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
- Hua Xing YuSuggested from co-authorship
Is this you? Claim this profile to confirm or dismiss it.
Works265 from public data
- 1,273
- Image-Image Domain Adaptation with Preserved Self-Similarity and Domain-Dissimilarity for Person Re-identification1,011
Through domain adaptation experiment, it is shown that images generated by SPGAN are more suitable for domain adaptation and yield consistent and competitive re-ID accuracy on two large-scale datasets.
- Conformer: Local Features Coupling Global Representations for Visual Recognition827
Experiments show that Conformer, under the comparable parameter complexity, outperforms the visual transformer (DeiT-B) by 2.3% on ImageNet and on MSCOCO, it outperforms ResNet-101 by 3.7% and 3.6% mAPs for object detection and instance segmentation, respectively, demonstrating the great potential to be a general backbone network.
- 780
- 712
- Towards Optimal Structured CNN Pruning via Generative Adversarial Learning585
This paper proposes an effective structured pruning approach that jointly prunes filters as well as other structures in an end-to-end manner and effectively solves the optimization problem by generative adversarial learning (GAL), which learns a sparse soft mask in a label-free and an end to end manner.
- Prototype Mixture Models for Few-Shot Semantic Segmentation499
Proposed prototype mixture models (PMMs), which correlate diverse image regions with multiple prototypes to enforce the prototype-based semantic representation, incorporate rich channel-wised and spatial semantics from limited support images.
- 467
- BEiT v2: Masked Image Modeling with Vector-Quantized Visual Tokenizers456
This work proposes to use a semantic-rich visual tokenizer as the reconstruction target for masked prediction, providing a systematic way to promote MIM from pixel-level to semantic-level, and proposes vector-quantized knowledge distillation to train the tokenizer, which discretizes a continuous semantic space to compact codes.
- Component Divide-and-Conquer for Real-World Image Super-Resolution419
This paper proposes a Component Divide-and-Conquer model and a Gradient-Weighted loss for SR, exploring the utility of guiding SR model with low-level image components, and proposes a feasible way to balance the difficulties of image components for SR.
- Scale Match for Tiny Person Detection400
A simple yet effective Scale Match approach to align the object scales between the two datasets for favorable tiny-object representation and the significant performance gain of this proposed approach over state-of-the-art detectors is shown.
- Fast and robust text detection in images and video frames380
A novel coarse-to-fine algorithm that is able to locate text lines even under complex background is proposed and Experimental results show that this approach can fast and robustly detect text lines under various conditions.
- SIXray: A Large-Scale Security Inspection X-Ray Benchmark for Prohibited Item Discovery in Overlapping Images362
A large-scale dataset and establish a baseline for prohibited item discovery in Security Inspection X-ray images, in which 6 classes of 8,929 prohibited items are manually annotated, and proposes an approach named class-balanced hierarchical refinement (CHR) to deal with these difficulties.
- AD-Cluster: Augmented Discriminative Clustering for Domain Adaptive Person Re-Identification318
A novel augmented discriminative clustering (AD-Cluster) technique that estimates and augments person clusters in target domains and enforces the discrimination ability of re-ID models with the augmented clusters.
- Weakly Supervised Instance Segmentation Using Class Peak Response298
This paper addresses the challenging image-level supervised instance segmentation task by exploiting class peak responses to enable a classification network for instance mask extraction, and reports state-of-the-art results on popular benchmarks, including PASCAL VOC 2012 and MS COCO.
- Oriented Response Networks288
Over multiple state-of-the-art DCNN architectures, such as VGG, ResNet, and STN, it is consistently observed that replacing regular filters with the proposed ARFs leads to significant reduction in the number of network parameters and improvement in classification performance.
- Timestep Embedding Tells: It’s Time to Cache for Video Diffusion Model286
This study introduces Timestep Embedding Aware Cache (TeaCache), a training-free caching approach that estimates and leverages the fluctuating differences among model outputs across timesteps to improve inference efficiency and visual quality.
- 278
- TS-CAM: Token Semantic Coupled Attention Map for Weakly Supervised Object Localization274
This paper introduces the token semantic coupled attention map (TS-CAM) to take full advantage of the self-attention mechanism in visual transformer for long-range dependency extraction and achieves state-of-the-art performance.
- C-MIL: Continuation Multiple Instance Learning for Weakly Supervised Object Detection261
A continuation optimization method is introduced into MIL and thereby creating continuation multiple instance learning (C-MIL), with the intention of alleviating the non-convexity problem in a systematic way.
- 261
- Beyond Bounding-Box: Convex-hull Feature Adaptation for Oriented and Densely Packed Object Detection258
- 235
- Min-Entropy Latent Model for Weakly Supervised Object Detection232
A min-entropy latent model (MELM) is proposed for weakly supervised object detection, unified with feature learning and optimized with a recurrent learning algorithm, which progressively transfers the weak supervision to object locations.
- Adaptive Keyframe Sampling for Long Video Understanding217
Experiments on two long video understanding benchmarks validate that AKS improves video QA accuracy (beyond strong baselines) upon selecting informative keyframes, and reveals the importance of information pre-filtering in video-based MLLMs.
