A5064738683- Video Surveillance and Tracking Methods
- Human Pose and Action Recognition
- Advanced Image and Video Retrieval Techniques
- Multimodal Machine Learning Applications
- Advanced Neural Network Applications
- Anomaly Detection Techniques and Applications
- Video Analysis and Summarization
- Gait Recognition and Analysis
- Domain Adaptation and Few-Shot Learning
- Face recognition and analysis
- Image Enhancement Techniques
- Remote-Sensing Image Classification
- Image Retrieval and Classification Techniques
- Spectroscopy and Laser Applications
- Cancer-related molecular mechanisms research
- Fire Detection and Safety Systems
- Text and Document Classification Technologies
- Automated Road and Building Extraction
- Advanced Computational Techniques and Applications
- Autonomous Vehicle Technology and Safety
- Remote Sensing in Agriculture
- Recommender Systems and Techniques
- CO2 Reduction Techniques and Catalysts
- Imbalanced Data Classification Techniques
- Laser Design and Applications
Wuhan University of Technology
2016-2025
China Southern Power Grid (China)
2025
The First Affiliated Hospital, Sun Yat-sen University
2025
Sun Yat-sen University
2025
Sanya University
2025
Nanyang Technological University
2023-2024
University of Electronic Science and Technology of China
2015-2024
Peking University
2022-2023
Xidian University
2020-2022
Wuhan University
2020
Visible-infrared person re-identification (VI-ReID) is an emerging and challenging cross-modality image matching problem because of the explosive surveillance data in night-time applications. To handle large modality gap, various generative adversarial network models have been developed to eliminate variations based on a cross-modal generation framework. However, lack point-wise ground-truths makes it extremely learn such generator. address these problems, we correspondence between...
Entangled representation of clothing and identity (ID)-intrinsic clues are potentially concomitant in conventional person Re- IDentification (ReID). Nevertheless, eliminating the negative impact on ID remains challenging due to lack theory difficulty isolating exact implications. In this paper, a causality-based Auto-Intervention Model, referred as AIM <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup> Codes will publicly available at...
This paper studies the challenging person re-identification (Re-ID) task under cloth-changing scenario, where same identity (ID) suffers from uncertain cloth changes. To learn cloth- and ID-invariant features, it is crucial to collect abundant training data with varying clothes, which difficult in practice. alleviate reliance on rich collection, we reinforce feature learning process by designing powerful complementary augmentation strategies, including positive negative augmentation....
Few-shot learning is a tough topic to solve since obtaining large number of training samples in real applications challenging. It has attracted increasing attention recently. Meta-learning prominent way address this issue, intending adapt predictors as base-learners new tasks swiftly. However, key challenge meta-learning its lack expressive capacity, which stems from the difficulty extracting general information small samples. As result, generalizability meta-learners trained...
Video captioning aims to generate natural language sentences that describe the given video accurately. Existing methods obtain favorable generation by exploring richer visual representations in encode phase or improving decoding ability. However, long-tailed problem hinders these attempts at low-frequency tokens, which rarely occur but carry critical semantics, playing a vital role detailed generation. In this paper, we introduce novel Refined Semantic enhancement method towards Frequency...
Recent person Re-IDentification (ReID) systems have been challenged by changes in personnel clothing, leading to the study of Cloth-Changing ReID (CC-ReID). Commonly used techniques involve incorporating auxiliary information ( <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">e.g</i> ., body masks, gait, skeleton, and keypoints) accurately identify target pedestrian. However, effectiveness these methods heavily relies on quality comes at cost...
Selective electroreduction of CO2 to C1 feed gas provides an attractive avenue store intermittent renewable energy. However, most the CO2-to-CO catalysts are designed from perspective structural reconstruction, and it is challenging precisely design a meaningful confining microenvironment for active sites on support. Herein, we report local sulfur doping method tune electronic structure isolated asymmetric nickel–nitrogen–sulfur motif (Ni1-NSC). Our Ni1-NSC catalyst presents >99% faradaic...
