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
- Suggested from co-authorship
Is this you? Claim this profile to confirm or dismiss it.
Works175 from public data
- Video Summarization With Attention-Based Encoder–Decoder Networks392
This paper proposes a novel video summarization framework named attentive encoder–decoder networks forVideo summarization (AVS), in which the encoder uses a bidirectional long short-term memory (BiLSTM) to encode the contextual information among the input video frames.
- Consensus-Aware Visual-Semantic Embedding for Image-Text Matching196
A Consensus-aware Visual-Semantic Embedding model is proposed to incorporate the consensus information, namely the commonsense knowledge shared between both modalities, into image-text matching and learns the associations and alignments between image and text based on the exploited consensus.
- Episode-Based Prototype Generating Network for Zero-Shot Learning178
A novel generative model is proposed that synthesizes visual prototypes conditioned on the class semantic prototypes into an adversarial framework paired with a parameter-economic Multi-modal Cross-Entropy Loss to capture the discriminative information.
- Improved prototypical networks for few-Shot learning132
This work proposes an attention-analogous strategy to explore the class distribution information by distributing different weights to samples based on their representativeness and proposes a distance scaling strategy to reduce the intra- class difference while enlarge the inter-class difference.
- Step-Wise Hierarchical Alignment Network for Image-Text Matching122
A step-wise hierarchical alignment network (SHAN) that decomposes image-text matching into multi-step cross-modal reasoning process that supplies the model with more complementary and sufficient semantic clues to understand the hierarchical correlations between image and text.
- Saliency-Guided Attention Network for Image-Sentence Matching114
A Saliency-guided Attention Network (SAN) is introduced that is characterized by building an asymmetrical link between vision and language to efficiently learn a fine-grained cross-modal correlation.
- 97
- Transductive Zero-Shot Learning With a Self-Training Dictionary Approach91
A bidirectional mapping-based semantic relationship modeling scheme that seeks for cross-modal knowledge transfer by simultaneously projecting the image features and label embeddings into a common latent space is proposed.
- Attribute-Guided Network for Cross-Modal Zero-Shot Hashing88
A novel attribute-guided network (AgNet) is proposed, which can perform not only IBIR but also text-based image retrieval (TBIR) and bridges the gap caused by modality heterogeneity and zero-shot setting.
- Stacked Semantics-Guided Attention Model for Fine-Grained Zero-Shot Learning88
A novel stacked semantics-guided attention (S2GA) model to obtain semantic relevant features by using individual class semantic features to progressively guide the visual features to generate an attention map for weighting the importance of different local regions.
- Asymmetric Cross-Scale Alignment for Text-Based Person Search78
A transformer-based model to extract multi-scale representations, and an asymmetric cross-attention mechanism to dynamically align the cross-modal entities in region/image-phrase scales are presented.
- Memorizing Complementation Network for Few-Shot Class-Incremental Learning77
A Memorizing Complementation Network (MCNet) is proposed to ensemble multiple models that complements the different memorized knowledge with each other in novel tasks in few-shot Class-Incremental Learning.
- Zero-Shot Learning via Latent Space Encoding76
Extensive experimental results on four benchmark datasets clearly demonstrate the superiority of the proposed encoder–decoder approach, namely latent space encoding (LSE), on several ZSL tasks, including traditional ZSL, generalized Z SL, and zero-shot retrieval.
- USER: Unified Semantic Enhancement With Momentum Contrast for Image-Text Retrieval72
A Unified Semantic Enhancement Momentum Contrastive Learning (USER) method for ITR is proposed, which delicately design two simple but effective Global representation based Semantic Enhancement (GSE) modules and incorporates the training mechanism of MoCo into ITR.
- Deep attentive and semantic preserving video summarization67
This work incorporates the ideas of both the encoder-decoder attention and semantic preserving loss in a deep Seq2Seq framework for video summarization to enhance the robustness of the model to outliers and introduce Huber loss to replace the popular mean square error loss.