- 211
- 205
- Multiple Expert Brainstorming for Domain Adaptive Person Re-Identification202
A multiple expert brainstorming network (MEB-Net) for domain adaptive person re-ID is proposed, opening up a promising direction about model ensemble problem under unsupervised conditions.
- Beyond Max-Margin: Class Margin Equilibrium for Few-shot Object Detection199
A class margin equilibrium (CME) approach, with the aim to optimize both feature space partition and novel class reconstruction in a systematic way, achieving state-of-the-art performance.
- A configurable method for multi-style license plate recognition191
A new method for multi-style LP recognition by representing the styles with quantitative parameters, i.e., plate rotation angle, plate line number, character type and format, which will result in a faster and more precise recognition.
- Video Playback Rate Perception for Self-Supervised Spatio-Temporal Representation Learning189
A novel self-supervised method, referred to as video Playback Rate Perception (PRP), to learn spatio-temporal representation in a simple-yet-effective way and outperforms state-of-the-art self- supervised models with significant margins.
- 187
- Domain General Face Forgery Detection by Learning to Weight179
This paper argues that different faces contribute differently to a detection model trained on multiple domains, making the model likely to fit domain-specific biases, and proposes the LTW approach based on the meta-weight learning algorithm, which configures different weights for face images from different domains.
- Video Cloze Procedure for Self-Supervised Spatio-Temporal Learning173
With VCP, a novel self-supervised method to learn rich spatial-temporal representation models (3D-CNNs) and apply such models on action recognition and video retrieval tasks, the trained models outperform the state-of-the-art self- supervised models with significant margins.
- Multiple Instance Active Learning for Object Detection170
MI-AOD defines an instance uncertainty learning module, which leverages the discrepancy of two adversarial instance classifiers trained on the labeled set to predict instance uncertainty of the unlabeled set, and estimates the image uncertainty by re-weighting instances in a multiple instance learning (MIL) fashion.
- 152
- Soft Proposal Networks for Weakly Supervised Object Localization152
This paper designs a network component, Soft Proposal (SP), to be plugged into any standard convolutional architecture to introduce the nearly cost-free object proposal, orders of magnitude faster than state-of-the-art methods.
- SRN: Side-Output Residual Network for Object Symmetry Detection in the Wild110
A new benchmark and an end-to-end deep learning approach are presented, opening up a promising direction for symmetry detection in the wild, and the proposed symmetry detection approach, named Side-output Residual Network (SRN), exploits the flow of errors among multiple scales to ease the problems of fitting complex outputs with limited layers.
- Filter Sketch for Network Pruning106
FilterSketch encodes the second-order information of pretrained weights, which enables the representation capacity of pruned networks to be recovered with a simple fine-tuning procedure, leading to a several-orders-of-magnitude reduction of time cost in the optimization of pruning.
- Multiple Anchor Learning for Visual Object Detection106
A Multiple Instance Learning (MIL) approach that selects anchors and jointly optimizes the two modules of a CNN-based object detector, referred to as Multiple Anchor Learning (MAL), which improves the baseline RetinaNet with significant margins on the commonly used MS-COCO object detection benchmark and achieves new state-of-the-art detection performance compared with recent methods.
- Towards Compact CNNs via Collaborative Compression105
A Collaborative Compression scheme, which joints channel pruning and tensor decomposition to compress CNN models by simultaneously learning the model sparsity and low-rankness and proposes multi-step heuristic compression to remove redundant compression units step-by-step.
- Visual abnormal behavior detection based on trajectory sparse reconstruction analysis100
A novel abnormal behavior detection approach by introducing trajectory sparse reconstruction analysis (SRA), solved by L1-norm minimization, leading to that a few of dictionary samples are used when reconstructing a behavior trajectory, which guarantees that the proposed approach is valid even when the dictionary set is very small.
- 95
- 91
- 90
- An effective method to detect and categorize digitized traditional Chinese paintings85
A scheme to detect TCPs from general images and categorize them into Gongbi and Xieyi schools is proposed, using support vector machine (SVM) as the main classifier to obtain satisfactory classification results.
- 84
- Visual object tracking via sample-based Adaptive Sparse Representation (AdaSR)83
A new approach for visual object tracking based on Sample-Based Adaptive Sparse Representation (AdaSR), which ensures that the tracked object is adaptively and compactly expressed with predefined samples, which is better than those of several representative tracking methods.
- 81
- Multi-View correlation distillation for incremental object detection78
A novel MVCD based incremental object detection method, which explores the correlations in the feature space of the two-stage object detector (Faster R-CNN), and design correlation distillation losses from channel-wise, point-wise and instance-wise views to regularize the learning of the incremental model.
- Output Constraint Transfer for Kernelized Correlation Filter in Tracking77
Extensive experiments show that the proposed output constraint transfer (OCT) method significantly improves KCF, and achieves better performance than other state-of-the-art trackers.
- Text detection and restoration in natural scene images76
A new method for text detection and recognition in natural scene images is presented, based on plane-to-plane homography, and results tested from a large dataset have demonstrated that the proposed method is effective and practical.
- 74
- Network Pruning Using Adaptive Exemplar Filters74
Inspired by the face recognition community, a message-passing algorithm Affinity Propagation on the weight matrices is used to obtain an adaptive number of exemplars, which then act as the preserved filters, resulting in an automatic and efficient pruning approach called EPruner.
