Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileAcademic lineage
Possible advisorsa guess from early papers, not confirmed
- Yu‐Jin ZhangSuggested from co-authorship
Is this you? Claim this profile to confirm or dismiss it.
Works281 from public data
- Robust Text Detection in Natural Scene Images662
An accurate and robust method for detecting texts in natural scene images using a fast and effective pruning algorithm to extract Maximally Stable Extremal Regions (MSERs) as character candidates using the strategy of minimizing regularized variations is proposed.
- OCRBench: on the hidden mystery of OCR in large multimodal models574
A comprehensive evaluation of large multimodal models, such as GPT4V and Gemini, in various text-related visual tasks, revealing both the strengths and weaknesses of these models, particularly in handling multilingual text, handwritten text, non-semantic text, and mathematical expression recognition.
- 277
- Deep Relational Reasoning Graph Network for Arbitrary Shape Text Detection255
This paper proposes a novel unified relational reasoning graph network for arbitrary shape text detection through an innovative local graph that bridges a text proposal model via Convolutional Neural Network and a deep relational reasoning network via Graphconvolutional Network, making the network end-to-end trainable.
- A Short Survey of Recent Advances in Graph Matching223
The aim is to provide a systematic and compact framework regarding the recent development and the current state-of-the-arts in graph matching.
- 207
- Adaptive Boundary Proposal Network for Arbitrary Shape Text Detection138
A novel adaptive boundary proposal network for arbitrary shape text detection, which can learn to directly produce accurate boundary for arbitraryshape text without any post-processing, is proposed.
- 82
- Arbitrary Shape Text Detection via Boundary Transformer80
A unified coarse-to-fine framework via boundary learning for arbitrary shape text detection, which can accurately and efficiently locate text boundaries without post-processing, and proposes a novel boundary energy loss that introduces an energy minimization constraint and an energy monotonically decreasing constraint to further optimize and stabilize the learning of boundary refinement.
- Arbitrary Shape Text Detection via Segmentation with Probability Maps64
This paper proposes an innovative and robust segmentation-based detection method via probability maps for accurately detecting text instances through a group of probability maps computed by a series of Sigmoid Alpha Functions to describe the possible probability distributions.
- 63
- 58
- Feature combination using boosting58
Experiments on classification of three UCI data sets and handwritten digit recognition indicate that this variant of boosting based feature combination is a promising learning algorithm for feature combination.
- 57
- VLPD: Context-Aware Pedestrian Detection via Vision-Language Semantic Self-Supervision55
A novel approach via Vision-Language semantic self-supervision for context-aware Pedestrian Detection (VLPD) to model explicitly semantic contexts without any extra annotations is proposed and achieves superior performances over the previous state-of-the-arts.
- 50
- Convex ensemble learning with sparsity and diversity48
This paper forms the classifier ensemble problem with the sparsity and diversity learning in a general mathematical framework, and designs a convex ensemble diversity measure, derived from the error-ambiguity decomposition.
- Kernel Proposal Network for Arbitrary Shape Text Detection47
This work is the first to introduce the dynamic convolution kernel strategy to efficiently and effectively tackle the adhesion problem of neighboring text instances in text detection and improve the robustness against unclear boundaries.
- DPFlow: Adaptive Optical Flow Estimation with a Dual-Pyramid Framework46
This paper proposes DPFlow, an adaptive optical flow architecture capable of generalizing up to 8K resolution inputs while trained with only low-resolution samples, and introduces Kubric-NK, a new benchmark for evaluating optical flow methods with input resolutions ranging from 1K to 8K.
- A novel classifier ensemble method with sparsity and diversity46
This work forms the classifier ensemble problem with the sparsity or/and diversity learning in a general framework, and proposes a heuristic algorithm, capable of obtaining ensemble classifiers with consideration of both sparsity and diversity.
- LSTM $$^{2}$$ 2 : Multi-Label Ranking for Document Classification45
An Long Short Term Memory-based multi-label ranking model for document classification, namely LSTM$$^2$$2 consisting of repLSTM—an adaptive data representation process and rankL STM—a unified learning-ranking process is proposed.
- RUArt: A Novel Text-Centered Solution for Text-Based Visual Question Answering44
Experimental results demonstrate that the proposed RUArt (Reading, Understanding and Answering the Related Text) method can effectively explore the contextual information of the text and mine the stable relationships between theText and objects.
- 42
- 41
- Recurrent Partial Kernel Network for Efficient Optical Flow Estimation38
This paper designs a new spatial recurrent encoder that extracts discriminative features at a significantly reduced size and utilizes Partial Kernel Convolution (PKConv) layers to produce variable multi-scale features with a single shared block.
- Open-Set Text Recognition via Character-Context Decoupling36
A Character-Context Decoupling framework is proposed to alleviate the problem of limited performance for open-set text recognition methods by separating contextual information and character-visual information.
- 35
- Chinese Short Text Classification with Mutual-Attention Convolutional Neural Networks35
This work proposes a novel framework called Mutual-Attention Convolutional Neural Networks, which integrates word and character-level features without losing too much feature information.