- GTNet: Generative Transfer Network for Zero-Shot Object Detection65
An IoU-Aware Generative Adversarial Network (IoUGAN) is designed as the feature synthesizer, which can be easily integrated into GTNet, and performs favorably against the state-of-the-art ZSD approaches.
- 58
- Ranking Graph Embedding for Learning to Rerank58
It is shown that introducing ranking information into dimensionality reduction significantly increases the performance of image search reranking.
- 56
- Query-aware sparse coding for web multi-video summarization49
A novel query-aware approach is proposed by formulating the multi-video summarization in a sparse coding framework, where the web images searched by a query are taken as the important preference information to reveal the query intent.
- SGD: Street View Synthesis with Gaussian Splatting and Diffusion Prior47
A novel approach is proposed that enhances the capacity of 3DGS by leveraging prior from a Diffusion Model along with complementary multi-modal data and exploiting depth data from LiDAR point clouds to supply additional spatial information.
- 46
- Transductive Zero-Shot Learning With Adaptive Structural Embedding45
Two corresponding methods named Adaptive STructural Embedding (ASTE) and Self-PAced Selective Strategy (SPASS) for visual-semantic embedding and domain adaptation in cross-modality learning and unseen class prediction steps are presented.
- Relevance Preserving Projection and Ranking for Web Image Search Reranking45
An HRPP is a spectral embedding algorithm to transform an original high-dimensional feature space into an intrinsically low-dimensional hypersphere space by preserving the manifold structure and a relevance relationship among the images and an H-Rank is a simple but effective ranking algorithm to sort the images by their distances to the hyperspheres center.
- 41
- Deep Ranking for Image Zero-Shot Multi-Label Classification40
This work investigates and formalizes a flexible framework consisting of two components, i.e., visual-semantic embedding and zero-shot multi-label prediction, and presents a deep regression model to project the visual features into the semantic space, which explicitly exploits the correlations in the intermediate semantic layer of word vectors and makes label prediction possible.
- Manifold regularized cross-modal embedding for zero-shot learning37
A Manifold regularized Cross-Modal Embedding (MCME) approach for ZSL is proposed by formulating the manifold constraint for intrinsic structure of the visual features as well as aligning pairwise consistency in order to preserve the locally visual structure in the embedding process.
- Hypergraph dominant set based multi-video summarization36
A novel framework with Hypergraph Dominant Set (HDS), termed MVS-HDS is presented, which clearly outperforms the state-of-the-art methods on the multi-Video Summarization (MVS) task.
- 35
- Information Symmetry Matters: A Modal-Alternating Propagation Network for Few-Shot Learning34
A Modal-Alternating Propagation Network (MAP-Net) is proposed to supplement the absent semantic information of unlabeled samples and design a Relation Guidance (RG) strategy to guide the visual relation vectors via semantics so that the propagated information is more beneficial.
- 34
- Pedestrian attribute recognition based on multiple time steps attention32
A new multiple time steps attention mechanism is proposed to boost the modeling of the relations between images and attributes and it is demonstrated to be effective on two benchmark datasets, i.e., PETA and RAP.
- 31
- Learning from Students: Online Contrastive Distillation Network for General Continual Learning31
An Online Contrastive Distillation Network (OCD-Net) is proposed, which explores the merit of the student model in each time step to guide the training process of theStudent model to consolidate the learned knowledge.
- Class-specific synthesized dictionary model for Zero-Shot Learning31
This work proposes a novel synthesized approach for addressing ZSL within a dictionary learning framework that learns both a dictionary matrix and a class-specific encoding matrix for each seen class to synthesize pseudo instances for unseen classes with auxiliary of seen class prototypes.
- 31
- 28
- Triple discriminator generative adversarial network for zero-shot image classification28
This paper presents a new triple discriminator GAN for ZSC called TDGAN, which incorporates a text-reconstruction network into a dual discriminators GAN (D2GAN), allowing to realize cross-modal mapping from text descriptions to their visual representations.
