Multi-Channel Multi-Resolution RGB-D Light Field Video with Convolutional Neural Networks – Constraint-based image segmentation is a key challenge for many computer vision problems. Most existing methods either use an RGB-D image as a pre-processing step, or directly feed the RGB image into a convolutional neural network (CNN). Previous work has explored the idea of adapting CNN’s structure to make use of the features of the input image. This work is based on learning a CNN model of the input image. In this paper, to overcome these two shortcomings, we propose a novel deep learning-based method to segment the input image with a CNN. Using the deep CNN model, we extend the existing CNN segmentation approach to the task of fine-tuning the image features. Results demonstrate that our proposed CNN model achieves a better performance on our segmentation task than the existing CNN model with respect to the performance of other existing deep learning-based CNN models.
We describe a new model-based generative adversarial network (GAN) for semantic object classification. We propose a novel method which uses a convolutional neural network to learn a new set of discriminative representations for the domain model which can be used to learn models for a variety of different semantic object categories. As we study the task of classification of semantic objects, we propose an unsupervised CNN to learn discriminative representations of semantic words. We validate our work by analyzing various benchmarks including MNIST and the state of the art state of the art SPMVM for semantic object classification. After training a discriminative learning network we are able to classify object classes from the input semantic sentence. The discriminative representations from the CNN have also been used to predict the object class from the learned representations. We test our method and compare it to other existing state of the art supervised object classification methods on the MNIST and SPMVM datasets. The classification accuracies are higher for the MNIST and SPMVM than for other state of the art supervised classification algorithms but only a few metrics are evaluated to show the performance.
Learning to detect and eliminate spurious events from unstructured analysis of time series
Deep Reinforcement Learning for Constrained Graph Reasoning
Multi-Channel Multi-Resolution RGB-D Light Field Video with Convolutional Neural Networks
Morphon: a collection of morphological and semantic wordsWe describe a new model-based generative adversarial network (GAN) for semantic object classification. We propose a novel method which uses a convolutional neural network to learn a new set of discriminative representations for the domain model which can be used to learn models for a variety of different semantic object categories. As we study the task of classification of semantic objects, we propose an unsupervised CNN to learn discriminative representations of semantic words. We validate our work by analyzing various benchmarks including MNIST and the state of the art state of the art SPMVM for semantic object classification. After training a discriminative learning network we are able to classify object classes from the input semantic sentence. The discriminative representations from the CNN have also been used to predict the object class from the learned representations. We test our method and compare it to other existing state of the art supervised object classification methods on the MNIST and SPMVM datasets. The classification accuracies are higher for the MNIST and SPMVM than for other state of the art supervised classification algorithms but only a few metrics are evaluated to show the performance.