- DE2: Dynamic ensemble of ensembles for learning nonstationary data34
Experimental results confirm that the novel Dynamic Ensemble of Ensembles (DE2) approach consistently leads to better performance and has promising generalization ability for learning in nonstationary environments.
- Accurate and robust text detection32
A robust text detection system which is a prominent step-in for text retrieval in natural scene images or videos and is evaluated on the ICDAR 2011 Robust Reading Competition dataset and a publicly available multilingual dataset.
- Learning Aligned Cross-Modal Representation for Generalized Zero-Shot Classification31
A novel Vision-Semantic Alignment (VSA) method to strengthen the alignment of cross-modal latent features on the latent subspaces guided by a learned classifier for Generalized Zero-Shot Classification (GZSC).
- Scene text detection in video by learning locally and globally29
In this system, scene text is first detected locally in individual frames and an optimal tracking trajectory is learned and linked globally with all detection, recognition and prediction information by dynamic programming.
- End-to-end trainable network for degraded license plate detection via vehicle-plate relation mining28
A novel and applicable method for degraded license plate detection via vehicle-plate relation mining, which localizes the license plate in a coarse-to-fine scheme and predicts the quadrilateral bounding box in the local region by regressing the four corners of thelicense plate to robustly detect oblique license plates.
- 28
- 28
- Transformer-based Reasoning for Learning Evolutionary Chain of Events on Temporal Knowledge Graph27
A novel Transformer-based reasoning model is proposed for TKG to learn the Evolutionary Chain of Events (ECE) and an additional time prediction task is devised to complete effective temporal information within the learned unified representation.
- 27
- 26
- 26
- Inverse-Like Antagonistic Scene Text Spotting via Reading-Order Estimation and Dynamic Sampling25
A unified end-to-end trainable inverse-like antagonistic text spotting framework dubbed IATS, which can effectively spot inverse-like scene texts without sacrificing general ones and achieves superior performance both on irregular and inverse-like text spotting.
- Self-Adaptive Aspect Ratio Anchor for Oriented Object Detection in Remote Sensing Images25
A novel Self-Adaptive Aspect Ratio Anchor (SARA) is proposed to explicitly explore aspect ratio variations of objects in remote sensing images and achieves a promising mAP value of 79.91% on the DOTA dataset.
- 25
- 23
- Adaptive Pattern-Parameter Matching for Robust Pedestrian Detection23
This paper proposes a novel detection approach via adaptive pattern-parameter matching that achieves superior performance compared with other state-of-the-art methods on subsets of different scales and occlusion types.
- An Overview of Bayesian Network Applications in Uncertain Domains23
A discussion on general BN representation, draw inferences, learning and prediction is followed by applications of BN in some specific domains, and some future research directions are highlighted.
- Change Detection in SAR Images Based on Deep Learning22
The objective in this paper is to combine both the change detection of SAR images with the deep neural networks to compare its efficiency with fuzzy clustering method and deep belief network and results appear that proposed deep-learning algorithms can further improve the change-detection process.
- Biomedical literature classification with a CNNs-based hybrid learning network21
This paper proposes a Boltzmann Convolutional neural network framework (B-CNN) for biomedicine semantic indexing, and designs a hierarchical coarse to fine style indexing structure for learning and classifying documents, and a novel feature extension approach with word sequence embedding and Wikipedia categorization.
- Scene Video Text Tracking With Graph Matching21
The experimental results demonstrate that the tracker with the graph matching method tends to increase the valid correspondence of trajectories and candidate objects.
- Visual Text Processing: A Comprehensive Review and Unified Evaluation20
This survey presents a comprehensive, multi-perspective analysis of recent advancements in visual text processing, focusing on two key questions: what textual features are most suitable for different visual text processing tasks and how can these distinctive text features be effectively incorporated into processing frameworks.
- 20
- A Denoising Based Autoassociative Model for Robust Sensor Monitoring in Nuclear Power Plants20
A denoised autoassociative sensor model (DAASM) based on deep learning framework is proposed which comprises multiple hidden layers which are pretrained greedily in an unsupervised fashion under denoising autoencoder architecture to address poor regularization and robust learning issues.
- FMI image based rock structure classification using classifier combination20
An automatic rock structure classification system using image processing and pattern recognition technologies that can significantly reduce the complexity and difficulty in the rock structure analysis task for the oil and gas exploration.
- 19
- Learning Document Semantic Representation with Hybrid Deep Belief Network19
A new Hybrid Deep Belief Network (HDBN) is proposed which uses Deep Boltzmann Machine (DBM) on the lower layers together with Deep belief Network (DBN) onThe upper layers and different input strategies for semantic distributed representation are explored.
- Joint Rotation-Invariance Face Detection and Alignment with Angle-Sensitivity Cascaded Networks18
This paper proposes a simple but effective architecture, named Angle-Sensitivity Cascaded Networks (ASCN), for jointly conducting rotation-invariance face detection and alignment and introduces a pose-equitable loss to balance the faces with large poses.
