A5100452405- Advanced Vision and Imaging
- Neural Networks Stability and Synchronization
- Neural Networks and Applications
- Computer Graphics and Visualization Techniques
- stochastic dynamics and bifurcation
- Higher Education and Teaching Methods
- Nonlinear Dynamics and Pattern Formation
- Image Enhancement Techniques
- Advanced Computational Techniques and Applications
- 3D Shape Modeling and Analysis
- Distributed and Parallel Computing Systems
- Stability and Controllability of Differential Equations
- Service-Oriented Architecture and Web Services
- Advanced Decision-Making Techniques
- Face recognition and analysis
- Video Surveillance and Tracking Methods
- Web and Library Services
- Advanced Mathematical Modeling in Engineering
- Physical Unclonable Functions (PUFs) and Hardware Security
- Cloud Computing and Resource Management
- Digital Media and Visual Art
- Cooperative Communication and Network Coding
- Optical measurement and interference techniques
- Technology and Security Systems
- Model Reduction and Neural Networks
Chongqing University of Posts and Telecommunications
2017-2024
Tsinghua University
2021-2024
Ocean University of China
2009-2024
Louisiana State University
2023-2024
Purple Mountain Laboratories
2023
Shandong University
2007-2022
Agency for Science, Technology and Research
2021
Shanghai Institute of Microsystem and Information Technology
2019-2020
ShanghaiTech University
2016-2020
University of Chinese Academy of Sciences
2020
Visually exploring in a real-world 4D spatiotemporal space freely VR has been long-term quest. The task is especially appealing when only few or even single RGB cameras are used for capturing the dynamic scene. To this end, we present an efficient framework capable of fast reconstruction, compact modeling, and streamable rendering. First, propose to decompose according temporal characteristics. Points associated with probabilities belonging three categories: static, deforming, new areas....
We present a novel type of neural fields that uses general radial bases for signal representation. State-of-the-art typically rely on grid-based representations storing local features and N-dimensional linear kernels interpolating at continuous query points. The spatial positions their are fixed grid nodes cannot well adapt to target signals. Our method instead builds upon with flexible kernel position shape, which have higher adaptivity can more closely fit To further improve the...
We present a novel Relightable Neural Renderer (RNR) for simultaneous view synthesis and relighting using multi-view image inputs. Existing neural rendering (NR) does not explicitly model the physical process hence has limited capabilities on relighting. RNR instead models formation in terms of environment lighting, object intrinsic attributes, light transport function (LTF), each corresponding to learnable component. In particular, incorporation physically based only enables but also...
Despite the impressive performance obtained by recent single-image hand modeling techniques, they lack capability to capture sufficient details of 3D mesh. This deficiency greatly limits their applications when high-fidelity is required, e.g., personalized modeling. To address this problem, we design a frequency split network generate meshes using different bands in coarse-to-fine manner. high-frequency details, transform mesh into domain, and proposed novel decomposition loss supervise each...
<p style='text-indent:20px;'>This paper is concerned with the existence and uniqueness of invariant measures for infinite-dimensional stochastic delay lattice systems defined on entire integer set. For Lipschitz drift diffusion terms, we prove by showing tightness a family probability distributions solutions in space continuous functions from finite interval to an space, based idea uniform tail-estimates, technique diadic division Arzela-Ascoli theorem. We also show when coefficients...
We present a novel 3D face reconstruction technique that leverages sparse photometric stereo (PS) and latest advances on registration / modeling from single image. observe morphable faces approach [21] provides reasonable geometry proxy for light position calibration. Specifically, we develop robust optimization can calibrate per-pixel lighting direction illumination at very high precision without assuming uniform surface albedos. Next, apply semantic segmentation input images the to refine...
Abstract In this paper, existence of invariant measure is mainly investigated for a fractional stochastic delay reaction–diffusion equation defined on unbounded domains. We first establish the mean-square uniform smallness tails solutions in order to overcome non-compactness standard Sobolev embeddings then show weak compactness family probability distributions by combining Ascoli–Arzelà theorem, tail-estimates as well technique dyadic division.
This paper deals with invariant measures of fractional stochastic reaction–diffusion equations on unbounded domains locally Lipschitz continuous drift and diffusion terms. We first prove the existence regularity measures, then show tightness set all equation when noise intensity varies in a bounded interval. also that every limit perturbed systems is an measure corresponding limiting system. Under further conditions, we establish ergodicity exponentially mixing property measures.
In this paper, we address the problem of simultaneous relighting and novel view synthesis a complex scene from multi-view images with limited number light sources. We propose an analysis-synthesis approach called Relit-NeuLF. Following recent neural 4D field network (NeuLF)[22], Relit-NeuLF first leverages two-plane representation to parameterize each ray in coordinate system, enabling efficient learning inference. Then, recover spatially-varying bidirectional reflectance distribution...
Helmholtz stereopsis (HS) exploits the reciprocity principle of light propagation ( <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">i.e.</i> , reciprocity) for 3D reconstruction surfaces with arbitrary reflectance. In this paper, we present polarimetric (polar-HS), which extends classical HS by considering polarization state in reciprocal paths. With additional phase information from polarization, polar-HS requires only one image pair. We...
Reinforcement Learning from Human Feedback (RLHF) has proven to be a strong method align Pretrained Large Language Models (LLMs) with human preferences. But training models RLHF is computationally expensive, and an overall complex process. In this work, we study where the underlying are trained using parameter efficient of Low-Rank Adaptation (LoRA) introduced by Hu et al. [2021]. We investigate setup "Parameter Efficient Learning" (PERL), in which perform reward model reinforcement learning...
Recovering 3D geometry of underwater scenes is challenging because the non-linear refraction light at water-air interface, which caused by camera housing. We present a field-based approach that leverages properties angular samples for high-quality reconstruction from single viewpoint. Specifically, we re-sample field image to patches. As exhibit weak view-dependent specularity, an patch tends have uniform intensity when sampled correct depth. thus impose this uniformity as constraint depth...
Particle Imaging Velocimetry (PIV) is a classical method that estimates fluid flow by analyzing the motion of injected particles. To reconstruct and track swirling particles difficult computer vision problem, as are dense in volume have similar appearances. Further, tracking large number particularly challenging due to heavy occlusion. Here we present low-cost PIV solution uses compact lenslet-based light field cameras imaging device. We develop novel optimization algorithms for particle 3D...
With the trend of outsourcing fabrication, split manufacturing is regarded as a promising way to both provide high-end nodes in untrusted external foundries and protect design from potential attackers. However, this work, we show that not inherently secure. A hardware trojan attacker can still discover necessary information with simulated annealing based attack approach at placement level. We further propose defense by moving insecure gates away their easily-attacked candidate locations....
A key challenge in software development process is to detect errors earlier phases of the life cycle. For this purpose, verification UML diagrams plays an important role detecting flaws at analysis and design phase. To enhance correctness one most popular diagrams: sequence diagram (SD), model checking propositional projection temporal logic (PPTL) adopted. With method, event deterministic finite automata are used describe formal models SD, PPTL selected a desired property. Experimental...