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
- Tat‐Seng ChuaSuggested from co-authorship
Is this you? Claim this profile to confirm or dismiss it.
Works553 from public data
- Looking for the Devil in the Details: Learning Trilinear Attention Sampling Network for Fine-Grained Image Recognition488
TASN consists of a trilinear attention module, which generates attention maps by modeling the inter-channel relationships, an attention-based sampler which highlights attended parts with high resolution, and a feature distiller, which distills part features into an object-level feature by weight sharing and feature preserving strategies.
- Visual-Textual Joint Relevance Learning for Tag-Based Social Image Search465
An approach that simultaneously utilizes both visual and textual information to estimate the relevance of user tagged images is proposed, and the relevance estimation is determined with a hypergraph learning approach.
- 405
- Learning to Assemble Neural Module Tree Networks for Visual Grounding327
A novel modular network called Neural Module Tree network (NMTree) is developed that regularizes the visual grounding along the dependency parsing tree of the sentence, where each node is a neural module that calculates visual attention according to its linguistic feature, and the grounding score is accumulated in a bottom-up direction where as needed.
- Object Relational Graph With Teacher-Recommended Learning for Video Captioning325
This paper proposes an object relational graph (ORG) based encoder, which captures more detailed interaction features to enrich visual representation and designs a teacher-recommended learning method to make full use of the successful external language model (ELM) to integrate the abundant linguistic knowledge into the caption model.
- 305
- Aspect Ranking: Identifying Important Product Aspects from Online Consumer Reviews287
This paper develops an aspect ranking algorithm, which aims to automatically identify important product aspects from online consumer reviews by simultaneously considering the aspect frequency and the influence of consumers' opinions given to each aspect on their overall opinions.
- Towards a new generation of artificial intelligence in China285
This Perspective briefly introduces the New Generation Artificial Intelligence Development Plan of China (2015–2030) from the point of view of the authors, a group of AI experts from academia and industry who have been involved in various stages of the plan.
- 281
- ContourNet: Taking a Further Step Toward Accurate Arbitrary-Shaped Scene Text Detection257
The ContourNet is proposed, which effectively handles the problems taking a further step toward accurate arbitrary-shaped text detection and suppresses false positives by only outputting predictions with high response value in both orthogonal directions.
- 248
- 248
- Disentangle Your Dense Object Detector234
This paper investigates three important conjunctions of dense detectors and proposes Disentangled Dense Object Detector (DDOD), in which simple and effective disentanglement mechanisms are designed and integrated into the current state-of-the-art dense object detectors.
- Parsing-Based View-Aware Embedding Network for Vehicle Re-Identification229
This paper proposes a parsing-based view-aware embedding network (PVEN), which helps capture the stable discriminative information of vehicle under different views and design a common-visible attention to focus on the common visible views, which not only shortens the distance among intra-instances, but also enlarges the discrepancy of inter-instance.
- Mining Travel Patterns from Geotagged Photos229
This study aims to leverage the wealth of these enriched online photos to analyze people’s travel patterns at the local level of a tour destination by building a statistically reliable database of travel paths from a noisy pool of community-contributed geotagged photos on the Internet.
- Group-aware Label Transfer for Domain Adaptive Person Re-identification217
A group-aware Label Transfer (GLT) algorithm is proposed, which enables the online interaction and mutual promotion of pseudo-label prediction and representation learning and can better correct the noisy pseudo label in an online fashion and narrow down the search space of the target identity.
- Self-Sustaining Representation Expansion for Non-Exemplar Class-Incremental Learning212
A novel self-sustaining representation expansion scheme that fuses main-br branch expansion and side-branch updating to maintain the old features and transfer the invariant knowledge to the new classes is proposed.
- 203
- Visual query suggestion198
This paper proposes a new query suggestion scheme named Visual Query Suggestion (VQS), which provides a more effective query interface to formulate an intent-specific query by joint text and image suggestions, and shows that VQS outperforms these engines in terms of both the quality of query suggestion and search performance.
- Multi-Scale Triplet CNN for Person Re-Identification197
A multi-scale triplet convolutional neural network which captures visual appearance of a person at various scales is proposed, addressing the problem of small training set in person re-identification.
- Filtration and Distillation: Enhancing Region Attention for Fine-Grained Visual Categorization196
A novel “Filtration and Distillation Learning” (FDL) model is proposed to enhance the region attention of discriminate parts for FGVC that utilizes the proposing-predicting matchability as the performance metric of Region Proposal Network (RPN), thus enable a direct optimization of RPN to filtrate most discriminative regions.
- Exploiting Sample Uncertainty for Domain Adaptive Person Re-Identification194
This paper proposes to estimate and exploit the credibility of the assigned pseudo-label of each sample to alleviate the influence of noisy labels, by suppressing the contribution of noisy samples, and introduces uncertainty-guided optimization, which brings significant improvement and achieves the state-of-the-art performance on benchmark datasets.
- Self-Promoted Prototype Refinement for Few-Shot Class-Incremental Learning193
This work proposes a novel incremental prototype learning scheme which consists of a random episode selection strategy that adapts the feature representation to various generated incremental episodes to enhance the corresponding extensibility, and a self-promoted prototype refinement mechanism which strengthens the expression ability of the new classes by explicitly considering the dependencies among different classes.
- 183
- Domain-Aware Visual Bias Eliminating for Generalized Zero-Shot Learning181
A novel Domain-aware Visual Bias Eliminating (DVBE) network is proposed that constructs two complementary visual representations, i.e., semantic-free and semantic-aligned, to treat seen and unseen domains separately and outperforms existing methods by averaged 5.7% improvement.
- Nighttime Dehazing with a Synthetic Benchmark181
A novel synthetic method to simulate nighttime hazy images from daytime clear images, which first reconstructs the scene geometry, then simulates the light rays and object reflectance, and finally renders the haze effects, which demonstrates their superiority over state-of-the-art methods in terms of both image quality and runtime.
- DADNet174
A novel deep model called Dilated-Attention-Deformable ConvNet (DADNet), which consists of two schemes: multi-scale dilated attention and deformable convolutional DME (Density Map Estimation).
- Camera Lens Super-Resolution170
This paper investigates SR from the perspective of camera lenses, named as CameraSR, which aims to alleviate the intrinsic tradeoff between resolution (R) and field-of-view (V) in realistic imaging systems and quantitatively analyzes the performance of commonly-used synthetic degradation models.
- Rain Streak Removal via Dual Graph Convolutional Network168
This work proposes a simple yet effective dual graph convolutional network (GCN) for single image rain removal that is easy to implement, end-to-end trainable and computationally efficient and achieves significant improvements over the recent state-of-the-art methods.
- Graph-based semi-supervised learning with multiple labels164
A novel graph-based learning framework in the setting of semi-supervised learning with multiple labels is proposed, characterized by simultaneously exploiting the inherent correlations among multiple labels and the label consistency over the graph.
- 163
- State-Relabeling Adversarial Active Learning150
This paper proposes a state relabeling adversarial active learning model (SRAAL), that leverages both the annotation and the labeled/unlabeled state information for deriving the most informative unlabeled samples.
- Context-Aware Visual Policy Network for Fine-Grained Image Captioning149
A Context-Aware Visual Policy network (CAVP) is proposed for fine-grained image-to-language generation: image sentence captioning and image paragraph captioning, which can attend to complex visual compositions over time.
- Attribute-augmented semantic hierarchy147
Experimental results show that the proposed A2 SH can characterize the semantic affinities among images accurately and can shape user search intent precisely and quickly, leading to more accurate search results as compared to state-of-the-art CBIR solutions.
- 144
- Oracle in Image Search144
This article demonstrates that, in order to estimate the mathematical expectations of Average Precision (AP) and Normalized Discounted Cumulative Gain (NDCG), it only need to predict the relevance probability of each image.
- 144
- 136
- 131
- Structure-Aware Residual Pyramid Network for Monocular Depth Estimation131
A Structure-Aware Residual Pyramid Network (SARPN) to exploit multi-scale structures for accurate depth prediction and an Adaptive Dense Feature Fusion (ADFF) module, which adaptively fuses effective features from all scales for inferring structures of each scale, is introduced.
- 128
- Multimedia answering127
This paper introduces a scheme that is able to enrich text answers with image and video information, built based on the community-contributed textual answers and thus it is more feasible and able to deal with more complex questions.
- Progressive Retinex125
A novel progressive Retinex framework is presented, in which illumination and noise of low-light image are perceived in a mutually reinforced manner, leading to noise reduction low- light enhancement results.
- Adaptive Frequency Filters As Efficient Global Token Mixers124
This work applies conventional convolution theorem to deep learning and reveals that adaptive frequency filters can serve as efficient global token mixers, and takes AFF token mixers as primary neural operators to build a lightweight neural network, dubbed AFFNet.
- 120
- An efficient sparse metric learning in high-dimensional space via l 1 -penalized log-determinant regularization119
An efficient sparse metric learning algorithm in high dimensional space via an l1-penalized log-determinant regularization that can be efficiently optimized in a block coordinate descent fashion which is much faster than the standard semi-definite programming which has been widely adopted in many other advanced distance learning algorithms.
- 118
- 117
- 115
- 114
- 113
- Deep Adversarial Graph Attention Convolution Network for Text-Based Person Search109
A novel deep adversarial graph attention convolution network (A-GANet) for text-based person search that learns an effective joint textual-visual latent feature space in adversarial learning manner, bridging modality gap and facilitating pedestrian matching.
- 108
- Connectionist Temporal Fusion for Sign Language Translation108
A hybrid deep architecture which consists of a temporal convolution module (TCOV), a bidirectional gated recurrent unit module (BGRU), and a fusion layer module (FL) to address the CSLT problem and achieves comparable performance to other existing methods with multiple EM iterations.
- PRRNet: Pixel-Region relation network for face forgery detection106
A novel network, called Pixel-Region Relation Network (PRRNet), to capture pixel-wise and region-wise relations respectively for face forgery detection, which achieves the state-of-the-art detection performance on three recent proposed face forgeries detection datasets.
- Hierarchical Gumbel Attention Network for Text-based Person Search106
A novel hierarchical Gumbel attention network for text-based person search via Gumbels top-k re-parameterization algorithm that adaptively selects the strong semantically relevant image regions and words/phrases from images and texts for precise alignment and similarity calculation.
- Context-Aware Visual Policy Network for Sequence-Level Image Captioning104
A Context-Aware Visual Policy network (CAVP) for sequence-level image captioning that explicitly accounts for the previous visual attentions as the context, and then decides whether the context is helpful for the current word generation given the current visual attention.
- 103
- Research and applications on georeferenced multimedia: a survey103
A comprehensive survey on recent research and applications on online georeferenced media based on the current technical achievements, open research issues and challenges are identified, and directions that can lead to compelling applications are suggested.
- Illumination-Invariant Person Re-Identification100
This work proposes a illumination-invariant person re-identification framework which is able to simultaneously achieve Retinex illumination decomposition and personRe-id, and designs a bottom-up attention network to remove the effect of weak illumination and obtain the enhanced image without introducing over-enhancement.
- 98
- Adaptive Reconstruction Network for Weakly Supervised Referring Expression Grounding98
A novel end-to-end adaptive reconstruction network (ARN) that builds the correspondence between image region proposal and query in an adaptive manner: adaptive grounding and collaborative reconstruction.
- Few Shot Generative Model Adaption via Relaxed Spatial Structural Alignment96
This work proposes a relaxed spatial structural alignment (RSSA) method to calibrate the target generative models during the adaption and designs a cross-domain spatial structural consistency loss comprising the self-correlation and disturbance correlation consistency loss.