- Robust Feature Extraction on Vibration Data under Deep-Learning Framework: An Application for Fault Identification in Rotary Machines18
Convolutional Neural Network was explored for unsupervised feature learning on vibration signals and it was hypothesized that feature learning on raw vibration signal possibly will extract vibration-features that can improve fault identification performance of subsequent classifier.
- Joint Antenna Position and Beamforming Optimization with Self-Interference Mitigation in Movable Antenna Aided ISAC System17
An MA-ISAC model within an FD monostatic framework, where the self-interference channel is modeled as a function of the antenna position vectors under the near-field channel condition, enables antenna position optimization for maximizing the weighted sum of communication capacity and sensing mutual information.
- Graph fusion network for multi-oriented object detection17
This work proposes a novel graph fusion network, named GFNet, which is extensible and adaptively fuse dense detection boxes to detect more accurate and holistic multi-oriented object instances and verifies the effectiveness and robustness of the method against general NMS methods in multi- oriented object detection.
- Multi-orientation scene text detection with scale-guided regression17
An innovative Scale-Guided Regression Module (SRM), specially for multi-orientation scene text detection, which adopts IoU loss to train the network and then integrates IoU lost and l 1 -smooth loss for fine-tuning so that the procedures of classification and regression can be coherently integrated.
- Exploring the Application of Large Language Models Based AI Agents in Leakage Detection of Natural Gas Valve Chambers16
Preliminary explorations into the in-depth applications of AI agents based on large language model (LLM) technology in oil and gas production leakage detection are conducted, demonstrating certain positive outcomes.
- 16
- 16
- Semantic indexing with deep learning: a case study16
This paper introduces a novel CNN-based semantic indexing method for biomedical abstract document collections and designs a hierarchical CNN indexing architecture for learning documents from a coarse- to fine-grained level with several multi-label training techniques.
- Automatic segmentation of cervical vertebrae in X-ray images16
An applicable automatic segmentation system for medical image of cervical spine that includes a parallel cascade structure based Haar-like features and the AdaBoost learning algorithm used to detect the location of cervical vertebrae as a initial position of Active Appearance Model (AAM).
- From Token to Word: OCR Token Evolution via Contrastive Learning and Semantic Matching for Text-VQA15
A novel Text-VQA method to alleviate OCR errors via OCR token evolution is proposed, which introduces a vocabulary predictor with character-level semantic matching, which enables the model to recover the correct word from the vocabulary even with misspelled OCR tokens.
- Semantic Bilinear Pooling for Fine-Grained Recognition15
A novel framework with a two-branch network (coarse branch and fine branch), i.e., semantic bilinear pooling, for fine-grained recognition with a hierarchical label tree that can adaptively learn the semantic information from the hierarchical levels.
- 15
- Social Book Search Reranking with Generalized Content-Based Filtering15
This paper proposes a general search-recommendation hybrid framework for Social Book Search, where learning-to-rank is used to adaptively combine all reranking results, and has the best performance on both datasets compared to other state-of-the-art systems.
- 15
- Rapid detection of Klebsiella pneumoniae based on one-tube RPA-CRISPR/Cas12a system14
A novel One-Tube RPA- CRISPR/Cas12a system, integrating recombinase polymerase amplification (RPA) method with the CRISPR/Cas12a diagnostic platform, to facilitate the detection of K. pneumoniae strains presents a streamlined, expeditious, and highly specific diagnostic approach.
- 14
- Unsupervised Multi-view Pedestrian Detection14
Extensive experiments on popular multi-view pedestrian detection benchmarks Wildtrack, Terrace, and MultiviewX show that the proposed UMPD, as the first fully-unsupervised method to the best knowledge, performs competitively to the previous state-of-the-art supervised methods.
- Ranking via partial ordering for answer selection14
A ranking method via partial ordering for answer selection, which is applicable to various tasks, such as question answering and reading comprehension, and which can significantly improve the answer selection performance.
- Semantic Sequential Query Expansion for Biomedical Article Search14
The semantic sequential dependence model is proposed, which provides an innovative combination of semantic information and the conventional SDM and the experimental results show that the query expansion approach outperforms the baseline and other participants in the BioASQ competitions.
- 14
- A VidEo-Based Intelligent Recognition and Decision System for the Phacoemulsification Cataract Surgery13
A Video-Based Intelligent Recognitionand Decision (VeBIRD) system, which breaks new ground by providing a generic framework for automatically tracking the operation process and classifying theCataract grade in microscope videos of the phacoemulsification cataract surgery, is developed.
- 13
- USTB at INEX2014: Social Book Search Track13
This paper performs re-ranking on Galago searching results on enriched XML index by 11 dierent s- trategies and combines the results with learning to rank to find that enriched index improves the eectiveness and tag is the best-performed social feature on re- ranking.
- 12
- 12
- 12
- 12
- Obtaining Super-Resolution Satellites Images Based on Enhancement Deep Convolutional Neural Network12
Enhanced Deep convolutional neural networks (EDCNN) learn through a large number of training samples, obtain relevant information within the image, and then use the information to achieve specific functions to achieve super-resolution reconstruction.
