Bundle Optimization for Multi-aspect Embedding

• DOI: 10.48550/arxiv.1703.09928
• Publication Date: 2017-01-01
• Authors: Qiong Zeng; Wenzheng Chen; Zhuo Han; Mingyi Shi; Yanir Kleiman; Daniel Cohen‐Or; Baoquan Chen; Yangyan Li
• Keywords: FOS: Computer and information sciences; Computer Vision and Pattern Recognition (cs.CV); Computer Science - Computer Vision and Pattern Recognition; 0202 electrical engineering, electronic engineering, information engineering; 02 engineering and technology

Understanding semantic similarity among images is the core of a wide range computer vision applications. An important step towards this goal to collect and learn human perceptions. Interestingly, context often ambiguous as can be perceived with emphasis on different aspects, which may contradictory each other. In paper, we present method for learning images, inferring their latent aspects embedding them into multi-spaces corresponding aspects. We consider multi-embedding problem an optimization function that evaluates embedded distances respect qualitative clustering queries. The key idea our approach embed measures share same in bundles. To ensure aspect sharing multiple measures, image classification queries are presented to, solved by users. collected clusters then converted bundles tuples, fed bundle algorithm jointly infers multi-aspect embedding. Extensive experimental results show significantly outperforms state-of-the-art approaches various datasets, scales well large measures.