- Multi-Objective Matrix Normalization for Fine-Grained Visual Recognition96
Experiments demonstrate that the proposed Multi-Objective Matrix Normalization (MOMN) method is superior to existing normalization-based methods in terms of both accuracy and efficiency.
- I2V-Adapter: A General Image-to-Video Adapter for Diffusion Models94
I2V-Adapter adeptly propagates the unnoised input image to subsequent noised frames through a cross-frame attention mechanism, maintaining the identity of the input image without any changes to the pretrained T2V model.
- The Ninth NTIRE 2024 Efficient Super-Resolution Challenge Report91
This paper provides a comprehensive review of the NTIRE 2024 challenge, focusing on efficient single-image super-resolution (ESR) solutions and their outcomes, and gauge the state-of-the-art in efficient single-image super-resolution.
- Grounding 3D Object Affordance from 2D Interactions in Images90
A novel Interaction-driven 3D Affordance Grounding Network (IAG), which aligns the region feature of objects from different sources and models the interactive contexts for 3D object affordance grounding and collects a Point-Image Affordance Dataset (PIAD) to support the proposed task.
- Domain-Assisted Product Aspect Hierarchy Generation: Towards Hierarchical Organization of Unstructured Consumer Reviews90
A domain-assisted approach to organize various aspects of a product into a hierarchy by integrating domain knowledge, as well as consumer reviews, and applies the hierarchy to the task of implicit aspect identification.
- Pose-Guided Feature Learning with Knowledge Distillation for Occluded Person Re-Identification87
A network named Pose-Guided Feature Learning with Knowledge Distillation (PGFL-KD), where the pose information is exploited to regularize the learning of semantics aligned features but is discarded in testing to achieve high accuracy while preserving low inference complexity.
- A Two-Stream Mutual Attention Network for Semi-Supervised Biomedical Segmentation with Noisy Labels87
A Two-Stream Mutual Attention Network (TSMAN) that weakens the influence of back-propagated gradients caused by incorrect labels, thereby rendering the network robust to unclean data.
- 85
- 85
- Efficient Model-Driven Network for Shadow Removal84
This work proposes a new shadow illumination model, which ensures the identity mapping among unshaded regions, and adaptively performs fine grained spatial mapping between shadow regions and their references, and reformulates the shadow removal task as a variational optimization problem.
- Real-World Person Re-Identification via Degradation Invariance Learning84
By introducing a self-supervised disentangled representation learning strategy, this framework is able to simultaneously extract identity-related robust features and remove real-world degradations without extra supervision and is ableTo achieve state-of-the-art performance on several Re-ID benchmarks.
- Dense 3D-Convolutional Neural Network for Person Re-Identification in Videos84
A Dense 3D-Convolutional Network (D3DNet) is proposed to jointly learn spatio-temporal and appearance representation for person re-identification in videos to address the challenge of large intra- class variance and small inter-class variance.
- Convolutional Attention Networks for Scene Text Recognition81
The experimental results validate the effectiveness of different components and show that the convolutional-based method achieves state-of-the-art or competitive performance over prior works, even without the use of RNN.
- Spatial-Temporal Correlation and Topology Learning for Person Re-Identification in Videos80
A novel Spatial-Temporal Correlation and Topology Learning framework is proposed to pursue discriminative and robust representation by modeling cross-scale spatial-temporal correlation by utilizing a CNN backbone and a key-points estimator to extract semantic local features from human body at multiple granularities as graph nodes.
- Learning to Discretely Compose Reasoning Module Networks for Video Captioning79
A novel visual reasoning approach for video captioning, named Reasoning Module Networks (RMN), to equip the existing encoder-decoder framework with the above reasoning capacity, and employs three sophisticated spatio-temporal reasoning modules.
- Robust Deep Co-Saliency Detection with Group Semantic79
A novel end-to-end deep learning approach for robust co-saliency detection by simultaneously learning highlevel group-wise semantic representation as well as deep visual features of a given image group by exploiting the inter-image interaction at semantic-level and the complementarity between group semantics and visual features.
- Entity-enhanced Adaptive Reconstruction Network for Weakly Supervised Referring Expression Grounding77
Qualitative results demonstrate that the proposed EARN can better handle the situation where multiple objects of a particular category are situated together and outperforms existing state-of-the-art methods.
- Learning Conditional Knowledge Distillation for Degraded-Reference Image Quality Assessment77
This paper proposes a practical solution named degraded-reference IQA (DR-IQA), which exploits the inputs of IR models, degraded images, as references to extract reference information from degraded images by distilling knowledge from pristine-quality images.
- 76
- Temporal Hierarchical Attention at Category- and Item-Level for Micro-Video Click-Through Prediction74
This paper investigates how to model user's historical behaviors so as to predict the user's click-through of micro-videos, and proposes a Temporal Hierarchical Attention at Category- and Item-Level (THACIL) network for user behavior modeling.
- LinesToFacePhoto73
This paper introduces a conditional self-attention mechanism to cGANs to capture long-range dependencies between different regions in faces, and builds a multi-scale discriminator that enforces the completeness of global structures and encourages fine details, thereby enhancing the realism of generated face images.
- Learning Compact Appearance Representation for Video-Based Person Re-Identification73
This paper presents a novel approach for video-based person re-identification using multiple convolutional neural networks (CNNs) to extract a compact yet discriminative appearance representation from several frames rather than the whole sequence.
- 72
- 72
- 71
- 71
- 71
- Dual Path Interaction Network for Video Moment Localization70
A unified top-down and bottom-up approach to moment localization called Dual Path Interaction Network (DPIN), where the alignment and discrimination information are closely connected to jointly make the prediction.
- 70
- Knowledge-guided Pairwise Reconstruction Network for Weakly Supervised Referring Expression Grounding69
A knowledge-guided pairwise reconstruction network (KPRN) is proposed, which models the relationship between the target entity (subject) and contextual entity (object) as well as grounds these two entities.
- 68
- 66
- 66
- 66
- 65
- Training Spiking Neural Networks with Accumulated Spiking Flow65
This paper proposes a new backpropagation method for SNNs based on the accumulated spiking flow (ASF), i.e. ASF-BP, and presents an adaptive linear estimation model to approach the dynamic characteristics of spiking neurons statistically.
- Rethinking Graph Neural Architecture Search from Message-passing65
The proposed Graph Neural Architecture Search (GNAS) can automatically learn better architecture with the optimal depth of message passing on the graph and achieves remarkable improvement over state-of-the-art manual designed and search-based GNNs on five large-scale datasets at three classical graph tasks.
- Learning and Fusing Multiple User Interest Representations for Micro-Video and Movie Recommendations65
- Optimizing Large Language Model Training Using FP4 Quantization64
This work introduces the first FP4 training framework for LLMs, addressing two key challenges with two key innovations: a differentiable quantization estimator for precise weight updates and an outlier clamping and compensation strategy to prevent activation collapse.
- 63
- Attention and Language Ensemble for Scene Text Recognition with Convolutional Sequence Modeling63
An attention-based architecture which is completely based on CNNs, and instead of using a single loss from language aspect, multiple losses from attention and language are accumulated for training the networks in an end-to-end way.
- Towards Realistic Data Generation for Real-World Super-Resolution62
A novel Realistic Decoupled Data Generator (RealDGen), an unsupervised learning data generation framework designed for real-world super-resolution, which excels in generating large-scale, high-quality paired data that mirrors real-world degradations.
- Revisiting Single Image Reflection Removal in the Wild62
An advanced reflection collection pipeline that is highly adaptable to a wide range of real-world reflection scenarios and incurs reduced costs in collecting large-scale aligned reflection pairs is devised, and a reflection location-aware cascaded framework specifically tailored for SIRR is designed.
- 62
- Robust Distance Metric Learning with Auxiliary Knowledge ∗62
Two algorithms are derived on the basis of manifold regularization and log-determinant divergence regularization technique, respectively, which can simultaneously exploit label information, unlabeled examples, and the metrics derived from auxiliary data sets and are efficient and flexible.
- FourierMamba: Fourier Learning Integration with State Space Models for Image Deraining61
FourierMamba, a new framework that correlates Fourier space information in the spatial and channel dimensions with distinct designs, and can directly use Mamba to perform frequency correlation and improve the channel information representation.
- LEMON: Learning 3D Human-Object Interaction Relation from 2D Images61
LEMON (LEarning 3D huMan-Object iNteraction relation), a unified model that mines interaction intentions of the counter-parts and employs curvatures to guide the extraction of ge-ometric correlations, combining them to anticipate the interaction elements.
- 61
- 61
- 59
- Deep Residual Attention Network for Spectral Image Super-Resolution59
A novel deep residual attention network is proposed for the spatial super-resolution (SR) of spectral images and is one of the winning solutions in the PIRM 2018 Spectral Super-resolution Challenge.
- GPSView59
This study proposes an augmented GPS navigation system, GPSView, to incorporate a scenic factor into the routing to allow travelers to enjoy sightseeing on the drive, and builds a database of scenic roadways with vistas of landscapes and sights along the roadside.
- 58
- 58
- 57
- Self-Supervised Visual Representations Learning by Contrastive Mask Prediction57
A novel contrastive mask prediction (CMP) task for visual representation learning and a mask contrast (MaskCo) framework to implement the idea, which provides a promising alternative to the ID-based methods for self-supervised learning in the wild.
- 57
- Memory-Augmented Relation Network for Few-Shot Learning57
This work investigates a new metric-learning method to explicitly exploit its relationships with the others in the working context, and formulate the distance metric as a learnable relation module which learns to compare for similarity measurement, and equip theWorking context with memory slots, both contributing to generality.
- Towards Human-Level License Plate Recognition57
A novel LPR framework consisting of semantic segmentation and character counting, towards achieving human-level performance significantly outperforms the previous state-of-the-art methods, and achieves the accuracies of more than 99% for almost all settings.
- Product comparison using comparative relations57
Experiments on popular electronic products show that the proposed approach outperforms the state-of-the-art methods.
- DeepFacePencil56
DeepFacePencil is proposed, an effective tool that is able to generate photo-realistic face images from hand-drawn sketches, based on a novel dual generator image translation network during training, designed to adaptively handle stroke distortions which are spatially varying to support various stroke styles and different level of details.
- 56
- Improved Video VAE for Latent Video Diffusion Model55
A keyframe-based temporal compression (KTC) architecture and a group causal convolution (GCConv) module to further improve video VAE (IV-VAE) and demonstrate the SOTA video reconstruction and generation abilities of the IV-VAE.
- Successive Graph Convolutional Network for Image De-raining55
This paper proposes a graph convolutional networks (GCNs)-based model and introduces a simple yet effective recurrent operations to perform the de-raining process in a successive manner and achieves state-of-the-art results on both synthetic and real-world data sets.
- Iterative Context-Aware Graph Inference for Visual Dialog55
This work proposes a novel Context-Aware Graph (CAG) neural network, where each node in the graph corresponds to a joint semantic feature, including both object-based (visual) and history-related (textual) context representations.
- Learning to Dub Movies via Hierarchical Prosody Models54
This work proposes a novel movie dubbing architecture to tackle these problems via hierarchical prosody modeling, which bridges the visual information to corresponding speech prosody from three aspects: lip, face, and scene.
- 54
- Space-Time Video Super-Resolution Using Temporal Profiles54
A novel space-time video super-resolution method, which aims to recover a high-frame-rate and high-resolution video from its low- frame- rate and low-resolution observation, followed by a refining module for artifacts alleviation and detail enhancement, which does not require any explicit or implicit motion estimation.
- A generalized least-squares approach regularized with graph embedding for dimensionality reduction54
The PCA (Principal Component Analysis) idea of minimizing least-squares reconstruction errors is regularized with graph embedding, to unify various local manifold embedding methods within a generalized framework to keep global and local low dimensional subspace.
