A5100351699- Multimodal Machine Learning Applications
- Risk and Safety Analysis
- Human Pose and Action Recognition
- Advanced Neural Network Applications
- Domain Adaptation and Few-Shot Learning
- Adversarial Robustness in Machine Learning
- Fault Detection and Control Systems
- Robot Manipulation and Learning
- Nuclear Engineering Thermal-Hydraulics
- Video Surveillance and Tracking Methods
- Anomaly Detection Techniques and Applications
- Robotic Path Planning Algorithms
- Natural Language Processing Techniques
- Nuclear reactor physics and engineering
- Robotics and Sensor-Based Localization
- Hand Gesture Recognition Systems
- Medical Image Segmentation Techniques
- Probabilistic and Robust Engineering Design
- Reliability and Maintenance Optimization
- Topic Modeling
- Computer Graphics and Visualization Techniques
- 3D Shape Modeling and Analysis
- COVID-19 diagnosis using AI
- Engineering Diagnostics and Reliability
- Speech and Audio Processing
Peking University
2022-2025
Harbin Engineering University
2009-2024
The University of Texas MD Anderson Cancer Center
2024
Zhengzhou University of Light Industry
2024
The University of Texas Health Science Center at Houston
2024
King University
2023
Chinese Academy of Sciences
2020-2023
Institute of Information Engineering
2023
Zhejiang University of Science and Technology
2023
Northeast Electric Power University
2023
Building home assistant robots has long been a goal for vision and robotics researchers. To achieve this task, simulated environment with physically realistic simulation, sufficient articulated objects, transferability to the real robot is indispensable. Existing environments these requirements simulation different levels of simplification focus. We take one step further in constructing an that supports household tasks training learning algorithm. Our work, SAPIEN, physics-rich hosts...
We propose a novel attention-based framework for 3D human pose estimation from monocular video. Despite the general success of end-to-end deep learning paradigms, our approach is based on two key observations: (1) temporal incoherence and jitter are often yielded single frame prediction; (2) error rate can be remarkably reduced by increasing receptive field in Therefore, we design an attentional mechanism to adaptively identify significant frames tensor outputs each neural net layer, leading...
This paper addresses the task of category-level pose estimation for articulated objects from a single depth image. We present novel approach that correctly accommodates object instances previously unseen during training. introduce Articulation-aware Normalized Coordinate Space Hierarchy (ANCSH) - canonical representation different in given category. As key to achieve intra-category generalization, constructs space as well set part spaces. The normalizes orientation, scales and articulations...
3D object detection is an important yet demanding task that heavily relies on difficult to obtain annotations. To reduce the required amount of supervision, we propose 3DIoUMatch, a novel semi-supervised method for applicable both indoor and outdoor scenes. We leverage teacher-student mutual learning framework propagate information from labeled unlabeled train set in form pseudo-labels. However, due high complexity, observe pseudo-labels suffer significant noise are thus not directly usable....
Fine-grained capture of 3D Human-Object Interactions (HOIs) enhances human activity comprehension and supports various downstream visual tasks. However, previous models often assume that humans interact with rigid objects using only a few body parts, constraining their applicability. In this paper, we address the intricate challenge Full-Body Articulated Interaction (f-AHOI), where complete bodies articulated having interconnected movable joints. We introduce CHAIRS, an extensive...
We propose a novel, object-agnostic method for learning universal policy dexterous object grasping from realistic point cloud observations and proprioceptive information under table-top setting, namely UniDexGrasp++. To address the challenge of vision-based across thousands instances, we Geometry-aware Curriculum Learning (GeoCurriculum) iterative Generalist-Specialist (GiGSL) which leverage geometry feature task significantly improve generalizability. With our proposed techniques, final...
Domain adaptive semantic segmentation methods commonly utilize stage-wise training, consisting of a warm-up and self-training stage. However, this popular approach still faces several challenges in each stage: for warm-up, the widely adopted adversarial training often results limited performance gain, due to blind feature alignment; self-training, finding proper categorical thresholds is very tricky. To alleviate these issues, we first propose replace stage by novel symmetric knowledge...