- 28
- Teachers cooperation: team-knowledge distillation for multiple cross-domain few-shot learning27
This paper distill knowledge from the cooperation of teacher networks to a single student network in a meta-learning framework and incorporates task-oriented knowledge distillation and multiple cooperation among teachers to train an efficient student with better generalization ability on unseen tasks.
- Lightweight group convolutional network for single image super-resolution27
Experimental results on four popular datasets show that the proposed LGCN not only outperforms the state-of-the-art SISR methods, but also achieves faster speed.
- Image-attribute reciprocally guided attention network for pedestrian attribute recognition27
This work proposes an image-attribute reciprocal guidance representation method, which investigates image-guided feature and attribute- guided feature to learn the comprehensive features for different attributes, and develops a fusion attention mechanism.
- Visual-quality-driven unsupervised image dehazing26
This paper proposes a novel unsupervised dehazing (lightweight) network without any reference images to directly predict clear images from the original hazy images, which consists of an interactive fusion module (IFM) and an iterative optimization module (IOM).
- Complementary Calibration: Boosting General Continual Learning With Collaborative Distillation and Self-Supervision26
A new collaborative distillation approach for addressing the relation deviation is proposed by utilizing ensemble dark knowledge of new model’s outputs and reserved outputs, which maintains the performance of old tasks as well as balancing the relationship among all classes.
- Self-taught cross-domain few-shot learning with weakly supervised object localization and task-decomposition25
This work proposes a task-expansion-decomposition framework for CD-FSL, called Self-Taught (ST) approach, which alleviates the problem of non-target guidance by constructing task-oriented metric spaces and employs Weakly Supervised Object Localization and self-supervised technologies to enrich task- oriented samples by exchanging and rotating the discriminative regions.
- Multi-modal generative adversarial network for zero-shot learning25
This paper proposes a novel approach for Zero-Shot Learning (ZSL), where the test instances are from the novel categories that no visual data are available during training, and fuse various types of class semantic prototypes resided in different semantic spaces with a feature fusion network to generate pseudo visual features.
- Reweighting and information-guidance networks for Few-Shot Learning24
Two strategies on the basis of Prototypical Networks are proposed to improve the discriminativeness and representativeness of the visual prototypes for few-shot learning task and could be seamlessly combined into some existing approaches to substantially improve their performances on few- shot classification.
- 24
- Semi-supervised LPP algorithms for learning-to-rank-based visual search reranking24
Two feature dimensionality reduction methods for search reranking are proposed and a novel visual search re ranking scheme is presented based on the proposed SSLPP/PCC-SSLPP and learning-to-rank algorithms.
- Non-linear perceptual multi-scale network for single image super-resolution23
A novel non-linear perceptual multi- scale module (NLPMSM) is developed to learn more discriminative multi-scale feature correlation by using high-order channel attention mechanism, so as to adaptively extract image features at different scales.
- Small and Dense Commodity Object Detection with Multi-Scale Receptive Field Attention23
A one-stage detecting framework, dubbed CommodityNet, is proposed that achieves a superior performance on small dense object detection and a new loss function named ω-focal loss is presented, which significantly improves the detection accuracy of the categories with few objects.
- Fusion-Attention Network for person search with free-form natural language23
This work develops an end-to-end fusion-attention structure, called Description-Strengthened Fusion-Attention Network (DSFA-Net) to tackle the challenging task of searching persons from surveillance videos or large scale image dataset.
- Dual Contrastive Network for Few-Shot Remote Sensing Image Scene Classification22
A transfer-based dual contrastive network (DCN) is proposed, which incorporates two auxiliary supervised contrastive learning branches during the training process, which focus on interclass discriminability and intraclass invariance.
- Multi-video summarization with query-dependent weighted archetypal analysis22
This work develops a Multi-Video Summarization via Multi-modal Weighted Archetypal Analysis (MVS-MWAA) method to extract a concise summarization that is both representative and informative and presents a Ranking from Bottom to Top (RBT) approach to make it understandable.
