2008-IUPR-05Jun_1853.pdf

0.0005 0.00075 0.001 0.01 0.1 0.10 0.2 0.25 0.3 0.4 0.45 0.5 0.50 0.55 0.6 0.60 0.62 0.64 0.65 0.66 0.68 0.70 0.72 0.74 0.75 0.76 0.78 0.8 0.80 0.82 0.84 0.85 0.90 1-NN 1.0 1.00 1.2 1.6 10-fold 11.0 1100 1300 150 1500 157 162 177 1816 1823 196 1977 1978 2000 2001 2002 2003 2005 2006 2007 2008 213 2169 2178 230 238 257 2600 263 264 271 2nd 3.0 3.1 3.2 3.3 3.4 30.7 370 377 4.1 4.2 4.5 416 431 460 472 5.0 5.00 5.5 56.3 7.8 7200 79.5 791 8.2 8.9 82.6 83.6 9.5 961-386261 97.000 978-1-60558-070-8 ABSTRACT ACKNOWLEDGEMENTS ACM APPROACH Adrian After Again Algorithms All Also Anal Analysis Annotations Apostolof Artificial Auer Averaged BMVC Bags Basic Bauckhage Bayes Beyond Boolean Boosting Borth Bosch Breuel Budapest CIVR CVPR Campbell Canada Categories Center Challenge3 Christian Chum Class Classification Closest Comprehensive Compt Computer Concept Conf Consequently Content-Based Content-based Copyright Corresponding Correspondingly Cybern D-67663 DFKI DISCUSSION Daniel Data Database Dempster Department Descriptors Deselaers Detection Detectors Discovering Discriminative Distributions Duda During EXPERIMENTS Edition Efros Eng Engine Epanechnikov Equation Equations Evaluation Everts Expectation Experiment Experimentation Extraction Falls Features Fei-Fei Fergus Ferrari Figure Finally For Frames FtydCER2l4k Further Fussenegger GI-Informatiktage Gaithersburg Gemert General Generalization Generic German Germany Geusebroek Google H.3.3 Hart Haubold Hebert High-Level Hofmann Huurnink IBM ICCV ICVS IEEE INTRODUCTION Identifying Image Image-CLEF Images InViRe Incomplete Indexing Information Innovation Instead Int Intell Intelligence Interest Intern International Interscience J.-M Journal July June Kaiserslautern Kernels Keyframe Keyframes Keysers Keywords Knowing Koelma Labeled Laird Latent Lazebnik Learning Learns Left Liempt Like Likelihood Liu Local Locations Long-term Lozano-Perez MIL MUSCLE Mach Machine Man Marin Marszalek Matching Maximization Maximum Measurement MediaMill Mikolajczyk Models Mohr More Mori Multiple-Instance Natsev Natural Ney Niagara Non-Relevant Notation Notebook November Object Objects Obviously Once Online Only Ontario Opelt Other Otherwise Our Oxford PASCAL PLSA Patches Pattern Perception Performance Permission Perona Philbin Point Ponce Portals Possible Posterior Probabilistic Publications Pyramid REFERENCES RELATED Recently Recognition Recognizing Relevance Relevant Research Results Retrieval RheinlandPfalz Right Rooij Rosenberg Royal Rubin Russell Sande Scale-Invariant Scene Schmid Schulze Science Search Section Semantic Since Sivic Smeulders Smith Snoek Society Spatial Statistical Stiftung Storage Stork Study Subject Such Summarization Sys System TRECVID TRECVID-2007 Table Tag Tagging Tamura Technical Techniques Terms Tesic Tests Textural Texture The Thereby Therefore These They This Thomas Thus Training Trans Two USA Uijlings Ulges University Unsupervised Using VOC Video Videos Vis Vision Visual WORK Watching Weakly While Wiley Workshop Worring Yamawaki Yang Youtube Zhang Zisserman abbreviated accepted access achieve achieved achieves acquire acquisition action actions actually adapted additional addressed adequate advantage aforementioned algorithm allow alse alternating analogy analysis annotate annotated annotation annotations annual appear appearance appears appli applicability applied applying approach approaches arbitrarily archives arouse assessed assessing assigned associated assume assumed assumes assumption attention automatic automatically autonomously average away background bags bandwidth bandwidths based baseline basic basketball beach bear behaves belongs benefit best better binary blinkx.com boosting bootstrap bound boundary broadcast broader built cability called campaign candidates captured case cases cat category certain challenge challenges check choice citation classification classified classifiers classifies classroom clear close clustering clutter collections color column combined commercial community comparable compared competitor complement complex compute computed concatenated concept concepts conclusion considered consisting consists construction contain containing content content-based contest contest2 context contributions convergence copies copy corpus corre correct correctly corresponds cross-validation data database databases dataset datasets decades decreases decreasing defined degenerates demo demo.iupr.org demonstrate demonstrated denote denotes densities density described descriptions descriptor desert detect detection detector detectors determined