- 70
- 68
- Rethinking Sampling Strategies for Unsupervised Person Re-Identification67
It is suggested that deteriorated over-fitting is an important factor causing poor performance, and enhancing statistical stability can rectify this problem, and a simple yet effective approach is proposed, termed group sampling, which gathers samples from the same class into groups.
- A novel laser vision sensor for weld line detection on wall-climbing robot63
Experiments show that the CSL sensor can capture 3D information of the weldment with very low measurement error, and the weld line detection approach is effective in wall-climbing robotic platform navigation.
- 61
- Anti-aliasing Semantic Reconstruction for Few-Shot Semantic Segmentation61
This paper reformulates few-shot segmentation as a semantic reconstruction problem, and converts base class features into a series of basis vectors which span a class-level semantic space for novel class reconstruction, which provides a systematic yet interpretable solution for few- shot learning problems.
- 57
- Jersey number detection in sports video for athlete identification56
A method for athlete identification by combing the segmentation, tracking and recognition procedures into a coarse-to-fine scheme for jersey number detection by using the Zernike moment features, which are invariant to rotation and scale for digital shape recognition.
- 55
- A framework for flexible summarization of racquet sports video using multiple modalities54
This work proposes a novel flexible video content summarization framework by taking advantages of the periodicity of video shot content and audio keywords in the racquet sports video, and combines the structure event detection method with the highlight ranking algorithm.
- 53
- 53
- 52
- Discrepant multiple instance learning for weakly supervised object detection51
D-MIL adopts multiple MIL learners to pursue discrepant yet complementary solutions indicating object parts, which are fused with a collaboration module for precise object localization.
- A new method to segment playfield and its applications in match analysis in sports video51
A novel playfield segmentation method based on Gaussian mixture models (GMMs) is proposed, which is robust to various sports videos even for very poor grass field conditions and match situation analysis is investigated.
- HiViT: Hierarchical Vision Transformer Meets Masked Image Modeling50
A new design of hierarchical vision transformers named HiViT (short for Hierarchical ViT) is offered that enjoys both high efficiency and good performance in MIM and is able to remove the unnecessary"local inter-unit operations", deriving structurally simple hierarchical visiontransformers in which mask-units can be serialized like plain vision transformer.
- 50
- Domain Contrast for Domain Adaptive Object Detection49
Domain Contrast is presented, a simple yet effective approach inspired by contrastive learning for training domain adaptive detectors that improves the baseline and state-of-the-art by significant margins, while demonstrating great potential for large domain divergence.
- 49
- 48
- 47
- A Unified View of Masked Image Modeling46
A simple yet effective method, termed as MaskDistill, which reconstructs normalized semantic features from teacher models at the masked positions, conditioning on corrupted input images, achieves comparable or superior performance than state-of-the-art methods.
- 45
- Generative Prompt Model for Weakly Supervised Object Localization44
A generative prompt model (GenPromp) is proposed, defining the first generative pipeline to localize less discriminative object parts by formulating WSOL as a conditional image denoising procedure.
- 44
- Pedestrian Detection with Deep Convolutional Neural Network44
This paper proposes to cascade simple Aggregated Channel Features (ACF) and rich Deep Convolutional Neural Network (DCNN) features for efficient and effective pedestrian detection in complex scenes.
- Fast-iTPN: Integrally Pre-Trained Transformer Pyramid Network With Token Migration43
Fast-iTPN can accelerate the inference procedure by up to up to 70% with negligible performance loss, demonstrating the potential to be a powerful backbone for downstream vision tasks.
- Exciting event detection in broadcast soccer video with mid-level description and incremental learning43
A method for exciting event detection in broadcast soccer video with mid-level description and SVM-based incremental learning is proposed to improve the extensibility of view classification and event detection.
- 42
- 41
- 41
- Linear Span Network for Object Skeleton Detection40
A Linear Span Network (LSN) is proposed which introduces Linear Span Units (LSUs) to minimizes the reconstruction error and utilizes subspace linear span besides the feature linear span to increase the independence of convolutional features and the efficiency of feature integration, which enhances the capability of fitting complex ground-truth.
- 40
- Learnable Distribution Calibration for Few-Shot Class-Incremental Learning39
A learnable distribution calibration (LDC) approach, built upon a parameterized calibration unit (PCU), which initializes biased distributions for all classes based on classifier vectors (memory-free) and a single covariance matrix.
- Object Localization under Single Coarse Point Supervision39
A coarse point refinement (CPR) approach, which to the best knowledge is the first attempt to alleviate semantic variance from the perspective of algorithm, which ensures training high-performance object localizer under coarse point supervision.
- Combined feature evaluation for adaptive visual object tracking38
A combined feature evaluation approach in filter frameworks for adaptive object tracking by combining color histogram and gradient orientation histogram, which gives a representation of both color and contour.
- Generic-to-Specific Distillation of Masked Autoencoders37
This study proposes generic-to-specific distillation (G2SD), to tap the potential of small ViT models under the supervision of large models pretrained by masked autoencoders.
- Discretization-aware architecture search37
Discretization-aware architecture search is presented, with the core idea being adding a loss term to push the super-network towards the configuration of desired topology, so that the accuracy loss brought by discretization is largely alleviate.
