GPU-Based for accelerating the BF-SIFT method for large scale 3D shape retrieval

Abstract

This paper addresses the problem of 3D shape retrieval in large databases of 3D objects (large scale retrieval). While this problem is emerging and interesting as the size of 3D object databases grows rapidly, the main two issues the community has to focus on are: computational efficiency of 3D object retrieval and the quality of retrieved results. In this work we are interested by the problem of the computational efficiency where we propose to accelerate the BF-SIFT method by exploiting the potential of the GPU to reduce the computation times of the shape indexing of the query and the shape matching using the GPU. Experimental results show that the execution time is significantly reduced, this promises that the large scale retrieval can be achieved using the GPU.

GPU-based Acceleration of Methods based on Clock Matching Metric for Large Scale 3D Shape Retrieval

“…In order to accelerate the retrieval process, various content-based retrieval methods and approaches have been proposed in the literature [1,2,3,4,8,14,16]. Including sequential solutions [1,4] and those based on high-performance computing, like multi-core [3]and GPU [2,8], ... Despite the high-efficiency of HPC solutions, the problem is that there are a few works in the literature that implement the 3D shape retrieval under GPU environment, most of them are partial since they only concern the shape indexing phase [8].…”

“…To accelerate the shape matching phase we have already proposed a GPU-based implementation [2], of a given 3D shape retrieval method called BF-SIFT [11]. The solution proposed in [2] is general and it can be applied to several methods based on the same similarity metric which is KLD (Kullback-Leibler divergence), such as Euclidean distance.…”

“…The challenges come from storage, computation speed, and features representation. Various methods and techniques have been proposed in the literature to accelerate the process of retrieving [1,2,3,4,8,14,16]. The different solutions proposed can be classified into two categories:…”

“…For GPU solutions, Wu [8] have proposed an implementation of SIFT algorithm, this solution is partial because it concerns only a step of shape indexing process and not the shape matching process. To our knowledge, the only work that dealt with the shape matching process on GPU is [2], in which we have proposed a GPU-based implementation of an existing method called BF-SIFT [11]. This solution is general it can be applied to different well-known dissimilarity measurements such as Euclidean L2, Minkowski.…”