Patch Hierarchy Research Articles

HiCT: Hierarchical Comprehend of Transformer for Weakly Supervised Object Localization

The weakly supervised object localization (WSOL) has always been a very challenging research subject in the field of computer vision, which aims to predict the localization of objects in an image using only an image-level class labeling approach. The traditional convolutional neural network (CNN) based-approaches utilize the local class activation and discrimination for classification guidance, and the biggest drawback of CNN is that it cannot capture the remote feature dependencies between pixels. Recently, the transformer architecture has been deployed in the WSOL, but the transformer cannot well capture local features. To address the above problems, we propose HiCT (Hierarchical comprehend of transformer), a simple and effective visual converter variation method. Moreover, we also propose a discriminative-based attention layer (DAL), which aims to mine the local feature information by utilizing the global token attention graph mechanism. To further improve the coverage of object localization, we introduce the spatial aware digging module (SADM). In addition, a set of complementarity loss calculators to patch hierarchy (CPH) is proposed to improve the sample class aggregation capability of our model. Finally, we conducted experiments on two commonly used datasets of CUB-200-2011 and ILSVRC, so as to verify the effectiveness of our method.

JASMIN-based Two-dimensional Adaptive Combined Preconditioner for Radiation Diffusion Equations in Inertial Fusion Research

AbstractWe present a JASMIN-based two-dimensional parallel implementation of an adaptive combined preconditioner for the solution of linear problems arising in the finite volume discretisation of one-group and multi-group radiation diffusion equations. We first propose the attribute of patch-correlation for cells of a two-dimensional monolayer piecewise rectangular structured grid without any suspensions based on the patch hierarchy of JASMIN, classify and reorder these cells via their attributes, and derive the conversion of cell-permutations. Using two cell-permutations, we then construct some parallel incomplete LU factorisation and substitution algorithms, to provide our parallel -GMRES solver with the help of the default BoomerAMG in the HYPRE library. Numerical results demonstrate that our proposed parallel incomplete LU preconditioner (ILU) is of higher efficiency than the counterpart in the Euclid library, and that the proposed parallel -GMRES solver is more robust and more efficient than the default BoomerAMG-GMRES solver.

Spatial, Temporal and Hierarchical Variability of the Factors Driving Urban Growth: A Case Study of the Treasure Valley of Idaho, USA

Urban landscape is a system of hierarchically nested spatial structures or land patches with their distinct dynamics and causal factors. The lower level structures evolve together to give out different forms, patterns and extents during the process of urbanization. An understanding of driving factors of these hierarchical structures has important implications for future land use planning and urban management. Adopting an innovative framework of urban patch hierarchy, this study investigates the drivers and spatiotemporal variability of their explanatory power at urban and intra-urban patch levels, with Treasure Valley of Idaho, USA as a case of study. We calibrated global and local logistic regression models against a set of 18 spatialized variables. Results show that growth drivers and their impact vary both across the study site and along the cycles of urbanization at different levels of urban landscape. Urban agglomeration factors including the proximity to urbanized area had the highest impact on new development, urban growth forms and land uses. Factors were less stable at the lower level because of their subtle interconnectedness with local structures. The study further revealed an existence of distinct patterns of association among the growth forms, land use types, and the included factors. Specific growth forms and land use classes demonstrated identical temporal patterns with certain driving factors.

The fully connectedN-dimensional skeleton: probing the evolution of the cosmic web

A method to compute the full hierarchy of the critical subsets of a density field is presented. It is based on a Watershed technique and uses a probability propagation scheme to improve the quality of the segmentation by circumventing the discreteness of the sampling. It can be applied within spaces of arbitrary dimensions and geometry. This recursive segmentation of space yields, for a d-dimensional space, a d− 1 succession of n-dimensional subspaces that fully characterize the topology of the density field. The final one-dimensional manifold of the hierarchy is the fully connected network of the primary critical lines of the field: the skeleton. It corresponds to the subset of lines linking maxima to saddle points, and provides a definition of the filaments that compose the cosmic web as a precise physical object, which makes it possible to compute any of its properties such as its length, curvature, connectivity etc. When the skeleton extraction is applied to initial conditions of cosmological N-body simulations and their present-day non-linear counterparts, it is shown that the time evolution of the cosmic web, as traced by the skeleton, is well accounted by the Zel'dovich approximation (ZA) provided the small-scale artificial smoothing introduced by the displacement field computation is taken into account. This scale is shown to be well modelled as in units of the non-linear scale, a being the expansion factor. Comparing this skeleton to the initial skeleton undergoing the Zel'dovich mapping shows that two effects are competing during the formation of the cosmic web: a general dilation of the larger filaments that is captured by a simple deformation of the skeleton of the initial conditions on the one hand, and the shrinking, fusion and disappearance of the more numerous smaller filaments on the other hand. The net result corresponds to a decrease of the cosmic skeleton's length with time. Other applications of the N-dimensional skeleton and its peak patch hierarchy are discussed.