- Balance between object and background: Object-enhanced features for scene image classification22
An unsupervised object-enhanced feature generation mechanism is proposed, which balances the different effects of object regions and background regions for scene image classification and can be easily embedded in the extraction process of existing prominent histogram-based feature representations.
- Eliminate Before Align: A Remote Sensing Image-Text Retrieval Framework with Keyword Explicit Reasoning21
This work devise an innovative Eliminate Before Align (EBA) strategy to filter out the weakly correlated sample pairs to mitigate their deviations from optimal embedding space during alignment, and introduces a Keyword Explicit Reasoning (KER) module to facilitate the positive role of subtle key concept differences.
- NTK-Guided Few-Shot Class Incremental Learning21
A novel conceptualization of anti-amnesia in terms of mathematical generalization, leveraging the Neural Tangent Kernel (NTK) perspective is presented, ensuring optimal NTK convergence and minimizing NTK-related generalization loss, which serve as the theoretical foundation for cross-task generalization.
- Coordinating Experience Replay: A Harmonious Experience Retention approach for Continual Learning21
A Coordinating Experience Replay approach consisting of a Harmonious Experience Retention approach and a Vital Exemplar Reserved Sampling strategy, which constrains the rehearsal process and develops two criteria to select examples for better replaying the beneficial experience is proposed.
- 20
- Dual triplet network for image zero-shot learning20
A novel deep metric learning framework called Dual-Triplet Network (DTNet) for image ZSL, which first projects the semantic information into the visual space with a mapping network and then employs two triplet networks for learning the visual-semantic mapping.
- 19
- Tolerant Self-Distillation for image classification19
Two metrics from the intra-class distribution of correct-predicted and incorrect-predicted samples are introduced to provide a new perspective on the overfitting issue and a knowledge distillation approach without pretraining a teacher model in advance named Tolerant Self-Distillation (TSD) is proposed for alleviating the overfitting issue.
- Hierarchical matching and reasoning for multi-query image retrieval19
A novel Hierarchical Matching and Reasoning Network (HMRN) for MQIR is proposed, which substantially surpasses the current state-of-the-art methods on the benchmark dataset.
- 19
- 19
- Zero-shot classification with unseen prototype learning19
This paper proposes a novel Unseen Prototype Learning (UPL) model, which is a simple yet effective framework to learn visual prototypes for unseen categories from the corresponding class-level semantic information, and the learned features are treated as latent classifiers directly.
- A decadal survey of zero-shot image classification18
The significance and practical application value of zero-shot image classification and the relation between zero-shot image classification and other related techniques are introduced.
- Dual-Path in Dual-Path Network for Single Image Dehazing18
A DPDP-Net (Dual-Path in Dual-Path network) framework by employing a hierarchical dual path network that achieves competitive performance against the state-of-the-art methods on both synthetic and real-world images.
- Rank canonical correlation analysis and its application in visual search reranking18
Experimental results show that Rank-CCA is superior to standard CCA and semi-supervised CCA (Semi- CCA) algorithm, and achieves comparable performance with several state-of-the-art reranking methods while preserving the superiority of low dimensional features.
- A semi-supervised zero-shot image classification method based on soft-target17
A novel Soft-Target Semi-supervised Classification (STSC) model with an autoencoder network is leveraged, where both labeled seen data from the seen categories and unlabeled ancillary data collected from Internet or other datasets are employed as two branches, respectively.
- Cross-modal guidance based auto-encoder for multi-video summarization17
This work employed sparse auto-encoder as the main model to generate the final summarization of Multi-Video Summarization, where the input was the multiple videos, the output was the keyframes set and the CAE-MVS clearly outperforms the state-of-the-art methods.
- Multi-Modal Memory Enhancement Attention Network for Image-Text Matching17
This paper proposes to achieve the fine-grained visual-textual alignment from two aspects: exploiting attention mechanism to locate the semantically meaningful portion and leveraging the memory network to capture the long-term contextual knowledge.
