Combining CPU and GPU architectures for fast similarity search

Abstract

The Signature Quadratic Form Distance on feature signatures represents a flexible distance-based similarity model for effective content-based multimedia retrieval. Although metric indexing approaches are able to speed up query processing by two orders of magnitude, their applicability to large-scale multimedia databases containing billions of images is still a challenging issue. In this paper, we propose a parallel approach that balances the utilization of CPU and many-core GPUs for efficient similarity search with the Signature Quadratic Form Distance. In particular, we show how to process multiple distance computations and other parts of the search procedure in parallel, achieving maximal performance of the combined CPU/GPU system. The experimental evaluation demonstrates that our approach implemented on a common workstation with 2 GPU cards outperforms traditional parallel implementation on a high-end 48-core NUMA server in terms of efficiency almost by an order of magnitude. If we consider also the price of the high-end server that is ten times higher than that of the GPU workstation then, based on price/performance ratio, the GPU-based similarity search beats the CPU-based solution by almost two orders of magnitude. Although proposed for the SQFD, our approach of fast GPU-based simiCommunicated by: Kaushik Chakrabarti.M. Kruliš · T. Skopal ( ) · J. Lokoč

Similarity Search and Applications

“…These experiments, as well as further investigations (not shown here due to space constraints) seem to show for smaller datasets, single-tree search may be fastest; for sufficiently high dimensions, LSH is faster. This corroborates existing results [32]; as the dimension of data gets higher, pruning rules become less effective. Regardless, in low-to-medium dimensions, the improved dual-tree traversal is dominant.…”

“…As GPU speed and memory size continues to increase -AMD recently released a 32GB GPU -the problem sizes appropriate for the GPU increase as well. Large problems will always have to be handled from disk [28], but even there, hybrid CPU-GPU implementations [32,46] rely on the GPU to solve large subproblems by brute-force.…”

Visual Descriptors in Methods for Video Hyperlinking

“…Position-color-texture Feature Signatures [23,24,44] were utilized to approximate a distribution of color and texture in each keyframe. This descriptor can be utilized in image retrieval tasks, where color and texture is meaningful for retrieval.…”

Indexing of Spatiotemporal Trajectories for Efficient Distance Threshold Similarity Searches on the GPU

“…This is in part because branches in the instruction flow cause thread serialization and thus loss of parallel efficiency [17]. The kNN query (not on trajectories) has been studied in the context of the GPU [24,20] and on hybrid CPU-GPU environments [21]. In this work we focus on indexing techniques for distance threshold similarity searches on trajectories for the GPU, which to our knowledge has only been explored in our previous work [11].…”