Learning Feature for RGB-D based Action Recognition and Detection – Object detection from a single image of an object is one of the key challenges of many industrial environments. In this paper we are interested in applying deep learning to a large-scale object recognition task. Deep learning architecture for object recognition is a popular approach to solve various object recognition problems. However, deep learning is usually limited to a single type of object. Since deep learning can solve many different object recognition problems from an image to a video, in this paper, we propose a new deep learning architecture which employs two complementary layers — the convolutional layer and the convolutional layer. Unlike current architectures, our architecture maintains a simple mapping between layers to achieve an efficient and accurate object recognition. Besides, our method is capable of recovering the object of interest given the object’s visual appearance, therefore can be used for different applications. Using the proposed architecture, more than 4.25 million frames of objects with their visual appearance were annotated. Our evaluation using both real images and online video datasets demonstrates our method to perform better than state-of-the-art object recognition methods.

A new and simple method, called Theta-Riemannian Metrics (Theta-Riemannian Metrics) is proposed for generating Riemannian metrics. Theta-Riemannian Metrics provides new methods for estimating the correlation distances between Riemannian metrics, and a new method for optimizing the relationship between correlation distances and the metric coefficients. We show that theta-Riemannian Metric can be decomposed into a hierarchical and multi-decompositions metric, and then use them to generate new metrics. We have shown that theta-Riemannian Metrics can be derived using a new model called Theta Riemannian Metrics which is optimized using Riemannian metric models. Results of our numerical experiments show that theta-Riemannian Metrics can outperform the state-of-the-art approaches for generating Riemannian metrics in terms of the expected regret.

Deep Convolutional Neural Network: Exploring Semantic Textural Deepness for Person Re-Identification

Graph learning via adaptive thresholding

# Learning Feature for RGB-D based Action Recognition and Detection

On the Reliable Detection of Non-Linear Noise in Continuous Background Subtasks

Learning Stochastic Gradient Temporal Algorithms with Riemannian MetricsA new and simple method, called Theta-Riemannian Metrics (Theta-Riemannian Metrics) is proposed for generating Riemannian metrics. Theta-Riemannian Metrics provides new methods for estimating the correlation distances between Riemannian metrics, and a new method for optimizing the relationship between correlation distances and the metric coefficients. We show that theta-Riemannian Metric can be decomposed into a hierarchical and multi-decompositions metric, and then use them to generate new metrics. We have shown that theta-Riemannian Metrics can be derived using a new model called Theta Riemannian Metrics which is optimized using Riemannian metric models. Results of our numerical experiments show that theta-Riemannian Metrics can outperform the state-of-the-art approaches for generating Riemannian metrics in terms of the expected regret.