View-Hosting: Streaming views on a single screen – Scene localization is a key component of many applications, including computer vision and image retrieval, as the goal is to identify a scene from a set of available view-aware sensors. In this work, we propose an iterative algorithm for scene localization under various camera viewpoint parameters. The proposed method is based on a low-bandwidth feature representation framework and it computes the optimal number of parameters by solving an optimization problem over the feature vectors. For this purpose, we adopt a new convolutional neural network to compute an optimal number of parameters while minimizing the cost associated with using the feature representation. Finally, we propose a deep learning model to handle the challenging scene localization problem. Experimental results on image retrieval, scene localization and object tracking show that the proposed method can be a highly promising step for scene localization.
Recently, a large amount of work has been performed on semantic graph embedding, including cross-domain and multidimensional embedding. However, the use of a single semantic graph embedding metric is not well-suited for the task of semantic graph embedding problem (QGSP). In this work, we propose a novel semantic graph embedding method based on semantic graph embeddings for QGSP. The underlying metric embedding method is used to embed two semantic groups with two semantic graph embeddings (1) semantic graph embeddings of a single domain for classification, and (2) two semantic graph embeddings of a different domain for labeling. We then show that the semantic embedding metric used in this work can be used to encode a combination of semantic graph embeddings and semantic graph embeddings in a unified framework. Experimental results on both synthetic and real datasets demonstrate the use of the proposed method improves the classification recognition performance.
Bayesian Neural Networks with Gaussian Process Models for Analysis of Multi-Dimensional Shapes
Unsupervised Active Learning with Partial Learning
View-Hosting: Streaming views on a single screen
Tensor learning for learning a metric of bandwidthRecently, a large amount of work has been performed on semantic graph embedding, including cross-domain and multidimensional embedding. However, the use of a single semantic graph embedding metric is not well-suited for the task of semantic graph embedding problem (QGSP). In this work, we propose a novel semantic graph embedding method based on semantic graph embeddings for QGSP. The underlying metric embedding method is used to embed two semantic groups with two semantic graph embeddings (1) semantic graph embeddings of a single domain for classification, and (2) two semantic graph embeddings of a different domain for labeling. We then show that the semantic embedding metric used in this work can be used to encode a combination of semantic graph embeddings and semantic graph embeddings in a unified framework. Experimental results on both synthetic and real datasets demonstrate the use of the proposed method improves the classification recognition performance.