2024-03-29T18:17:25Zhttps://ipsj.ixsq.nii.ac.jp/ej/?action=repository_oaipmhoai:ipsj.ixsq.nii.ac.jp:001450092024-03-29T05:26:34Z01164:04619:07845:08330
MRF-Based Multi-Label Classification using Label RelationsMRF-Based Multi-Label Classification using Label Relationsenghttp://id.nii.ac.jp/1001/00144976/Technical Reporthttps://ipsj.ixsq.nii.ac.jp/ej/?action=repository_action_common_download&item_id=145009&item_no=1&attribute_id=1&file_no=1Copyright (c) 2015 by the Information Processing Society of JapanThe University of TokyoThe University of TokyoThe University of TokyoThe University of TokyoRyosuke, FurutaYusuke, FukushimaToshihiko, YamasakiKiyoharu, AizawaMulti-label classification and multi-classifier fusion have been independently explored as different problems. We propose a re-labeling method that can simultaneously treat these two problems in an unified framework. Our method considers (a) correlations between different labels, and (b) correlations between different feature types. In particular, the proposed method models both label and feature correlations in a single Markov random field (MRF), and jointly optimizes the label assignment problem. We apply our method to impression prediction of oral presentations. We train and evaluate the proposed method using a collection of 1,646 TED talk videos for 14 different impression types. Experimental results on this dataset show that the proposed method obtains a statistically significant macro-average accuracy of 93.3%, outperforming several competitive baseline methods.Multi-label classification and multi-classifier fusion have been independently explored as different problems. We propose a re-labeling method that can simultaneously treat these two problems in an unified framework. Our method considers (a) correlations between different labels, and (b) correlations between different feature types. In particular, the proposed method models both label and feature correlations in a single Markov random field (MRF), and jointly optimizes the label assignment problem. We apply our method to impression prediction of oral presentations. We train and evaluate the proposed method using a collection of 1,646 TED talk videos for 14 different impression types. Experimental results on this dataset show that the proposed method obtains a statistically significant macro-average accuracy of 93.3%, outperforming several competitive baseline methods.AA11131797研究報告コンピュータビジョンとイメージメディア(CVIM)2015-CVIM-19816182015-09-072188-87012015-09-03