A5090481419- Domain Adaptation and Few-Shot Learning
- Advanced Neural Network Applications
- Multimodal Machine Learning Applications
- Hydrocarbon exploration and reservoir analysis
- Coal Properties and Utilization
- Radiomics and Machine Learning in Medical Imaging
- COVID-19 diagnosis using AI
- Electromagnetic Simulation and Numerical Methods
- Model Reduction and Neural Networks
- Medical Image Segmentation Techniques
- Numerical methods in engineering
- Colorectal Cancer Screening and Detection
- AI in cancer detection
- Advanced Image and Video Retrieval Techniques
- Environmental remediation with nanomaterials
- Cancer-related molecular mechanisms research
- Brain Tumor Detection and Classification
- NMR spectroscopy and applications
- Image Processing and 3D Reconstruction
- Hydraulic Fracturing and Reservoir Analysis
- Advanced Numerical Methods in Computational Mathematics
- Catalytic Processes in Materials Science
- Photochemistry and Electron Transfer Studies
- Covalent Organic Framework Applications
- Radioactive element chemistry and processing
City University of Hong Kong
2021-2024
Nanchang Hangkong University
2023-2024
Shandong First Medical University
2024
Chinese University of Hong Kong
2024
Northeast Normal University
2021-2022
Dalian University of Technology
2022
China Academy of Space Technology
2022
University of Chinese Academy of Sciences
2018-2020
Huazhong University of Science and Technology
2018-2020
EURECOM
2015
Domain Adaptive Object Detection (DAOD) leverages a labeled domain to learn an object detector generalizing novel free of annotations. Recent advances align class-conditional distributions by narrowing down cross-domain prototypes (class centers). Though great success, they ignore the significant within-class variance and domain-mismatched semantics within training batch, leading sub-optimal adaptation. To overcome these challenges, we propose SemantIc-complete Graph MAtching (SIGMA)...
Abstract Polyps are well-known cancer precursors identified by colonoscopy. However, variability in their size, appearance, and location makes the detection of polyps challenging. Moreover, colonoscopy surveillance removal highly operator-dependent procedures occur a complex organ topology. There exists high missed rate incomplete colonic polyps. To assist clinical reduce rates, automated methods for detecting segmenting using machine learning have been achieved past years. major drawback...
Object detection and depth perception are key foundations of object tracking machine navigation, facilitating a thorough understanding the surrounding environment. Currently, autonomous vehicles employ complex bulky systems with high cost energy consumption to achieve demanding multimodal vision. An imperative exists for development compact reliable technology enhance cost-effectiveness efficiency driving systems. Meta-lens, novel flat optical device, has an artificial nanoantenna array...
U-Net has become a cornerstone in various visual applications such as image segmentation and diffusion probability models. While numerous innovative designs improvements have been introduced by incorporating transformers or MLPs, the networks are still limited to linearly modeling patterns well deficient interpretability. To address these challenges, our intuition is inspired impressive results of Kolmogorov-Arnold Networks (KANs) terms accuracy interpretability, which reshape neural network...
The domain gap severely limits the transferability and scalability of object detectors trained in a specific when applied to novel one. Most existing works bridge by minimizing discrepancy category space aligning category-agnostic global features. Though great success, these methods model with prototypes within batch, yielding biased estimation domain-level distribution. Besides, alignment leads disagreement class-specific distributions two domains, further causing inevitable classification...
Domain Adaptive Object Detection (DAOD) generalizes the object detector from an annotated domain to a label-free novel one. Recent works estimate prototypes (class centers) and minimize corresponding distances adapt cross-domain class conditional distribution. However, this prototype-based paradigm 1) fails capture variance with agnostic structural dependencies, 2) ignores domain-mismatched classes sub-optimal adaptation. To address these two challenges, we propose improved SemantIc-complete...