- SGAP-Net: Semantic-Guided Attentive Prototypes Network for Few-Shot Human-Object Interaction Recognition17
A Semantic-Guided Attentive Prototypes Network (SGAP-Net) is proposed to learn a semantic-guided metric space where HOI recognition can be performed by computing distances to attentive prototypes of each class.
- Hierarchical and complementary experts transformer with momentum invariance for image-text retrieval16
- Boosting Video-Text Retrieval with Explicit High-Level Semantics16
A novel visual-linguistic aligning model named HiSE for VTR is proposed, which improves the cross-modal representation by incorporating explicit high-level semantics and achieves the superior performance over state-of-the-art methods on three benchmark datasets, including MSR-VTT, MSVD and DiDeMo.
- 16
- 15
- Heterogeneous memory enhanced graph reasoning network for cross-modal retrieval15
This paper proposes a heterogeneous memory enhanced graph reasoning network, named HMGR, to connect the semantic correlations between vision and language, and designs a novel dual-path network architecture to generate relationship enhanced global representations by employing modality-specific graph reasoning on extracted local features for each instance.
- 15
- 15
- Stacked squeeze-and-excitation recurrent residual network for visual-semantic matching15
A Stacked Squeeze-and-Excitation Recurrent Residual Network (SER2-Net) for visual-textual matching is introduced and a novel objective namely Cross-modal Semantic Discrepancy (CMSD) loss, which is characterized by exploiting the interdependency among different semantic levels to narrow the cross- modal distribution discrepancy is proposed.
- Semantic softmax loss for zero-shot learning15
This letter proposes a nonlinear approach to impose ZSL as a multi- class classification problem via a Semantic Softmax Loss by embedding the class semantic descriptors into the softmax layer of multi-class classification network.
- Zero-shot learning with regularized cross-modality ranking15
This paper employs the hinge ranking loss to exploit the structures among different modalities and devise efficient regularizers to constrain the variation of the samples in the identical modality.
- 15
- 15
- 15
- 14
- 14
- 14
- Zero-shot learning with Multi-Battery Factor Analysis14
This paper develops an algorithm named Multi-Battery Factor Analysis (MBFA) to build a unified semantic space, and then employs multiple types of side information in it to achieve the ZSL.
- 13
- 13
- 12
- 11
- 11
- Scene image classification with biased spatial block and pLSA11
A novel scene image classification method based on biased spatial block information and an improved coding approach in bag-of-visual-words (BOW) model is proposed to achieve better discrimination power for image classification.
- 10
- 10
- 10
- Modality-experts coordinated adaptation for large multimodal models10
A modality-experts coordinated adaptation (ModeX) method for the multimodal domain is proposed, offering an effective, plug-and-play, and lightweight adaptation architecture for diverse LMMs, which demonstrates superior stability and offers higher training efficiency, both in terms of training parameters and training duration.
- 10
- Ranking Fisher discriminant analysis10
A novel dimensionality reduction algorithm based on Fisher discriminant analysis is proposed, in which relevance degree information—the data label in ranking applications, is introduced to a semi-supervised form of FDA, inWhich both local information and unlabeled data are employed.
- 9
- Uncertainty-aware enhanced dark experience replay for continual learning9
A novel sam- pling strategy dubbed Uncertainty-Aware Sampling (UAS) strategy, which employs model and data uncertainties as criteria to select samples that are stable to the model and have low noise for rehearsal, and a dual network to acquire new knowledge while maintaining old knowledge is designed.
- 9
- 9
- Conformal Loss-Controlling Prediction8
This work proposes a learning framework named conformal loss-controlling prediction, which extends CP to the situation where the value of a loss function needs to be controlled, and focuses on the loss for any test object.
- On the Approximation Risk of Few-Shot Class-Incremental Learning8
This work thoroughly explore the approximation risk of FSCIL, encompassing both transfer and consistency risks, and derives practical guidelines for designing and training FSCIL models.
