Recently, object detection in aerial images has gained much attention in computer vision. Different from objects in natural images, aerial objects are often distributed with arbitrary orientation. Therefore, the detector requires more parameters to encode the orientation information, which are often highly redundant and inefficient. Moreover, as ordinary CNNs do not explicitly model the orientation variation, large amounts of rotation augmented data is needed to train an accurate object detector. In this paper, we propose a Rotation-equivariant Detector (ReDet) to address these issues, which explicitly encodes rotation equivariance and rotation invariance. More precisely, we incorporate rotation-equivariant networks into the detector to extract rotation-equivariant features, which can accurately predict the orientation and lead to a huge reduction of model size. Based on the rotation-equivariant features, we also present Rotation-invariant RoI Align (RiRoI Align), which adaptively extracts rotation-invariant features from equivariant features according to the orientation of RoI. Extensive experiments on several challenging aerial image datasets DOTA-v1.0, DOTA-v1.5 and HRSC2016, show that our method can achieve state-of-the-art performance on the task of aerial object detection. Compared with previous best results, our ReDet gains 1.2, 3.5 and 2.6 mAP on DOTA-v1.0, DOTA-v1.5 and HRSC2016 respectively while reducing the number of parameters by 60% (313 Mb vs. 121 Mb). The code is available at: url{https://github.com/csuhan/ReDet}.
This paper presents GraphFederator, a novel approach to construct joint representations of multi-party graphs and supports privacy-preserving visual analysis of graphs. Inspired by the concept of federated learning, we reformulate the analysis of multi-party graphs into a decentralization process. The new federation framework consists of a shared module that is responsible for joint modeling and analysis, and a set of local modules that run on respective graph data. Specifically, we propose a federated graph representation model (FGRM) that is learned from encrypted characteristics of multi-party graphs in local modules. We also design multiple visualization views for joint visualization, exploration, and analysis of multi-party graphs. Experimental results with two datasets demonstrate the effectiveness of our approach.
The past decade has witnessed significant progress on detecting objects in aerial images that are often distributed with large scale variations and arbitrary orientations. However most of existing methods rely on heuristically defined anchors with different scales, angles and aspect ratios and usually suffer from severe misalignment between anchor boxes and axis-aligned convolutional features, which leads to the common inconsistency between the classification score and localization accuracy. To address this issue, we propose a Single-shot Alignment Network (S$^2$A-Net) consisting of two modules: a Feature Alignment Module (FAM) and an Oriented Detection Module (ODM). The FAM can generate high-quality anchors with an Anchor Refinement Network and adaptively align the convolutional features according to the anchor boxes with a novel Alignment Convolution. The ODM first adopts active rotating filters to encode the orientation information and then produces orientation-sensitive and orientation-invariant features to alleviate the inconsistency between classification score and localization accuracy. Besides, we further explore the approach to detect objects in large-size images, which leads to a better trade-off between speed and accuracy. Extensive experiments demonstrate that our method can achieve state-of-the-art performance on two commonly used aerial objects datasets (i.e., DOTA and HRSC2016) while keeping high efficiency. The code is available at https://github.com/csuhan/s2anet.
Thermal energy agitates all matter and its competition with ordering tendencies is one of the most fundamental organizing principles in the physical world. Thus, it is natural to enquire if an effective temperature could result when external energy input enhances agitation. Potentially this could extend the insights of statistical thermodynamics to nonequilibrium systems, but despite proposals that the effective temperature concept may apply to synthetic active matter, biological motors, granular materials and turbulent fluids, its predictive value remains unclear. Here, combining computer simulations and imaging experiments, we design a two-component system of driven Janus colloids such that collisions produced by external energy sources play the role of temperature, and in this system we demonstrate quantitative agreement with hallmarks of statistical thermodynamics for binary phase behavior: the archetypal phase diagram with equilibrium critical exponents, Gaussian displacement distributions, fluctuation-dissipation relations, and capillarity. These quantitative analogies to equilibrium expectations, observed in this decidedly nonequilibrium system, constitute an existence proof from which to compare future theories of nonequilibrium, but limitations of this concept are also highlighted.
Anisotropic colloidal particles constitute an important class of building blocks for self-assembly directed by electrical fields. The aggregation of these building blocks is driven by induced dipole moments, which arise from an interplay between dielectric effects and the electric double layer. For particles that are anisotropic in shape, charge distribution, and dielectric properties, calculation of the electric double layer requires coupling of the ionic dynamics to a Poisson solver. We apply recently proposed methods to solve this problem for experimentally employed colloids in static and time-dependent electric fields. This allows us to predict the effects of field strength and frequency on the colloidal properties.
We present a comprehensive study of the optical properties of InAs/InP self-assembled quantum dots (QDs) using an empirical pseudopotential method and configuration interaction treatment of the many-particle effects. The results are compared to those of InAs/GaAs QDs. The main results are: (i) The alignment of emission lines of neutral exciton, charged exciton and biexciton in InAs/InP QDs is quite different from that in InAs/GaAs QDs. (ii) The hidden correlation in InAs/InP QDs is 0.7 - 0.9 meV, smaller than that in InAs/GaAs QDs. (iii) The radiative lifetimes of neutral exciton, charged exciton and biexciton in InAs/InP QDs are about twice longer than those in InAs/GaAs QDs. (v) The phase diagrams of few electrons and holes in InAs/InP QDs differ greatly from those in InAs/GaAs QDs. The filling orders of electrons and holes are shown to obey the Hunds rule and Aufbau principle, and therefore the photoluminescence spectra of highly charged excitons are very different from those of InAs/GaAs QDs.
