A5066572238- Advanced Image and Video Retrieval Techniques
- Image Retrieval and Classification Techniques
- Advanced Neural Network Applications
- Robotics and Sensor-Based Localization
- Multimodal Machine Learning Applications
- Domain Adaptation and Few-Shot Learning
- Video Surveillance and Tracking Methods
- Image and Signal Denoising Methods
- Remote-Sensing Image Classification
- Smart Grid Energy Management
- Fault Detection and Control Systems
- Microgrid Control and Optimization
- Augmented Reality Applications
- Advanced Algorithms and Applications
- 3D Surveying and Cultural Heritage
- Topic Modeling
- Microplastics and Plastic Pollution
- Industrial Vision Systems and Defect Detection
- Misinformation and Its Impacts
- Human Pose and Action Recognition
- Machine Learning and ELM
- Advanced Vision and Imaging
- Robot Manipulation and Learning
- biodegradable polymer synthesis and properties
- Advanced Image Fusion Techniques
Northwest University
2016-2025
Hebei University of Science and Technology
2023-2024
Gansu Desert Control Research Institute
2024
Xinjiang University
2024
Newcastle University
2024
Yangtze University
2024
Chinese Academy of Medical Sciences & Peking Union Medical College
2023
State Administration of Cultural Heritage
2023
Beijing Sport University
2023
Many recent 6D pose estimation methods exploited object 3D models to generate synthetic images for training because labels come free. However, due the domain shift of data distributions between real and images, network trained only on fails capture robust features in estimation. We propose solve this problem by making insensitive different domains, rather than taking more difficult route forcing be similar images. Inspired adaption methods, a Domain Adaptive Keypoints Detection Network...
The existing publicly available datasets with pixel-level labels contain limited categories, and it is difficult to generalize the real world containing thousands of categories. In this paper, we propose an approach generate object masks detailed structures/boundaries automatically enable semantic image segmentation targets in without manually labelling. A Guided Filter Network (GFN) first developed learn knowledge from existed dataset, such GFN then transfers learned initial coarse for...
We are concerned with using user-tagged images to learn proper hashing functions for image retrieval. The benefits two-fold: (1) we could obtain abundant training data deep models; (2) tagging possesses richer semantic information which help better characterize similarity relationships between images. However, suffers from noises, vagueness and incompleteness. Different previous unsupervised or supervised learning, propose a novel weakly-supervised framework consists of two stages:...
Generative models are widely used to produce synthetic images with annotations, alleviating the burden of image collection and annotation for training deep visual models. However, challenges such as limited diversity, noisy pseudo labels, domain gaps between real often undermine their effectiveness in downstream tasks. This paper introduces Iterative Self-Training Class-Aware Text-to-Image Synthesis (IST-CATS) framework, which addresses these by integrating a class-aware text-to-image...
We are concerned with using user-tagged images to learn proper hashing functions for image retrieval. The benefits two-fold: (1) we could obtain abundant training data deep models; (2) tagging possesses richer semantic information which help better characterize similarity relationships between images. However, suffers from noises, vagueness and incompleteness. Different previous unsupervised or supervised learning, propose a novel weakly-supervised framework consists of two stages:...
Online Shopping has become a part of our daily routine, but it still cannot offer intuitive experience as store shopping. Nowadays, most e-commerce Websites Question Answering (QA) system that allows users to consult other who have purchased the product. However, need wait patiently for others’ replies. In this paper, we investigate how provide quick response asker by plausible answer identification from product reviews. By analyzing similarity and discrepancy between explicit answers...
The state-of-art 6D object pose detection methods use convolutional neural networks to estimate objects' poses from RGB images. However, they require huge numbers of images with explicit 3D annotations such as poses, bounding boxes and keypoints, either obtained by manual labeling or inferred synthetic generated CAD models. Manual for a large number is laborious task, we usually do not have the corresponding models objects in real environment. In this paper, develop keypoint-based method...
Progressive remodeling of cardiac gene expression underlies decline in function, eventually leading to heart failure. However, the major determinants transcriptional network switching from normal failed hearts remain be determined.
The global energy landscape is undergoing a transformation towards decarbonization, sustainability, and cost-efficiency. In this transition, microgrid systems integrated with renewable sources (RES) storage (ESS) have emerged as crucial component. However, optimizing the operational control of such an system lacks holistic view multiple environmental, infrastructural economic considerations, not to mention need factor in uncertainties from both supply demand. This paper presents data-driven...
Multi-keyword query is widely supported in text search engines. However, an analogue image retrieval systems, multi-object query, rarely studied. Meanwhile, traditional object-based methods often involve multiple steps separately and need expensive location labeling for detecting objects. In this work, we propose a weakly-supervised Deep Multiple Instance Hashing (DMIH) framework retrieval. DMIH integrates object detection hashing learning on the basis of popular CNN model to build...
Multi-keyword query is widely supported in text search engines. However, an analogue image retrieval systems, multi-object query, rarely studied. Meanwhile, traditional object-based methods often involve multiple steps separately. In this work, we propose a weakly-supervised Deep Multiple Instance Hashing (DMIH) approach for retrieval. Our DMIH approach, which leverages popular CNN model to build the end-to-end relation between raw and binary hash codes of its objects, can support queries...
Person Re-identification (Re-ID) aims to accurately identify the same person in images from a large dataset captured by non-overlapping cameras. Recently, local-scale features of representation have been shown be effective improving performance Re-ID. However, most previous methods overlooked inherent and potential relationships among joint parts human skeleton structure. There are differences structure, such as bone length between joints, which can considered highly distinguishable feature...