MSR-Bing Image Retrieval Challenge is one of the ACM Multimedia Grand Challenges held by the Microsoft. This year, two tasks have been put forward in the competition, asking participants to design a query-image system and a dog-breed system based on a wide range of categories of Bing dataset. As a real world large-scale image retrieval competition, MSR-Bing Image Retrieval Challenge attracted quantities of teams to participate.
The final competition result of 2015 MSR-Bing Image Retrieval Challenge was announced in Brisbane, Australia on Oct 28th, 2015. Qiang Song’s team (supervised by Prof. Jian Cheng) stood out among all the participants and won the 1st prize and the 3rd prize respectively in two tasks.
Specifically, a deep learning based fine-grained method and a search based framework proposed by Qiang Song and his team have effectively addressed the problems in image classification and retrieval tasks. For instance, the top 5 precision of fine-grained task by the above method has reached 85% accuracy, which surpasses the result of the second place nearly 15%. Overall, with great performance in generalization, our method was highlighted by the Institute of Microsoft Research.
Qiang Song presents in the MSR-Bing Image Retrieval Challenge