- 12
- Aligning enhanced feature representation for generalized zero-shot learning11
This paper proposes an innovative asymmetric VAE network by aligning enhanced feature representation (AEFR) for GZSL and proposes a novel distributional decoupling-based contrastive learning (D2-CL) to guide learning classification-relevant information while aligning the representations at the taxonomy level in the latent representation space.
- MaterialSeg3D: Segmenting Dense Materials from 2D Priors for 3D Assets11
This work proposes MaterialSeg3D, a 3D asset material generation framework to infer underlying material from the 2D semantic prior, and devise a mechanism to parse material in 3D space.
- Non-Autoregressive Transformer with Unified Bidirectional Decoder for Automatic Speech Recognition11
A new non-autoregressive transformer with a unified bidirectional decoder (NAT-UBD), which can simultaneously utilize left- to-right and right-to-left contexts for ASR, and can run 49.8× faster than the AR transformer baseline when decoding in a single step.
- Weakly Correlated Knowledge Integration for Few-shot Image Classification11
A novel weakly correlated knowledge integration (WCKI) framework to integrate knowledge transferred from different sources (i.e., visual domain and textual domain) and demonstrates significant improvements on multiple few-shot image classification datasets.
- 11
- 11
- ISART: A Generic Framework for Searching Books with Social Information11
Experiments show that this technology permits embedding social information to promote book search effectiveness, and IsArt, by making use of it, has the best performance on CLEF/INEX Social Book Search Evaluation datasets of all 4 years, compared with some other state-of-the-art methods.
- 11
- 11
- Image retrieval framework driven by association feedback with feature element evaluation built in11
An original image retrieval framework which breaks features into feature elements - FEs which have meaningful visual sense instead of combining them to get semantic meanings, and which can grasp the user's target more accurately.
- 10
- CAliC: Accurate and Efficient Image-Text Retrieval via Contrastive Alignment and Visual Contexts Modeling10
The visual unimodal encoder is pretrained with cross-modal contrastive learning to extract easily aligned visual features, which improves the retrieval accuracy and the inference speed, and outperforms existing state-of-the-art retrieval methods on image-text retrieval tasks over Flickr30K and COCO benchmarks.
- 10
- Fast alignment for sparse representation based face recognition10
This paper proposes a fast alignment strategy for sparse representation based face recognition algorithms and demonstrates that the proposed alignment strategy greatly improves the efficiency of original algorithms without losing accuracy and robustness.
- A generic pseudo relevance feedback framework with heterogeneous social information10
This paper proposes a novel and generic PRF framework for searching and recommendation with heterogeneous social information, and constructs an effective searching system, where learning-to-rank is utilized adaptively to combine results from various PRF strategies with different social information.
- 9
- 9
- GCCNet: Grouped channel composition network for scene text detection9
This work proposes a novel Grouped Channels Composition (GCC) block to achieve the data-driven anchor design and adaptive anchor assignment and proposes a Unified Loss Weighting module to alleviate the inconsistency between classification score and localization accuracy.
- Mutual-Supervised Feature Modulation Network for Occluded Pedestrian Detection9
A novel Mutual-Supervised Feature Modulation (MSFM) network is proposed, to better handle occluded pedestrian detection, and achieves superior performance compared to other state-of-the-art methods on both datasets, especially in heavy occlusion cases.
- USTB at Social Book Search 2016 Suggestion Task: Active Books Set and Re-ranking.9
This paper describes its participation in the Social Book Search (SBS) Suggestion Task and developed some new re-ranking models, based on what was applied in INEX2014, SBS2015 and combined those re- ranking models according to a specific order shown the best performance.
- A generic retrieval system for biomedical literatures: USTB at BioASQ2015 Question Answering Task9
A generic retrieval model based on the sequential dependence model, Word Embedding and Ranking Model, and from the view of the special signicance of titles(Title Signicance Validation), is built to re-rank the top-K results by counting the meaningful nouns in the titles.
- Pedestrian Analysis and Counting System with Videos9
A new and efficient framework for pedestrian analysis and counting, which consists of a rule induction classifier with optical-flow feature and a linear regression model, which can count pedestrians precisely in general scenes without the influence of abnormal behaviors.
- Learning Based Visibility Measuring with Images9
A method of visibility calculation based on image analysis and learning that makes use of the existing video cameras to calculate visibility in real time and has the characteristic of low cost, fast calculation, and convenience.
- 8
- Diversity-Based Random Forests with Sample Weight Learning8
The proposed DRFS method can select and combine tree classifiers adaptively and improves the performance on a variety of classification tasks, and is proposed to solve the convex optimization iteratively and learn sample weights automatically.
- Classification of EgyptSat-1 Images Using Deep Learning Methods8
This research finds that CNN outperforms ML by 9%.
- DeTEXT: A Database for Evaluating Text Extraction from Biomedical Literature Figures8
How figures were selected from open-access full-text biomedical articles and how annotation guidelines and annotation tools were developed are described and the inter-annotator agreement and the reliability of the annotations are discussed.
