arXiv subjects

Xinjie Wang

Publications and source records attributed to Xinjie Wang.

29 records · Page 2Linked to original sources

arXiv2025

DIPO: Dual-State Images Controlled Articulated Object Generation Powered by Diverse Data

We present DIPO, a novel framework for the controllable generation of articulated 3D objects from a pair of images: one depicting the object in a resting state and the other in an articulated state. Compared to the single-image approach, our dual-image input imposes only a modest overhead for data collection, but at the same time provides important motion information, which is a reliable guide for predicting kinematic relationships between parts. Specifically, we propose a dual-image diffusion model that captures relationships between the image pair to generate part layouts and joint parameters. In addition, we introduce a Chain-of-Thought (CoT) based graph reasoner that explicitly infers part connectivity relationships. To further improve robustness and generalization on complex articulated objects, we develop a fully automated dataset expansion pipeline, name LEGO-Art, that enriches the diversity and complexity of PartNet-Mobility dataset. We propose PM-X, a large-scale dataset of complex articulated 3D objects, accompanied by rendered images, URDF annotations, and textual descriptions. Extensive experiments demonstrate that DIPO significantly outperforms existing baselines in both the resting state and the articulated state, while the proposed PM-X dataset further enhances generalization to diverse and structurally complex articulated objects. Our code and dataset will be released to the community upon publication.

cs.CV↗

arXiv2025

GaussTR: Foundation Model-Aligned Gaussian Transformer for Self-Supervised 3D Spatial Understanding

3D Semantic Occupancy Prediction is fundamental for spatial understanding, yet existing approaches face challenges in scalability and generalization due to their reliance on extensive labeled data and computationally intensive voxel-wise representations. In this paper, we introduce GaussTR, a novel Gaussian-based Transformer framework that unifies sparse 3D modeling with foundation model alignment through Gaussian representations to advance 3D spatial understanding. GaussTR predicts sparse sets of Gaussians in a feed-forward manner to represent 3D scenes. By splatting the Gaussians into 2D views and aligning the rendered features with foundation models, GaussTR facilitates self-supervised 3D representation learning and enables open-vocabulary semantic occupancy prediction without requiring explicit annotations. Empirical experiments on the Occ3D-nuScenes dataset demonstrate GaussTR's state-of-the-art zero-shot performance of 12.27 mIoU, along with a 40% reduction in training time. These results highlight the efficacy of GaussTR for scalable and holistic 3D spatial understanding, with promising implications in autonomous driving and embodied agents. The code is available at https://github.com/hustvl/GaussTR.

cs.CV↗

arXiv2024

Gaussian Object Carver: Object-Compositional Gaussian Splatting with surfaces completion

3D scene reconstruction is a foundational problem in computer vision. Despite recent advancements in Neural Implicit Representations (NIR), existing methods often lack editability and compositional flexibility, limiting their use in scenarios requiring high interactivity and object-level manipulation. In this paper, we introduce the Gaussian Object Carver (GOC), a novel, efficient, and scalable framework for object-compositional 3D scene reconstruction. GOC leverages 3D Gaussian Splatting (GS), enriched with monocular geometry priors and multi-view geometry regularization, to achieve high-quality and flexible reconstruction. Furthermore, we propose a zero-shot Object Surface Completion (OSC) model, which uses 3D priors from 3d object data to reconstruct unobserved surfaces, ensuring object completeness even in occluded areas. Experimental results demonstrate that GOC improves reconstruction efficiency and geometric fidelity. It holds promise for advancing the practical application of digital twins in embodied AI, AR/VR, and interactive simulation environments.

cs.CV↗

arXiv2023

Octree-based hierarchical sampling optimization for the volumetric super-resolution of scientific data