Spatial context influences patch residence time in foraging hierarchies

Understanding responses of organisms to spatial heterogeneity in resources has emerged as a fundamentally important challenge in contemporary ecology. We examined responses of foraging herbivores to multi-scale heterogeneity in plants. We asked the question, "Is the behavior observed at coarse scales in a patch hierarchy the collective outcome of fine scale behaviors or, alternatively, does the spatial context at coarse scales entrain fine scale behavior?" To address this question we created a nested, two-level patch hierarchy. We examined the effects of the spatial context surrounding a patch on the amount of time herbivores resided in the patch. We developed a set of competing models predicting residence time as a function of the mass of plants contained in a patch and the distance between patches and examined the strength of evidence in our observations for these models. Models that included patch mass and inter-patch distance as independent variables successfully predicted observed residence times (bears: r (2)=0.67-0.76 and mule deer: r (2)=0.33-0.55). Residence times of grizzly bears (Ursus arctos) and mule deer (Odocoileus hemionus) responded to the spatial context surrounding a patch. Evidence ratios of Akaike weights demonstrated that models containing effects of higher levels in the hierarchy on residence time at lower levels received up to 34 times more support in the data than models that failed to consider the higher level context for grizzly bears and up to 48 times more support for mule deer. We conclude that foraging by large herbivores is influenced by more than one level of heterogeneity in patch hierarchies and that simple empirical models offer a viable alternative to optimal foraging models for the prediction of patch residence times.

A patch hierarchy approach to modeling surface and subsurface hydrology in complex flood‐plain environments

AbstractGeomorphology interacts with surface‐ and ground‐water hydrology across multiple spatial scales. Nonetheless, hydrologic and hydrogeologic models are most commonly implemented at a single spatial scale. Using an existing hydrogeologic computer model, we implemented a simple hierarchical approach to modeling surface‐ and ground‐water hydrology in a complex geomorphic setting. We parameterized the model to simulate ground‐ and surface‐water flow patterns through a hierarchical, three‐dimensional, quantitative representation of an anabranched montane alluvial flood plain (the Nyack Flood Plain, Middle Fork Flathead River, Montana, USA). Comparison of model results to field data showed that the model provided reasonable representations of spatial patterns of aquifer recharge and discharge, temporal patterns of flood‐water storage on the flood plain, and rates of ground‐water movement from the main river channel into a large lateral spring channel on the flood plain, and water table elevation in the alluvial aquifer. These results suggest that a hierarchical approach to modeling ground‐ and surface‐water hydrology can reproduce realistic patterns of surface‐ and ground‐water flux on alluvial flood plains, and therefore should provide an excellent ‘quantitative laboratory’ for studying complex interactions between geomorphology and hydrology at and across multiple spatial scales. Copyright © 2004 John Wiley & Sons, Ltd.

Effect of scale on plant disease incidence and heterogeneity in a spatial hierarchy

The effect of patch scale on disease incidence and heterogeneity was assessed in a three-scaled spatial hierarchy using data collected from strawberry leaf blight epidemics and with Monte-Carlo simulation studies. Cluster sampling was used to collect disease incidence data from several strawberry fields at two commercial farms in Ohio over 2 years. Sampling units consisted of five leaves of three leaflets each. Leaflets represented the lowest scale of the hierarchy, leaves represented the intermediate scale, and the sampling units represented the upper scale. Results from data randomizations indicated that diseased leaflets were nearly maximally aggregated among leaves within sampling units, resulting in close to the minimum possible incidence and heterogeneity of disease among sampling units at the leaf scale. Based on Monte-Carlo simulations, incidence and heterogeneity of disease at the leaflet (i.e. lowest) scale had the largest effect on incidence and heterogeneity of disease at the leaf scale, respectively. However, analysis of variance of simulation results indicated that arrangement of diseased leaflets among leaves within sampling units, sampling unit size (=number of leaves or leaflets per sampling unit), leaf complexity (leaflets per leaf), and all interactions affected disease incidence and heterogeneity at the leaf scale. For a given level of intensity of leaflet disease incidence and heterogeneity, incidence and heterogeneity of disease at the leaf scale were most affected by sampling unit size when leaflets were arranged in an aggregated fashion among leaves; increasing the number of leaves per sampling unit was associated with decreasing disease incidence and increasing heterogeneity at the leaf scale. However, leaf complexity had the greater effect on leaf disease incidence and heterogeneity when diseased leaflets were arranged randomly or uniformly among leaves within sampling units; greater leaf complexity was associated with greater incidence and heterogeneity. Combined results indicated that incidence and heterogeneity of incidence of Phomopsis leaf blight varied between scales in a patch hierarchy. This variability was shown to be a function of lower-scale disease incidence and heterogeneity, sampling unit size, and leaf complexity.

Riparian vegetation‐environment relationships: complimentarity of gradients versus patch hierarchy approaches

Abstract. Two prominent conceptual frameworks, environmental gradients and patch hierarchies, are used in combination to describe vegetation patterns along a riparian corridor in a semi‐arid South African system. We adopt both approaches, since riparian corridors are characterized by both strong environmental gradients above, away from and along the river, as well as a mosaic of patches in the geomorphology at multiple hierarchical scales. Constrained and unconstrained ordinations were used to determine the variability in vegetation pattern accounted for by the gradient and the geomorphic patch hierarchy data sets. The gradient data set consisted of vertical, lateral and longitudinal dimensions of the macro‐channel, while the patch hierarchy data set consisted of substratum type, morphological unit and channel type. Elevation up the macro‐channel bank, of the gradient data set, explained the main variation in vegetation pattern, and alluded to overriding processes of flooding frequency and water availability as determinants of vegetation pattern. Along the fluvially dynamic macro‐channel floor (lower elevation range), patchiness at the scale of the morphological unit best explained vegetation pattern. This relationship with morphological units suggests that the formation of well developed alluvial bars, and the degree of bedrock influence are important processes. The nested hierarchical framework used provided a good basis for identifying scale specific pattern in a relational manner. In systems characterized by strong environmental gradients as well as a patch mosaic at different spatial and temporal scales, the combined use of both perspectives to develop a fuller understanding of vegetation pattern is imperative and is encouraged.