- Masked Feature Generation Network for Few-Shot Learning8
This paper presents a feature-augmentation approach called Masked Feature Generation Network (MFGN) for Few-Shot Learning (FSL), a challenging task that attempts to recognize the novel classes with a few visual instances for each class.
- 7
- 7
- Spatial attention-guided deformable fusion network for salient object detection7
A spatial-attention-guided deformable fusion network for salient object detection, which consists of a contour enhancement module (CEM), an SADFM and a gate module (GM), aiming to reduce the loss of high-level features in the transfer process.
- 7
- Visual search reranking with RElevant Local Discriminant Analysis7
The proposed RELDA algorithm preserves the local manifold structure of the whole data as well as controls the relevance between labeled examples and has an analytic form of the globally optimal solution and can be computed based on eigen-decomposition.
- Relevance and irrelevance graph based marginal Fisher analysis for image search reranking7
A novel "Ranking Dimensionality Reduction" scheme specifically designed for learning-to-rank based image ranking, which aims at not only discovering the intrinsic structure of data but also keeping the ordinal information in ranking and retrieval is presented.
- 7
- 7
- Parameter-Efficient Fine-Tuning for Continual Learning: A Neural Tangent Kernel Perspective6
This work recast the challenge of test-time forgetting into the quantifiable generalization gaps during training, identifying three key factors that influence these gaps and the performance of PEFT-CL: training sample size, task-level feature orthogonality, and regularization.
- Raformer: Redundancy-Aware Transformer for Video Wire Inpainting6
The research proposes the Redundancy-Aware Transformer (Raformer) method, which employs a novel strategy to selectively bypass redundant parts, such as static background segments devoid of valuable information for inpainting, and outperforms other state-of-the-art methods.
- 6
- 6
- 6
- 6
- Single underwater image restoration based on adaptive color correction and adaptive transmission fusion6
Experimental results show that the proposed method can effectively remove color cast and improve contrast and visibility, especially for underwater images with high turbidity or artificial light sources.
- 6
- Interpretable Few-Shot Image Classification via Prototypical Concept-Guided Mixture of LoRA Experts5
A Few-Shot Prototypical Concept Classification (FSPCC) framework that systematically mitigates two key challenges under low-data regimes: parametric imbalance and representation misalignment and introduces a geometry-aware concept discrimination loss that enforces orthogonality among concepts, paving the way for more transparent visual recognition systems.
- 5
- 5
- Dual Distillation Discriminator Networks for Domain Adaptive Few-Shot Learning5
This work proposes Dual Distillation Discriminator Networks (D3Net) from the perspective of the lack of labeled target domain style samples in DA-FSL and realizes the distribution alignment between the source domain and the target domain and constraints the FSL task distribution by prototype distributions on the mixed domain.
- Balancing Feature Alignment and Uniformity for Few-Shot Classification5
This paper proposes a unified framework that perturbs the feature embedding space using two low-bias estimators and effectively combines knowledge distillation between class-wise pairs and enlarges the diversity in feature representations.
- COREN: Multi-Modal Co-Occurrence Transformer Reasoning Network for Image-Text Retrieval5
This paper proposes a unified multi-modal Co-Occurrence transformer Reasoning Network, dubbed as COREN, to comprehensively discover the semantic correlations of the two modalities and proposes a novel Adaptive Similarity Aggregation (ASA) module to achieve a more accurate cross- modal similarity measurement.
- A Novel Fusion Forecast Model for Hail Weather in Plateau Areas Based on Machine Learning5
Experimental results show that the fusion strategy is particularly notable for reducing the number of false alarms on the premise of ensuring the hit rate, and sheds new light on using Bayesian classification approaches for high-dimensional solutions.
- Detail-preserving single nighttime image dehazing5
An effective single nighttime image haze removal approach based on multiscale fusion that recovers a high-quality, haze-free, and noise-free image with vivid details.
