Stereo matching algorithms usually consist of four steps, including matching cost calculation, matching cost aggregation, disparity calculation, and disparity refinement. Existing CNN-based methods only adopt CNN to solve parts of the four steps, or use different networks to deal with different steps, making them difficult to obtain the overall optimal solution. In this paper, we propose a network architecture to incorporate all steps of stereo matching. The network consists of three parts. The first part calculates the multi-scale shared features. The second part performs matching cost calculation, matching cost aggregation and disparity calculation to estimate the initial disparity using shared features. The initial disparity and the shared features are used to calculate the feature constancy that measures correctness of the correspondence between two input images. The initial disparity and the feature constancy are then fed to a subnetwork to refine the initial disparity. The proposed method has been evaluated on the Scene Flow and KITTI datasets. It achieves the state-of-the-art performance on the KITTI 2012 and KITTI 2015 benchmarks while maintaining a very fast running time.

No abstract

No abstract

Learning visual features from unlabeled image data is an important yet challenging task, which is often achieved by training a model on some annotation-free information. We consider spatial contexts, for which we solve so-called jigsaw puzzles, i.e., each image is cut into grids and then disordered, and the goal is to recover the correct configuration. Existing approaches formulated it as a classification task by defining a fixed mapping from a small subset of configurations to a class set, but these approaches ignore the underlying relationship between different configurations and also limit their application to more complex scenarios.This paper presents a novel approach which applies to jigsaw puzzles with an arbitrary grid size and dimensionality. We provide a fundamental and generalized principle, that weaker cues are easier to be learned in an unsupervised manner and also transfer better. In the context of puzzle recognition, we use an iterative manner which, instead of solving the puzzle all at once, adjusts the order of the patches in each step until convergence. In each step, we combine both unary and binary features on each patch into a cost function judging the correctness of the current configuration. Our approach, by taking similarity between puzzles into consideration, enjoys a more reasonable way of learning visual knowledge. We verify the effectiveness of our approach in two aspects. First, it is able to solve arbitrarily complex puzzles, including high-dimensional puzzles, that prior methods are difficult to handle. Second, it serves as a reliable way of network initialization, which leads to better transfer performance in a few visual recognition tasks including image classification, object detection, and semantic segmentation.

The recently developed droplet-based single-cell transcriptome sequencing (scRNA-seq) technology makes it feasible to perform a population-scale scRNA-seq study, in which the transcriptome is measured for tens of thousands of single cells from multiple individuals. Despite the advances of many clustering methods, there are few tailored methods for population-scale scRNA-seq studies. Here, we develop a Bayesian mixture model for single-cell sequencing (BAMM-SC) method to cluster scRNA-seq data from multiple individuals simultaneously. BAMM-SC takes raw count data as input and accounts for data heterogeneity and batch effect among multiple individuals in a unified Bayesian hierarchical model framework. Results from extensive simulation studies and applications of BAMM-SC to in-house experimental scRNA-seq datasets using blood, lung and skin cells from humans or mice demonstrate that BAMM-SC outperformed existing clustering methods with considerable improved clustering accuracy, particularly in the presence of heterogeneity among individuals.

No abstract

No abstract

Analytical target cascading is a hierarchical multilevel multidisciplinary design methodology. In analytical target cascading, top-level design targets (i.e., specifications) are propagated to lower-level design problems in a consistent and efficient manner. In this paper, a modified Lagrangian dual formulation and coordination for analytical target cascading are developed to enhance a formulation and coordination proposed earlier in the literature. The proposed approach guarantees all the properties established earlier and additionally offers new significant advantages. As established previously for the convex case, the proposed analytical target cascading coordination converges to a global optimal solution with corresponding optimal Lagrange multipliers in the dual space. The Lagrange multipliers can be viewed as the weights for deviations in analytical target cascading formulations. Thus the proposed coordination algorithm finds the optimal solution and the optimal weights for the deviation terms simultaneously. The enhancement allows for target cascading between levels, for the use of augmented Lagrangian to improve convergence of the coordination algorithm, and for prevention of unboundedness. A guideline to set the step size for subgradient optimization when solving the Lagrangian dual problem is also proposed. Nomenclature C sys = children subproblem set with respect to a system problem c = penalty constants for the augmented terms R sub sub = subsystem responses at the subsystem level R sys sub = subsystem responses at the system level T sys = system targets y sub sub = subsystem linking variables at the subsystem level y sys = system linking variables y sys sub = subsystem linking variables at the system level k = step size for subsystem k R k = Lagrange multipliers associated with the deviations of responses of subsystem k y k = Lagrange multipliers associated with the deviations of linking variables of subsystem k k = subgradient for subsystem k