- MSRA-USTC-SJTU AT TRECVID 2007: HIGH-LEVEL FEATURE EXTRACTION AND SEARCH54
This paper describes the MSRA-USTC-SJTU experiments for TRECVID 2007 and investigates the benefit of unlabeled data by semi-supervised learning, and the multi-layer (ML) multi-instance (MI) relation embedded in video by MLMI kernel, as well as the correlations between concepts by correlative multi-label learning.
- Lifelong Unsupervised Domain Adaptive Person Re-identification with Coordinated Anti-forgetting and Adaptation53
An effective scheme for this task, dubbed CLUDA-ReID, where the anti-forgetting is harmoniously coordinated with the adaptation, and a meta-based Coordinated Data Replay strategy is proposed to replay old data and update the network with a coordinated optimization direction for both adaptation and memorization.
- EMScore: Evaluating Video Captioning via Coarse-Grained and Fine-Grained Embedding Matching53
This work proposes EMScore (Embedding Matching-based score), a novel reference-free metric for video captioning, which directly measures similarity between video and candidate captions and can be flexibly extended to the conditions where human-labeled references are available.
- Multimedia Question Answering53
QA's evolution from text to multimedia and the challenges in the field are identified by surveying recent research in multimedia question answering.
- 52
- 52
- Self-supervised Cross-view Representation Reconstruction for Change Captioning51
A self-supervised cross-view representation reconstruction (SCORER) network that reconstruct the representations of unchanged objects by cross-attention, thus learning a stable difference representation for caption generation and devise a cross-modal backward reasoning to improve the quality of caption.
- 51
- Automatic labeling hierarchical topics51
This paper proposes two effective algorithms that automatically assign concise labels to each topic in a hierarchy by exploiting sibling and parent-child relations among topics and shows that the inter-topic relation is effective in boosting topic labeling accuracy.
- Building a comprehensive ontology to refine video concept detection51
This paper presents a comprehensive representation scheme for video semantic ontology in which all the three components are well studied, and leverage LSCOM to construct the concept lexicon, describe concept property as the weights of different modalities which are obtained manually or by data-driven approach.
- Single Image Shadow Detection via Complementary Mechanism50
A novel shadow detection framework by investigating the mutual complementary mechanisms contained in this specific task, which achieves superior shadow detection performance against state-of-the-art methods with a relatively low computational cost.
- Attribute-assisted reranking for web image retrieval50
This paper proposes to exploit semantic attributes for image search reranking by using a hypergraph to model the relationship between images by integrating low-level visual features and attribute features.
- 50
- Hierarchical Global-Local Temporal Modeling for Video Captioning49
A Hierarchical Temporal Model is proposed for the video captioning task, based on exploring the global and local temporal structure to better recognize fine-grained objects and actions and could generate more accurate descriptions in handling shot-switching problems.
- ViViD: Video Virtual Try-on using Diffusion Models48
This work designs the Garment Encoder to extract fine-grained clothing semantic features, guiding the model to capture garment details and inject them into the target video through the proposed attention feature fusion mechanism, to ensure spatial-temporal consistency.
- Background Activation Suppression for Weakly Supervised Object Localization and Semantic Segmentation48
A background activation suppression (BAS) method designed to facilitate the learning of generator by suppressing the background activation value and which achieves state-of-the-art weakly supervised semantic segmentation performance on the PASCAL VOC 2012 and MS COCO 2014 datasets.
- Spatiotemporal-Textual Co-Attention Network for Video Question Answering48
A novel Spatiotemporal-Textual Co-Attention Network (STCA-Net) for video question answering jointly learns spatially and temporally visual attention on videos as well as textual attention on questions.
- Integrating rich information for video recommendation with multi-task rank aggregation48
A multi-task rank aggregation approach is proposed to integrate the ranking lists for different users in a joint manner and can easily incorporate other methods by adding their generated ranking lists into the multi- task learning algorithm.
- Exploring Fourier Prior for Single Image Rain Removal47
A two-stage model where the first stage restoring the amplitude of rainy images to clean rain streaks, and the second stage restores the phase information to refine fine-grained background structures to improve the generalization ability of an image rain removal model.
- 47
- Human activity recognition by manifold regularization based dynamic graph convolutional networks47
A manifold regularized dynamic graph convolutional network (MRDGCN) that automatically updates the structure information by manifold regularization until model fitting and builds an optimization convolution layer formulation to acquire the optimal structure information.
- 47
- Dense Residual Network: Enhancing global dense feature flow for character recognition46
This paper proposes an effective convolutional deep model termed Dense Residual Network (DRN) for the task of optical character recognition and proposes a refined residual dense block (r-RDB) to retain the ability of local feature fusion and local residual learning of original RDB to enhance the local and global feature learning abilities of DenseNet.
- A robust vision inspection system for detecting surface defects of film capacitors46
A robust vision inspection system for detecting the surface defects of film capacitors using a novel Non-subsampled Contourlet Transform (NSCT) based algorithm, which can improve the detection efficiency and reduce production costs.
- 45
- 45
- 45
- 45
- 44
- A Structured Graph Attention Network for Vehicle Re-Identification44
A Structured Graph ATtention network (SGAT) is proposed to fully exploit these relationships and allow the message propagation to update the features of graph nodes and achieve significant improvements over the state-of-the-art methods.
- 44
- 43
- 42
- 41
- BERT4SessRec41
This paper describes the solution for the Content-Based Video Relevance Prediction (CBVRP) challenge, where the task is to predict user click-through behavior on new TV series or new movies according to the user's historical behavior and uses the Bidirectional Encoder Representations from Transformer methodology.
- Benchmarking Large Vision-Language Models via Directed Scene Graph for Comprehensive Image Captioning40
A detailed caption benchmark, termed as CompreCap, to evaluate the visual context from a directed scene graph view, to assess the generated detailed captions from LVLMs on multiple levels, including the object-level coverage, the accuracy of attribute descriptions, the score of key relationships, etc.
- 40
- Learning Video-Text Aligned Representations for Video Captioning40
Experiments on two public benchmarks, i.e., VATEX and MSR-VTT, demonstrate that the proposed video-text alignment module with a retrieval unit and an alignment unit to learn video- texts aligned representations outperforms state-of-the-art performances by a large margin.
- Attribute feedback40
This work presents a new interactive Content Based Image Retrieval (CBIR) scheme, termed Attribute Feedback, which shapes users' information needs more precisely and quickly by collecting feedbacks on intermediate level semantic attributes.
- k-Partite graph reinforcement and its application in multimedia information retrieval39
This paper proposes a k-partite graph reinforcement approach to fuse these sub-queries based on the to-be-retrieved database and demonstrates that this method effectively boosts retrieval performance.
- Efficient Real-World Image Super-Resolution Via Adaptive Directional Gradient Convolution38
A novel equivalent parameter fusion method for DGConv is devised that maintains its rich representational capabilities while keeping computational costs consistent with a single Vanilla Convolution (VConv), enabling DGConv to improve the performance of existing super-resolution networks without incurring additional computational expenses.
- Event-Based Eye Tracking. AIS 2024 Challenge Survey38
This survey reviews the AIS 2024 Event-Based Eye Tracking (EET) Challenge and examines novel and diverse methods from the submitted factsheets to advance future event-based eye tracking research.
- Leveraging Deep Statistics for Underwater Image Enhancement37
A convolutional neural network (CNN)-based framework to learn hierarchical statistical features related to color cast and contrast degradation and to leverage them for underwater image enhancement is proposed and achieves state-of-the-art performance.
- Boosting Image De-Raining via Central-Surrounding Synergistic Convolution36
Evaluations of twelve de-raining methods on nine public datasets demonstrate that the proposed Synergistic Convolution can comprehensively improve the performance of twelve de-raining networks under various rainy conditions without changing the original network structure or introducing extra computational costs.
- 36
- 35
- Answering Opinion Questions on Products by Exploiting Hierarchical Organization of Consumer Reviews35
A new framework for opinion Questions Answering is developed, which enables accurate question analysis and effective answer generation by making use of the hierarchy of consumer reviews.
- 35
- 34
- 34
- Learning Discriminative Noise Guidance for Image Forgery Detection and Localization33
This study develops a two-step discriminative noise-guided approach to explicitly enhance the representation and use of noise inconsistencies, thereby fully exploiting noise information to improve the accuracy and robustness of forgery detection.
- Spatiotemporal Fusion in 3D CNNs: A Probabilistic View33
This paper proposes to convert the spatiotemporal fusion strategies into a probability space, which allows us to perform network-level evaluations of various fusion strategies without having to train them separately, and generates new fusion strategies which achieve the state-of-the-art performance on four well-known action recognition datasets.
- 33
- Mining Travel Patterns from GPS-Tagged Photos33
This study builds a statistically reliable database of travel paths, and investigates the tourist traffic flow among different RoAs, by exploiting Markov chain model and testings on four major cities demonstrate promising results.
- QMambaBSR: Burst Image Super-Resolution with Query State Space Model32
A novel Query Mamba Burst Super-Resolution (QMambaBSR) network is introduced and an Adaptive Upsampling (AdaUp) module is proposed that dynamically adjusts the upsampling kernel to suit the characteristics of different burst scenes, achieving superior detail reconstruction.
- 32
- ShotTagger32
This paper leverages user's tags as a source to analyze the content within the video and develops a novel system named ShotTagger to assign tags at the shot level, based on a multiple instance learning framework.
- GREAT: Geometry-Intention Collaborative Inference for Open-Vocabulary 3D Object Affordance Grounding31
Great (GeometRy-intEntion collAboraTive inference) is proposed for Open-Vocabulary 3D Object Affordance Grounding, a novel framework that mines the object invariant geometry attributes and performs analogically reason in potential interaction scenarios to form affordance knowledge, fully combining the knowledge with both geometries and visual contents to ground 3D object affordance.
- 31
- 31
- ECENet: Explainable and Context-Enhanced Network for Muti-modal Fact verification31
A novel Explainable and Context-Enhanced Network (ECENet) for multi-modal fact verification is proposed, making the first attempt to integrate multi-clue feature extraction, multi-level feature reasoning, and justification (explanation) generation within a unified framework.
- Towards Generalized UAV Object Detection: A Novel Perspective from Frequency Domain Disentanglement30
This paper proposes a novel frequency domain disentanglement method to improve UAV-OD generalization, which consistently and significantly outperforms baseline and state-of-the-art domain generalization methods.
- 30
- VideoGen-Eval: Agent-based System for Video Generation Evaluation29
VideoGen-Eval is proposed, an agent evaluation system that integrates LLM-based content structuring, MLLM-based content judgment, and patch tools designed for temporal-dense dimensions to achieve a dynamic, flexible, and expandable video generation evaluation.
- 29
- 29
- NTIRE 2025 Challenge on Event-Based Image Deblurring: Methods and Results28
This paper presents an overview of NTIRE 2025, the First Challenge on Event-Based Image Deblurring, detailing the proposed methodologies and corresponding results, and anticipate that this challenge will drive further advancements in event-based vision research.
- NTIRE 2025 Challenge on Raw Image Restoration and Super-Resolution28
The NTIRE 2025 RAW Image Restoration and Super-Resolution Challenge is reviewed, highlighting the proposed solutions and results, and presents the current state-of-the-art in RAW Restoration.
- 28
- JPEG Artifacts Removal via Contrastive Representation Learning28
This work proposes an unsupervised compression quality representation learning strategy for the blind JPEG artifacts removal, which utilizes contrastive learning to obtain discriminative compression quality representations in the latent feature space and integrates them in an information lossless way.
- A Feature-Adaptive Semi-Supervised Framework for Co-saliency Detection28
A novel feature-adaptive semi-supervised (FASS) framework for co-saliency detection, which seamlessly integrates multi-view feature learning, graph structure optimization and co- saliency prediction in a unified solution.