The task of Vision-Language Navigation (VLN) is for an embodied agent to reach the global goal according instruction. Essentially, during navigation, a series sub-goals need be adaptively set and achieved, which naturally hierarchical navigation process. However, previous methods leverage single-step planning scheme, i.e., directly performing action at each step, unsuitable such In this paper, we propose Adaptive Zone-aware Hierarchical Planner (AZHP) explicitly divides process into two...
Spatial intelligence is a critical component of embodied AI, promoting robots to understand and interact with their environments. While recent advances have enhanced the ability VLMs perceive object locations positional relationships, they still lack capability precisely orientations-a key requirement for tasks involving fine-grained manipulations. Addressing this limitation not only requires geometric reasoning but also an expressive intuitive way represent orientation. In context, we...
Fruit-picking robot is an important form of intelligent agricultural machinery and has great impact in smart agriculture. Motion planning for its multi-Degree Freedoms (DOFs) manipulator significant to picking efficiency fruit quality. However, it difficult plan appropriate trajectory such a complex system. The purpose this article generate smooth trajectories fruit-picking manipulators using shortcuts that are constrained velocity, acceleration jerk. proposed algorithm smooths the motion as...
We study the problem of multi-robot active mapping, which aims for complete scene map construction in minimum time steps. The key to this lies goal position estimation enable more efficient robot movements. Previous approaches either choose frontier as via a myopic solution that hinders efficiency, or maximize long-term value reinforcement learning directly regress position, but does not guarantee construction. In paper, we propose novel algorithm, namely NeuralCoMapping, takes advantage...
Aiming at the problems of poor self-adaptive ability in traditional feature extraction methods and weak generalization single classifier under big data, an internal parameter-optimized Deep Belief Network (DBN) method based on grasshopper optimization algorithm (GOA) is proposed. First, minimum Root Mean Square Error (RMSE) network training taken as fitness function, which GOA used to search for optimal parameter combination DBN. After that learning rate number batch DBN have great influence...
Abstract The nuclear power plant (NPP) plays a crucial role in providing clean energy, significantly contributing to mitigating global warming. However, this advantage is accompanied by public concerns regarding potential accidents. imperative for robust safety system has led the exploration of artificial intelligence (AI) integration NPP minimize human error and enhance efficiency during reactor This research conducts comparative study between two deep learning (DL) models classify five...
Face recognition service providers protect face privacy by extracting compact and discriminative facial features (representations) from images, storing the for real-time recognition. However, such can still be exploited to recover appearance of original building a reconstruction network. Although sev-eral privacy-preserving methods have been proposed, enhancement offace protection is at expense accuracy degradation. In this paper, we propose an adver-sarial features-based (AdvFace) approach...
Most Siamese-based trackers adopt correlation operation to perform similarity matching on feature fusion of template branch and search branch. However, the directly uses slide window area feature, it is difficult distinguish target background information when encountering similar interference clutter, which can easily lead tracking failure. In this paper, a mixed attention Siamese Network (MASNet) proposed, neatly integrates modelling global mutual into framework. Specifically, through...
Best-Estimate plus uncertainty analysis (BEPU) method has been widely used to analyze various transient accident of PWR. However, the traditional BEPU some limitations: 1) The input parameters are not clearly defined, resulting in inaccurate conclusions sensitivity analysis. 2) Uncertainty quantification and usually share same set samples, but they have different requirements for sample size. In this work, an improved is proposed, which can alleviate above defects. possesses following two...
Most of Path Planning Algorithms are applied independently when solving real-time path planning problems. However, utilizing multiple algorithms simultaneously can avoid the disadvantages their each as much possible. This paper divides mission into two grades employing different algorithms. A* algorithm is used to plan a avoiding threats which divided several great circle cruise parts and penetration parts. A kind Terrain Following/Terrain Avoidance (TF/TA) create track based on planned by...
Machines that can predict the effect of physical interactions on dynamics previously unseen object instances are important for creating better robots and interactive virtual worlds. In this work, we focus predicting 3D objects a plane have just been subjected to an impulsive force. particular, changes in state-3D position, rotation, velocities, stability. Different from previous our approach generalize predictions shapes initial conditions were during training. Our method takes object's...