The ever-growing environmental concern has yielded the electrocatalytic carbon dioxide reduction (eCO2R) center of research attention, as it offers a possible pathway to achieve net-zero emission and realize post-fossil-fuel society. production multicarbon (C2+) species with higher energy density is more desirable but unfortunately presents greater challenge. Other than catalyst itself, electrolyte also been demonstrated exhibit nontrivial impacts on eCO2R, systematic understanding effect...
In recent years, the convolutional neural network (CNN) has made remarkable achievements in semantic segmentation. The method of segmentation a desirable application prospect. Nowadays, methods mostly use an encoder-decoder architecture as way generating pixel by prediction. encoder is for extracting feature maps and decoder recovering map resolution. An improved on basis proposed. We can get better accuracy several hard classes reduce computational complexity significantly. This possible...
Campylobacter jejuni is a major foodborne pathogen worldwide. As it forms biofilms, can become persistent contaminant in the food and pharmaceutical industries. In this study, was demonstrated that C. could make more biofilm aerobic conditions than microaerobic conditions, only 13.9% entered coccus (a VBNC state) under conditions; however, rate increased to 95.5% conditions. form mixed culture with Escherichia coli or Pseudomonas aeruginosa pure culture. Scanning electron microscope results...
Scene understanding in adverse weather conditions (e.g. rainy and foggy days) has drawn increasing attention, arising some specific benchmarks algorithms. However, scene segmentation under is still challenging under-explored due to the following limitations on datasets methods: 1) Manually synthetic samples with empirically settings human subjective assumptions; 2) Limited conditions, including rain patterns, intensity, degradation factors; 3) Separated training manners for image deraining...
Single image deraining aims to remove rain perturbation while restoring the clean background scene from a image. However, existing methods tend produce blurry and over-smooth outputs, lacking some textural details. Wavelet transform can depict contextual information of an at different levels, showing impressive capability learning structural in images avoid artifacts, thus has been recently explored consider inherent overlap both pixel domain frequency embedding space. wavelet-based ignore...
Non-autoregressive video captioning methods generate visual words in parallel but often overlook semantic correlations among them, especially regarding verbs, leading to lower caption quality. To address this, we integrate action information of highlighted objects enhance connections words. Our proposed Action-aware Language Skeleton Optimization Network (ALSO-Net) tackles the challenge extracting across frames, improving understanding complex context-dependent actions and reducing sentence...
The manufacturing and processing of metal materials have greatly promoted the development chemical industry. However, resulting corrosion phenomenon also comes one after another. Metal not only does great harm to production equipment, but poses a burden on environment. Coating hydrophobic compound surface as coating for anti-corrosion is major way protect in recent years. After adsorbed surface, protective film formed layer metal, so delay metal. substances separate from corrosive medium by...
In class-agnostic object counting, the goal is to estimate total number of instances in an image without distinguishing between specific categories. Existing methods often predict this count considering class-specific outputs, leading inaccuracies when such outputs are required. These stem from two key challenges: 1) prevalence single-category images datasets, which leads models generalize categories as representative all objects, and 2) use mean squared error loss during training, applies...
The depth and width of the network have been investigated to influence performance image classification during resent research. Wide residual networks (WRNs) proved that can be improved by networks. With consideration significance, expanding is increase number channels. However, not all channels are needed. Meanwhile, much channel information will lost while exploiting global average pooling at end WRNs for representations because mean value only related first order information. two...
Multi-view action recognition aims to identify categories from given clues. Existing studies ignore the negative influences of fuzzy views between view and in disentangling, commonly arising mistaken results. To this end, we regard observed image as composition components, give full play advantages multiple via adaptive cooperative representation among these two forming a Dual-Recommendation Disentanglement Network (DRDN) for multi-view recognition. Specifically, 1) For action, leverage...