- 27
- Unveiling Hidden Details: A RAW Data-Enhanced Paradigm for Real-World Super-Resolution27
This paper argues that key image processing steps in Image Signal Processing, such as denoising and demosaicing, inherently result in the loss of fine details in LR images, making LR RAW a valuable information source, and pioneers the use of details hidden in RAW data to complement existing RGB-only methods, yielding superior outputs.
- 27
- 27
- Hybrid Image Enhancement With Progressive Laplacian Enhancing Unit27
A novel hybrid network with Laplacian enhancing unit that can adjustably enhance the detail information by adding the residual of previous feature maps is proposed, which achieves superior image enhancement results compared with the state-of-the-arts, while retaining competitive implementation efficiency.
- 26
- 26
- 26
- 26
- Fine-grained Feature Alignment with Part Perspective Transformation for Vehicle ReID26
This paper proposes part perspective transformation module (PPT) to map the different parts of vehicle into a unified perspective respectively and proposes a dynamically batch hard triplet loss to select the common visible regions of the compared vehicles.
- 26
- 26
- 25
- JPEG Compression-aware Image Forgery Localization25
This work introduces a JPEG compression-aware image forgery localization network to flexibly handle forged images compressed with various JPEG quality factors, and introduces a boundary correction branch to alleviate the edge artifacts caused by JPEG compression.
- 25
- Ground-Aware Point Cloud Semantic Segmentation for Autonomous Driving25
The proposed ground-aware attention module captures long-range dependence between ground and objects, which significantly facilitates the segmentation of small objects that only consist of a few points in extremely sparse point clouds.
- 25
- 24
- MambaPupil: Bidirectional Selective Recurrent model for Event-based Eye tracking24
A bidirectional long-term sequence modeling and time-varying state selection mechanism to fully utilize contextual temporal information in response to the variability of eye movements is proposed and achieves state-of-the-art performance.
- Fast and Accurate Electron Microscopy Image Registration with 3D Convolution24
This work proposes an unsupervised deep learning method for serial electron microscopy image registration with fast speed and high accuracy, and introduces the 3D convolution layer to learn the relationship between several adjacent images, which effectively reduces error accumulation in serial EM image registration.
- Extract Bone Parts Without Human Prior: End-to-end Convolutional Neural Network for Pediatric Bone Age Assessment24
This paper proposes a novel single-stage Attention-Recognition Convolutional Neural Network (AR-CNN), believed to be the first end-to-end structure to extract bone parts for BAA without segmentation, detection and human prior.
- PIRM2018 Challenge on Spectral Image Super-Resolution: Methods and Results24
The Perceptual Image Restoration and Manipulation (PIRM) workshop challenge on spectral image super-resolution is described, the methods used by the participants, summarise the results and discuss their rankings.
- CA 3 Net24
A novel Contextual-Attentional Attribute-Appearance Network ($\rm CA^3Net) for person re-identification simultaneously exploits the complementarity between semantic attributes and visual appearance, the semantic context among attributes, visual attention on attributes as well as spatial dependencies among body parts, leading to discriminative and robust pedestrian representation.
- 23
- Domain-Specific Embedding Network for Zero-Shot Recognition23
A novel Domain-Specific Embedding Network (DSEN) that can apply specific projections to different domains for unbiased embedding, as well as several domain constraints.
- Question-Aware Tube-Switch Network for Video Question Answering23
A novel Question- Aware Tube-Switch Network (TSN) for video question answering which contains a Mix module to synchronously combine the appearance and motion representation at time slice level and a Switch module to adaptively choose appearance or motion tube as primary at each reasoning step, guiding the multi-hop reasoning process.
- 23
- Ntire 2025 Image Shadow Removal Challenge Report22
This work examines the findings of the NTIRE 2025 Shadow Removal Challenge, with 17 teams successfully submitting their solutions during the final evaluation phase, and one focusing on reconstruction fidelity and the other on visual perception through a user study.
- 22
- 22
- 22
- Multi-Scale Group Transformer for Long Sequence Modeling in Speech Separation22
This paper proposes a novel variation of Transformer, named multi-scale group Transformer (MSGT), which significantly reduces the complexity, and multi- scale fusion, which retains Transform's ability to capture long-term dependency.
- Hierarchical organization of unstructured consumer reviews22
This paper generates a hierarchical organization of the consumer reviews based on various aspects of the product, and aggregate consumer opinions on the aspects, and demonstrates the effectiveness of this approach.
- AIM 2025 Low-Light RAW Video Denoising Challenge: Dataset, Methods and Results21
This paper introduces a new benchmark of 756 ten-frame sequences captured with 14 smartphone camera sensors across nine conditions, with high-SNR references obtained via burst av-eraging, and describes the dataset, challenge protocol, and submitted approaches.
- Spatial-Aware Token for Weakly Supervised Object Localization21
This work proposes to learn a task-specific spatial-aware token (SAT) to condition localization in a weakly supervised manner and achieves state-of-the-art performance on both CUB-200 and ImageNet, with 98.45% and 73.13% GT-known Loc, respectively.
- Streaming Video Model21
This work proposes to unify video understanding tasks into one novel streaming video architecture, referred to as Streaming Vision Transformer (S-ViT), which produces frame-level features with a memory-enabled temporally-aware spatial encoder to serve the frame-based video tasks.
- Learning Degradation-Invariant Representation for Robust Real-World Person Re-Identification21
A degradation invariance learning framework for robust person re-identification that is able to continuously maintain degradation-invariant priors from one or more datasets to improve the robustness of identity features, achieving state-of-the-art Re-ID performance on several challenging real-world benchmarks with a unified model.
- Mutually Reinforced Spatio-Temporal Convolutional Tube for Human Action Recognition21
A novel and efficient Mutually Reinforced Spatio-Temporal Convolutional Tube (MRST) is proposed for human action recognition that decomposes 3D inputs into spatial and temporal representations, mutually enhances both of them by exploiting the interaction of spatial andporal information and selectively emphasizes informative spatial appearance and temporal motion.
- Learning Multi-view Deep Features for Small Object Retrieval in Surveillance Scenarios21
This paper proposes to leverage discriminative Convolutional Neural Network to learn deep structure and color feature to form an efficient multi-view object representation and utilizes CNN trained on ImageNet to abstract rich semantic structure information.
- Learning concept bundles for video search with complex queries21
The proposed concept bundle learning approach outperforms the state-of-the-art methods by at least 19% and 29% on TRECVID 2008 and YouTube datasets, respectively; and the use of concept bundles can improve the search performance for complex queries by at at least 37.5% on SOTA 2008 and 52% on YouTube datasets.
- MMAR: Towards Lossless Multi-Modal Auto-Regressive Probabilistic Modeling20
A novel Multi-Modal Auto-Regressive (MMAR) probabilistic modeling framework, which disentangle the diffusion process from auto-regressive backbone model by employing a lightweight diffusion head on top each auto-regressed image patch embedding.
- Towards Sequence Modeling Alignment between Tokenizer and Autoregressive Model20
AliTok is a novel Aligned Tokenizer that alters the dependency structure of the token sequence and achieves high fidelity and predictability simultaneously by incorporating prefix tokens and employing a two-stage tokenizer training process.
- ESCNet: Entity-enhanced and Stance Checking Network for Multi-modal Fact-Checking20
This work constructs the first large-scale Chinese Multi-modal Fact-Checking dataset which encompasses 46,000 claims and proposes a novel Entity-enhanced and Stance Checking Network (ESCNet), which jointly models stance semantic reasoning features and knowledge-enhanced entity pair features, in order to simultaneously learn effective semantic-level and knowledge-level claim representations.
- 20
- Synergy between Semantic Segmentation and Image Denoising via Alternate Boosting20
It is observed that not onlyDenoising helps combat the drop of segmentation accuracy due to the noisy input, but also pixel-wise semantic information boosts the capability of denoising.
- 19
- 19
- VisualCritic: Making LMMs Perceive Visual Quality Like Humans19
This paper presents VisualCritic, the first LMM for broad-spectrum image subjective quality assessment, which enables new capabilities of quantitatively measuring the perceptual quality of given images in terms of their Mean Opinion Score, noisiness, colorfulness, sharpness, and other numerical indicators.
- Enhancement by Your Aesthetic: An Intelligible Unsupervised Personalized Enhancer for Low-Light Images19
An intelligible unsupervised personalized enhancer (iUP-Enhancer) for low-light images, which establishes the correlations between the low- light and the unpaired reference images with regard to three user-friendly attributions (brightness, chromaticity, and noise).
- 19
- 19
- JPEG Artifacts Removal via Compression Quality Ranker-Guided Networks19
This work designs a quality ranker to measure the compression degree and proposes compression quality ranker-guided networks to guide de-blocking and perceptual quality optimization, which can implicitly use the information contained in the compression factors to produce better results.
- Cross-Fiber Spatial-Temporal Co-enhanced Networks for Video Action Recognition19
A novel Cross-Fiber Spatial-Temporal Co-enhanced (CFST) architecture aiming to reduce the number of parameters tremendously while achieve accurate recognition of actions is proposed, significantly boosts the performance of existing convolution networks and achieves state-of-the-art accuracy on three challenging benchmarks.
- Cross-Modality Feature Learning via Convolutional Autoencoder19
Compared to the conventional solutions relying on hand-crafted features, the proposed MUCAE approach encodes features from image pixels and text characters directly and produces more representative and robust features.
- MSRA atT TRECVID 2008: High-Level Feature Extraction and Automatic Search.19
This paper describes the MSRA experiments for TRECVID 2008, which representatively investigated the benefit of global and local low-level features by a variety of learning-based methods, including supervised and semi-supervised learning algorithms.
- Learning Semantic-aware Normalization for Generative Adversarial Networks18
This paper proposes a novel image synthesis approach by learning adversarial training, a channel grouping loss, and a mutual information loss for joint optimization, which not only enables high-fidelity image synthesis but leads to superior interpretable properties.
- Inferential Machine Comprehension: Answering Questions by Recursively Deducing the Evidence Chain from Text18
A new network is proposed to solve the inference problem by decomposing it into a series of attention-based reasoning steps, and an operational cell with prior structure is designed to make each step can be directly inferred from the text.
- Exploring the Task Cooperation in Multi-goal Visual Navigation18
This work introduces two new designs to the reinforcement learning scheme: inverse dynamics model (InvDM) and multi-goal co-learning (MgCl), proposed to capture the navigation-relevant association between state and goal, and provide additional training signals to relieve the sparse reward issue.
- 18
- Depth map super-resolution using stereo-vision-assisted model18
A novel Stereo-Vision-Assessment model for depth map super-resolution that includes a non-local prior constraint, a local prior constraint and a stereo disparity prior constraint is proposed.
- Refining video annotation by exploiting pairwise concurrent relation18
This paper proposes to refine video annotation by leveraging the pairwise concurrent relation among video concepts, explicitly modeled by a concurrent matrix and then a propagation strategy is adopted to refine the annotations.
- Event-Based Eye Tracking. 2025 Event-Based Vision Workshop17
This survey serves as a review for the 2025 Event-Based Eye Tracking Challenge organized as part of the 2025 CVPR event-based vision workshop and summarizes the innovative methods from teams rank the top in the challenge.
- SIGMAN: Scaling 3D Human Gaussian Generation with Millions of Assets17
Experimental results demonstrate that the proposed latent space generation paradigm for 3D human digitization, which involves compressing multi-view images into Gaussians via a UVstructured VAE, produces highquality 3D human Gaussians with intricate textures, facialdetails, and loose clothing deformation.