- Semantic Retrieval Based on Feature Element Constructional Model and Bias Competition Mechanism8
An original image retrieval framework is proposed and developed, and a novel cognitive model - feature element constructional model is proposed, which provides great power for semantic retrieval.
- Multi-Scale Texture Fusion for Reference-Based Image Super-Resolution: New Dataset and Solution7
This work proposes an innovative multi-scale texture fusion for reference-based super-resolution via a state-space model, which enables efficient multi-scale feature fusion and long-term dependency modeling for better texture restoration.
- Self-supervised learning enhances periapical films segmentation with limited labeled data7
This two-stage, domain-specific self-supervised framework effectively learns robust anatomical features and can improve diagnostic accuracy through consistent segmentation and enhance workflow efficiency by reducing manual analysis time, especially in resource-constrained dental practices.
- Profiling brain morphology for autism spectrum disorder with two cross-culture large-scale consortia7
Two ASD subgroups with distinct brain morphological abnormalities are identified: subgroup “L” is characterized by generally smaller brain region volumes and higher rates of abnormality, while subgroup “H” exhibits larger volumes with less pronounced deviations in specific areas.
- 7
- 7
- SSDM2: a Two-Stage Semantic Sequential Dependence Model Framework for Biomedical Question Answering7
The proposed SSDM2 framework based on a cognitive-inspired model and sequential dependence model to answer biomedical questions with relevant snippets in academic papers is proposed and outperforms several state-of-the-art baselines and BioASQ participants.
- 7
- Building Your Own Reading List Anytime via Embedding Relevance, Quality, Timeliness and Diversity7
This paper proposes a searching framework for building a topical reading list anytime, where the Relevance (between topics and books), Quality, Quality, quality, timeliness and diversity are embedded into vector representations respectively based on user-generated contents and statistics on social media.
- Feature element theory for image recognition and retrieval7
A new hypothesis - element presence theory is proposed to explain the truth of the whole visual cognition and the evaluation on feature elements is discussed and the illustration on feature element theory based image retrieval system is given.
- 6
- 6
- 6
- 6
- 6
- Health assistant: answering your questions anytime from biomedical literature6
A biomedical question answering framework is presented and a system, Health Assistant, is implemented to enable the search process and demonstrate the effectiveness and robustness of the system, compared to BioASQ participant systems and some state-of-the-art methods on both document retrieval and snippet retrieval tasks.
- 6
- 6
- Classifier Ensemble Using a Heuristic Learning with Sparsity and Diversity6
This paper investigated classifier ensemble with learning both sparsity and diversity using a heuristic method, and introduced the diversity contribution ability to select proper classifier components and evolve classifier weights.
- A Rock Structure Recognition System Using FMI Images6
An efficient and intelligent rock structure recognition system is developed by engaging image processing and pattern recognition technologies and chooses the most effective color and shape features for rock images.
- Association Feedback: A Novel Tool for Feature Elements Based Image Retrieval6
A new approach: association feedback is proposed, which analyzes synthetically the feedback data, retrieval history and existing result to find out the associated elements that potentially hit the retrieval target.
- Association of lung health and cardiovascular health (Life’s Essential 8)5
Overall lung health was better in the high LE8 group than in the low LE8 group and RCS analyses demonstrated a non-linear negative correlation between LE8 and cough, sputum, and wheeze.
- 5
- 5
- 5
- Bloody Image Classification with Global and Local Features5
This paper proposes region-based techniques to identify a color image being bloody or not and designs a bloody image classification method with global visual features using Support Vector Machines and constructs a novel bloody region identification approach using Convolutional Neural Networks.
- A central tendency-based privacy preserving model for sensitive XML association rules using Bayesian networks5
A model for identifying sensitive items (nodes) to declare sensitive XML association rules and then to hide them is proposed to minimize an incredible disclosure risk involved in XML association rule mining from external parties in a competitive business environment.
- Learning Imbalanced Classifiers Locally and Globally with One-Side Probability Machine5
This paper proposes a new model called One-Side Probability Machine (OSPM), which is the first model capable of learning data both locally and globally and can lead to rigorous treatment on biased classification tasks.
- Exchange rate prediction with non-numerical information5
This paper engages a novel integrated approach which successfully combines information obtained from both numerical and non-numerical factors for exchange rate prediction, and can achieve the Directional Symmetry accuracy of 86.96%, which is much higher than only exploiting numerical information.
- Exchange Rate Forecasting Using Classifier Ensemble5
Experimental results showed that the ensemble method has a higher degree of forecasting accuracy after adding the non-numerical information.
- 5
- 4
- 4
- 4
- SD-GAN: Semantic Decomposition for Face Image Synthesis with Discrete Attribute4
This work proposes an innovative framework to tackle challenging facial discrete attribute synthesis via semantic decomposing, dubbed SD-GAN, and constructs a large and valuable dataset MEGN for completing the lack of discrete attributes in the existing dataset.