- Multi-Modality Adversarial Auto-Encoder for Zero-Shot Learning5
An auto-encoder framework paired with multi-modality adversarial networks respectively for visual and semantic modalities to reinforce the visual-semantic interactions with a bidirectional alignment, which ensures the generated visual features to fit the real visual distribution and to be highly related to the semantics.
- Manifold embedding for zero-shot recognition5
Two manifold embedding-based ZSR approaches are proposed to capture the intrinsic structures of both the visual space and the class semantic space, i.e., ME-ZSR and MCCA- ZSR, which build embedding from visual space to semantic space and to embed both visual and semantic features into a common space.
- A Fresh Look at Generalized Category Discovery Through Non-Negative Matrix Factorization4
A Non-Negative Generalized Category Discovery (NN-GCD) framework is proposed, which outperforms state-of-the-art methods on GCD benchmarks, and introduces a hybrid sparse regularization approach to impose sparsity constraints.
- 4
- Multi-feature self-attention super-resolution network4
A multi-feature self-attention super-resolution network (MFSN) to embed multi-scale encoding information into the attention mechanism and demonstrates the superiority of the MFSN network in terms of both quantitative metrics and visual quality.
- 4
- Zero-Shot Learning with Deep Canonical Correlation Analysis4
A novel deep embedding model for ZSL is proposed, which formulates the embedding space with Deep Canonical Correlation Analysis (DCCA) and transforms the side information and the visual representation via two independent deep neural networks, and then they are highly linearly correlated in the final output layer.
- 3
- 3
- Multi-task classification network for few-shot learning2
A Multi-task Classification Network (MCN) is proposed to decompose the current classification problem into a image-image classification problem and a semantic-image classification problem and an Uncertainty-Aware Decision Module (UADM) is introduced which biases the final classification result towards the result with lower uncertainty in the two types of classification.
- Relevance equilibrium network for cross-domain few-shot learning2
This work introduces a novel Relevance Equilibrium Network (ReqNet) to enhance the generalizability of few-shot models on target domain tasks and designs a Style Augmentation module to diversify low-level visual styles of feature representations, alleviating the model’s over-sensitivity to class- or task-irrelevant changes.
- Collaborative Global-Local Networks for Fine-Grained Few-shot Learning2
This paper proposes a Collaborative Global-Local Network (CGLNet) to exploit both global and local visual features for few-shot fine-grained classification and shows that the proposed model achieves superior performance to state-of-the-art methods.
- 2
- Image Search Reranking with Semi-supervised LPP and Ranking SVM2
This paper studies the task of image search reranking, and proposes a novel system scheme based on Locality Preserving Projections (LPP) and RankingSVM and improves LPP by incorporating the relevance degree information into it, and names it as semi-supervised LPP (Semi-LPP).
- 2
- Overlaid caption extraction in news video based on SVM2
An effective overlaid caption extraction approach for news video by first scanning the video key frames using a small window, and then classify the blocks into the text and non-text ones via support vector machine (SVM), with statistical features extracted from the gray level co-occurrence matrices.
- Underlying Semantic Diffusion for Effective and Efficient In-Context Learning1
This work proposes Underlying Semantic Diffusion (US-Diffusion), an enhanced diffusion model that improves underlying semantics learning, computational efficiency, and in-context learning capabilities on multi-task scenarios, and introduces Separate & Gather Adapter (SGA), which decouples input conditions for different tasks while sharing the architecture, enabling better in-context learning and generalization across diverse visual domains.
- Radlora: a smart low-rank adaptive approach for radiological image classification1
This work introduces a method that integrates architectural enhancements with strategic regularization and employs proximal gradient descent to enhance optimization efficiency, thereby improving training effectiveness within a fixed architectural framework and rendering RadLoRA more suitable for clinical applications.
- Attribute-wise reasoning reinforcement learning for pedestrian attribute retrieval1
An attribute-wise reasoning reinforcement learning framework (AwRL) is introduced to address the overlooked intra-attribute imbalance problem of pedestrian attribute retrieval, and a novel supervised reinforcement loss function (SR-Loss) is developed to enhance the robustness of the retrieval model.
