A5100749600- Infrared Target Detection Methodologies
- Synthetic Aperture Radar (SAR) Applications and Techniques
- Advanced SAR Imaging Techniques
- Advanced Neural Network Applications
- Video Surveillance and Tracking Methods
- Remote Sensing and LiDAR Applications
- Infrastructure Maintenance and Monitoring
- Image Enhancement Techniques
- Automated Road and Building Extraction
- Computer Graphics and Visualization Techniques
- Advanced Vision and Imaging
- Robotics and Sensor-Based Localization
- Traffic Prediction and Management Techniques
- Advanced Measurement and Detection Methods
- Advanced Optical Sensing Technologies
- Speech Recognition and Synthesis
- Sarcoidosis and Beryllium Toxicity Research
- Prosthetics and Rehabilitation Robotics
- Social Robot Interaction and HRI
- Building Energy and Comfort Optimization
- Engineering and Test Systems
- AI in Service Interactions
- Cold Atom Physics and Bose-Einstein Condensates
- Vascular Malformations Diagnosis and Treatment
- Geophysical Methods and Applications
Zhejiang University of Technology
2024-2025
Shanghai Sixth People's Hospital
2024
The First People's Hospital of Wenling
2024
Southern Medical University
2024
Chinese Academy of Sciences
2021-2024
Tsinghua University
2023-2024
Huazhong University of Science and Technology
2024
Institute of Automation
2024
Peking University People's Hospital
2024
Peking University
2024
We propose Universal Document Processing (UDOP), a foundation AI model which unifies text, image, and layout modalities together with varied task formats, including document understanding generation. UDOP leverages the spatial correlation between textual content image to one uniform representation. With novel Vision-Text-Layout Transformer, pretraining multi-domain downstream tasks into prompt-based sequence generation scheme. is pretrained on both large-scale unlabeled corpora using...
Currently prevalent multi-modal 3D detection methods rely on dense detectors that usually use Bird's-Eye-View (BEV) feature maps. However, the cost of such BEV maps is quadratic to range, making it not scalable for long-range detection. Recently, LiDAR-only fully sparse architecture has been gaining attention its high efficiency in perception. In this paper, we study how develop a detector. Specifically, our proposed detector integrates well-studied 2D instance segmentation into LiDAR side,...
Accurate and efficient mapping of road networks is crucial for evaluating urban development, transportation accessibility, environmental impact. However, existing extraction methods utilizing remote sensing images suffer from limited generalization ability object occlusion, resulting in fragmented discontinuous segmentation. Consequently, these limitations impede the practical applicability multi-city diverse-scenario applications. To address challenges, we propose SWCARE, a method with...
We studied quantum depletion in a gaseous Bose-Einstein condensate. An optical lattice enhanced the atomic interactions and modified dispersion relation resulting strong depletion. The depleted fraction was directly observed as diffuse background time-of-flight images. Bogoliubov theory provided semi-quantitative description for our observations of fractions excess 50%.
Phase unwrapping is a challenging task for interferometry based techniques in the presence of noise. The majority existing phase are path-following methods, which explicitly or implicitly define an intelligent path and integrate difference along to mitigate effect erroneous pixels. In this paper, path-independent method proposed where unwrapped gradient determined from wrapped subsequently denoised by TV minimization method. Unlike map 2π jumps present, smooth slowly-varying at noise-free...
ABSTRACT Avians can change their wing shape in a fast and efficient manner resulting high manoeuvrabilities. The is composed of several bones dozens muscles, enabling it to have morphing abilities. However, various postures require complex control strategies, but no method or device has yet been designed. Therefore, an arm‐wing cooperative achieved for the first time current research. In method, similarities between human arm skeletal avian were figured out, then direct mapping motions was...
Infrared target detection is of great significance to various critical fields, however, it still a crucial problem robustly and quickly detect infrared small targets under complicated circumstances. In this letter, novel method based on resampling-guided-image model proposed. Initially, new image guided filtering introduced as an improved supplant for normally used patch-image (IPI) model. Our approximates the effect IPI in strengthening low-rankness background components provides favorable...
Accurate prediction results can provide an excellent reference value for the prevention of large-scale flight delays. Most currently available regression algorithms use a single time series network to extract features, with less consideration spatial dimensional information contained in data. Aiming at above problem, delay method based on Att-Conv-LSTM is proposed. First, order fully both temporal and dataset, long short-term memory used getting characteristics, convolutional neural adopted...