- 4
- 4
- Detecting Multi-Resolution Pedestrians Using Group Cost-Sensitive Boosting with Channel Features4
A robust multi-resolution detection approach with a novel group cost-sensitive boosting algorithm derived from the standard AdaBoost algorithm to further explore different costs for different resolution groups of the samples in the boosting process, and to place greater emphasis on low-resolution groups in order to better handle the detection of multi- resolution targets.
- 4
- Effective human detection via multi-model classification and adaptive late fusion4
Human detection serves as an important basis to achieve certain video surveillance-oriented biometrics such as gait, face and actions since the first step is to find and locate human targets inSurface-based surveillance systems.
- Multi-label Ranking with LSTM $$^2$$ for Document Classification4
A LSTM\(^2\) (Long short term memory) model for document classification which can be trained as a whole by predicting label serially and which is a good solution to deal with the error propagation for variable length output.
- 4
- USTB at Social Book Search 2015 Suggestion Task: Metadata Expansion and Reranking.4
Focus on the lack of important information of the majority books, document expansion by crawling book information from other two web sites which can enrich the index is used and re- ranking and Random Forests combining those re-ranking models show better performance.
- Dynamic Ensemble of Ensembles in Nonstationary Environments4
Experimental results confirm that the novel Dynamic Ensemble of Ensembles DE2 approach consistently leads to better performance and has promising generalization ability for learning in non-stationary environments.
- 4
- A Segmentation and Recognition System for Touching and Broken Numeral Strings Based on Viterbi Algorithms4
Experiments indicate that this Viterbi algorithms based method is efficient for segmentation and recognition of touching and broken numeral strings, and enhances the accuracy and robustness of the recognition system.
- 4
- China Autism Brain Imaging Consortium: Charting Brain Growth in Chinese Children with Autism3
Brain growth charts are constructed that reveal a developmental shift in autistic children, transitioning from early overgrowth to delayed maturation, to push forward a more comprehensive understanding of the brain development in Chinese autism children.
- 3
- MarketGen: A Scalable Simulation Platform with Auto-Generated Embodied Supermarket Environments3
MarketGen provides a comprehensive framework to accelerate research in embodied AI for complex commercial applications and proposes a novel benchmark for assessing supermarket agents, featuring two daily tasks in a supermarket.
- Decoupling and Interaction: task coordination in single-stage object detection3
A Task-coordinated Single-stage Object Detector (TSOD) is proposed to enhance the coordination of multiple tasks by introducing a Task-decoupled Feature Alignment Mechanism (TFAM) and a Task-interactive Enhancement Mechanism (TEM), which explicitly combines different task-sensitive features for joint classification score prediction and selects samples with high task consistency for training.
- HQOD: Harmonious Quantization for Object Detection3
The Harmonious Quantization for Object Detection (HQOD) framework is proposed, which consists of a task-correlated loss to encourage detectors to focus on improving samples with lower task harmony quality during QAT, and a harmonious Intersection over Union (IoU) loss is incorporated to balance the optimization of the regression branch across different IoU levels.
- A lightweight convolutional network for infrared object detection and tracking3
A method for target detection and tracking, with a deeper and lightweight MobileNet V2 structure as the backbone network is proposed, which shows that the tracker can get a balance between tracking accuracy and inference speed, which is crucial for deployment on mobile devices.
- Pyramid Memory Block and Timestep Attention for Speech Emotion Recognition3
This work proposes an improved speech emotion recognition (SER) end-to-end system that achieves 2.67% improvement compared to the commonly used CNN-biLSTM model which requires much more computing resource.
- Combined Correlation Filters with Siamese Region Proposal Network for Visual Tracking3
An ensemble tracking framework based on SiamRPN tracker, consisting of two components: (1) Correlation Filter module with hierarchical features fusion; and (2) Siam RPN module, trained in an end-to-end manner, which enhances both robustness and module adaptability effect.
- Chinese text-line detection from web videos with fully convolutional networks3
The proposed system can directly detect the English word and Chinese characters in the caption text-lines without word or character segmentation with the high performance on real-world web video datasets.
- Selection of optimal denoising-based regularization hyper-parameters for performance improvement in a sensor validation model3
It is found that the greedy-learning procedure improved the overall robustness of the sensor model and the effects of denoising-based regularization hyper-parameters such as noise-type and noise-level on sensor model performance were explored.
- 3
- K2 Algorithm based Text Detection with an Adaptive Classifier Threshold3
A Bayesian network scores are used to classify candidate character regions by computing posterior probabilities to define an adaptive threshold to detect text in scene images with accuracy.
- Social Book Search with Pseudo-Relevance Feedback3
This paper proposes a search system based on pseudo-relevance feedback (PRF) for expanding and enriching the social information of queries and has an encouraged performance compared to several state-of-the-art (contest) systems.
- Visibility Measurement with Image Understanding3
In this paper, machine learning is introduced into the study of visibility measurement and a method of daytime visibility measurement is proposed based on image understanding that has both high visibility measuring precision and good flexibility.
