Adversarial-inspired Semi-supervised Adversarial Image Segmentation – We present a new method for visual semantic segmentation of images at a low-dimensional plane with respect to non-convex parameters. Our method, named Deep Visual Semidefinite Network (DVS-NN), models the input vector as a non-convex function on the manifold, and extracts a non-convex function from the manifold to compute high-dimensional parametrized subgraphs for each quadrant. In this paper, an adversarial training is performed with a low-dimensional parametric model and the discriminative information is computed jointly from the manifold and the parametric model. The learned semantic segmentation aims at obtaining more precise parameters, and the discriminative information can be decoded for a lower-dimensional space. Experiments on three benchmark datasets demonstrate the potential of our method in terms of image segmentation performance on different semantic segmentations tasks.

The most popular segmentation models for image classification are based on unsupervised deep architectures. Recently, it has been shown that the traditional deep architecture models have not been well suited to real-world tasks such as human evaluation of images. In this paper, we propose a new deep architecture called PNS (Deep Network for Image Prediction). PNS is a combination of two different learning approaches. First, it learns to predict image segments accurately while the network learns to estimate an image’s shape. The network is trained to learn the optimal pose, and then it is able to predict each segment in a supervised manner. We test PNS on two datasets and show that it outperforms most existing approaches for human evaluation of image segmentation.

Predicting the outcomes of games

A Generative Model and Algorithm for Bayesian Nonlinear Eigenproblems with Implicit Conditional Effects

Adversarial-inspired Semi-supervised Adversarial Image Segmentation

A Simple, Yet Efficient, Method for Learning State and Action Graphs from Demographic Information: Distribution Data

Image Segmentation using Sparsity-based Densely Connected Convolutional Neural Networks with OutliersThe most popular segmentation models for image classification are based on unsupervised deep architectures. Recently, it has been shown that the traditional deep architecture models have not been well suited to real-world tasks such as human evaluation of images. In this paper, we propose a new deep architecture called PNS (Deep Network for Image Prediction). PNS is a combination of two different learning approaches. First, it learns to predict image segments accurately while the network learns to estimate an image’s shape. The network is trained to learn the optimal pose, and then it is able to predict each segment in a supervised manner. We test PNS on two datasets and show that it outperforms most existing approaches for human evaluation of image segmentation.