Domain Adaptive Object Detection (DAOD) models a joint distribution of images and labels from an annotated source domain learns domain-invariant transformation to estimate the target with given images. Existing methods assume that are completely clean, yet large-scale datasets often contain error-prone annotations due instance ambiguity, which may lead biased severely degrade performance adaptive detector de facto. In this paper, we represent first effort formulate noisy DAOD propose Noise...
Open Set Domain Adaptation (OSDA) transfers the model from a label-rich domain to label-free one containing novel-class samples. Existing OSDA works overlook abundant semantics hidden in source domain, leading biased learning and transfer. Although causality has been studied remove semantic-level bias, non-available samples result failure of existing causal solutions OSDA. To break through this barrier, we propose novel causality-driven solution with unexplored front-door adjustment theory,...
Point cloud segmentation is fundamental in many medical applications, such as aneurysm clipping and orthodontic planning. Recent methods mainly focus on designing powerful local feature extractors generally overlook the around boundaries between objects, which extremely harmful to clinical practice degenerates overall performance. To remedy this problem, we propose a GRAph-based Boundary-aware Network (GRAB-Net) with three paradigms, Graph-based Boundary-perception Module (GBM),...
The degradation pathway of the antibiotic metronidazole (MNZ) in wastewater was investigated computationally with a physical statistical method and quantum chemical approach. In both cases, density functional theory (DFT) at M06-2X/6-311+G(d,p) level used to calculate structures property parameters all molecules. On one hand, decay isolated MNZ molecule excited given excitation energy studied using molecular fragmentation (SMF) model. other reaction mechanisms oxidized by hydroxyl radicals...
Modern deep learning techniques on automatic multi-modal medical diagnosis rely massive expert annotations, which is time-consuming and prohibitive. Recent masked image modeling (MIM)-based pre-training methods have witnessed impressive advances for meaningful representations from unlabeled data transferring to downstream tasks. However, these focus natural images ignore the specific properties of data, yielding unsatisfying generalization performance diagnosis. In this paper, we aim...
Recent advancements in large generative models and real-time neural rendering using point-based techniques pave the way for a future of widespread visual data distribution through sharing synthesized 3D assets. However, while standardized methods embedding proprietary or copyright information, either overtly subtly, exist conventional content such as images videos, this issue remains unexplored emerging formats like Gaussian Splatting. We present GaussianStego, method steganographic...
The event-based Vision-Language Model (VLM) recently has made good progress for practical vision tasks. However, most of these works just utilize CLIP focusing on traditional perception tasks, which obstruct model understanding explicitly the sufficient semantics and context from event streams. To address deficiency, we propose EventVL, first generative MLLM (Multimodal Large Language Model) framework explicit semantic understanding. Specifically, to bridge data gap connecting different...
Domain Adaptive Object Detection (DAOD) transfers knowledge from a labeled source domain to an unannotated target under closed-set assumption. Universal DAOD (UniDAOD) extends handle open-set, partial-set, and adaptation. In this paper, we first unveil two issues: domain-private category alignment is crucial for global-level features, the probability heterogeneity of features across different levels. To address these issues, propose novel Dual Probabilistic Alignment (DPA) framework model as...
U-Net has become a cornerstone in various visual applications such as image segmentation and diffusion probability models. While numerous innovative designs improvements have been introduced by incorporating transformers or MLPs, the networks are still limited to linearly modeling patterns well deficient interpretability. To address these challenges, our intuition is inspired impressive results of Kolmogorov-Arnold Networks (KANs) terms accuracy interpretability, which reshape neural network...
Domain Adaptive Object Detection (DAOD) transfers an object detector to a novel domain free of labels. However, in the real world, besides encountering scenes, domains always contain novel-class objects de facto, which are ignored existing research. Thus, we formulate and study more practical setting, Open-set (AOOD), considering both scenes classes. Directly combing off-the-shelled cross-domain open-set approaches is sub-optimal since their low-order dependence, e.g., confidence score,...