- Keyword Extraction Based on Multi-feature Fusion for Chinese Web Pages3
Comparative experiments show that extracting keywords generated by the proposed keyword extraction method has higher precision and recall compared with the classical TF-IDF algorithm.
- 2
- 2
- 2
- Semi‐supervised domain adaptation via subspace exploration2
This study proposes to explore a robust subspace for Semi‐Supervised Domain Adaptation (SSDA) explicitly, and a novel denoising strategy via class‐agnostic disturbance to improve the discriminative ability of subspace is adopted.
- 2
- 2
- Boosting Multi-Modal E-commerce Attribute Value Extraction via Unified Learning Scheme and Dynamic Range Minimization2
A novel approach to boost multi-modal e-commerce attribute value extraction via unified learning scheme and dynamic range minimization is proposed, which achieves superior performance over the other state-of-the-art techniques.
- 2
- A Hybrid Self-Attention Model for Pedestrians Detection2
A novel hybrid attention module is proposed which is capable of obtaining inter-dependencies between features from both channel and spatial dimensions through local convolution and dual-pass pooling, and guiding the network to focus on better pedestrians’ feature representation while suppressing background noise.
- CARNet: Densely Connected Capsules with Capsule-Wise Attention Routing2
This paper proposes a novel routing algorithm named capsule-wise attention routing based on attention mechanism, which is proven experimentally to outperform other state-of-the-art capsule networks on SVHN and CIFAR-10 benchmarks while reducing the amount of parameters by 62% at most.
- Chinese word segmentation with local and global context representation learning2
Experimental results show that Chinese character representations can effectively learn the semantic information and achieves a pretty good improvement on precision, recall and f-measure.
- Query-Based Automatic Multi-document Summarization Extraction Method for Web Pages2
The comparative experimental results show that the automatic multi-document summarization extraction method proposed is more effective on the Precision and Recall than others in abstract extraction.
- 2
- Probabilistic Concept Graph Reasoning for Multimodal Misinformation Detection1
Experiments demonstrate that PCGR achieves state-of-the-art MMD accuracy and robustness to emerging manipulation types, outperforming prior methods in both coarse detection and fine-grained manipulation recognition.
- Dialectical Structured Reasoning for Explainable Multimodal Fake News Detection1
Experiments demonstrate that Dialectical Structured Reasoning achieves state-of-the-art performance while producing transparent, di-alectically grounded explanations that closely mirror human reasoning process.
- 1
- SelectTSL: Prompt-Guided Selective Target Sound Localization in Complex Scenarios1
This work designs a target-aware selective localization strategy that employs a Prompt-Guided Selective Attention Module (PGSA) to generate prompt-informed embeddings and proposes SelectTSL, an end-to-end architecture that localizes only the user-specified target in multi-source acoustic scenes.
- Breaking Through the Spike: Spike Window Decoding for Accelerated and Precise Automatic Speech Recognition1
The Spike Window Decoding algorithm is proposed, which greatly improves the inference speed by making the number of frames decoded in WFST linearly related to the number of spiking frames in the CTC output, while guaranteeing the recognition performance.
- 1
- 1
- 1
- 1
- 1
- Exploring Stable Meta-Optimization Patterns via Differentiable Reinforcement Learning for Few-Shot Classification1
This work proposes an innovative generic differentiable Reinforcement Learning (RL) strategy for few-shot classification that aims to explore stable meta-optimization patterns in meta-training by learning generalizable optimizations for producing task-adaptive embeddings.
- 1
- 1
- Understanding Dynamic Auditory and Tactile Perception for Water Filling Level Estimation1
This work examines the acoustic features and develops a novel deep learning-based regression method for dynamic auditory perception for water Filling Level Estimation (FLE) in HRI, and demonstrates the viability of dynamic auditory perception as a reliable and complementary modality in FLE.
- 1
- 1
- Dual-Path Transformer-Based GAN for Co-speech Gesture Synthesis1
This paper facilitates parallel directional cross-modal transformers, and an interactive and cascaded 2D attention module, to achieve selective fusion of the gesture-related cues, and proposes new metrics to evaluate gesture diversity and speech-gesture correspondence, without 3D pose annotation requirements.
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- Robust Segmentation for Video Captions with Complex Backgrounds1
An effective caption text segmentation approach to improve OCR accuracy is proposed and can effectively remove the complex backgrounds with similar intensity to text.
- Text Categorization with Diversity Random Forests1
Diversity Random Forests is proposed, which diversely and adaptively select and combine tree classifiers with diversity learning and sample weighting, and a new self-training algorithm to iteratively run the convex optimization and automatically learn the sample weights.
- Weakly Supervised Compressive Tracking with Effective Prediction Model1
This paper introduces an effective prediction model based on optical flow, which could achieve reliable estimation of the position and the scale of current object and guides the compressive classifier's updating to alleviate the phenomenon of object drift.
- 1
- Retrieval model of social books based on multi-feature fusion1
The results show that feature extraction methods and the social re-ranking model presented in this paper can effectively improve the performance of traditional retrieval method.