When introducing physics-constrained deep learning solutions to the volumetric super-resolution of scientific data, the training is challenging to converge and always time-consuming. We propose a new hierarchical sampling method based on octree to solve these difficulties. In our approach, scientific data is preprocessed before training, and a hierarchical octree-based data structure is built to guide sampling on the latent context grid. Each leaf node in the octree corresponds to an indivisible subblock of the volumetric data. The dimensions of the subblocks are different, making the number of sample points in each randomly cropped training data block to be adaptive. We reconstruct the octree at intervals according to loss distribution to perform the multi-stage training. With the Rayleigh-Bénard convection problem, we deploy our method to state-of-the-art models. We constructed adequate experiments to evaluate the training performance and model accuracy of our method. Experiments indicate that our sampling optimization improves the convergence performance of physics-constrained deep learning super-resolution solutions. Furthermore, the sample points and training time are significantly reduced with no drop in model accuracy. We also test our method in training tasks of other deep neural networks, and the results show our sampling optimization has extensive effectiveness and applicability. The code is publicly available at https://github.com/xinjiewang/octree-based_sampling.

physics.comp-ph↗

arXiv2020

Deep Reinforcement Learning Based Mobile Edge Computing for Intelligent Internet of Things

In this paper, we investigate mobile edge computing (MEC) networks for intelligent internet of things (IoT), where multiple users have some computational tasks assisted by multiple computational access points (CAPs). By offloading some tasks to the CAPs, the system performance can be improved through reducing the latency and energy consumption, which are the two important metrics of interest in the MEC networks. We devise the system by proposing the offloading strategy intelligently through the deep reinforcement learning algorithm. In this algorithm, Deep Q-Network is used to automatically learn the offloading decision in order to optimize the system performance, and a neural network (NN) is trained to predict the offloading action, where the training data is generated from the environmental system. Moreover, we employ the bandwidth allocation in order to optimize the wireless spectrum for the links between the users and CAPs, where several bandwidth allocation schemes are proposed. In further, we use the CAP selection in order to choose one best CAP to assist the computational tasks from the users. Simulation results are finally presented to show the effectiveness of the proposed reinforcement learning offloading strategy. In particular, the system cost of latency and energy consumption can be reduced significantly by the proposed deep reinforcement learning based algorithm.

eess.SP↗

arXiv2008

Ideal barriers to polarization reversal and domain-wall motion in strained ferroelectric thin films

The ideal intrinsic barriers to domain switching in c-phase PbTiO_3 (PTO), PbZrO_3 (PZO), and PbZr_{1-x}Ti_xO_3 (PZT) are investigated via first-principles computational methods. The effects of epitaxial strain on the atomic structure, ferroelectric response, barrier to coherent domain reversal, domain-wall energy, and barrier to domain-wall translation are studied. It is found that PTO has a larger polarization, but smaller energy barrier to domain reversal, than PZO. Consequentially the idealized coercive field is over two times smaller in PTO than PZO. The Ti--O bond length is more sensitive to strain than the other bonds in the crystals. This results in the polarization and domain-wall energy in PTO having greater sensitivity to strain than in PZO. Two ordered phases of PZT are considered, the rock-salt structure and a (100) PTO/PZO superlattice. In these simple structures we find that the ferroelectric properties do not obey Vergard's law, but instead can be approximated as an average over individual 5-atom unit cells.

cond-mat.mtrl-sci↗

arXiv2007

Fermi-surface calculation of the anomalous Hall conductivity

While the intrinsic anomalous Hall conductivity is normally written in terms of an integral of the electronic Berry curvature over the occupied portions of the Brillouin zone, Haldane has recently pointed out that this quantity (or more precisely, its ``non-quantized part'') may alternatively be expressed as a Fermi-surface property. Here we present an {\it ab-initio} approach for computing the anomalous Hall conductivity that takes advantage of this observation by converting the integral over the Fermi sea into a more efficient integral on the Fermi surface only. First, a conventional electronic-structure calculation is performed with spin-orbit interaction included. Maximally-localized Wannier functions are then constructed by a post-processing step in order to convert the {\it ab-initio} electronic structure around the Fermi level into a tight-binding-like form. Working in the Wannier representation, the Brillouin zone is sampled on a large number of equally spaced parallel slices oriented normal to the total magnetization. On each slice, we find the intersections of the Fermi-surface sheets with the slice by standard contour methods, organize these into a set of closed loops, and compute the Berry phases of the Bloch states as they are transported around these loops. The anomalous Hall conductivity is proportional to the sum of the Berry phases of all the loops on all the slices. Illustrative calculations are performed for Fe, Co and Ni.

