- Advanced Vision and Imaging
- Advanced Image and Video Retrieval Techniques
- 3D Surveying and Cultural Heritage
- Domain Adaptation and Few-Shot Learning
- Robotics and Sensor-Based Localization
- Visual Attention and Saliency Detection
- Image Retrieval and Classification Techniques
- Industrial Vision Systems and Defect Detection
- Indoor and Outdoor Localization Technologies
- Multimodal Machine Learning Applications
- Speech and Audio Processing
- Natural Language Processing Techniques
- Video Surveillance and Tracking Methods
- Advanced Neural Network Applications
- Bone and Joint Diseases
- Frequency Control in Power Systems
- COVID-19 diagnosis using AI
- Retinoids in leukemia and cellular processes
- Image and Object Detection Techniques
- Machine Learning and Data Classification
- Integrated Energy Systems Optimization
- Bone Metabolism and Diseases
- Biomarkers in Disease Mechanisms
- Microbial Metabolism and Applications
- Magnetic Field Sensors Techniques
Huawei Technologies (China)
2025
Shandong University
2012-2023
Xi’an University of Posts and Telecommunications
2020-2023
NetEase (China)
2023
Shandong Provincial Hospital
2012-2015
University of Pittsburgh
2010
Visual cognition of the indoor environment can benefit from spatial layout estimation, which is to represent an scene with a 2-D box on monocular image. In this paper, we propose fully exploit edge and semantic information room image for estimation. More specifically, present encoder–decoder network shared encoder two separate decoders, are composed multiple deconvolution (transposed convolution) layers, jointly learn maps labels We combine these predictions in scoring function evaluate...
The goal of room layout estimation is to predict the three-dimensional box that represents spatial structure from a monocular image. In this paper, deconvolution network trained first edge map Compared previous fully convolutional networks, proposed has multilayer process can refine estimate layer by layer. also connected layers aggregate information every region throughout entire During generation process, an adaptive sampling strategy introduced based on obtained high-quality maps....
Considerable progress has recently been made in leveraging CLIP (Contrastive Language-Image Pre-Training) models for text-guided image manipulation. However, all existing works rely on additional generative to ensure the quality of results, because alone cannot provide enough guidance information fine-scale pixel-level changes. In this paper, we introduce CLIPVG, a manipulation framework using differentiable vector graphics, which is also first CLIP-based general that does not require any...
The inaccurate translation of numbers can lead to significant security issues, ranging from financial setbacks medical inaccuracies. While large language models (LLMs) have made advancements in machine translation, their capacity for translating has not been thoroughly explored. This study focuses on evaluating the reliability LLM-based systems when handling numerical data. In order systematically test capabilities currently open source LLMs, we constructed a dataset between Chinese and...
Room layout estimation seeks to infer the overall spatial configuration of indoor scenes using perspective or panoramic images. As is determined by dominant planes, this problem inherently requires reconstruction these planes. Some studies reconstruct planes from images learning pixel-level instance-level plane parameters. However, directly parameters has problems susceptibility occlusions and position dependency. In paper, we introduce Comprehensive depth map Planar (C2P) conversion, which...
What is the direct genetic contribution of polycystic ovary syndrome (PCOS) susceptibility single nucleotide polymorphisms (SNPs), identified by previous genome-wide association studies (GWAS) to definitive clinical features syndrome?Each PCOS feature had a specific association, and rs4385527 in chromosome 9 open reading frame 3 (C9orf3) conferred particular risk three defined this study, which suggested its fundamental role etiology PCOS.PCOS heterogeneous disorder characterized anovulation...
In this paper, we propose a normalized difference vector (NDV) for texture representation. Compared to local-binary-pattern-based descriptors, the proposed NDV takes full advantage of local difference, and size can be extended flexibly cover large region. We further employ bag-of-words model integrate descriptors into global feature representation an image. addition, two strategies are introduced achieve rotation invariance. test descriptor on benchmark datasets, such as AniTex, VehApp,...
Background: Diabetes-associated periodontitis (DPD) is an inflammatory and destructive disease of periodontal tissues in the diabetic population. The manifested as more severe destruction difficult to treat when compared with (PD). Eldecalcitol (ELD) a novel active vitamin D3 analog; however, little clinical evidence available on its role improving PD DPD, specific mechanisms remain unclear. In this study, we evaluated preventative effects ELD toward DPD explored underlying molecular...
Key technologies in empty bottle inspection systems are studied this paper to solve detecting error and poor adaptability problems. Those have two different approaches: the ones first group locate track mouth, bottom walls while other involve defect detecting. Such vision required perform with high accuracy under speed mechanical vibration working conditions. This study proposes distinctive algorithms for locating, tracking based on requirements images of walls. On premise satisfying...
In this paper, we propose a Wi-Fi positioning method based on Deep Learning (DL). To deal with the variant and unpredictable wireless signals, is casted in four-layer Neural Network (DNN) structure that capable of learning reliable features from large set noisy samples avoids need for hand-engineering. Also, to maintain temporal coherence, Hidden Markov Model (HMM) fine localizer introduced smooth result obtained immediate estimation DNN-based coarse localizer. The data required experiments...
The task of spatial layout estimation monocular image is to segment an RGB indoor scenes with semantic surface labels (i.e., ceiling, floor, front wall, left and right wall). Most recent methods have produce hypotheses based on the estimated edge map or labels, then rank hypotheses. In this paper, we present end-to-end framework that can directly output type keypoint coordinates (defined in LSUN challenge). proposed method takes advantage transfer learning via fake samples, i.e., plenty...
The task of 3D layout estimation in an indoor scene is to predict the holistic structural information from RGB image. It costly obtain ground truth layout, and this issue severely restricts learning based approaches. In paper, we present a novel weakly supervised framework that able learn effectively with 2D segmentation mask as supervision. We employ deep neural network plane parameters camera intrinsic Based on predicted instances, well corresponding depth map can be generated. key...
Image classification based on Deep learning usually requires a large amount of labeled data for model training. However, the labeling cost is expensive. The aim active to reduce manual by selecting lesser through query sampling strategy. In this paper, framework improved saliency guided augmentation, while enhancing generality algorithm and reducing cost. Firstly, augmentation map used expand sample data, which can retain most important information in data. addition, low-computation neural...
An efficient approach to detect and localize dining plates in images videos based on human vision is reported this paper. Rapid processing achieved by using random checkpoints edges only, simulating the knowledge target search biological visual systems. A new grouping algorithm convexity presented for a top-down segmentation of candidate objects. Our experiments demonstrate effectiveness approach.
Visual cognition of the indoor environment can benefit from spatial layout estimation, which is to represent an scene with a 2D box on monocular image. In this paper, we propose fully exploit edge and semantic information room image for estimation. More specifically, present encoder-decoder network shared encoder two separate decoders, are composed multiple deconvolution (transposed convolution) layers, jointly learn maps labels We combine these predictions in scoring function evaluate...