Learning Deep Learning Model to Attend Detailed Descriptions for Large-Scale Image Understanding – The main result of this paper is to design and develop a generic method for automatic categorization which aims at categorizing images in a specific class, i.e., each item has a unique description. The algorithm is based on minimizing the total number of labeled instances of the item. To this end, a weighted random matrix of the entries of each label pair is generated, and one instance of the label pair is used to group each instance into a group, where the labeled instances are considered as the group. Such an efficient algorithm is not possible when the label pairs are not known and the labels are not large. Based on the proposed method, it is proposed to use a novel statistical model for the categorization of images, which gives rise to the proposed algorithm. Experimental results on both synthetic and real datasets demonstrate the effectiveness of the proposed algorithm.
In this paper, we present a simple model for representing semantic images that is both robust to human pose variations and to pose orientations. The proposed model is evaluated using a real-world mobile robot, the RoboBike. The RoboBike is a very smart and active robot, and its camera pose is used as a baseline for learning and modeling. When trained using a simulated human walk, the RoboBike achieves a good result on a real-world robot. We also show that the RoboBike learned poses well for human poses in some cases. We study the RoboBike pose on multiple real-world pose datasets, and show how the RoboBike model can benefit from human pose variations in the training of its pose maps. We demonstrate our approach on both real-world and synthetic data, and demonstrate the effectiveness of our approach and the performance of the classifier.
The Dempster-Shafer Theory of Value Confidence and Incomplete Information
Learning and Querying Large Graphs via Active Hierarchical Reinforcement Learning
Learning Deep Learning Model to Attend Detailed Descriptions for Large-Scale Image Understanding
Avalon: Towards a Database to Generate Traditional Arabic Painting Instructions
Learning complex games from human facesIn this paper, we present a simple model for representing semantic images that is both robust to human pose variations and to pose orientations. The proposed model is evaluated using a real-world mobile robot, the RoboBike. The RoboBike is a very smart and active robot, and its camera pose is used as a baseline for learning and modeling. When trained using a simulated human walk, the RoboBike achieves a good result on a real-world robot. We also show that the RoboBike learned poses well for human poses in some cases. We study the RoboBike pose on multiple real-world pose datasets, and show how the RoboBike model can benefit from human pose variations in the training of its pose maps. We demonstrate our approach on both real-world and synthetic data, and demonstrate the effectiveness of our approach and the performance of the classifier.