- 17
- Hierarchical Semantic Enhancement Network for Multimodal Fake News Detection17
A novel Hierarchical Semantic Enhancement Network (HSEN) for multimodal fake news detection by learning text-related image semantic and precise news high-order knowledge semantic information to outperform state-of-the-art methods.
- Cross-modal Semantic Alignment Pre-training for Vision-and-Language Navigation17
CSAP is a new method of Cross-modal Semantic Alignment Pre-training for Vision-and-Language Navigation designed to learn the alignment from trajectory-instruction pairs through two novel tasks, including trajectory-conditioned masked fragment modeling and contrastive semantic-alignment modeling.
- 17
- 17
- 17
- Cluster and Scatter: A Multi-grained Active Semi-supervised Learning Framework for Scalable Person Re-identification17
A Multi-grained Active Semi-Supervised learning framework, named MASS, is proposed to address the scalable person re-identification problem existing in the practical scenarios and introduces a customized purification loss for the purified clustering, which utilizes the complementary information in both labeled and unlabeled data to optimize the model for solving the ineffectiveness problem.
- Dual Context-Aware Refinement Network for Person Search17
A novel dual context-aware refinement network (DCRNet) for person search is proposed, which jointly explores two kinds of contexts including intra-instance context and inter- instance context to learn discriminative representation.
- Mediapedia: Mining Web Knowledge to Construct Multimedia Encyclopedia17
A dynamic multimedia encyclopedia that takes advantage of, and in fact is built from the text and image resources on the Web, and is flexible and extensible such that it can easily incorporate new kinds of mediums such as video and languages into the framework.
- HERO: Human Reaction Generation from Videos16
This work presents HERO, a simple yet effective framework for Human rEaction geneRation from videOs that considers both global and frame-level local representations of the video to extract the interaction intention, and then uses the extracted interaction intention to guide the synthesis of the reaction.
- 16
- Exploring Tuning Characteristics of Ventral Stream’s Neurons for Few-Shot Image Classification16
This work computationally model two groups of neurons found in ventral stream which are respectively sensitive to shape cues and color cues and proposes the hierarchical feature regularization method with these neuron models to regularize the backbone of a few- shot model, thus making it produce more generic and robust features for few-shot classification.
- Structural Semantic Adversarial Active Learning for Image Captioning16
A structural semantic adversarial active learning (SSAAL) model that leverages both visual and textual information for deriving the most representative samples while maximizing the image captioning performance is proposed.
- Deep Self-representative Concept Factorization Network for Representation Learning16
DSCF-Net seamlessly integrates the robust deep concept factorization, deep self-expressive representation and adaptive locality preserving feature learning into a unified framework and delivers state-of-the-art performance on several public databases.
- 16
- 16
- LA-Net16
A novel Layout-Aware Convolutional Neural Network (LA-Net) is proposed for accurate monocular depth estimation by simultaneously perceiving scene layout and local depth details and formulate an order-sensitive softmax loss to well constrain the ill-posed depth inferring problem.
- Achieving dynamic load balancing through mobile agents in small world P2P networks16
A new load balancing method in unstructured P2P networks based on mobile agents and resource grouping techniques is introduced, which exhibits robustness and adaptability under external attacking, high query workload, and higher network churns.
- 16
- DemosaicFormer: Coarse-to-Fine Demosaicing Network for HybridEVS Camera15
A coarse-to-fine framework named DemosaicFormer which comprises coarse demosaicing and pixel correction and achieves the best performance in terms of all the evaluation metrics in the MIPI 2024 challenge on Demosaic for Hybridevs Camera.
- 15
- 15
- 15
- 15
- Multimedia encyclopedia construction by mining web knowledge15
A dynamic multimedia encyclopedia that takes advantage of, and in fact is built from the text and image resources on the Web, and is flexible and extensible such that it can easily incorporate new kinds of mediums such as video and languages into the framework.
- Efficient Test-Time Adaptive Object Detection via Sensitivity-Guided Pruning14
This paper introduces a sensitivity-guided channel pruning strategy that quantifies each channel based on its sensitivity to domain discrepancies at both image and instance levels, and applies weighted sparsity regularization to selectively suppress and prune these sensitive channels, focusing adaptation efforts on invariant ones.
- $\text{S}^{3}$Mamba: Arbitrary-Scale Super-Resolution via Scaleable State Space Model14
A Scalable State Space Model (SSSM) is proposed to modulate the state transition matrix and the sampling matrix of step size during the discretization process, achieving scalable and continuous representation modeling with linear computational complexity.
- Event-Driven Heterogeneous Network for Video Deraining14
This work meticulously design a bio-inspired Spiking Neural Network that adapts to the sparse event sequences, capturing features of falling rain and establishes a bimodal feature fusion module that combines dense convolutional features with sparse spiking features.
- Neuromorphic Event Signal-Driven Network for Video De-raining14
This study incorporates the neuromorphic event signal into the video de-raining to enhance the dynamic information perception, and designs an optimization algorithm to solve the objective, and constructs a de-raining network with CNNs as the backbone architecture using a modular strategy to mimic the optimization process.
- 14
- Regularized Mask Tuning: Uncovering Hidden Knowledge in Pre-trained Vision-Language Models14
This work designs a new type of tuning method, termed as regularized mask tuning, which masks the network parameters through a learnable selection, and is synergistic with most existing parameter-efficient tuning methods and can boost the performance on top of them.
- Location-Free Camouflage Generation Network14
A novel Location-free Camouflage Generation Network (LCG-Net) that fuse high-level features of foreground and background image, and generate result by one inference is proposed, which is hundreds of times faster than previous methods.
- Degradation-agnostic Correspondence from Resolution-asymmetric Stereo14
This paper finds that, although a stereo matching network trained with the photometric loss is not optimal, its feature extractor can produce degradation-agnostic and matching-specific features that can be utilized to formulate a feature-metric loss to avoid thePhotometric inconsistency.
- ASTA-Net: Adaptive Spatio-Temporal Attention Network for Person Re-Identification in Videos14
This work proposes a novel Adaptive Spatio-Temporal Attention Network (ASTA-Net) to adaptively aggregate the spatial and temporal attention features into discriminative pedestrian representation for person re-identification in videos.
- 14
- Robust Multiview Feature Learning for RGB-D Image Understanding14
A robust multiview feature learning approach is developed, which exploits the intrinsic relations among multiple views, leading to effective features and making the learning process robust to noises in RGB-D image understanding.
- Interactive social group recommendation for Flickr photos14
An interactive group recommendation framework with Human-in-the-Loop that is able to provide accurate group recommendations with quite a small amount of user efforts is proposed.
- 13
- AIM 2025 challenge on Inverse Tone Mapping Report: Methods and Results13
This paper presents a comprehensive review of the AIM 2025 Challenge on Inverse Tone Mapping, highlighting innovative strategies for enhancing HDR re-construction quality and establishing strong benchmarks to guide future research in inverse tone mapping.
- EventMamba: Enhancing Spatio-Temporal Locality with State Space Models for Event-Based Video Reconstruction13
EventMamba is introduced—a specialized model designed for EBVR task that innovates by incorporating random window offset (RWO) in the spatial domain, moving away from the restrictive fixed partitioning.
- Constructing Spatio-Temporal Graphs for Face Forgery Detection13
A novel forgery detector named Spatio-Temporal Graph Network (STGN) is proposed, which contains two kinds of graph-convolution-based units, the Spatial Relation Graph Unit (SRGU) and the Temporal Attention Graph unit (TAGU), which can combine the discriminative power of spatial inconsistency and the generalization capacity of temporal incoherence for face forgery detection.
- 13
- End-to-end Boundary Exploration for Weakly-supervised Semantic Segmentation13
This paper argues that using a single model to accomplish image- and pixel-level classification will fall into the balance of multi-target and consequently weakens the recognition capability, and proposes an effective encoder-decoder framework to explore object boundaries and solve the above dilemma.
- Frank-Wolfe Network: An Interpretable Deep Structure for Non-Sparse Coding13
An interpretable deep structure namely Frank- Wolfe Network (F-W Net), whose architecture is inspired by unrolling and truncating the Frank-Wolfe algorithm for solving the non-sparse coding problem, is proposed and the hyper-parameter can be made learnable instead of pre-chosen in F-WNet.
- Densely Supervised Hierarchical Policy-Value Network for Image Paragraph Generation13
A novel Densely Supervised Hierarchical Policy-Value (DHPV) network for effective paragraph generation is proposed which exploits compositionality at token-totoken and sentence-to-sentence levels simultaneously and can preserve the semantic and syntactic constituent integrity.
- Multimodal Semantic Attention Network for Video Captioning13
This work proposes a Multimodal Semantic Attention Network (MSAN), which is a new encoder-decoder framework incorporating multimodal semantic attributes for video captioning, and adds auxiliary classification loss to this model that can obtain more effective visual features and high-level multimodAL semantic attribute distributions for sufficient video encoding.
- Linear Distance Preserving Pseudo-Supervised and Unsupervised Hashing13
A novel and general hashing framework, which simultaneously considers a new linear pair-wise distance preserving objective and point-wise constraint, and a pseudo-supervised and unsupervised hashing method, in which quantization loss is considered.
- CCLBR: Congestion Control-Based Load Balanced Routing in Unstructured P2P Systems13
A new churn-resilient system to find alternative routing paths for the purpose of balancing the query loads under higher network churns and heavy workloads, ultimately to improve the search efficiency.
- Text Mining in Multimedia13
This chapter provides a comprehensive survey of recent research efforts on text mining in multimedia, and focuses on four aspects: surrounding text mining; tag mining; joint text and visual content mining; and cross text andVisual content mining.
- 12
- Grounding 3D Scene Affordance From Egocentric Interactions12
The Egocentric Interaction-driven 3D Scene Affordance Grounding (Ego-SAG) framework is proposed, which utilizes interaction intent to guide the model in focusing on interaction-relevant sub-regions and aligns affordance features from different sources through a bidirectional query decoder mechanism.
- 12
- 12
- Weakly Supervised High-Fidelity Clothing Model Generation12
Experiments show that projecting the rough alignment of clothing and body onto the StyleGAN space can yield photo-realistic wearing results, and demonstrate the superiority of DGP over several state-of-the-art supervised methods when generating clothing model images.
- Multifocal Attention-Based Cross-Scale Network for Image De-raining12
An effective multifocal attention-based cross-scale network is constructed, which exhaustively utilizes the cross- scale correlations of both rain streaks and background, to achieve image de-raining.
- Multi-Scale Spatial-Temporal Integration Convolutional Tube for Human Action Recognition12
A novel and efficient Multi-Scale Spatial-Temporal Integration Convolutional Tube (MSTI) aiming at achieving accurate recognition of actions with lower computational cost is proposed and an end-to-end deep network, MSTI-Net, is presented based on the proposed MSTi tube for human action recognition.
- Co-Saliency Spatio-Temporal Interaction Network for Person Re-Identification in Videos12
A novel Co-Saliency Spatio-Temporal Interaction Network (CSTNet) for person re-identification in videos is proposed, which captures the common salient foreground regions among video frames and explores the spatial-temporal long-range context interdependency from such regions, towards learning discriminative pedestrian representation.
- Stacked Convolutional Deep Encoding Network For Video-Text Retrieval12
A stacked convolutional deep encoding network for video-text retrieval task, which considers to simultaneously encode long-range and short-range dependency in the videos and texts to obtain more robust textual representations.
- A generalized multi-dictionary least squares framework regularized with multi-graph embeddings12
A generalized framework, which extends the PCA idea of minimizing least squares reconstruction errors, to include data distribution and multiple dictionaries for preserving outliers-free global structures in multi-view datasets and to also preserve local manifold structures is proposed.
