A5064702771- Advanced Image and Video Retrieval Techniques
- Multimodal Machine Learning Applications
- Image Retrieval and Classification Techniques
- Mercury impact and mitigation studies
- Domain Adaptation and Few-Shot Learning
- Visual Attention and Saliency Detection
- Natural Language Processing Techniques
- Environmental Toxicology and Ecotoxicology
- Generative Adversarial Networks and Image Synthesis
- Advanced Vision and Imaging
- Luminescence and Fluorescent Materials
- Computer Graphics and Visualization Techniques
- Topic Modeling
- Image and Video Quality Assessment
- Advanced Text Analysis Techniques
- Corporate Taxation and Avoidance
- COVID-19 impact on air quality
- Essential Oils and Antimicrobial Activity
- Metaheuristic Optimization Algorithms Research
- Handwritten Text Recognition Techniques
- Moringa oleifera research and applications
- Space exploration and regulation
- Ecology and Conservation Studies
- Heavy Metal Exposure and Toxicity
- User Authentication and Security Systems
Southwest University
2022-2025
University of International Business and Economics
2022-2024
Ministry of Education of the People's Republic of China
2023
University of Toronto
2023
Fudan University
2016-2021
University of Oxford
2015
The 5th generation wireless system 5G will support Internet of Things IoT by increasing the interconnectivity electronic devices to a variety new and promising networked applications such as home future, environmental monitoring networks, infrastructure management systems. potential benefits are profound they diverse. However, come with some significant challenges. Not least these is that increased integral an network increases its vulnerability malevolent attacks. There still no proven...
Previous researches of sketches often considered in pixel format and leveraged CNN based models the sketch understanding. Fundamentally, a is stored as sequence data points, vector representation, rather than photo-realistic image pixels. SketchRNN studied generative neural representation for by Long Short Term Memory networks (LSTM). Unfortunately, learned primarily generation tasks, other tasks recognition retrieval sketches. To this end inspired recent BERT model, we present model...
Sketch has been employed as an effective communication tool to express the abstract and intuitive meaning of object. While content-based sketch recognition studied for several decades, instance-level Based Image Retrieval (iSBIR) task attracted significant research attention recently. In many previous iSBIR works -- TripletSN, DSSA, edge maps were intermediate representations in bridging cross-domain discrepancy between photos sketches. However, it is nontrivial efficiently train effectively...
Image aesthetic assessment is an important issue in multimedia, but most existing studies employ supervised learning methods that rely on large-scale annotated data. However, scoring annotations are difficult to obtain large quantities. Therefore, this paper explores zero-shot image assessment. We predict scores by introducing knowledge of different attributes (e.g., Focus). First, we use prompt tuning a unique for each attribute as external knowledge. Second, leverage relations considering...
LLMs acquire a wide range of abilities during pre-training, but aligning under Reinforcement Learning with Human Feedback (RLHF) can lead to forgetting, which is also known as the alignment tax. To empirically verify this hypothesis, we conducted experiments existing RLHF algorithms using OpenLLaMA-3B, revealed pronounced tax in NLP tasks. On other hand, despite various techniques mitigate they are often at odds performance, leading trade-off between reward maximization and forgetting...
This paper proposes a novel approach for Sketch-Based Image Retrieval (SBIR), which the key is to bridge gap between sketches and photos in terms of data representation. Inspired by channel-wise attention explored recent years, we present Domain-Aware Squeeze-and-Excitation (DASE) network, seamlessly incorporates prior knowledge sample sketch or photo into SE module make capable emphasizing appropriate channels according domain signal. Accordingly, proposed network can switch its mode...
The process of learning good representations for machine tasks can be very computationally expensive. Typically, the model learned on training set is leveraged to infer labels testing data. Interestingly, this and inference paradigm, however, quite different from typical scheme human biological visual systems. Essentially, neuroscience studies have shown that right hemisphere brain predominantly makes a fast processing low-frequency spatial signals, while left more focuses analyzing...
Verbs are important in semantic understanding of natural language. Traditional verb representations, such as FrameNet, PropBank, VerbNet, focus on verbs' roles. These roles too coarse to represent semantics. In this paper, we introduce patterns semantics, that each pattern corresponds a single the verb. First analyze principles for patterns: generality and specificity. Then propose nonparametric model based description length. Experimental results prove high effectiveness patterns. We...