- Integrally Migrating Pre-trained Transformer Encoder-decoders for Visual Object Detection34
This study proposes to integrally migrate pre-trained transformer encoder-decoders (imTED) to a detector, constructing a feature extraction path which is "fully pre-trained" so that detectors’ generalization capacity is maximized.
- Feature Calibration Network for Occluded Pedestrian Detection34
A novel feature learning method in the deep learning framework, referred to as Feature Calibration Network (FC-Net), to adaptively detect pedestrians under various occlusions, which improves detection performance on occluded pedestrians up to 10% while maintaining excellent performance on non-occluded instances.
- 34
- Self-Guided Adaptation: Progressive Representation Alignment for Domain Adaptive Object Detection33
This work proposes a Self-Guided Adaptation model, targeting at aligning feature representation and transferring object detection models across domains while considering the instantaneous alignment difficulty, and shows that SGA improves the state-of-the-art methods with significant margins especially on large domain shift cases.
- The 1st Tiny Object Detection Challenge: Methods and Results33
A brief summary of the 1st Tiny Object Detection (TOD) Challenge is provided including brief introductions to the top three methods.
- Self-Learning Scene-Specific Pedestrian Detectors Using a Progressive Latent Model33
Extensive experiments demonstrate that even without annotation the proposed self-learning approach outperforms weakly supervised learning approaches, while achieving comparable performance with transfer learning and fully supervised approaches.
- Regressor-Segmenter Mutual Prompt Learning for Crowd Counting32
This study proposes mutual prompt learning (mPrompt), which leverages a regressor and a segmenter as guidance for each other, solving bias and inaccuracy caused by annotation variance while distinguishing foreground from back-ground.
- Rethinking Performance Estimation in Neural Architecture Search32
This paper provides a novel yet systematic rethinking of PE in a resource constrained regime, termed budgeted PE (BPE), which precisely and effectively estimates the performance of an architecture sampled from an architecture space.
- Building Vision Models upon Heat Conduction31
The Heat Conduction Operator (HCO) built upon the physical heat conduction principle is introduced, enabling robust visual representations and achieving up to a 3× throughput, 80% less GPU memory allocation, and 35% fewer computational FLOPs compared to the Swin-Transformer.
- 31
- 30
- ChatterBox: Multi-round Multimodal Referring and Grounding30
A new benchmark and an efficient vision-language model is presented, named ChatterBox, which outperforms existing models in MRG both quantitatively and qualitatively, paving a new path towards multimodal dialogue scenarios with complicated and precise interactions.
- 29
- Adversarial Reinforced Instruction Attacker for Robust Vision-Language Navigation28
A Dynamic Reinforced Instruction Attacker (DR-Attacker), which learns to mislead the navigator to move to the wrong target by destroying the most instructive information in instructions at different timesteps, is proposed.
- Image captioning via semantic element embedding27
This paper integrates the CNN feature with the semantic feature into the proposed Element Embedding LSTM (EE-LSTM) model to predict a language description, which bridges the modality gap between visual images and semantic captions.
- 26
- End-to-End Weakly Supervised Object Detection with Sparse Proposal Evolution25
Experiments on VOC and COCO object detection datasets show that SPE outperforms the state-of-the-art end-to-end methods by 7.0% mAP and 8.1% AP50, and is an order of magnitude faster than the two-stage methods, setting the first solid baseline for end- to-end WSOD with sparse proposals.
- 25
- Beyond Masking: Demystifying Token-Based Pre-Training for Vision Transformers24
This paper designs five other learning objectives that follow the same procedure as MIM but degrade the input image in different ways, and achieves superior performance over MIM in a series of downstream recognition tasks without extra computational cost.
- 24
- 24
- Context-aware network for RGB-D salient object detection23
By incorporating two variant LSTM units for the exploration of contextual dependencies, this paper presents a context-aware network (CAN) to detect salient objects in RGB-D images and shows that the proposed CAN can achieve state-of-the-art performance forRGB-D stereoscopic salient object detection.
- 23
- Nearest Neighbor Classifier Embedded Network for Active Learning22
This paper attempts to replace the softmax classifier in deep neural network with a nearest neighbor classifier, considering its progressive generalization ability within the unknown sub-space, and proposes NCE-Net, a conceptually simple but surprisingly powerful approach to active learning.
- 22
- Architecture Disentanglement for Deep Neural Networks21
This paper introduces neural architecture disentanglement (NAD), which learns to disentangle a pre-trained DNN into sub-architectures according to independent tasks, forming information flows that describe the inference processes.
- CrossRectify: Leveraging disagreement for semi-supervised object detection20
This paper proposes an effective detection framework named CrossRectify, to obtain accurate pseudo labels by simultaneously training two detectors with different initial parameters, which leverages the disagreements between detectors to discern the self-errors and refines the pseudo label quality by the proposed cross-rectifying mechanism.
- 20
- 19
- Scene Text Detection via Integrated Discrimination of Component Appearance and Consensus19
Experiments show that the proposed approach to scene text detection reaches the state of the art in both detection accuracy and computational efficiency, and in particularly, it performs best when dealing with low-resolution text in clutter backgrounds.
- Language-Driven Visual Consensus for Zero-Shot Semantic Segmentation18
Equipped with a vision-language prompting strategy, this approach significantly boosts the generalization capacity of segmentation models for unseen classes and introduces route attention into the transformer decoder to find visual consensus, thereby enhancing semantic consistency within the same object.