developed did different differs difficult digital discover discretizations discriminative distance distinguish distinguished distributed distribution distributions document documents does doing domain dotted downloaded downloading downside draw drawing drawn draws e.g early easy efficient eiffeltower emerged empirically entities entity entropy equivalent estim estimate estimated estimates estimating estimation evaluate evaluation evaluations evidence example exit expected experiment experiments expert explicit explicitly extra extraction face fact fair false fashion feature features fee filtered final finally fixed flag focus following formulation forward fraction frame frames framework frequently function fusing fusion future gave general general1 generalization generalize generalizing generative given gives giving golf good goto grant granted graphical ground groups growing hand hard headed hidden high high-level higher highest hiking histogram histograms hope horizontal hosts hrs http human i.e identified identifies identify identifying illustrated illustrates illustration image images impact impr impression improve improved improvement improvements in-depth includes including incomplete incorporating increased indicate indicated industry inferred influence influenced informatik.uni-kl.de information instance instances instead intensively interesting interview intra-shot introduce introduced introduction investigate involves involving irrelevant iteration iterative iupr.dfki.de judged just kernel key keyframe keyframes keysers knowledge known labeled labeling labels large large-scale latent leads learn learning left level like likelihood limited lines list lists local location locations lost low low-level lowest maintain majority make makes manner manual manually marginalize market matching material max maximum measured meta-data methods mining model modeled modeling modelled modelling models moderate modern motion motion-based motivation multimodal multiple nearest need needs negative negatives neighbor news non-parametric non-relevant notation notice number object objects observations observed obtain obtained obvious ones online optimal optimizes order organized outlined overall page pages paper parameter parameter-free parameters parametric participating particular parts patch-based patches peak perfect performance permission personal pieces pixel place plan plot plots plotted plugging point portal portals position positive positives possible post posterior posteriors potentially practice precisely precision presence present presented preserved previous previously prior priors probabilistic probabilities probability problem procedure profit project projects proof properly properties proposed prototype provide provided publication published purpose quality quantitative quantities queries question random range rapidly real-world realworld reason recent recently recognition redistribute reduce reduced refer referred region rel related relative releno relevance relevant relied rely represent representative represented republish required requires research researched researchers resembles respect respectively restrictive result resulting results retrieval revealed revealing reveals rightmost rigorous robust robustness row rule run sample sampled scale scenes scheme schulze score scored scores search second section seen segmented semantic separated serve servers set sets setup shot shots showed showing shown shows significant significantly similar simple simpler simulating single situation socalled soccer source sources space specialized specific specifically split sponds sports standard state-of-theart stations statistical step steps stored strategies strategy streams strong strongest strongly structure study studying subjective subsumed successful successfully suitability suitable sum superior supervision support supported swimming systems tackled tag tagged tagger taggers tagging tags takes targeted task tell temporal term terms test tested testing text-based texture throws tiled tiles time-consuming tjk tmb topic topics traditional trained training trecvid trends tries true truth try turns ulges underlying unknown unseen unsegmented untypical updated upload uploading use used user-generated users uses using usually valuable value values van vance variable variables varied varies vast vector vectors video videos videotagging view viewed views visible vision visual visualized visually voting way weak weakly weighted well-known widely work works world-wide www-nlpir.nist.gov www.pascal-network.org yields yj1 yjk yjmj youtube youtube.com