Search

This paper presents UniPortrait, an innovative human image personalization framework that unifies single- and multi-ID customization with high face fidelity, extensive facial editability, free-form input description, and diverse layout generation. UniPortrait consists of only two plug-and-play modules: an ID embedding module and an ID routing module. The ID embedding module extracts versatile editable facial features with a decoupling strategy for each ID and embeds them into the context space of diffusion models. The ID routing module then combines and distributes these embeddings adaptively to their respective regions within the synthesized image, achieving the customization of single and multiple IDs. With a carefully designed two-stage training scheme, UniPortrait achieves superior performance in both single- and multi-ID customization. Quantitative and qualitative experiments demonstrate the advantages of our method over existing approaches as well as its good scalability, e.g., the universal compatibility with existing generative control tools. The project page is at https://aigcdesigngroup.github.io/UniPortrait-Page/ ., Comment: Tech report; Project page: https://aigcdesigngroup.github.io/UniPortrait-Page/

In this paper we consider adaptive deep neural network approximation for stochastic dynamical systems. Based on the Liouville equation associated with the stochastic dynamical systems, a new temporal KRnet (tKRnet) is proposed to approximate the probability density functions (PDFs) of the state variables. The tKRnet gives an explicit density model for the solution of the Liouville equation, which alleviates the curse of dimensionality issue that limits the application of traditional grid based numerical methods. To efficiently train the tKRnet, an adaptive procedure is developed to generate collocation points for the corresponding residual loss function, where samples are generated iteratively using the approximate density function at each iteration. A temporal decomposition technique is also employed to improve the long-time integration. Theoretical analysis of our proposed method is provided, and numerical examples are presented to demonstrate its performance., Comment: 24 pages, 13 figures

Exploring ultrafast magnetization control in two-dimensional (2D) magnets through optically driven coherent phonons has been well-established. Yet, the microscopic interplay between spin dynamics and lattice degrees of freedom remains less explored. Employing real-time time-dependent density functional theory (rt-TDDFT) coupled with Ehrenfest dynamics, we systematically investigate laser-induced spin-nuclei dynamics with coherent phonon excitation in the 2D ferromagnet Fe3GeTe2. We found that selectively pre-exciting three typical coherent phonon modes results in up to a 53% additional spin moment loss in an out-of-plane A2 1g mode within ~50 fs. Coherent phonon control of spin dynamics is closely linked to laser pulse parameters. The underlying microscopic mechanism of this phenomenon is primarily governed by coherent phonon-induced asymmetric spin-resolved charge transfer following the disappearance of the laser pulse, thereby enabling effective control of the spin moment loss. Our findings offer a novel insight into the coupling of coherent phonons with spin systems in 2D limits on femtosecond timescales.

We introduce an integrated precise LiDAR, Inertial, and Visual (LIV) multimodal sensor fused mapping system that builds on the differentiable \pre{surface splatting }\now{Gaussians} to improve the mapping fidelity, quality, and structural accuracy. Notably, this is also a novel form of tightly coupled map for LiDAR-visual-inertial sensor fusion. This system leverages the complementary characteristics of LiDAR and visual data to capture the geometric structures of large-scale 3D scenes and restore their visual surface information with high fidelity. The initialization for the scene's surface Gaussians and the sensor's poses of each frame are obtained using a LiDAR-inertial system with the feature of size-adaptive voxels. Then, we optimized and refined the Gaussians using visual-derived photometric gradients to optimize their quality and density. Our method is compatible with various types of LiDAR, including solid-state and mechanical LiDAR, supporting both repetitive and non-repetitive scanning modes. Bolstering structure construction through LiDAR and facilitating real-time generation of photorealistic renderings across diverse LIV datasets. It showcases notable resilience and versatility in generating real-time photorealistic scenes potentially for digital twins and virtual reality, while also holding potential applicability in real-time SLAM and robotics domains. We release our software and hardware and self-collected datasets to benefit the community.

Despite spin (SAM) and orbital (OAM) angular momentum dynamics are well-studied in demagnetization processes, their components receive less focus. Here, we utilize the non-collinear spin version of real-time time-dependent density functional theory (rt-TDDFT) to unveil significant x and y components of SAM and OAM induced by circularly left and right polarized laser pulse in Ferromagnetic Fe, Co and Ni. Our results show that the magnitude of OAM is an order of magnitude larger than that of SAM, highlighting a stronger optical response from the orbital degrees of freedom of electrons compared to spin. Additionally, we observe a marked dependency of the oscillations of the x and y components on the helicity of the pulse. Intriguingly, left and right polarized pulses induce chirality in the precession of SAM and OAM, respectively, with clear associations with laser frequency and duration. Finally, we demonstrate that the time scale of OAM and SAM precession occurs even earlier than that of the demagnetization process and OISTR effect. Our investigation provides detailed insight into the dynamics of SAM and OAM of electrons during and shortly after a circularly polarized laser excitation.