Objects in remote sensing images are typically characterized with various appearances composed of complex spatial and spectral information, making the stable feature representation objects a difficult task. To address above issues, we propose new object detection method by combining CNN (Convolutional Neural Network) Swin Transformer. Specifically, first SCCA (Spatial-Channel Coordinate Attention) module to highlight essential features an image fusing spatial, channel, location information....
At present, the emphasis of Inverse Synthetic Aperture Radar (ISAR) systems on characteristics high carrier frequency, wide bandwidth, multi-polarization capability, distribution, and networking has led to development progress ISAR imaging technology. The changes technology can be summarized into two aspects: fine improve image quality multidimensional enrich information. methods radar (such as echo pulse compression, system distortion correction, velocity motion compensation, range profile...
Text-to-image diffusion models have achieved tremendous success in the field of controllable image generation, while also coming along with issues privacy leakage and data copyrights. Membership inference arises these contexts as a potential auditing method for detecting unauthorized usage. While some efforts been made on models, they are not applicable to text-to-image due high computation overhead enhanced generalization capabilities. In this paper, we first identify conditional...
We propose a new image registration method based on joint respective window sampling (RWS) and modified motion compensation (MMC) in an interferometric inverse synthetic aperture radar (InISAR) imaging system using two antennas. The causation quantitative analysis of the offset between ISAR images different antennas along baseline are analyzed. In proposed method, RWS according to measured distance target antennas, compensates range direction. MMC is adopted eliminate Doppler Simulation...
The Alexa Prize program has empowered numerous university students to explore, experiment, and showcase their talents in building conversational agents through challenges like the SocialBot Grand Challenge TaskBot Challenge. As increasingly appear multimodal embodied contexts, it is important explore affordances of interaction augmented with computer vision physical embodiment. This paper describes SimBot Challenge, a new challenge which teams compete build robot assistants that complete...
Abstract Background Depression is a common emotional and psychiatric complication of traumatic brain injury (TBI) that has significant negative impacts on patient recovery. Despite the importance identifying treating depression in TBI patients, there currently no simple standardized system available for assessing likelihood post-TBI depression. In this study we are aim to explore clinical value peripheral blood inflammatory markers predicting mental disorders after TBI. Methods A total 67...
We present a method for enabling Reinforcement Learning of motor control policies complex skills such as dexterous manipulation. posit that key difficulty training is the exploring problem state space, accessible and useful regions this space form structure along manifolds original high-dimensional space. This work presents to enable support exploration with Sampling-based Planning. use generally applicable non-holonomic Rapidly-exploring Random Trees algorithm multiple methods resulting...
Automated segmentation of vestibular schwannoma (VS) using magnetic resonance imaging (MRI) can enhance clinical efficiency. Though many advanced methods exist for automated VS segmentation, the accuracy is hindered by ambivalent tumor borders and cystic regions in some patients. In addition, these provide results that do not indicate uncertainty, making their translation into workflows difficult due to potential errors. Providing a definitive result along with uncertainty or self-confidence...
Accurate prediction of metro traffic is crucial for optimizing scheduling and enhancing overall transport efficiency. Analyzing fine-grained comprehensive relations among stations effectively imperative Origin-Destination (OD) prediction. However, existing OD models either mix information from multiple pairs the station's perspective or exclusively focus on a subset pairs. These approaches may overlook pairs, leading to difficulties in predicting potential anomalous conditions. To address...
During activities of daily living (ADLs), the wrist is mainly engaged in positioning and directing hand. Researches have demonstrated that restoring mobility can significantly enhance manipulation ability, reduce body distortion caused by motion compensation, improve quality life for amputees. However, most activities, particularly delicate ones, place high demands on ability to maintain a certain rotation angle, also known as non-back-drivable which poses challenge design prosthetic wrists....
The objective of machine unlearning (MU) is to eliminate previously learned data from a model. However, it challenging strike balance between computation cost and performance when using existing MU techniques. Taking inspiration the influence label smoothing on model confidence differential privacy, we propose simple gradient-based approach that uses an inverse process smoothing. This work introduces UGradSL, simple, plug-and-play smoothed labels. We provide theoretical analyses...