- 12
- Manifold Alignment via Global and Local Structures Preserving PCA Framework12
A novel semi-supervised manifold alignment method that combines, in each manifold, both global and local linear reconstructions and can concurrently match the pair-wise correspondence and preserve both theglobal and local structures of each dataset to obtain a latent low-dimensional space.
- Adaptive Structure-Constrained Robust Latent Low-Rank Coding for Image Recovery12
A robust representation learning model called Adaptive Structure-constrained Low-Rank Coding (AS-LRC) for the latent representation of data and selects the L2, 1-norm on the projection for extracting group sparse features rather than learning low-rank features by Nuclear-norm regularization, which can make learnt features robust to noise and outliers in samples.
- Gradient-domain-based enhancement of multi-view depth video12
A gradient-domain based enhancement method that exploits the coherence of both temporal and inter-view dimensions in addition of the spatial one to enhance multi-view depth with desired sharpness and consistency.
- 11
- CCM: Adding Conditional Controls to Text-to-Image Consistency Models11
This technical report considers alternative strategies for adding ControlNet-like conditional control to CMs and presents three significant findings.
- FAMLP: A Frequency-Aware MLP-Like Architecture For Domain Generalization11
This work proposes a novel frequency-aware MLP architecture, in which the domain-specific features are filtered out in the transformed frequency domain, augmenting the invariant descriptor for label prediction, and is the first to propose a MLP-like backbone for domain generalization.
- 11
- Deep Palette-Based Color Decomposition for Image Recoloring with Aesthetic Suggestion11
A deep color decomposition based framework for image recoloring is presented, allowing users to achieve professional color edition through simple interactive operations, and a novel aesthetic constraint derived from color harmony theory is proposed to augment the color reconstruction from user-specified colors.
- Instance Segmentation from Volumetric Biomedical Images Without Voxel-Wise Labeling11
This paper proposes a weakly supervised instance segmentation method that needs no voxel-wise labeling and takes advantage of the popular proposal-based framework for instance detection and the peak response mapping for finding visual cues of instances.
- Adaptive Alignment Network for Person Re-identification11
A new Adaptive Alignment Network (AAN) is proposed, towards robust and accurate person re-identification, which automatically aligns pedestrian images from coarse to fine by learning both patch-wise and pixel-wise alignments, leading to effective pedestrian representation invariant to the variance of human pose and location across images.
- The AdaBoost algorithm for vehicle detection based on CNN features11
The AdaBoost algorithm for vehicle detection based on CNN features is proposed and the result of the experiments proves that the algorithm not only improves the real-time performance but also the recognition rate.
- Difficulty guided image retrieval using linear multiview embedding11
This paper proposes a linear multiview embedding algorithm which learns a linear transformation from a small set of data and can effectively infer the subspace features of new data and suggest the effectiveness of the proposed difficulty guided retrieval system with LME.
- TRECVID 2010 Known-item Search by NUS11
This paper describes the system for auto search and inter- active search in the known-item search (KIS) task in TRECVID 2010, and introduces three enhance- ments to the VisioGo system that allow users to indicate related (but not relevant) shots to the query.
- 11
- 10
- VisionSelector: End-to-End Learnable Visual Token Compression for Efficient Multimodal LLMs10
VisionSelector is proposed, a scorer module decoupled from the MLLM backbone that incorporates a differentiable Top-K mechanism and a curriculum annealing strategy to bridge the training-inference gap, enabling efficient and adaptive token selection across various compression rates.
- WeGen: A Unified Model for Interactive Multimodal Generation as We Chat10
Extensive experiments demonstrate the effectiveness of unifying multimodal understanding and generation in WeGen and show it achieves state-of-the-art performance across various visual generation benchmarks, and demonstrate the potential of WeGen as a user-friendly design copilot as desired.
- 10
- 10
- Automatic Relation-aware Graph Network Proliferation10
This work proposes Automatic Relation-aware Graph Network Proliferation (ARGNP) for efficiently searching GNNs with a relation-guided message passing mechanism and designs a network proliferation search paradigm to progressively determine the GNN architectures by iteratively performing network division and differentiation.
- Principled Knowledge Extrapolation with GANs10
It is shown that an adversarial game with a closed-form discriminator can be used to address the knowledge extrapolation problem, and a novel principal knowledge descent method can efficiently estimate the extrapolated distribution through the adversarial game.
- 10
- 10
- 10
- BEVTrack: A Simple and Strong Baseline for 3D Single Object Tracking in Bird's-Eye View9
BEVTrack directly estimates object motion in Bird's-Eye View (BEV) using a single regression loss to enhance accuracy for targets with diverse attributes, and learns adaptive likelihood functions tailored to individual targets, avoiding the limitations of fixed distribution assumptions in previous methods.
- 9
- MIPI 2024 Challenge on Demosaic for Hybridevs Camera: Methods and Results9
A total of 110 participants from both industrial and academic backgrounds contributed many valuable solutions to address the difficulty of the restoration of HybridEVS’s raw data, thus raising the reconstructed performance to a new height.
- Event-driven Video Deblurring via Spatio-Temporal Relation-Aware Network9
A new Spatio-Temporal Relation-Attention network (STRA), which model the brightness changes as an extra prior to aware blurring contexts in each frame to recover spatial texture from events constantly and develops a temporal memory block to restore long-range dependencies of event sequences continuously.
- Towards Neuron Segmentation from Macaque Brain Images: A Weakly Supervised Approach9
This paper builds the first dual-channel three-dimensional image dataset of macaque brain for neuron segmentation and proposes two kinds of “weak” labels, i.e. central points and rough masks, to prepare training data with an affordable cost.
- Making History Matter: Gold-Critic Sequence Training for Visual Dialog.9
This work intentionally imposes wrong answers in the history, obtaining an adverse reward, and sees how the historic error impacts the codec's future behavior by subtracting the gold-critic baseline --- reward obtained by using ground-truth history --- from the adverse reward.
- Joint Sketch-Attribute Learning for Fine-Grained Face Synthesis9
A hybrid generator is proposed to learn a unified embedding of shape from sketches and appearance from attributes for synthesizing images with conditional GANs and an attribute modulation module is proposed, which transfers user-preferred attributes to reinforce sketch representation with appearance details.
- 9
- Corrections to “Exploiting Web Images for Semantic Video Indexing Via Robust Sample-Specific Loss” [Oct 14 1677-1689]9
In the above paper [ibid, vol. 16, no. 6, p. 1679, Oct. 2014], the sentence "Yang et al. [?] integrated semisupervised learning and transfer learning techniques to exploit manually-labeled images for video tagging" should have appeared as "Yang and colleagues [34].
- A novel segmentation based video-denoising method with noise level estimation9
A fast bilateral-median-filter-based method is proposed herein, which creates a segmentation-based, customised BM3D method that can be used to remove coloured multiplicative noise from videos.
- A Pattern Matching Based Model for Implicit Opinion Question Identification9
An effective framework based on mutual information and sequential pattern mining is proposed to construct an opinion lexicon that not only contains opinion words but also patterns and is combined with a machine learning technique to identify opinion questions.
- Active learning for social image retrieval using Locally Regressive Optimal Design9
A novel active learning algorithm, called Locally Regressive Optimal Design (LROD), to improve the effectiveness of relevance feedback-based social image retrieval by adopting a local linear regression model to predict the label value of each data point in a local patch.
- Query expansion by spatial co-occurrence for image retrieval9
A novel query expansion method called Spatial Co-occurrence Query Expansion (SCQE) is proposed, by utilizing the spatial co-occurring information of visual words mined from the database images to boost the retrieval performance.
- 9
- 8
- 8
- RelationVLM: Making Large Vision-Language Models Understand Visual Relations8
RelationVLM is presented, a large vision-language model capable of comprehending various levels and types of relations whether across multiple images or within a video and has strong capability in understanding semantic relations, temporal associations and geometric transforms.
- CLaM: An Open-Source Library for Performance Evaluation of Text-driven Human Motion Generation8
An open-source library with a powerful Contrastive Language-and-Motion (CLaM) pre-training evaluator, which can be employed for evaluating a variety of text-driven human motion generation algorithms, and builds a large-scale HumanML3D-synthesis dataset.
- 8
- 8
- Alleviating Spatial Misalignment and Motion Interference for UAV-based Video Recognition8
A novel framework named Attentional Spatial and Adaptive Temporal Relations Modeling is presented, which can effectively and accurately capture spatiotemporal relations for UAV videos and demonstrates the superiority of this method over state-of-the-art approaches.
- 8
- Learning Dual Convolutional Dictionaries for Image De-raining8
A new dual dictionary learning objective function is proposed, and then unfold it into the form of neural networks to learn prior knowledge from the data automatically to solve the de-raining problems.
- MKEL: Multiple Kernel Ensemble Learning via Unified Ensemble Loss for Image Classification8
A novel ensemble model, called Multiple Kernel Ensemble Learning (MKEL), is developed by introducing a unified ensemble loss.
- General-Purpose Speech Representation Learning through a Self-Supervised Multi-Granularity Framework8
- CircleNet for Hip Landmark Detection8
A much simpler and more efficient framework called CircleNet is proposed to improve the accuracy of landmark detection by predicting landmark and corresponding radius and not only constrain the relationship between landmarks but also integrate landmark detection and object detection into an end-to-end framework.
- Posterior-Guided Neural Architecture Search8
A Bayesian approach to the NAS problem, the posterior-guided NAS (PGNAS) avoids tuning a number of hyper-parameters and enables a very effective architecture sampling in posterior probability space and naturally alleviates the mismatch between the sampled architecture and weights caused by the weight sharing.
- SLiKER: Sparse loss induced kernel ensemble regression8
To overcome the influence of existing outliers, the regressor loss is modeled as a sparse signal, thus a Sparse Loss induced Kernel Ensemble Regression (SLiKER) method is obtained.
- 8
- Progressive Learning for Neuronal Population Reconstruction from Optical Microscopy Images8
A progressive learning strategy to take advantages of both traditional neuron tracing methods and deep learning techniques to reconstruct more complete neuronal populations without using manual annotations is proposed.
- 8
- Co-occurrent Structural Edge Detection for Color-Guided Depth Map Super-Resolution8
An edge detection convolutional neural network is designed to obtain the co-occurrent structural edge in depth map and its corresponding color image and is packed into another customized CNN for depth map super-resolution.
- Content-Based Video Relevance Prediction with Second-Order Relevance and Attention Modeling8
The proposed method for the Content-Based Video Relevance Prediction (CBVRP) challenge is based on deep learning, i.e. a deep network is trained to predict the relevance between two video sequences from their features, and the attention mechanism is designed for second-order relevant video sequences.
- Learning attribute-aware dictionary for image classification and search8
An Attribute aware Dictionary Learning (AttrDL) scheme to learn multiple sub-dictionaries with specific semantic meanings and shows that the proposed scheme outperforms the state-of-the-art algorithms in both image classification and search tasks.
- Visual query attributes suggestion8
A novel query suggestion scheme named Visual Query Attributes Suggestion (VQAS) for image search with QBE, where informative attributes are suggested to the user as complements to the query.
- Learnable Frequency Decomposition for Image Forgery Detection and Localization7
A forensic frequency decomposition network (F2D-Net) is proposed, which incorporates deep Fourier transforms and leverages both phase information and high and low-frequency components to enhance IFDL.
- Diverter-Guider Recurrent Network for Diverse Poems Generation from Image7
An end-to-end probabilistic Diverter-Guider Recurrent Network (DG-Net), which is a context-based encoder-decoder generative model with the hierarchical stochastic variables, which can generate diverse poems with the poeticness and relevance.