Disentangling orbital (OAM) and spin (SAM) angular momenta in the ultrafast spin dynamics of two-dimensional (2D) ferromagnets on subfemtoseconds is a challenge in the field of ultrafast magnetism. Herein, we employed non-collinear spin version of real-time time-dependent density functional theory to investigate the orbital and spin dynamics of 2D ferromagnets Fe3GeTe2 (FGT) induced by circularly polarized light. Our results show the demagnetization of Fe sublattice in FGT is accompanied by helicity-dependent precession of OAM and SAM excited by circularly polarized lasers. We further identify that precession of OAM and SAM in FGT is faster than the demagnetization within a few femtoseconds. Remarkably, circularly polarized lasers can significantly induce a periodically transverse response of OAM and SAM on very ultrafast timescales of ~250 attoseconds. Our finding suggests a powerful new route for attosecond regimes of the angular momentum manipulation to coherently control helicity-dependent orbital and spin dynamics in 2D limits.

Recent attention in instance segmentation has focused on query-based models. Despite being non-maximum suppression (NMS)-free and end-to-end, the superiority of these models on high-accuracy real-time benchmarks has not been well demonstrated. In this paper, we show the strong potential of query-based models on efficient instance segmentation algorithm designs. We present FastInst, a simple, effective query-based framework for real-time instance segmentation. FastInst can execute at a real-time speed (i.e., 32.5 FPS) while yielding an AP of more than 40 (i.e., 40.5 AP) on COCO test-dev without bells and whistles. Specifically, FastInst follows the meta-architecture of recently introduced Mask2Former. Its key designs include instance activation-guided queries, dual-path update strategy, and ground truth mask-guided learning, which enable us to use lighter pixel decoders, fewer Transformer decoder layers, while achieving better performance. The experiments show that FastInst outperforms most state-of-the-art real-time counterparts, including strong fully convolutional baselines, in both speed and accuracy. Code can be found at https://github.com/junjiehe96/FastInst ., Comment: CVPR 2023

In this paper, we present a deep neural network based adaptive learning (DNN-AL) approach for switched systems. Currently, deep neural network based methods are actively developed for learning governing equations in unknown dynamic systems, but their efficiency can degenerate for switching systems, where structural changes exist at discrete time instants. In this new DNN-AL strategy, observed datasets are adaptively decomposed into subsets, such that no structural changes within each subset. During the adaptive procedures, DNNs are hierarchically constructed, and unknown switching time instants are gradually identified. Especially, network parameters at previous iteration steps are reused to initialize networks for the later iteration steps, which gives efficient training procedures for the DNNs. For the DNNs obtained through our DNN-AL, bounds of the prediction error are established. Numerical studies are conducted to demonstrate the efficiency of DNN-AL.

Manipulating spin in antiferromagnetic (AFM) materials has great potential in AFM opto-spintronics. Laser pulses can induce a transient ferromagnetic (FM) state in AFM metallic systems, but have never been proven in two-dimensional (2D) AFM semiconductors and related van der Waals (vdW) heterostructures. Here, using 2D vdW heterostructures of FM MnS2 and AFM MXenes as prototypes, we investigated optically induced interlayer spin transfer dynamics based on the real-time time-dependent density functional theory (rt-TDDFT). We observed that laser pulses induce significant spin injection and the interfacial atom-mediated spin transfer from MnS2 to Cr2CCl2. In particular, we first demonstrated the transient FM state in semiconducting AFM/FM heterostructures during photoexcited processes. Because the proximity magnetism breaks the magnetic symmetry of Cr2CCl2 in heterostructures. Our results provide the microscopic understanding for optically controlled interlayer spin dynamics in 2D magnetic heterostructures and open a new way to manipulate magnetic orders in ultrafast opto-spintronics.

$\alpha$-clustering structure is a significant topic in light nuclei. A Bayesian convolutional neural network (BCNN) is applied to classify initial non-clustered and clustered configurations, namely Woods-Saxon distribution and three-$\alpha$ triangular (four-$\alpha$ tetrahedral) structure for $^{12}$C ($^{16}$O), from heavy-ion collision events generated within a multi-phase transport (AMPT) model. Azimuthal angle and transverse momentum distributions of charged pions are taken as inputs to train the classifier. On multiple-event basis, the overall classification accuracy can reach $95\%$ for $^{12}$C/$^{16}$O + $^{197}$Au events at $\sqrt{S_{NN}} =$ 200 GeV. With proper constructions of samples, the predicted deviations on mixed samples with different proportions of both configurations could be within $5\%$. In addition, setting a simple confidence threshold can further improve the predictions on the mixed dataset. Our results indicate promising and extensive possibilities of application of machine-learning-based techniques to real data and some other problems in physics of heavy-ion collisions., Comment: 12 pages, 11 figures; Accepted by Physical Review C