- ControlCap: Controllable Region-Level Captioning18
A controllable region-level captioning approach, which introduces control words to a multimodal model to address the caption degeneration issue, and leverages a discriminative module to generate control words within the caption space to partition it to multiple sub-spaces.
- 18
- Progressive Cluster Purification for Unsupervised Feature Learning18
This work proposes a novel clustering based method, which, by iteratively excluding class inconsistent samples during progressive cluster formation, alleviates the impact of noise samples in a simple-yet-effective manner.
- Unsupervised sports video scene clustering and its applications to story units detection18
A new and efficient clustering approach for scene analysis in sports video that performs in an unsupervised manner and relies on the scene likeness analysis of the shots in the video, which is called J-based Scene Clustering.
- A new method to calculate the camera focusing area and player position on playfield in soccer video18
An algorithm to cope with the former problem, i.e. to compute players' position on court and the homographies between image and playfield, and the result is encouraging.
- A New Text Detection Algorithm in Images/Video Frames18
A new text detection algorithm for images/video frames in a coarse-to-fine framework with multiscale wavelet energy feature employed and a SVM classifier employed to identify texts from the candidate ones is proposed.
- CoCNN: RGB-D deep fusion for stereoscopic salient object detection17
A coupled CNN (CoCNN) is proposed to fuse color and disparity features from low to high layers in a unified deep model and formulate a constraint in the form of a conditional random field model to refine the saliency map based on the a priori assumption that objects with similar saliency values have similar colors and disparities.
- 17
- 17
- 17
- Pedestrian detection in images via cascaded L1-norm minimization learning method17
The proposed CLML method, which is designed from the perspective of Vapnic's theory in the statistical learning, integrates feature selection with classifier construction via solving meaningful optimization models.
- 17
- 16
- Cogradient Descent for Bilinear Optimization15
A Cogradient Descent algorithm to address the bilinear problem, based on a theoretical framework to coordinate the gradient of hidden variables via a projection function, which improves the state-of-the-art by a significant margin.
- Fault detection and classification in chemical processes using NMFSC and structural SVMs15
A novel fault detection and classification approach via Non‐negative Matrix Factorisation with Sparseness Constraints (NMFSC) and Structural Support Vector Machines (Structural SVMs) and the results show the superiority of the new method compared with the state‐of‐the‐art approaches.
- 15
- Virtual Classification: Modulating Domain-Specific Knowledge for Multidomain Crowd Counting14
This study proposes a simple-yet-effective modulating domain-specific knowledge network (MDKNet) to handle the domain bias issue in multidomain crowd counting and proves the superiority of MDKNet in tackling multidomain crowd counting and the effectiveness for multidomain learning.
- Agreement-Discrepancy-Selection: Active Learning with Progressive Distribution Alignment14
This paper proposes an agreement-discrepancy-selection approach, and target at unifying distribution alignment with sample selection by introducing adversarial classifiers to the convolutional neural network (CNN).
- iffDetector: Inference-Aware Feature Filtering for Object Detection14
This article introduces a generic inference-aware feature filtering (IFF) module that can be easily combined with existing detectors, resulting in the iffDetector, which consistently outperforms state-of-the-art methods with significant margins.
- 14
- Depth-Guided Texture Diffusion for Image Semantic Segmentation13
A depth-guided texture diffusion approach that effectively tackles the outlined challenge of enhancing structural information vital for precisely extracting object outlines, and effectively bridges the disparity between depth and RGB modalities, enabling more accurate semantic segmentation.
- CC-Diff: Enhancing Contextual Coherence in Remote Sensing Image Synthesis13
CC-Diff is introduced, a novel Dual Re-sampler for feature extraction, with a built-in `Context Bridge' to explicitly capture the intricate interdependency between foreground and background, and outperforms state-of-the-art methods across critical quality metrics.
- 13
- IOS-Net: An inside-to-outside supervision network for scale robust text detection in the wild13
A novel inside-to-outside supervision network (IOS-Net) that can well tackle two types of scale variance: various size and diverse aspect ratios of the text regions and multiple-scale supervision on the stacked HSMs.
- 13
- Consolidating Reinforcement Learning for Multimodal Discrete Diffusion Models12
This study establishes MaskGRPO as a systematic policy optimization approach and the first practical way for discretized visual diffusion, and delicately tailored the rollout method for visual sequences, which yields diverse completions and reliable optimization gradients.
- Proposal Distribution Calibration for Few-Shot Object Detection12
A simple yet effective proposal distribution calibration (PDC) approach to neatly enhance the localization and classification abilities of the RoI head by recycling its localization ability endowed in base training and enriching high-quality positive samples for semantic fine-tuning.
- 12
- 12
- Beyond Instance Discrimination: Relation-Aware Contrastive Self-Supervised Learning11
This paper presents relation-aware contrastive self-supervised learning (ReCo) to integrate instance relations, i.e., global distribution relation and local interpolation relation, into the CSL framework in a plug-and-play fashion and consistently gains remarkable performance improvements.
