Visual grounding, a task to ground (i.e., localize) natural language in images, essentially requires composite visual reasoning. However, existing methods over-simplify the composite nature of language into a monolithic sentence embedding or a coarse composition of subject-predicateobject triplet. In this paper, we propose to ground natural language in an intuitive, explainable, and composite fashion as it should be. In particular, we develop a novel modular network called Neural Module Tree network (NMTREE) that regularizes the visual grounding along the dependency parsing tree of the sentence, where each node is a neural module that calculates visual attention according to its linguistic feature, and the grounding score is accumulated in a bottom-up direction where as needed. NMTREE disentangles the visual grounding from the composite reasoning, allowing the former to only focus on primitive and easy-to-generalize patterns. To reduce the impact of parsing errors, we train the modules and their assembly endto-end by using the Gumbel-Softmax approximation and its straight-through gradient estimator, accounting for the discrete nature of module assembly. Overall, the proposed NMTREE consistently outperforms the state-of-the-arts on several benchmarks. Qualitative results show explainable grounding score calculation in great detail. * Corresponding Author. a pink umbrella carried by a girl in pink boots (a) Holistic a pink umbrella carried by a girl in pink boots (b) Tripletpink umbrella carried by girl in pink boots ADJ NN VB ADP acl agent pobj prep pobj amod amod ADP NN NN ADJ Sum[by] Comp[in] Comp[carried] Sum[umbrella] Sum[girl] Sum[boots] Single[pink] Single[pink] Single[ROOT] NMTREE Dependency Parsing Tree (c) NMTREE Region Proposals man VB ADP pobj prep dobj amod nsubj NN ADJ NN NN riding VB dobj Dep. label Embedding Word Embedding POS tag Embedding Concat Fc layer Gumbel Softmax Sum[riding]? Comp[riding]? the man with helmet riding a brown horse Single[ROOT] Sum[man] Comp[with] Comp[riding] Sum[horse] Single[brown]
The dispersion and loss in microstructured fibers are studied using a full-vectorial compact-2D finite-difference method in frequency-domain. This method solves a standard eigen-value problem from the Maxwell's equations directly and obtains complex propagation constants of the modes using anisotropic perfectly matched layers. A dielectric constant averaging technique using Ampere's law across the curved media interface is presented. Both the real and the imaginary parts of the complex propagation constant can be obtained with a high accuracy and fast convergence. Material loss, dispersion and spurious modes are also discussed.
A finite-difference frequency-domain (FDFD) method is applied for photonic band gap calculations. The Maxwell's equations under generalized coordinates are solved for both orthogonal and non-orthogonal lattice geometries. Complete and accurate band gap information is obtained by using this FDFD approach. Numerical results for 2D TE/TM modes in square and triangular lattices are in excellent agreements with results from plane wave method (PWM). The accuracy, convergence and computation time of this method are also discussed.
A node non-uniform deployment based on clustering algorithm for underwater sensor networks (UWSNs) is proposed in this study. This algorithm is proposed because optimizing network connectivity rate and network lifetime is difficult for the existing node non-uniform deployment algorithms under the premise of improving the network coverage rate for UWSNs. A high network connectivity rate is achieved by determining the heterogeneous communication ranges of nodes during node clustering. Moreover, the concept of aggregate contribution degree is defined, and the nodes with lower aggregate contribution degrees are used to substitute the dying nodes to decrease the total movement distance of nodes and prolong the network lifetime. Simulation results show that the proposed algorithm can achieve a better network coverage rate and network connectivity rate, as well as decrease the total movement distance of nodes and prolong the network lifetime.
Existing move-restricted node self-deployment algorithms are based on a fixed node communication radius, evaluate the performance based on network coverage or the connectivity rate and do not consider the number of nodes near the sink node and the energy consumption distribution of the network topology, thereby degrading network reliability and the energy consumption balance. Therefore, we propose a distributed underwater node self-deployment algorithm. First, each node begins the uneven clustering based on the distance on the water surface. Each cluster head node selects its next-hop node to synchronously construct a connected path to the sink node. Second, the cluster head node adjusts its depth while maintaining the layout formed by the uneven clustering and then adjusts the positions of in-cluster nodes. The algorithm originally considers the network reliability and energy consumption balance during node deployment and considers the coverage redundancy rate of all positions that a node may reach during the node position adjustment. Simulation results show, compared to the connected dominating set (CDS) based depth computation algorithm, that the proposed algorithm can increase the number of the nodes near the sink node and improve network reliability while guaranteeing the network connectivity rate. Moreover, it can balance energy consumption during network operation, further improve network coverage rate and reduce energy consumption.
We propose a full-vectorial Galerkin method for the analysis of circular symmetric fibers with arbitrary index profiles. A set of orthogonal Laguerre-Gauss functions is used to calculate the dispersion relation and mode fields of TE and TM modes. Examples are given for both standard step-index fibers and Bragg fibers. For standard step-index fiber with low or high index contrast, the Galerkin method agrees well with the analytical results. In the case of the TE mode of a Bragg fiber it agrees well with the asymptotic results.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.