- 7
- One-Shot Texture Retrieval with Global Context Metric7
This paper presents an OS-TR network to encode both reference and query image, leading to texture segmentation towards the reference category, and proposes a directionality-aware module to capture the texture variations at each direction, resulting in spatially invariant representation.
- Referring Expression Grounding by Marginalizing Scene Graph Likelihood7
This paper uses scene graph: a graphical representation parsed from the referring expression, where the nodes are objects with attributes and the edges are relationships, and can ground every object to its corresponding region, and perform reasoning with the unlabeled contexts by marginalizing out them using the sum-product belief propagation.
- Temporal-Contextual Attention Network for Video-Based Person Re-identification7
A new Temporal-Contextual Attention Network (TCA-Net) for person re-identification in videos that exploits temporally local context among consecutive frames to concentrate selectively on crucial frames within a video sequence.
- 7
- Semi-automatic Flickr Group Suggestion7
A semiautomatic group suggestion approach with Human-in-the-Loop is proposed, which is able to provide accurate group suggestions with quite a small amount of user effort.
- Utilizing related samples to learn complex queries in interactive concept-based video search7
This paper learns a detector for the query by simultaneously leveraging the related concept detectors, as well as users' feedbacks including relevant, irrelevant, and related samples, and uses this detector to predict the presence of the query in new video segments.
- 7
- 6
- 6
- 6
- 6
- 6
- MaTCR: Modality-Aligned Thought Chain Reasoning for Multimodal Task-Oriented Dialogue Generation6
A Modality-aligned Thought Chain Reasoning (MaTCR) framework to insert explicit reasoning process for multimodal task-oriented dialogue generation is proposed and extensive experiments over a benchmark dataset demonstrate that MaTCR outperforms the existing methods and provides stronger interpretability.
- 6
- 6
- Structure injected weight normalization for training deep networks6
The proposed deep structural weight normalization methods can reduce the number of trainable parameters while guaranteeing high accuracy, whereas DSWN-NM can accelerate the convergence while improving the performance of deep networks.
- 6
- Semantic Image Analogy with a Conditional Single-Image GAN6
This work proposes a novel method to model the patch-level correspondence between semantic layout and appearance of a single image by training a single-image GAN that takes semantic labels as conditional input.
- Hierarchical Granularity Transfer Learning6
A novel Bi-granularity Semantic Preserving Network (BigSPN) is proposed to bridge the granularity gap for robust knowledge transfer and experiments show that BigSPN is an effective framework for Hierarchical Granularity Transfer Learning.
- Explainability by Parsing: Neural Module Tree Networks for Natural Language Visual Grounding.6
A novel modular network called Neural Module Tree network (NMTree) is developed that regularizes the visual grounding along the dependency parsing tree of the sentence, where each node is a module network that calculates or accumulates the grounding score in a bottom-up direction where as needed.
- Click-boosting random walk for image search reranking6
The use of click-through data, which can be viewed as the footprints of user searching behavior, is explored as an effective means of understanding query, for providing the basis on identifying the recurrent patterns that are potentially helpful for reranking.
- Robust Semantic Video Indexing by Harvesting Web Images6
A robust image-to-video indexing approach to learn reliable classifiers from a limited number of training videos together with abundant user-tagged images is developed and an optimal kernel space is automatically discovered by the approach to tackle the domain difference problem.
- Which Tags Are Related to Visual Content?6
A data-driven method is developed to investigate the relatedness between a tag and the image visual content, and extensive experiments demonstrate the effectiveness of the method.
- 5
- 5
- MLP Embedded Inverse Tone Mapping5
This work devise a meta-learning strategy to pre-train a lightweight multilayer perceptron (MLP) model that maps SDR pixels to HDR/WCG ones on an external dataset, resulting in a domain-wise iTM model, and embedding image-wise metadata into their SDR counterparts for later iTM reconstruction.
- 5
- Exert Diversity and Mitigate Bias: Domain Generalizable Person Re-identification with a Comprehensive Benchmark5
A novel DG-ReID framework based on diverse feature space learning with domain factorization is proposed to effectively learn rich domain-adaptive discriminative features through the two designed blocks with fairly limited additional cost in both memory and computation.
- 5
- Label Noise-Resistant Mean Teaching for Weakly Supervised Fake News Detection5
A novel label noise-resistant mean teaching approach (LNMT) for weakly supervised fake news detection that leverages unlabeled news and feedback comments of users to enlarge the amount of training data and facilitates model training by generating refined labels as weak supervision.
- Local-binarized very deep residual network for visual categorization5
The results show that the proposed local binary residual block (LBB) model can accelerate the training and inference of the network with only a slight performance degradation.
- Transferrable Referring Expression Grounding with Concept Transfer and Context Inheritance5
A Concept-Context Disentangled network (CCD) is proposed which transfers concepts from auxiliary classification data with new categories meanwhile inherits context from REG data to ground new objects and outperforms the compared approach on the new-category test datasets.
- 5
- Collaborative Q-Learning Based Routing Control in Unstructured P2P Networks5
The proposed collaborative Q-learning method learns the networks parameters such as processing capacity, number of connections, and number of resources in the peers, along with their state of congestion, to assure alternative routing path to balance the query loads among the peers under higher network churns.
- 5
- 5
- 4
- GRACE: Estimating Geometry-level 3D Human-Scene Contact from 2D Images4
GRACE (Geometry-level Reasoning for 3D Human-scene Contact Estimation), a new paradigm for 3D human contact estimation that incorporates a point cloud encoder-decoder architecture along with a hierarchical feature extraction and fusion module, enabling the effective integration of 3D human geometric structures with 2D interaction semantics derived from images.
- Anchoring Values in Temporal and Group Dimensions for Flow Matching Model Alignment4
Value-Anchored Group Policy Optimization (VGPO), a framework that redefines value estimation across both temporal and group dimensions, achieves state-of-the-art image quality while simultaneously improving task-specific accuracy, effectively mitigating reward hacking.
- MATE: Motion-Augmented Temporal Consistency for Event-Based Point Tracking4
This paper presents an event-based framework for tracking any point, which tackles the challenges posed by spatial sparsity and motion sensitivity in events through two tailored modules, and outperforms all existing methods, even those that combine events and video frames.
- 4
- A Lottery Ticket Hypothesis Approach with Sparse Fine-tuning and MAE for Image Forgery Detection and Localization4
This paper introduces an enhancement to the Masked Autoencoder (MAE), aptly termed the Forgery MAE, which retains the inherent characteristics of natural images while integrating multi-source forgery information and shows strong resilience against unseen forgeries.
- Downstream-Pretext Domain Knowledge Traceback for Active Learning4
A downstream-pretext domain knowledge traceback (DOKT) method that traces the data interactions of downstream knowledge and pre-training guidance for selecting diverse and instructive samples near the decision boundary and generalizes well to various application scenarios such as semantic segmentation and image captioning.
- 4
- 4
- Learning Semantics-Grounded Vocabulary Representation for Video-Text Retrieval4
A novel SemVTR framework designed to learn semantics-grounded video- text representations in a vocabulary space, in which each dimension corresponds to a semantic concept represented by a word, which significantly outperforms existing methods on four video-text retrieval benchmarks.
- 4
- 4
- TDI TextSpotter: Taking Data Imbalance into Account in Scene Text Spotting4
This paper proposes a novel scene text spotter TDI that outperforms state-of-the-art methods on three public datasets in both text detection and end-to-end text recognition, which fully proves the effectiveness and robustness of the method.
- March on Data Imperfections: Domain Division and Domain Generalization for Semantic Segmentation4
A novel model-agnostic training optimization algorithm which has two prominent components: Domain Division and Domain Generalization to deal with data imbalance and to alleviate the degradation caused by label noise is presented.
- 4
- Object Trajectory Proposal via Hierarchical Volume Grouping4
This paper proposes a novel object trajectory proposal method using hierarchical volume grouping, which represents a given video with hierarchical volumes by mapping hierarchical regions with optical flow and ranks the object candidates using a multi-modal fusion scoring mechanism.
- A comprehensive representation scheme for video semantic ontology and its applications in semantic concept detection4
This paper presents a comprehensive representation scheme for video semantic ontology in which all the three components are well studied, and leverage LSCOM to construct the concept lexicon, describe concept property as the weights of different modalities which are obtained manually or by data-driven approach.
- Video Browser Showdown by NUS4
A novel interactive video browsing system for KIS task that integrates visual content-based, text-based and concept-based search approaches, and allows users to flexibly choose the search approaches.
- Learning "verb-object" concepts for semantic image annotation4
This paper proposes a novel image annotation approach by learning "verb-object" concepts based on the assumption that the classifiers of the "verbs" concepts which contain the same object usually share a common structure.
- Likelihood-Aware Semantic Alignment for Full-Spectrum Out-of-Distribution Detection3
A Likelihood-Aware Semantic Alignment framework to promote the image-text alignment into semantically high-likelihood regions and demonstrates the remarkable OOD detection performance of the proposed LSA, especially on the intractable Near-OOD setting.
- 3
- Decouple to Reconstruct: High Quality UHD Restoration Via Active Feature Disentanglement and Reversible Fusion3
A Controlled Differential Disentangled VAE is proposed, which utilizes Hierarchical Contrastive Disentanglement Learning and an Orthogonal Gated Projection Module to guide the VAE to actively discard easily recoverable background information while encoding more difficult-to-recover degraded information into the latent space.
- AIM 2025 Challenge on High FPS Motion Deblurring: Methods and Results3
This paper thoroughly evaluates the state-of-the-art advances in high-FPS single image motion deblurring, showcasing the significant progress in the field, while leveraging samples of the novel dataset, MIORe, that introduces challenging examples of movement patterns.
- Neural Fractional Attention Differential Equations3
This paper proposes a generalized neural Fractional Attention Differential Equation (FADE), which combines the memory-retention capabilities of fractional calculus with contextual learnable attention mechanisms and demonstrates that the adaptive attention-based fractional framework outperforms both integer-order neural ODE models and existing fractional approaches.
- VanGogh: A Unified Multimodal Diffusion-based Framework for Video Colorization3
VanGogh is introduced, a unified multimodal diffusion-based framework for video colorization that achieves superior temporal consistency and color fidelity and can be exercised both global and local control over the generation process, resulting in higher-quality colorized videos.
- 3
- Noise-Resistance Learning via Multi-Granularity Consistency for Unsupervised Domain Adaptive Person Re-Identification3
A novel multi-granularity consistency network (MGCN) is proposed to perform the noise-resistance learning across different granularity consistency perspectives, including prototype-wise consistency, triplet-wise consistency and list-wise consistency, to suppress the contribution of noisy samples simultaneously.
- 3
- 3
- Knowledge-Enhanced Hierarchical Information Correlation Learning for Multi-Modal Rumor Detection3
This work proposes a novel knowledge-enhanced hierarchical information correlation learning approach (KhiCL) for multi-modal rumor detection by jointly modeling the basic semantic correlation and high-order knowledge- enhanced entity correlation.
- Decoupling Degradation and Content Processing for Adverse Weather Image Restoration3
This work introduces a novel adverse weather image restoration method, called DDCNet, which decouples the degradation removal and content reconstruction process at the feature level based on their channel statistics, and exploits the unique advantages of the Fourier transform in both processes.
- 3
- Cross-Domain Object Representation via Robust Low-Rank Correlation Analysis3
A novel cross-domain object representation algorithm (RLRCA) which not only explores the complexity of multiple relationships of variables by canonical correlation analysis (CCA) but also uses a low rank model to decrease the effect of noisy data.
- 3
- 3
- Dynamically building diversified classifier pruning ensembles via canonical correlation analysis3
A novel method of dynamically building diversified sparse ensembles by applying a technique known as the canonical correlation to model the relationship between the input data variables and output base classifiers and achieves better accuracy as compared to other ensemble methods.