- High performance person re-identification via a boosting ranking ensemble11
This paper proposes a novel framework for person re-identification called Boosting Ranking Ensemble (BRE), which adaptively assembles features and metrics from a ranking perspective and explores an effective rectifier loss in the proposed BRE framework to alleviate the influence of the misclassified samples.
- 11
- CircleNet: Reciprocating Feature Adaptation for Robust Pedestrian Detection11
CircleNet implements feature ensemble with the idea of hard negative boosting in an end-to-end manner to achieve feature adaptation by mimicking the process humans looking at low resolution and occluded objects.
- DQnet: Cross-Model Detail Querying for Camouflaged Object Detection10
It is argued that partial activation is caused by the intrinsic characteristics of CNN, where the convolution operations produce local receptive fields and experience difficulty to capture long-range feature dependency among image regions.
- SPSTracker: Sub-Peak Suppression of Response Map for Robust Object Tracking10
This paper proposes a rectified online learning approach for sub-peak response suppression and peak response enforcement and target at handling progressive interference in a systematic way, referred to as SPSTracker, which applies simple-yet-efficient Peak Response Pooling to aggregate and align discriminative features, as well as leveraging a Boundary Response Truncation to reduce the variance of feature response.
- Subjective evaluation criterion for selecting affective features and modeling highlights10
A subjective evaluation criterion based on the analysis of the average error method and pairwise comparisons method is proposed which proves that previous works on affective features and linear model highlights are effective and 82.0% affective accuracy is obtained fully automatically by computer which is a marvelous highlights ranking result.
- LongVideo-R1: Smart Navigation for Low-cost Long Video Understanding9
The LongVideo-R1 agent is fine-tuned upon the Qwen-3-8B model through a two-stage paradigm: supervised fine-tuning followed by reinforcement learning (RL), where RL employs a specifically designed reward function to maximize selective and efficient clip navigation.
- 9
- Correspondence-Guided SfM-Free 3D Gaussian Splatting for NVS9
This study proposes a correspondence-guided SfM-free 3D Gaussian splatting for NVS that uses correspondences between the target and the rendered result to achieve better pixel alignment, facilitating the optimization of relative poses between frames.
- Album Storytelling with Iterative Story-aware Captioning and Large Language Models9
This work proposes a new iterative album storytelling pipeline, which starts with an initial story and builds a story-aware caption model to refine the captions using the whole story as guidance, then feeds into the LLMs to generate a new refined story.
- Towards Deviation-Robust Agent Navigation via Perturbation-Aware Contrastive Learning9
This paper presents a model-agnostic training paradigm, called Progressive Perturbation-aware Contrastive Learning (PROPER) to enhance the generalization ability of existing VLN agents to the real world, by requiring them to learn towards deviation-robust navigation.
- 9
- API-Net: Robust Generative Classifier via a Single Discriminator9
An Anti-Perturbation Inference (API) method, which searches for anti-perturbations to maximize the lower bound of the joint log-likelihood of inputs and classes, and is constructed upon a generative classifiers upon a single discriminator, which takes advantage of the generative properties to tackle off-manifold examples while maintaining a succinct structure for effective optimization.
- 9
- Collective motion pattern inference via Locally Consistent Latent Dirichlet Allocation9
Experimental results on three public datasets indicate that LC-LDA based motion patterns can improve the trajectory clustering performance effectively and improve the unsupervised inference capability and compactness of learned collective motion patterns.
- Uncertainty-guided Optimal Transport in Depth Supervised Sparse-View 3D Gaussian8
This work introduces a novel method to supervise the depth distribution of 3D Gaussians, utilizing depth priors with integrated uncertainty estimates, and integrates a patch-wise optimal transport strategy to complement traditional L2 loss in depth supervision.
- Multi-Agent Automated Machine Learning8
Multi-agent automated machine learning (MA2ML) explicitly assigns credit to each agent according to its marginal contribution to enhance cooperation among modules, and incorporates off-policy learning to improve search efficiency.
- 8
- Calibrated Stochastic Gradient Descent for Convolutional Neural Networks8
A theorem is developed to prove that an unbiased estimator for the network variables can be obtained in a probabilistic way based on the Lipschitz hypothesis, and a generic gradient calibration layer is developed which can be easily used to build convolutional neural networks.
- Rotation Invariant Local Binary Convolution Neural Networks8
A new convolutional module, local binary orientation module (LBoM), which takes advantages of both local binary convolutional and active rotating filters to effectively deal with the rotation variations with fewer parameters is proposed.
- 8
- A scheme for racquet sports video analysis with the combination of audio-visual information8
A new scheme for structure indexing and highlight generating based on the combination of audio and visual information is proposed for racquet sports video, e.g. table tennis, tennis and badminton.
- ChatterBox: Multimodal Referring and Grounding with Chain-of-Questions7
Experiments show that ChatterBox demonstrates superiority in MCQ both quantitatively and qualitatively, paving a new path towards multimodal dialogue scenarios with logical interactions.
- CPR++: Object Localization via Single Coarse Point Supervision7
This study proposes coarse point refinement (CPR), which to the best knowledge is the first attempt to alleviate semantic variance from an algorithmic perspective in POL, and observes that CPR++ can obtain scale information and further reduce the semantic variance in a global region, thus guaranteeing high-performance object localization.