- Cross-Modal N-Pair Network for Generalized Zero-Shot Learning1
This paper introduces a novel Cross Modal N-Pairs Network (CMNPN), which firstly maps visual features and semantic prototypes into a common space with an embedding network, and then employs two N-pairs networks, VMNPN and SMNPN, to optimize an N-pair loss in both visual and semantic spaces.
- Zero-Shot Leaning with Manifold Embedding1
A Manifold Embedding based ZSL (ME-ZSL) approach is proposed by formulating the manifold structure for the visual to textual embedding with the intra-class compactness, the inter-class separability, and the locality preservation.
- Image classification with saliency region and multi-task learning1
This paper first detects the object region of an image with saliency detection technique, then extracts its visual features, and then cluster them into groups, and finally employ MTL and SVM to build the classifier.
- TJUT-TJU@TRECVID 2011: Surveillance Event Detection1
A spatio-temporal interest point detector is adopted, which explicitly encodes appearance features together with motion information, to extract robust point features in a sliding window and the bag-of-word (BoW) approach is employed.
- Image Search Reranking with Transductive Learning to Rank Framework1
A semi-supervised learning to rank framework based on Ranking SVM to better compute the similarity between images, a new similarity measure algorithm named SM-PCA, which is relying on Principle Component Analysis, is proposed and introduced into semi- supervised learning.
- –
- Probabilistic von Mises–Fisher representation learning for few-shot remote sensing scene classification–
A probabilistic reformulation of FS-RSISC that moves beyond rigid point-based prototypes by modeling each class as a probability distribution over the hyperspherical feature space and adopts the von Mises–Fisher (vMF) distribution to capture both semantic uncertainty and feature diversity.
- Bias Calibration for Semi-Supervised Continual Learning–
A Bias Calibration method based on nearest-neighbor semi-supervised continual learning, which integrates and adapts Confidence-Enhanced Learning and Guided Contrastive Learning, which aims to reduce competition among similar classes and penalizes low-confidence predictions, thereby generating high-confidence pseudo-labels for unlabeled data and mitigating confirmation bias.
- –
- –
- –
- Concept agent network for zero-base generalized few-shot learning–
The Concept Agent Network is proposed, a novel framework that interprets visual features as affinity features, thereby effectively diminishing feature overlap by aggregating feature embeddings of the novel classes according to their similarity with the base classes.
- –
- –
- Cross-Modal Retrieval with Discriminative Dual-Path CNN–
The approach, named Discriminative Dual-Path CNN (DDPC), achieves the state-of-the-art results on the MS-COCO dataset, improving IBTR by 4.9% and TBIR by 5.8% based on Recall@1 on the 5K test set.
- Marginal Fisher Regression Classification for Face Recognition–
The MFRC aims at minimizing the within-class compactness over the between-class separability to find an optimal embedding matrix for the LRC so that the L RC on that subspace achieves a high discrimination for classification.
- Balance between Diversity and Relevance for Image Search Results–
A DIR (DDrank-based Image search Rerank) algorithm, which can enrich the topic coverage while keeping the relevance with minimal impact, is proposed, which is based on vertex reinforced random walk and can automatically balance the relevance and diversity of the top ranked vertices in a principle way.
- –
- Face Auto-Location and Its System Implementation–
A new approach to face auto-location based on traditional face detection and new artificial neural network control strategy is proposed, which is simple and effective and implemented by object-oriented method.
- –
- Video shot classification with concept detection–
This research proposes a novel shot classification algorithm with concept detection for news video programs, in which anchorperson and reporter and monologue shots are distinguished by Conditional Random Fields (CRFs), and the last three categories are picked out by rule-based methods.
- –
- Application of Predictive Diamond Searching for Video Coding–
By introducing the predictive motion vector to motion estimation stage in video coding, predictive diamond searching (PDS) is used to get steady and smooth motion field, making full use of the correlation of adjacent blocks.
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 2026-10-14. 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.