Semantic Regularities in Textual-Visual Embedding – This paper investigates the ability of human beings to use visual language to describe the world. In a natural language, people are trained to describe events and events. In a language that is designed to be interpretable, humans may only describe events and events with complex syntactic structure. Humans are trained to describe objects and events in a human language. This paper provides a general framework for analyzing and developing natural language to describe the world by using a human language.

In this paper we describe a novel method for generating visualizations of human activities. To do so we first establish that a human action can be predicted into an approximate representation by a deep Convolutional Neural Network. In the future, we consider also a video action. This could be accomplished by constructing and training a deep model. In addition, we propose a novel algorithm called ResNet, which takes the input sequences and outputs the video frames of an action. By using this model and its predictions, we can then predict the sequence for human activity. This can be accomplished by applying the ResNet framework to a different task. In this work, we analyze the ResNet method and investigate how it is applied to image recognition, video object categorization, and video face recognition. The experiments show that the ResNet model significantly outperforms all three methods.

Deep Learning for Precise Action Prediction

Probabilistic Belief Propagation

Semantic Regularities in Textual-Visual Embedding

Learning Text and Image Descriptions from Large Scale Video Annotations with Semi-supervised Learning

Fast and accurate water detection with convolutional neural networkIn this paper we describe a novel method for generating visualizations of human activities. To do so we first establish that a human action can be predicted into an approximate representation by a deep Convolutional Neural Network. In the future, we consider also a video action. This could be accomplished by constructing and training a deep model. In addition, we propose a novel algorithm called ResNet, which takes the input sequences and outputs the video frames of an action. By using this model and its predictions, we can then predict the sequence for human activity. This can be accomplished by applying the ResNet framework to a different task. In this work, we analyze the ResNet method and investigate how it is applied to image recognition, video object categorization, and video face recognition. The experiments show that the ResNet model significantly outperforms all three methods.