PIR: A Domain Specific Language for Multimedia Information Retrieval

PIR: A Domain Specific Language for Multimedia Information Retrieval

Xiaobing Huang (University of Wisconsin, USA), Tian Zhao (University of Wisconsin, USA) and Yu Cao (University of Massachusetts, USA)
DOI: 10.4018/978-1-4666-8619-9.ch035
OnDemand PDF Download:
$37.50

Abstract

Multimedia Information Retrieval (MIR) is a problem domain that includes programming tasks such as salient feature extraction, machine learning, indexing, and retrieval. There are a variety of implementations and algorithms for these tasks in different languages and frameworks, which are difficult to compose and reuse due to the interface and language incompatibility. Due to this low reusability, researchers often have to implement their experiments from scratch and the resulting programs cannot be easily adapted to parallel and distributed executions, which is important for handling large data sets. In this paper, we present Pipeline Information Retrieval (PIR), a Domain Specific Language (DSL) for multi-modal feature manipulation. The goal of PIR is to unify the MIR programming tasks by hiding the programming details under a flexible layer of domain specific interface. PIR optimizes the MIR tasks by compiling the DSL programs into pipeline graphs, which can be executed using a variety of strategies (e.g. sequential, parallel, or distributed execution). The authors evaluated the performance of PIR applications on single machine with multiple cores, local cluster, and Amazon Elastic Compute Cloud (EC2) platform. The result shows that the PIR programs can greatly help MIR researchers and developers perform fast prototyping on single machine environment and achieve nice scalability on distributed platforms.
Chapter Preview
Top

1. Introduction

Multimedia Information Retrieval (MIR) (Datta, Joshi, Li, & Wang, 2008; Lew, 2012; Lew, Sebe, Djeraba, & Jain, 2006; Yoshitaka & Ichikawa, 1999) refers to the research endeavor that centers on searching knowledge from multimedia data. In the last decades, substantial progress has been made in different area of MIR research, such as multimedia feature extraction (Hu, Xie, Li, Zeng, & Maybank, 2011; Tuytelaars & Mikolajczyk, 2008), learning and semantics (Atrey, Hossain, El Saddik, & Kankanhalli, 2010; Clinchant, Ah-Pine, & Csurka, 2011; Wang & Hua, 2011), and high performance indexing and query (Moise, Shestakov, Gudmundsson, & Amsaleg, 2013; Scherp & Mezaris, 2013; Shestakov, Moise, Gudmundsson, & Amsaleg, 2013; Mohamed & Marchand Maillet, 2012). As shown by recent surveys (Datta et al., 2008; Lew, 2012; Lew et al., 2006; Yoshitaka & Ichikawa, 1999), since the year 2000, the MIR research efforts have grown tremendously in terms of the number of researchers and practitioners involved, as well as the research papers published. As a result of substantial progress of MIR research and applications, many related software packages, libraries, and systems have been developed and evaluated using a wide range of multimedia data. Some prominent examples include the GIFT (the GNU Image-Finding Tool) (CVML, 2007), FIRE (the Flexible Image Retrieval Engine) (Deselaers et al., 2010), Caliph & Emir (Lux, 2009), LIRE (Lucene Image Retrieval) (Savvas & Chatzichristofis, 2008), ImageTerrier and OpenIMAJ (Hare, Samangooei, Dupplaw, & Lewis, 2011). While significant progress in both MIR research and software development have been made, in practice, we have witnessed that code reuse and system composition for MIR research are still very difficult and the new system developed on top of existing MIR implementation are not optimized for efficiency and cannot be easily adapted for parallelization, which is essential for handling large multimedia data sets. In addition, there is often a steep learning curve for researchers to understand and appropriately use existing frameworks and packages that serve a wide range of MIR purposes before they can even write a single line of code. In fact, sometimes the learning cost is so high that researchers have to give up and turn to create their own software packages instead; such practices accumulatively worsen the current status. Moreover, a lot of components of the MIR software libraries are sequential programs that are designed to run on shared memory computer architectures. MIR experiments of large data sets are time consuming and resource intensive; they often take hours to days to complete and some may even fail after exhausting main memory.

Complete Chapter List

Search this Book:
Reset