- Unsupervised Domain Adaptation on Person Reidentification Via Dual-Level Asymmetric Mutual Learning7
A dual-level asymmetric mutual learning (DAML) method to learn discriminative representations from a broader knowledge scope with diverse embedding spaces by two heterogeneous networks mutually learn knowledge from asymmetric subspaces through the pseudo label generation in a hard distillation manner.
- 7
- 7
- SRN: Side-Output Residual Network for Object Reflection Symmetry Detection and Beyond7
Experimental results verify that the Sym-PASCAL benchmark is challenging related to real-world images, SRN achieves state-of-the-art performance, and MT-SRN has the capability to simultaneously predict edge and symmetry mask without loss of performance.
- 7
- 7
- Hierarchical residual stochastic networks for time series recognition7
Experiments show that end-to-end hierarchical residual stochastic networks to effectively and efficiently describe spatio-temporal variations significantly boost the performance of time series recognition and improve the state-of-the-art of skeleton, action, and radar behavior recognition performance.
- A scalable convolutional neural network for task-specified scenarios via knowledge distillation7
A task-specified knowledge distillation algorithm is proposed to derive a simplified model with pre-set computation cost and minimized accuracy loss, which suits the resource constraint front-end systems well and proves the feasibility of the proposed approach as well as the existence of task- specified redundancy.
- 7
- 6
- ClickTrack: Towards real-time interactive single object tracking6
A new paradigm for single object tracking algorithms, ClickTrack, a new paradigm using clicking interaction for real-time scenarios, and the Guided Click Refiner (GCR), which accepts point and optional textual information as inputs, transforming the point into the bounding box expected by the operator.
- Deep contour and symmetry scored object proposal6
This paper proposes leveraging the essential properties of images, i.e., contour and symmetry, to score the redundant region proposals, and outperforms the state-of-the-art on AUC and uses much fewer object proposals to achieve comparable recall rate.
- 6
- 6
- Rich Image Description Based on Regions6
A method of generating rich image descriptions from image regions by detecting regions with R-CNN and utilizing the RNN to generate sentences for image regions and proposing an optimization method to select one suitable region.
- Feature evaluation by particle filter for adaptive object tracking6
An online combined feature evaluation method for visual object tracking that adapts to changing appearances of both tracked object and background and extends the traditional filter framework from modeling motion states to modeling feature evaluation.
- Objectionable Image Recognition System in Compression Domain6
An intelligent recognition system for objectionable image in JPEG compression domain applies robust skin color model and skin texture analysis to detect the skin regions in an input image based on its DC and ACs and input into a decision-tree classifier for classification.
- 5
- 5
- Instance-level Visual Active Tracking with Occlusion-Aware Planning4
OA-VAT is proposed, a unified pipeline with three complementary modules that delivers robust, real-time performance for practical deployment of visual active tracking in real-world deployment.
- DynRefer: Delving into Region-level Multimodal Tasks via Dynamic Resolution4
A DynRefer approach, to pursue high-accuracy region-level referring through mimicking the resolution adaptability of human visual cognition, which achieves state-of-the-art results on multiple region-level multimodal tasks using a single model.
- Thinking with Images via Self-Calling Agent4
Self-Calling Chain-of-Thought with self-calling is proposed, a novel visual reasoning paradigm that reformulates iMCoT as a language-only CoT with self-calling that employs group-relative policy optimization to reinforce effective reasoning behavior to enhance optimization.
- ClawMachine: Learning to Fetch Visual Tokens for Referential Comprehension4
ClawMachine is presented, offering a new methodology that notates an entity directly using the visual tokens without additional syntax to unify the prompt and answer of visual referential tasks without additional syntax.
- 4
- 4
- 4
- Balancing Understanding and Generation in Discrete Diffusion Models3
XDLM offers a principled theoretical unification of MDLM and UDLM, recovering each paradigm as a special case, and an alleviated memory bottleneck enabled by an algebraic simplification of the posterior probabilities, which advances the Pareto frontier between understanding capability and generation quality.
- Delving Deep into Semantic Relation Distillation3
A novel methodology, Semantics-based Relation Knowledge Distillation (SeRKD), which reimagines knowledge distillation through a semantics-relation lens among each sample, which skillfully integrates superpixel-based semantic extraction with relation-based knowledge distillation for a sophisticated model compression and distillation.
- Discriminatively Matched Part Tokens for Pointly Supervised Instance Segmentation3
DMPT first allocates a token for each object part by finding a semantic extreme point, and then introduces part classifiers with deformable constraint to re-estimate part tokens which are utilized to guide and enhance the fine-grained localization capability of the self-attention mechanism.
- 3
- ReDDiT: Rehashing Noise for Discrete Visual Generation3
A rehashing noise approach for discrete diffusion transformer (termed ReDDiT), with the aim to extend absorbing states and improve expressive capacity of discrete diffusion models.
- Self-supervised feature-gate coupling for dynamic network pruning3
Experimental results validate that the proposed FGC method improves the baseline approach with significant margins, outperforming the state-of-the-arts with better accuracy-computation trade-off.
- Few-shot Learning for Multi-Modality Tasks3
To simulate the learning schema of humans and relieve the reliance on the large-scale annotation benchmarks, researchers start shifting towards the few-shot learning problem: they try to learn a model to correctly recognize novel categories with only a few annotated samples.