- Near-Duplicate Video Retrieval Through Toeplitz Kernel Partial Least Squares3
KPLS is used to find strong information correlation in near-duplicate videos and to solve the problem of “curse of kernelization” when querying a large-scale video database, a Toeplitz Kernel Partial Least Squares method is proposed.
- FEATURE DETECTION AND CORRESPONDENCE FOR CAMERA CALIBRATION3
The experimental results show that the geometrical constraint based method can improve remarkably the performance of the feature detection and camera calibration.
- Visual-Geometric Collaborative Guidance for Affordance Learning2
This work proposes a visual-geometric collaborative guided affordance learning network that jointly leverages visual and geometric cues to model the complex relationships in human–object interactions and the contact feature representation, and outperforms the representative models regarding objective metrics and visual quality.
- 2
- Towards Better De-raining Generalization via Rainy Characteristics Memorization and Replay2
A new framework that enables networks to progressively expand their de-raining knowledge base by tapping into a growing pool of datasets, significantly boosting their adaptability and surpasses state-of-the-art methods in generalizing to new real-world scenarios is introduced.
- PFSD: A Multi-Modal Pedestrian-Focus Scene Dataset for Rich Tasks in Semi-Structured Environments2
This work presents the multi-modal Pedestrian-Focused Scene Dataset (PFSD), rigorously annotated in semi-structured scenes with the format of nuScenes, and proposes a novel Hybrid Multi-Scale Fusion Network (HMFN), to detect pedestrians in densely populated and occluded scenarios.
- 2
- 2
- DCTMamba: Advancing JPEG Image Restoration Through Long-Sequence Modeling and Adaptive Frequency Strategy2
DCTMamba, a new framework designed to apply Mamba more effectively to JPEG image restoration, integrates the Discrete Cosine Transform into the Mamba to establish the sequential scanning from lower to higher frequencies, enabling the network to initially reconstruct coarse structures and progressively refine the image with more intricate details.
- SafeCFG: Controlling Harmful Features with Dynamic Safe Guidance for Safe Generation2
This work proposes SafeCFG to adaptively control harmful features with dynamic safe guidance by modulating the CFG generation process, which can simultaneously modulate different harmful CFG generation processes, so it could eliminate harmful elements while preserving high-quality generation.
- 2
- 2
- UniDense: Unleashing Diffusion Models with Meta-Routers for Universal Few-Shot Dense Prediction2
This paper constructs a versatile MoE (Mixture of Experts) architecture for dense prediction based on the Stable Diffusion model, and introduces a novel framework UniDense for universal few-shot dense prediction, which enables efficient few-shot adaptation of the entire model.
- RAIN: Real-time Animation of Infinite Video Stream2
RAIN is a pipeline solution capable of animating infinite video streams in real-time with low latency using a single RTX 4090 GPU, and can animate characters in real-time with much better quality, accuracy, and consistency than competitors while costing less latency.
- 2
- 2
- 2
- 2
- 2
- One-Shot Neural Architecture Search Through A Posteriori Distribution Guided Sampling.2
This paper proposes a novel one-shot NAS scheme to explicitly estimate the joint a posteriori distribution over network architecture and weights, and sample networks for evaluation according to it, and reduces the number of sampled sub-networks by orders of magnitude.
- 2
- Multi-Level Deep Cascade Trees for Conversion Rate Prediction.2
This paper proposes multi-Level Deep Cascade Trees (ldcTree), which is a novel decision tree ensemble approach that leverages deep cascade structures by stacking Gradient Boosting Decision Trees (GBDT) to effectively learn feature representation.
- 2
- Sparse canonical correlation analysis for recognition2
The proposed sparse canonical cor-relation analysis (SCCA) can find a reasonable trade-off between statistical fidelity and in-terpretability and outperforms the traditional CCA algorithm.
- 2
- Marginalized multi-layer multi-instance kernel for video concept detection2
This paper proposes a novel kernel, termed marginalized multi-layer multi-instance (MarMLMI) kernel for video concept detection, which exploits the hierarchical structure of video, i.e., both themulti-layer structure and the multi- instance relationship.
- Combining SIFT and Global Features for Web Image Classification2
Experimental results show that the proposed web images classification method by integrating SIFT features of the images with global features is able to improve the performance of web image classification.
- Locally regressive G-optimal design for image retrieval2
The proposed LRGOD algorithm is developed based on a locally regressive least squares model which makes use of the labeled and unlabeled images, as well as simultaneously exploits the local structure of each image.
- Optimizing multimodal reranking for web image search2
A web image search reranking approach with exploring multiple modalities that simultaneously integrate the learning of relevance scores, the weighting of different feature sets, the distance metric and the scaling for each feature set into a unified scheme.
- 1
- Reliable Lifelong Multimodal Editing: Conflict-Aware Retrieval Meets Multi-Level Guidance1
CARML is proposed, a novel retrieval-augmented editing framework that integrates conflict-aware dynamic retrieval with multi-level implicit and explicit guidance for reliable lifelong multimodal editing.
- 1
- WeMMU: Enhanced Bridging of Vision-Language Models and Diffusion Models via Noisy Query Tokens1
Noisy Query Tokens are proposed, which learn a distributed representation space between the VLM and Diffusion Model via end-to-end optimization, enhancing continual learning and enabling stable continual learning across diverse tasks.
- HOIMamba: Efficient Mamba-based Disentangled Progressive Learning for HOI Detection1
A novel Efficient Mamba-based Disentangled Progressive Learning (HOIMamba) for HOI Detection is proposed to absorb the advantages of the existing three approaches and adaptively aggregate multi-level interaction semantics guided by cross-task bidirectional information contexts.
- Efficient Prompt-based Multimodal Interaction for Audio-Visual Event Localization1
This work proposes an efficient prompt-based multimodal interaction approach for audio-visual event localization that freezes a pre-trained transformer model and designs query and global prompt to facilitate information exchange and fusion across modalities.
- 1
- 1
- SVE-Former: A fast fourier transformer via singular vector embedding1
SVE-Former is proposed, a transformer variant that employs singular vector embedding to comprehensively capture dependencies among tokens and outperforms state-of-the-art linear attention mechanisms such as Cosformer and Linformer, achieving over a 10 % reduction in training time.
- 1
- 1
- MGAW: An Effective Method for Geo-localization in Adverse Weather1
A new data-driven framework designed to adapt to various domain transformations by effectively extracting features from complex environments is introduced, achieving third place in the Multimedia Drone Satellite Matching Challenge across multiple environments in the University-160k-WX dataset.
- Hierarchical Information Enhancement Network for Cascade Prediction in Social Networks1
A novel Hierarchical Information Enhancement Network (HIENet) is proposed for cascade prediction that integrates fundamental cascade sequence, user social graphs, and sub-cascade graph into a unified framework and introduces a Multi-modal Cascade Transformer to powerfully fuse these clues, providing a comprehensive understanding of cascading process.
- Leverage Task Context for Object Affordance Ranking1
A novel Context-embed Group Ranking Framework with task relation mining module and graph group update module to deeply integrate task context and perform global relative relationship transmission and demonstrates the feasibility of the task context based affordance learning paradigm and the superiority of the model over state-of-the-art models in the fields of saliency ranking and multimodal object detection.
- 1
- 1
- 1
- 1
- Language-Conditioned Region Proposal and Retrieval Network for Referring Expression Comprehension1
A language-conditioned region proposal and retrieval network that first detects those regions only related to the language and then retrieves the target region by compositional reasoning on the language is proposed.
- 1
- Particle Swarm Programming-Based Interactive Content-Based Image Retrieval1
Tests demonstrate that the proposed PSP-based CBIR technique outperforms other linear or recent RF methods proposed for CBIR and can increase diversity of initial particles, enhance search power and improve convergence over PSO.
- 1
- 1
- Building Locally Discriminative Classifier Ensemble Through Classifier Fusion Among Nearest Neighbors1
A novel discriminative classifier fusion method is proposed, which applies local classification results of classifiers among nearest neighbors to build a local classifier ensemble from this dynamically selected process.
- 1
- Approximate distributed clustering by learning the confidence radius on Fisher discriminant ratio1
Experimental results show that the proposed approach can achieve better clustering accuracies than the DFEKM algorithm while preserving much lower bandwidth consumptions.
- Topology Adaptation Based on Mobile Agent in Unstructured P2P Networks1
A new topology adaptation approach based on mobile agent is proposed to resolve the congestion caused by resource location in unstructured P2P networks.
- Physics-informed single-frame end-to-end learning for denoising and background removal in fluorescence imaging–
A physics-informed end-to-end learning framework that enables denoising and background suppression from a single fluorescence image, integrating a microscope-parameterized forward model, and can be extended to remove out-of-focus background in thick samples are introduced.
- –
- –
- –
- Difference-Aware Multi-Grained Feature Learning With Memory-Attended Joint Interaction Transformer for Image Difference Captioning–
This paper designs a difference-aware multi-grained feature learning (DAMFEL) to adapt a pre-trained CLIP to IDC by jointly modeling coarse- and fine-grained contrastive alignments between image pairs and captions and devise a memory-attended joint interaction transformer (MAJIT) for caption generation.
- –
- Geometry-Aware Enhancement and Data Augmentation for Street-to-Satellite Geo-localization–
A lightweight and geometry-aware framework that enhances front-view street images via a calibration-free spherical transform that projects single-view images into a bird's-eye view (BEV), effectively aligning their perspective with satellite imagery.
- –
- –
- –
- Dual Uncertainty-Guided Feature Alignment Learning for Text-Based Person Retrieval–
DUAL designs an alignment loss that utilizes identity-level uncertainty coefficients, estimated via a Gaussian Mixture Model based on the distances between image and text features of the same identity, effectively mitigating the impact of semantic content diversity.
- –
- Learning Object Affordance Ranking with Task Context–
The first large-scale task-oriented affordance ranking dataset is constructed and the feasibility of the task context based affordance learning paradigm is demonstrated and the superiority of the model over state-of-the-art models in the fields of saliency ranking and multimodal object detection is demonstrated.
- –
- Multi-perspective Memory Enhanced Network for Identifying Key Nodes in Social Networks–
A novel Multi-perspective Memory Enhanced Network (MMEN) for identifying key nodes in social networks, which mines key nodes from multiple perspectives and utilizes memory networks to store historical information.
- Neural Dependencies Emerging from Learning Massive Categories–
It is demonstrated that identifying neural dependencies is equivalent to solving the Covariance Lasso (CovLasso) regression problem proposed in this paper, and the potential of neural dependencies in understanding internal data correlations, generalizing models to unseen categories, and improving model robustness with a dependency-derived regularizer is shown.
- –
- –
- –
- –
- –
- –
- –
- Collaborative Detection and Caption Network–
This paper proposes an effective end-to-end network, called a Collaborative Detection and Caption network, which takes a video caption network as video- to-sentence sub-network and principle syntactic components detector as video -to-words sub- network.
- –
- –
- –
- –
- –
- –
- –
- Sparse principle motion component for one-shot gesture recognition–
Experimental results show that the proposed sparse PCA based principle motion component (SPMC) method for one-shot gesture recognition can improve the accuracy of gesture recognition.
- –
- Improving Color Constancy with Internet Photo Collections–
An iterative weighted Specular-Edge color constancy scheme that leverages large database of images gathered from the web and uses the visual context of an input image as an image-specific prior to increase the accuracy of estimating the illuminant.
- –
- Hierarchical Organization of Collaboratively Constructed Content–
A domain-assisted approach is proposed to generate a hierarchical structure to organize consumer reviews on products, which provides a well-visualized way to browse consumer reviews at different granularity to meet various users’ needs, which can help to improve information dissemination and accessibility.
- –
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.