Multiple instance learning for labeling faces in broadcasting news video
Multiple instance learning for labeling faces in broadcasting news video
复制标题
DOI：
10.1145/1101149.1101155
复制
发表时间：
2005-11
期刊：
影响因子：
--
通讯作者：
Jun Yang;Rong Yan;Alexander Hauptmann
中科院分区：
文献类型：
--
作者：
Jun Yang;Rong Yan;Alexander Hauptmann
用名字标注新闻视频中的人脸是一个有趣的研究问题，以前是用监督方法解决的，这需要用户在标注训练数据上付出很大的努力。在本文中，我们研究了一个更具挑战性的问题设置，其中没有关于数据标签的完整信息。具体来说，通过利用人脸名字的唯一性，我们将该问题表述为一个特殊的多实例学习（MIL）问题，即独占MIL或eMIL问题，从而可以通过使用部分标记信息训练的模型作为人脸的匿名判断来解决该问题，这需要较少的用户收集精力。针对eMIL问题，提出了两种判别概率学习方法：独占密度（Exclusive Density， ED）和迭代密度（Iterative ED）。在人脸标记问题上的实验表明，所提方法的性能优于传统的MIL算法，并接近用完整数据标签训练的监督方法所取得的性能。
Labeling faces in news video with their names is an interesting research problem which was previously solved using supervised methods that demand significant user efforts on labeling training data. In this paper, we investigate a more challenging setting of the problem where there is no complete information on data labels. Specifically, by exploiting the uniqueness of a face's name, we formulate the problem as a special multi-instance learning (MIL) problem, namely exclusive MIL or eMIL problem, so that it can be tackled by a model trained with partial labeling information as the anonymity judgment of faces, which requires less user effort to collect. We propose two discriminative probabilistic learning methods named Exclusive Density (ED) and Iterative ED for eMIL problems. Experiments on the face labeling problem shows that the performance of the proposed approaches are superior to the traditional MIL algorithms and close to the performance achieved by supervised methods trained with complete data labels.