness of the proposed SICoT system in alleviating the long tail problem. In the visual search platform Pailitao\footnote{http://www.pailitao.com} at Alibaba, we settle a practical large scale product recognition application driven by the proposed SICoT system, and achieve a significant gain of unique visitor~(UV) conversion rate.