Multiple instance learning for model ensemble and meta data transfer

Abstract

Traditional Exemplar-SVMs (ESVM) require millions of negative samples to establish a linear exemplar detector. However, Exemplar Linear Discriminant Analysis (ELDA) can achieve similar performance while avoid negative samples mining. To construct a strong object classifier, Multiple Instance Learning (MIL) is used to combine exemplar detectors and reduce annotation ambiguity. By applying MIL to Exemplar-LDA (ELDA), we simplify the training process and achieve better performance than ESVM on object detection. Moreover, exemplar models can transfer the available meta-data (segmentation, geometric structure, etc.) of training samples directly onto the detected objects, which provide more accurate and richer attributions than the detection results of a bounding box.