A5100344999- Advanced Image and Video Retrieval Techniques
- Advanced Neural Network Applications
- Visual Attention and Saliency Detection
- Advanced Vision and Imaging
- Robotics and Sensor-Based Localization
- Face Recognition and Perception
- Video Surveillance and Tracking Methods
- Handwritten Text Recognition Techniques
- Consumer Perception and Purchasing Behavior
- Video Analysis and Summarization
- Remote Sensing in Agriculture
- Generative Adversarial Networks and Image Synthesis
- 3D Shape Modeling and Analysis
- Language, Metaphor, and Cognition
- Advanced Optical Imaging Technologies
- Gaze Tracking and Assistive Technology
- Image Retrieval and Classification Techniques
- Computer Graphics and Visualization Techniques
- Genetics, Bioinformatics, and Biomedical Research
- Second Language Acquisition and Learning
- Diverse Scientific Research Studies
- EFL/ESL Teaching and Learning
- Multimodal Machine Learning Applications
- Geochemistry and Geologic Mapping
- Image and Video Stabilization
Tsinghua University
2024
Northeastern University
2022-2023
Hebei Normal University of Science and Technology
2023
Technical University of Munich
2023
The University of Queensland
2018
Peking University Shenzhen Hospital
2017-2018
Peking University
2018
Dalian Jiaotong University
2018
The management of antibiotic resistance gene (ARG) contamination in the soil-plant system is a critical area research with significant implications for public health and environmental sustainability. Recently, engineered nanomaterials...
Existing lightweight salient object detection (SOD) methods aim to solve the problem of high computational costs that is prevalent with heavyweight methods. However, compared methods, accuracy greatly reduced while real-time performance not significantly improved. Therefore, we establish a trade off between cost and by improving network efficiency. We propose fast extremely end-to-end wavelet neural (ELWNet) for detection. ELWNet can achieve segmentation at approximately 70FPS (GPU), 19FPS...
The progress of LiDAR-based 3D object detection has significantly enhanced developments in autonomous driving and robotics. However, due to the limitations LiDAR sensors, shapes suffer from deterioration occluded distant areas, which creates a fundamental challenge perception. Existing methods estimate specific achieve remarkable performance. these rely on extensive computation memory, causing imbalances between accuracy real-time To tackle this challenge, we propose novel model named...
Existing point cloud based 3D detectors are designed for the particular scene, either indoor or outdoor ones. Because of substantial differences in object distribution and density within clouds collected from various environments, coupled with intricate nature metrics, there is still a lack unified network architecture that can accommodate diverse scenes. In this paper, we propose Uni3DETR, detector addresses detection same framework. Specifically, employ transformer point-voxel interaction...
Panoramic video provides an immersive experience by presenting a 360° spherical content. Due to the limitations of coding and storage technology, panoramic needs be projected onto two-dimensional plane for encoding. In this paper, we propose polar square projection scheme. We project area near poles sphere into two planes latitude circle on is squares plane, in addition, rest rectangle means equal projection. Experimental results show our proposed can obtain gain 11.63% BD-rate compared...
Deep learning based salient object detection methods have recently received significant attention. However, current still suffer from shortcomings such as informative background information being ignored which a problem for image saliency understanding. Additionally, it is also challenge to suppress the noisy features in network. By analyzing difference between high-level and low-level ResNet-50, we utilize Bilateral Feature Fusion (BFF) module deal with caused by ignoring information....
In the current state of 3D object detection research, severe scarcity annotated data, substantial disparities across different data modalities, and absence a unified architecture, have impeded progress towards goal universality. this paper, we propose \textbf{OV-Uni3DETR}, open-vocabulary detector via cycle-modality propagation. Compared with existing detectors, OV-Uni3DETR offers distinct advantages: 1) Open-vocabulary detection: During training, it leverages various accessible especially...
In this paper, we formally address universal object detection, which aims to detect every category in scene. The dependence on human annotations, the limited visual information, and novel categories open world severely restrict universality of detectors. We propose UniDetector, a detector that recognizes enormous world. critical points for UniDetector are: 1) it leverages images multiple sources heterogeneous label spaces training through image-text alignment, guarantees sufficient...
Image semantic segmentation technology is one of the core research contents in field computer vision. With improvement performance and continuous development deep learning technology, researchers have more enthusiasm to study actual effect image segmentation. The results allow computers a detailed accurate understanding images, wide range application needs fields autonomous driving, intelligent security, medical imaging, remote sensing etc. However, existing algorithms disadvantages easy...
A multi‐scale document image rectification method based on text‐features is presented. In contrast to most used text‐lines, the are more robust depict distortions of images with multi‐column layouts, multi‐type fonts or non‐textural objects. Then, problem formulated utilising a reverse strategy according text‐features. The experiments have demonstrated flexibility and high performance approach.
The characteristic similarity of clothing can describe the degree between and clothing, from this point view be excavated more collocation rules. In paper, garment algorithm is studied based on deep learning technology. On basis style expert opinions, rules are expanded through clothing. Convolution Neural Network (CNN) Long Short-Term Memory (LSTM) in used to find characteristics perspective features, compatibility features semantic A new combination found by calculation characteristics....