A GPU-based MapReduce framework for MSR-Bing Image Retrieval Challenge

This paper presents a large-scale image retrieval system based on an efficient Graphics Processing Units (GPU)-based MapReduce framework for the MSR-Bing Image Retrieval Challenge. The proposed system is designed for searching images and scoring image-query pairs based on their relevances efficiently and accurately. Unlike the former systems which usually start with text queries to select partial images and then process their visual contents, the proposed system attempts to search similar images directly from the entire dataset through visual content and then compare their text similarities, owing to the powerful computational capabilities of the proposed GPU-based MapReduce framework. It is shown that the proposed system achieves 0.492 in terms of DCG@25 on the final evaluation.