Previous researches of sketches often considered in pixel format and leveraged CNN based models the sketch understanding. Fundamentally, a is stored as sequence data points, vector representation, rather than photo-realistic image pixels. SketchRNN studied generative neural representation for by Long Short Term Memory networks (LSTM). Unfortunately, learned primarily generation tasks, other tasks recognition retrieval sketches. To this end inspired recent BERT model, we present model...
Cross-modality distillation arises as an important topic for data modalities containing limited knowledge such depth maps and high-quality sketches. Such techniques are of great importance, especially memory privacy-restricted scenarios where labeled training is generally unavailable. To solve the problem, existing label-free methods leverage a few pairwise unlabeled to distill by aligning features or statistics between source target modalities. For instance, one typically aims minimize L2...
Fish individual fecundity is an important factor governing fish replenishment and population dynamics. An in-depth understanding of the dynamics not only for study fisheries ecology but also great practical significance. From mid-February to early March 2023, we collected 99 samples Hemibarbus medius in Beiliu River analyzed their body length, weight, empty shell gonadal weight. Using weighing mass method, calculated absolute each individual, relative terms length gonadosomatic index,...
Sketch has been employed as an effective communicative tool to express the abstract and intuitive meanings of object. Recognizing free-hand sketch drawing is extremely useful in many real-world applications. While content-based recognition studied for several decades, instance-level Sketch-Based Image Retrieval (SBIR) tasks have attracted significant research attention recently. The existing datasets such QMUL-Chair QMUL-Shoe, focus on retrieval chairs shoes. However, there are key...
This paper proposes a novel approach for Sketch-Based Image Retrieval (SBIR), which the key is to bridge gap between sketches and photos in terms of data representation. Inspired by channel-wise attention explored recent years, we present Domain-Aware Squeeze-and-Excitation (DASE) network, seamlessly incorporates prior knowledge sample sketch or photo into SE module make capable emphasizing appropriate channels according domain signal. Accordingly, proposed network can switch its mode...
The research of generating images from scene graphs has become a hot topic, benefiting the success Generative Adversarial Network (GAN). Previous works in this field are still challenged by complexity texture pattern and object structure images. In fact, it is more desirable to generate sketches directly, rather than image synthesis graphs. sketch an abstract iconic representation, which describes layout well but ignoring complex patterns, leading reasonable generation task. Furthermore,...
Bovine coronavirus (BCoV) is a major viral pathogen linked to respiratory and enteric problems in newborn calves.The goal of this study look into the molecular occurrence, haemato-biochemical changes, risk factors associated with occurrence BCoV infection cattle calves at different dairy farms small households district Jhelum, Pakistan.From July 2020 June 2021, 200 faecal samples were collected from exhibiting symptoms diarrhoea dysentery.S&C Biotech Coronavirus Antigen Rapid Test Kits used...
During the pandemic period, import and export trade has been affected seriously. This study investigates different implications on importing exporting countries of international under COVID-19 pandemic. The data in this article include top five net exporters importers 2019 ranked by their account balance. After comparison, there are four findings have found. First, epidemic had a severe negative impact for both countries. Second, impacts relatively deeper consist longer than Third, unlike...
Tibetan sheep of type plateau, Immune genes, RNA-Seq, Real-Time PCRTotal RNA was extracted from healthy spleen tissues plateau and smalltailed Han sheep.After mRNA purification segmentation, cDNA synthesized for Illumina highthroughput sequencing.The quality the filtered raw data evaluated.Subsequently, cleaned were used sequence alignment, KEGG enrichment analysis, differential gene cluster analysis GO to screen disease resistance-related, immune-determinant genes.In addition, screened...
Verbs are important in semantic understanding of natural language. Traditional verb representations, such as FrameNet, PropBank, VerbNet, focus on verbs' roles. These roles too coarse to represent semantics. In this paper, we introduce patterns semantics, that each pattern corresponds a single the verb. First analyze principles for patterns: generality and specificity. Then propose nonparametric model based description length. Experimental results prove high effectiveness patterns. We...