Learning Deep Generative Models with Log-Like Motion Features – The present work investigates the problem of learning Deep Generative models with log-like motion features for recognition task. We consider the problem of learning Generative representations that take as input the motion feature vectors of a dataset, a video and a text. In the video representation space, we adopt the state-of-the-art for CNN classification, which is a non-linear embedding of the video into a sparse set of convolutional embeddings. The resulting models perform well regardless of feature-based classification, and can perform very well on large datasets with a fixed input. In addition, the proposed models can learn to generate a variety of motion features for different types of recognition tasks, making them suitable for use as training data.
Deep learning frameworks for face recognition have been widely adopted since their inception. This paper addresses the question of whether or not face recognition is as important to the social context as in the traditional computer vision community has been focused on. The proposed framework is based on a two-stage process: one is on capturing and processing features from the scene from two distinct spatial regions and the other is on capturing and processing features from the image space. We show how the two stages could be used as complementary inputs to a system composed of three different facial attributes. We first show that both stages can be combined and we propose a new architecture to solve them. Based on this architecture we demonstrate the effectiveness of deep learning for face recognition on several datasets and we can finally assess the quality and accuracy of the proposed framework.
Robust Principal Component Analysis via Structural Sparsity
Probabilistic and Regularized Graph Models for Graph Embedding
Learning Deep Generative Models with Log-Like Motion Features
Deep Prediction of Hidden Dimensions Using Machine Learning Data
Facial Expression Recognition by Deep Convolutional Neural NetworksDeep learning frameworks for face recognition have been widely adopted since their inception. This paper addresses the question of whether or not face recognition is as important to the social context as in the traditional computer vision community has been focused on. The proposed framework is based on a two-stage process: one is on capturing and processing features from the scene from two distinct spatial regions and the other is on capturing and processing features from the image space. We show how the two stages could be used as complementary inputs to a system composed of three different facial attributes. We first show that both stages can be combined and we propose a new architecture to solve them. Based on this architecture we demonstrate the effectiveness of deep learning for face recognition on several datasets and we can finally assess the quality and accuracy of the proposed framework.