- System for Online Patent Information Analysis Based on Data Mining1
The results show that the online patent analysis system designed and implemented can help to mine and analyze an immense amount of patent data rapidly, conveniently and effectively and the user can understand the patent value more profoundly to study, formulate and implement relative patent strategies.
- 1
- –
- CID-TKG: Collaborative Historical Invariance and Evolutionary Dynamics Learning for Temporal Knowledge Graph Reasoning–
A novel collaborative learning framework for TKGR that integrates evolutionary dynamics and historical invariance semantics as an effective inductive bias for reasoning and achieves state-of-the-art performance under extrapolation settings.
- –
- Dual-Geometry Graph Network: Unifying Local and Global Priors for Few-Shot Learning–
This work proposes a novel Dual-Geometry Graph Network (DGGN) to adaptively integrate the local and global geometric priors via two key pathways and introduces two new regularization loss functions to explicitly enhance the geometric representation ability of the local and global pathways respectively.
- Predicting Emergent Tool Use in LLMs Before It Emerges: A Proxy Perspective–
A proxy-task perspective that predicts tool-use capabilities by measuring early model performance on selected non-emergent proxy tasks, and formalizes how these weighted signals approximate emergent tool use through bounded extrapolation under relaxed assumptions is formalized.
- Remote Sensing Techniques Shaping the Future of Wildfire Early Detection–
This review systematically evaluates remote sensing technologies employed in early wildfire detection, classifying them into satellite systems, aerial platforms, and ground‐based sensors and critically examines detection methodologies, revealing that deep learning significantly improves detection accuracy; however, its effectiveness is limited by resolution constraints.
- CHE-TKG: Collaborative Historical Evidence and Evolutionary Dynamics Learning for Temporal Knowledge Graph Reasoning–
CHE-TKG is proposed, a novel collaborative dual-view learning framework for TKG reasoning that explicitly separates and jointly models historical evidence and evolutionary dynamics, aiming to learn and exploit their complementary predictive signals.
- Beyond Optical Flow: Latent Micro-Motion as Visual Evidence in UAV Video–
It is argued that video also contains a largely overlooked form of motion information, namely latent micro-motion, which arises from subtle, structure-constrained responses of rigid components to physical interaction with the environment.
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- Unsupervised Real-World Super-Resolution via Rectified Flow Degradation Modelling–
This paper proposes a novel Rectified Flow Degradation Module (RFDM) that introduces degradation-transformed LR (DT-LR) images as intermediaries that better captures real-world degradation and enhances the realism of generated LR images.
- –
- –
- –
- RoleMotion: A Large-Scale Dataset towards Robust Scene-Specific Role-Playing Motion Synthesis with Fine-grained Descriptions–
RoleMotion, a large-scale human motion dataset that encompasses a wealth of role-playing and functional motion data tailored to fit various specific scenes, is introduced and the interplay of motion generation of body and hands is explored.
- –
- –
- –
- –
- –
- I 2 TTS: Image-Indicated Immersive Text-to-Speech Synthesis with Spatial Perception–
Embedded Immersive Text-to-speech Synthesis is proposed, an end-to-end multimodal TTS framework that synthesizes high-quality, immersive speech from text and visual scene prompts that produces spatially accurate and natural-sounding speech, achieving superior performance in both subjective and objective evaluations.
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- Learning to Diversify for Ensembling of An Amount of Classifiers–
This paper introduces a novel approach for classifier ensemble, learning to diversify, which learns to adaptively combine classifiers by considering both accuracy and diversity, and proposes an iterative training algorithm for weights optimization.
- SUMAC 2020: The 2nd Workshop on Structuring and Understanding of Multimedia heritAge Contents–
A representative scope of Computer Science methodologies dedicated to the processing of multimedia heritage contents and their exploitation is covered by the works presented, with the ambition of advancing and raising awareness about this fully developing research field.
- –
- Detecting Advertising Materials via Multi-Scale Instance Segmentation Network.–
This work adopts a fully convolutional instance segmentation network to capture the semantic information and link information of pixels and jointly optimize the network with semantic loss, link loss, and contour loss, to obtain detection results.
- TED-KISS–
This work presents a new publicly available known-item speech video search benchmark, namely TED-KISS, which takes TED talks as an example, and analyzes the characteristics of the benchmark through evaluations of several existing text-based IR and Neural-IR models, which also can be served as baselines for this task.
- –
- –
- –
- –
- Medical Image Compression Based on Discrete Wavelet Translation and Vector Quantization–
An effective medial image compression approach with Discrete Wavelet Transform (DWT) and Vector Quantization (VQ) techniques is designed and several experimental results show that the method is useful for medical image compression.
- –
- –
Publication data from OpenAlex, with missing venues and authors filled in from Crossref; citation counts are the higher of OpenAlex and Semantic Scholar, last synced 2026-10-12. One-sentence summaries under some papers are written by Semantic Scholar’s model. Citation counts may be lower than on Google Scholar, which indexes more sources.
Report an error
Wrong papers, two people merged into one, or a profile that should not be here? Tell us and we will fix or hide it.