Contextual Spaces Re-Ranking: accelerating the Re-sort Ranked Lists step on heterogeneous systems

摘要

Re-ranking algorithms have been proposed to improve the effectiveness of content-based image retrievalrnsystems by exploiting contextual information encoded in distance measures and ranked lists. In this paper,rnwe show how we improved the efficiency of one of these algorithms, called Contextual Spaces Re-rnRanking (CSRR). One of our approaches consists in parallelizing the algorithm with OpenCL to use therncentral and graphics processing units of an accelerated processing unit. The other is to modify the algorithmrnto a version that, when compared with the original CSRR, not only reduces the total running timernof our implementations by a median of 1:6u0002 but also increases the accuracy score in most of our testrncases. Combining both parallelization and algorithm modification results in a median speedup of 5:4u0002rnfrom the original serial CSRR to the parallelized modified version. Different implementations for CSRR’srnRe-sort Ranked Lists step were explored as well, providing insights into graphics processing unit sorting,rnthe performance impact of image descriptors, and the trade-offs between effectiveness and efficiency.