cond-mat.mtrl-sci↗

arXiv2007

Spectral and Fermi surface properties from Wannier interpolation

We present an efficient first-principles approach for calculating Fermi surface averages and spectral properties of solids, and use it to compute the low-field Hall coefficient of several cubic metals and the magnetic circular dichroism of iron. The first step is to perform a conventional first-principles calculation and store the low-lying Bloch functions evaluated on a uniform grid of k-points in the Brillouin zone. We then map those states onto a set of maximally-localized Wannier functions, and evaluate the matrix elements of the Hamiltonian and the other needed operators between the Wannier orbitals, thus setting up an ``exact tight-binding model.'' In this compact representation the k-space quantities are evaluated inexpensively using a generalized Slater-Koster interpolation. Because of the strong localization of the Wannier orbitals in real space, the smoothness and accuracy of the k-space interpolation increases rapidly with the number of grid points originally used to construct the Wannier functions. This allows k-space integrals to be performed with ab-initio accuracy at low cost. In the Wannier representation, band gradients, effective masses, and other k-derivatives needed for transport and optical coefficients can be evaluated analytically, producing numerically stable results even at band crossings and near weak avoided crossings.

cond-mat.mtrl-sci↗

arXiv2006

First-principles perturbative computation of dielectric and Born charge tensors in finite electric fields

We present a perturbative treatment of the response properties of insulating crystals under a dc bias field, and use this to study the effects of such bias fields on the Born effective charge tensor and dielectric tensor of insulators. We start out by expanding a variational field-dependent total-energy functional with respect to the electric field within the framework of density-functional perturbation theory. The second-order term in the expansion of the total energy is then minimized with respect to the first-order wave functions, from which the Born effective charge tensor and dielectric tensor are easily computed. We demonstrate an implementation of the method and perform illustrative calculations for the III-V semiconductors AlAs and GaAs under finite bias field.

cond-mat.mtrl-sci↗

arXiv2006

Ab initio calculation of the anomalous Hall conductivity by Wannier interpolation

The intrinsic anomalous Hall effect in ferromagnets depends on subtle spin-orbit-induced effects in the electronic structure, and recent ab-initio studies found that it was necessary to sample the Brillouin zone at millions of k-points to converge the calculation. We present an efficient first-principles approach for computing the anomalous Hall conductivity. We start out by performing a conventional electronic-structure calculation including spin-orbit coupling on a uniform and relatively coarse k-point mesh. From the resulting Bloch states, maximally-localized Wannier functions are constructed which reproduce the ab-initio states up to the Fermi level. The Hamiltonian and position-operator matrix elements, needed to represent the energy bands and Berry curvatures, are then set up between the Wannier orbitals. This completes the first stage of the calculation, whereby the low-energy ab-initio problem is transformed into an effective tight-binding form. The second stage only involves Fourier transforms and unitary transformations of the small matrices set up in the first stage. With these inexpensive operations, the quantities of interest are interpolated onto a dense k-point mesh and used to evaluate the anomalous Hall conductivity as a Brillouin zone integral. The present scheme, which also avoids the cumbersome summation over all unoccupied states in the Kubo formula, is applied to bcc Fe, giving excellent agreement with conventional, less efficient first-principles calculations. Remarkably, we find that more than 99% of the effect can be recovered by keeping a set of terms depending only on the Hamiltonian matrix elements, not on matrix elements of the position operator.

cond-mat.mtrl-sci↗

arXiv2006

First-Principles Perturbative Computation of Phonon Properties of Insulators in Finite Electric Fields

We present a perturbative method for calculating phonon properties of an insulator in the presence of a finite electric field. The starting point is a variational total-energy functional with a field-coupling term that represents the effect of the electric field. This total-energy functional is expanded in small atomic displacements within the framework of density-functional perturbation theory. The linear response of field-polarized Bloch functions to atomic displacements is obtained by minimizing the second-order derivatives of the total-energy functional. In the general case of nonzero phonon wavevector, there is a subtle interplay between the couplings between neighboring k-points introduced by the presence of the electric field in the reference state, and further-neighbor k-point couplings determined by the wavevector of the phonon perturbation. As a result, terms arise in the perturbation expansion that take the form of four-sided loops in k-space. We implement the method in the {\tt ABINIT} code and perform illustrative calculations of the field-dependent phonon frequencies for III-V semiconductors.

cond-mat.mtrl-sci↗