A5114859293- 3D Shape Modeling and Analysis
- Computer Graphics and Visualization Techniques
- Advanced Vision and Imaging
- Robotics and Sensor-Based Localization
- 3D Surveying and Cultural Heritage
- Particle physics theoretical and experimental studies
- High-Energy Particle Collisions Research
- Advanced Image and Video Retrieval Techniques
- Particle Detector Development and Performance
- Parallel Computing and Optimization Techniques
- Robotic Path Planning Algorithms
- Robot Manipulation and Learning
- Distributed and Parallel Computing Systems
- Quantum Chromodynamics and Particle Interactions
- Image Processing and 3D Reconstruction
- Advanced Manufacturing and Logistics Optimization
- Optical measurement and interference techniques
- Interconnection Networks and Systems
- Image and Object Detection Techniques
- Advanced Neural Network Applications
- Advanced Numerical Analysis Techniques
- Biofuel production and bioconversion
- Dark Matter and Cosmic Phenomena
- Optimization and Packing Problems
- Advanced Data Storage Technologies
National University of Defense Technology
2014-2025
Beijing University of Chemical Technology
2023-2025
University of Shanghai for Science and Technology
2024
Fudan University
1990-2023
Collaborative Innovation Center of Advanced Microstructures
2023
University of Chinese Academy of Sciences
2021-2022
Chinese Academy of Sciences
2021-2022
Institute of High Energy Physics
2013-2022
Czech Technical University in Prague
2021
The University of Adelaide
2019-2021
Online reconstruction based on RGB-D sequences has thus far been restrained to relatively slow camera motions (<1m/s). Under very fast motion (e.g., 3m/s), the can easily crumble even for state-of-the-art methods. Fast brings two challenges depth fusion: 1) high nonlinearity of pose optimization due large inter-frame rotations and 2) lack reliably trackable features blur. We propose tackle difficulties fast-motion tracking in absence inertial measurements using random optimization,...
Deep learning approaches to 3D shape segmentation are typically formulated as a multi-class labeling problem. These models trained for fixed set of labels, which greatly limits their flexibility and adaptivity. We opt top-down recursive decomposition develop the first deep model hierarchical shapes, based on neural networks. Starting from full represented point cloud, our performs binary decomposition, where network at all nodes in hierarchy share weights. At each node, node classifier is...
We solve a challenging yet practically useful variant of 3D Bin Packing Problem (3D-BPP). In our problem, the agent has limited information about items to be packed into single bin, and an item must immediately after its arrival without buffering or readjusting. The item's placement also subjects constraints order dependence physical stability. formulate this online 3D-BPP as constrained Markov decision process (CMDP). To we propose effective easy-to-implement deep reinforcement learning...
A novel algorithm, called the lambda test, is presented for an efficient and accurate data dependence analysis of multidimensional array references. It extends numerical methods to allow all dimensions references be tested simultaneously. Hence, it combines efficiency accuracy both approaches. This algorithm has been implemented in Parafrase, a Fortran program parallelization restructurer developed at University Illinois Urbana-Champaign. Some experimental results are show its...
Online semantic 3D segmentation in company with real-time RGB-D reconstruction poses special challenges such as how to perform convolution directly over the progressively fused geometric data, and smartly fuse information from frame frame. We propose a novel fusion-aware point which operates on surface being reconstructed exploits effectively inter-frame correlation for high-quality feature learning. This is enabled by dedicated dynamic data structure that organizes online acquired cloud...
We introduce a contextual descriptor which aims to provide geometric description of the functionality 3D object in context given scene. Differently from previous works, we do not regard as an abstract label or represent it implicitly through agent. Our descriptor, called interaction ICON for short, explicitly represents geometry object-to-object interactions. approach analysis is based on key premise that should mainly be derived interactions between objects and isolation. Specifically,...
We introduce SCORES, a recursive neural network for shape composition. Our takes as input sets of parts from two or more source 3D shapes and rough initial placement the parts. It outputs an optimized part structure composed shape, leading to high-quality geometry construction. A unique feature our composition is that it not merely learning how connect goal produce coherent plausible despite large incompatibilities among The may significantly alter synthesize novel based on inputs, while...
Online reconstruction based on RGB-D sequences has thus far been restrained to relatively slow camera motions (<1m/s). Under very fast motion (e.g., 3m/s), the can easily crumble even for state-of-the-art methods. Fast brings two challenges depth fusion: 1) high nonlinearity of pose optimization due large inter-frame rotations and 2) lack reliably trackable features blur. We propose tackle difficulties fast-motion tracking in absence inertial measurements using random optimization,...
