Semantic segmentation via pixel-to-center similarity calculation

Since the fully convolutional network has achieved great success in semantic segmentation,lots of works have been proposed to extract discriminative pixel representations.However,the authors observe that existing methods still suffer from two typical challenges:(i)The intra-class feature variation between different scenes may be large,leading to the difficulty in maintaining the consistency between same-class pixels from different scenes;(ii)The inter-class feature distinction in the same scene could be small,resulting in the limited performance to distinguish different classes in each scene.The authors first rethink se-mantic segmentation from a perspective of similarity between pixels and class centers.Each weight vector of the segmentation head represents its corresponding semantic class in the whole dataset,which can be regarded as the embedding of the class center.Thus,the pixel-wise classification amounts to computing similarity in the final feature space between pixels and the class centers.Under this novel view,the authors propose a Class Center Similarity(CCS)layer to address the above-mentioned challenges by generating adaptive class centers conditioned on each scenes and supervising the similarities between class centers.The CCS layer utilises the Adaptive Class Center Module to generate class centers conditioned on each scene,which adapt the large intra-class variation between different scenes.Specially designed Class Distance Loss(CD Loss)is introduced to control both inter-class and intra-class distances based on the predicted center-to-center and pixel-to-center similarity.Finally,the CCS layer outputs the processed pixel-to-center similarity as the segmentation prediction.Extensive experiments demonstrate that our model performs favourably against the state-of-the-art methods.

Survey on depth and RGB image-based 3D hand shape and pose estimation

The field of vision-based human hand three-dimensional(3D)shape and pose estimation has attracted significant attention recently owing to its key role in various applications,such as natural human computer interactions.With the availability of large-scale annotated hand datasets and the rapid developments of deep neural networks(DNNs),numerous DNN-based data-driven methods have been proposed for accurate and rapid hand shape and pose estimation.Nonetheless,the existence of complicated hand articulation,depth and scale ambiguities,occlusions,and finger similarity remain challenging.In this study,we present a comprehensive survey of state-of-the-art 3D hand shape and pose estimation approaches using RGB-D cameras.Related RGB-D cameras,hand datasets,and a performance analysis are also discussed to provide a holistic view of recent achievements.We also discuss the research potential of this rapidly growing field.

Scalable and ultrafast epitaxial growth of single-crystal graphene wafers for electrically tunable liquid-crystal microlens arrays

The scalable growth of wafer-sized single-crystal graphene in an energy-efficient manner and compatible with wafer process is critical for the killer applications of graphene in high-performance electronics and optoelectronics. Here, ultrafast epitaxial growth of single-crystal graphene wafers is realized on singlecrystal Cu90Ni10(1 1 1) thin films fabricated by a tailored two-step magnetron sputtering and recrystallization process. The minor nickel(Ni) content greatly enhances the catalytic activity of Cu, rendering the growth of a 4 in. single-crystal monolayer graphene wafer in 10 min on Cu90Ni10(1 1 1), 50 folds faster than graphene growth on Cu(1 1 1). Through the carbon isotope labeling experiments, graphene growth on Cu90Ni10(1 1 1) is proved to be exclusively surface-reaction dominated, which is ascribed to the Cu surface enrichment in the Cu Ni alloy, as indicated by element in-depth profile. One of the best benefits of our protocol is the compatibility with wafer process and excellent scalability. A pilot-scale chemical vapor deposition(CVD) system is designed and built for the mass production of single-crystal graphene wafers, with productivity of 25 pieces in one process cycle. Furthermore, we demonstrate the application of single-crystal graphene in electrically controlled liquid-crystal microlens arrays(LCMLA), which exhibit highly tunable focal lengths near 2 mm under small driving voltages. By integration of the graphene based LCMLA and a CMOS sensor, a prototype camera is proposed that is available for simultaneous light-field and light intensity imaging. The single-crystal graphene wafers could hold great promising for highperformance electronics and optoelectronics that are compatible with wafer process.

Lyapunov-based switched-gain impact angle control guidance

To hit stationary ground targets in specified direction, a nonlinear impact angle control guidance law based on Lyapunov stability theory is proposed. The proposed law ensures the convergence of the heading angle and the miss distance to guarantee a successful engagement. The impact angle can be adjusted by varying a single parameter. And the maximum value of acceleration has been analyzed to get the proper range for control parameter. Considering the achievable impact angle set is limited, an additional phase is added to form a two-phase control strategy.The first phase is to establish a proper initial condition for the second phase, and the second phase is to hit the target with a certain impact angle. An analysis of the proper selection of control parameters to expand the achievable impact angle set is presented. The performance of the proposed guidance law is illustrated with simulations.

Ascent Trajectory Optimization for Air-Breathing Hypersonic Vehicles Based on IGS-MPSP

This paper proposes an improved Generalized Quasi-Spectral Model Predictive Static Programming(GS-MPSP)algorithm for the ascent trajectory optimization for hypersonic vehicles in a complex°ight environment.The proposed method guarantees the satisfaction of constraints related to the state and control vector while retaining its high computational e±ciency.The spectral representation technique is used to describe the control variables,which reduces the number of decision variables and makes the control input smooth enough.Through Taylor expansion,the constraints are transformed into an inequality containing only decision variables,such that it can be added into GS-MPSP framework.By Gauss quadrature collocation method,only a few collocation points are needed to solve the sensitivity matrix,which greatly accelerates the calculation.Subsequently,the analytical expression is obtained by combining the static optimization with the penalty function method.Finally,the simulation results demonstrate that the proposed improved GS-MPSP algorithm can achieve both high computational e±-ciency and high terminal precision under the constraints..