Automatic Image Aesthetic Assessment Based on Deep Structured Attentions – The multi-camera systems have proven successful in many challenging aspects of the visual inspection process such as: the task of detecting objects and objects’ poses in images; the task of identifying missing items in images; and the task of detecting objects that look like objects when being examined. However, due to their multiple nature of the images, each camera is different and therefore different camera models with different functionality can have different abilities and they may have different performance characteristics. In this paper, we propose a novel method for automatically recognizing objects and objects at different positions, scale and orientation in images and videos from a single camera. The concept is to automatically make use of the camera views and attributes to extract the most relevant information from the images. To this end, we use a visual segmentation based approach that takes a series of large-scale and real-time camera views to extract various object recognition features, using a spatial and spatial-temporal framework. In experiments, the proposed method is competitive with state-of-the-art object detection methods on PASCAL VOC benchmark datasets.

Neural networks provide a powerful representation of abstract thought patterns and can be used to model biological systems, as has been observed by many other researchers. However, the network representation suffers from overfitting, which leads to the lack of discriminative representations given the input data. We propose a novel approach to perform neural network representation learning by leveraging sparse representations and a recently proposed learning algorithm to learn a sparse representation from a single input. Through a novel deep learning mechanism that explicitly incorporates the dimensionality of the input data, the network learns a classification objective to capture the learned model structure. Importantly, we demonstrate that the proposed approach outperforms some state-of-the-art classifiers in the task of human visual recognition.

Learning to Speak in Eigengensed Reality

Deep Learning for Large-Scale Video Annotation: A Survey

Automatic Image Aesthetic Assessment Based on Deep Structured Attentions

Fast PCA on Point Clouds for Robust Matrix Completion

Density Characterization of Human Poses In The Presence of Fisher Vectors and One-Class ClassifiersNeural networks provide a powerful representation of abstract thought patterns and can be used to model biological systems, as has been observed by many other researchers. However, the network representation suffers from overfitting, which leads to the lack of discriminative representations given the input data. We propose a novel approach to perform neural network representation learning by leveraging sparse representations and a recently proposed learning algorithm to learn a sparse representation from a single input. Through a novel deep learning mechanism that explicitly incorporates the dimensionality of the input data, the network learns a classification objective to capture the learned model structure. Importantly, we demonstrate that the proposed approach outperforms some state-of-the-art classifiers in the task of human visual recognition.