Template matching is a fundamental task in computer vision and has been studied for decades. It plays an essential role manufacturing industry estimating the poses of different parts, facilitating downstream tasks such as robotic grasping. Existing methods fail when template source images have modalities, cluttered backgrounds or weak textures. They also rarely consider geometric transformations via homographies, which commonly exist even planar industrial parts. To tackle challenges, we...
This research investigated the effects of hydrothermal depolymerization with Fe/Ni loaded C catalysts on anaerobic digestion (AD) performance corn stover (CS). CS was depolymerized at 140 °C for 20 min Fe/C or Ni/C catalysts, and then anaerobically digested. The results showed that biomethane yield Fe/C-600 increased by 36.6% compared to control. increase could be attributed effective (DC), indicated modified structures solid fraction enriched available components liquid fraction. SEM...
Camouflaged Object Segmentation (COS) remains a challenging problem due to the subtle visual differences between camouflaged objects and backgrounds. Owing exceedingly limited cues available from visible spectrum, previous RGB single-modality approaches often struggle achieve satisfactory results, prompting exploration of multimodal data enhance detection accuracy. In this work, we present UniCOS, novel framework that effectively leverages diverse modalities improve segmentation performance....
A processor self-scheduling scheme is proposed for general parallel nested loops in multiprocessor systems. In this scheme, programs are instrumented to allow processors schedule loop iterations among themselves dynamically at run time without involving the operating system. The has two levels. At low level, it uses simple fetch-and-op operations take advantage of regular structure innermost nests; high irregular outer (parallel or serial) and IF-THEN-ELSE constructs handled by using dynamic...
We introduce focal points for characterizing, comparing, and organizing collections of complex heterogeneous data apply the concepts algorithms developed to 3D indoor scenes. represent each scene by a graph its constituent objects define as representative substructures in collection. To organize collection, we cluster scenes based on set extracted points: are closely connected when viewed from perspective that cluster. The key concept representativity requires occur frequently they result...
We introduce AdaCoSeg, a deep neural network architecture for adaptive co-segmentation of set 3D shapes represented as point clouds. Differently from the familiar single-instance segmentation problem, is intrinsically contextual: how shape segmented can vary depending on it in. Hence, our features an learning module to produce consistent which adapts set. Specifically, given input unsegmented shapes, we first employ offline pre-trained part prior propose per-shape parts. Then iteratively and...
Abstract We propose a novel approach to robot‐operated active understanding of unknown indoor scenes, based on online RGBD reconstruction with semantic segmentation. In our method, the exploratory robot scanning is both driven by and targeting at recognition segmentation objects from scene. Our algorithm built top volumetric depth fusion framework performs real‐time voxel‐based labeling over reconstructed volume. The guided an estimated discrete viewing score field (VSF) parameterized 3D...
We study the problem of reconstructing 3D feature curves an object from a set calibrated multi-view images. To do so, we learn neural implicit field representing density distribution edges which refer to as Neural Edge Field (NEF). Inspired by NeRF [20], NEF is optimized with view-based rendering loss where 2D edge map rendered at given view and compared ground-truth extracted image that view. The rendering-based differentiable optimization fully exploits detection, without needing...
We present a learning-based approach to relight single image of Lambertian and low-frequency specular objects. Our method enables inserting objects from photographs into new scenes relighting them under the environment lighting, which is essential for AR applications. To object, we solve both inverse rendering re-rendering. resolve ill-posed rendering, propose weakly-supervised by low-rank constraint. facilitate training, contribute Relit, large-scale (750K images) dataset videos with...
In this paper, we give an overview of the Cedar multiprocessor and present recent performance results. These include some computational kernels Perfect Benchmarks. We also a methodology for judging parallel system apply to Cedar, Cray YMP-8, Thinking Machines CM-5.
Many approaches to shape comparison and recognition start by establishing a correspondence. We "turn the table" show that quality correspondences can be obtained performing many tasks. What is more, method we develop computes fine-grained, topology-varying part correspondence between two 3D shapes where core evaluation mechanism only recognizes globally. This made possible casting problem in deformation-driven framework relying on data-driven "deformation energy" which rates visual...
Surface reconstruction from raw point clouds has been studied for decades in the computer graphics community, which is highly demanded by modeling and rendering applications nowadays. Classic solutions, such as Poisson surface reconstruction, require normals extra input to perform reasonable results. Modern transformer-based methods can work without normals, while results are less fine-grained due limited encoding performance local fusion discrete points. We introduce a novel normalized...