- Manifold constraint transfer for visual structure-driven optimization3
It is shown that manifold constraints can be transferred from the data to the optimized variables if these are linearly correlated and the resulting optimization problem can be solved with an efficient alternating direction method of multipliers that can consistently integrate the manifold constraints during the optimization process.
- 3
- 3
- 3
- 3
- Visual Object Tracking via One-Class SVM3
A new visual object tracking approach via one-class SVM (OC-SVM), inspired by the fact that OCSVM's support vectors can form a hyper-sphere, whose center can be regarded as a robust object estimation from samples, is proposed.
- 3
- Multi-posture Human Detection in Video Frames by Motion Contour Matching3
A thresholding method is used to classify a matching result into moving human or negatives by motion contour matching and in the matching process scale problem and interpersonal contour difference are considered.
- AceTone: Bridging Words and Colors for Conditional Image Grading2
AceTone is proposed, the first approach that supports multimodal conditioned color grading within a unified framework and achieves state-of-the-art performance on both text-guided and reference-guided grading tasks, improving LPIPS by up to 50% over existing methods.
- VideoAnchor: Reinforcing Subspace-Structured Visual Cues for Coherent Visual-Spatial Reasoning2
VideoAnchor is proposed, a plug-and-play module that leverages subspace affinities to reinforce visual cues across frames without retraining, effectively anchoring attention to shared visual structures.
- DWPIS: Dynamic-Weight Parallel Instance and Skeleton Network for Railway Centerline Detection2
A dynamic-weight parallel instance and skeleton network to directly extract the centerlines that can be viewed as skeletons and is useful to guide the autonomous flight of a UAV in railway inspection.
- Spectral Aware Softmax for Visible-Infrared Person Re-Identification2
The spectral-aware softmax (SA-Softmax) loss is proposed, which can fully explore the embedding space with the modality information and has clear interpretability and superior performance over the state-of-the-art methods in such a cross-modality condition.
- 2
- 2
- 2
- Visual trajectory analysis via Replicated Softmax-based models2
The Replicated Softmax model is applied to represent trajectories with “bag-of-word” features from a spatially distributed codebook and then used to characterize trajectory with latent topic units by stacking an additional label layer or representation layers.
- 2
- 2
- Fast and Robust Pedestrian Detection Algorithm with Multi-Scale Orientation Features2
The occluded features are completed with average features of training positives, given an occlusion model, which enable the proposed approach to work in crowd scenes.
- Image feature matching with network flow: a global optimization method2
A new approach is presented for obtaining feature matching based on the identified features from images by introducing the notion of a Min-cost K-flow Problem (MKP), which consists in finding a min-cost flow subject to the constraint that the flow value is K.
- 1
- Expandable Residual Approximation for Knowledge Distillation1
Inspired by the progressive approximation principle in the Stone–Weierstrass theorem, expandable residual approximation (ERA) is proposed, a novel KD method that decomposes the approximation of residual knowledge into multiple steps, reducing the difficulty of mimicking the teacher's representation through a divide-and-conquer approach.
- Overcomplete-to-sparse representation learning for few-shot class-incremental learning1
Overcomplete-to-sparse representation learning (O2SRL) solves the ‘new class generalization’ and ‘old class forgetting’ problems systematically by regularizing both feature completeness and sparsity.
- Mixture of Physical Priors Adapter for Parameter-Efficient Fine-Tuning1
A novel approach that models network weights by leveraging a combination of physical priors, enabling more accurate approximations and serving as a lightweight, plug-and-play module that seamlessly integrates into transformer architectures, with adaptable complexity depending on the local context.
- 1
- Global2Local: A Joint-Hierarchical Attention for Video Captioning1
A novel joint-hierarchical attention model is proposed, which embeds the key clips, the key frames and the key regions jointly into the captioning model in a hierarchical manner and demonstrates the superiority of the proposed method over the state-of-the-art methods.
- 1
- 1
- 1
- 1
- Dimension Measuring System of Round Parts Based on Machine Vision1
Experimental results proved that the measuring scheme can realize real time detection with high accuracy and has broad prospects for the application of dimension measurement.
- A Robust Caption Detecting Algorithm on MPEG Compressed Video1
An algorithm is proposed to detect captions in MPEG compressed video by using energy features extracted from DCT coefficients and a new region growing method named "density-based region growing" to connect blocks into candidate text regions.
- dRAE: Representation Autoencoder with Hyper-Spherical Codes–
This work proposes Hyper-Spherical Quantization (HSQ), which decouples semantic content from feature magnitude via angular routing, preventing code assignment from being dominated by scale rather than meaning.
- –
- –
- –
- Adaptively regularized SAR imaging network with efficient unfolding–
This study introduces a learnable regularizer to the unfolding network and proposes an adaptively regularized SAR imaging network (ARSAR-Net) for scene-agnostic imaging (imaging across heterogeneous scenes of varying sparsity levels).
- –
- –
- –
- Exploring Complicated Search Spaces With Interleaving-Free Sampling–
This study explores the search algorithm upon a complicated search space with long-distance connections and shows that existing weight-sharing search algorithms fail due to the existence of interleaved connections, and presents a simple-yet-effective algorithm, termed interleaving-free neural architecture search (IF-NAS).
- –
- –
- –
- –
- –
- –
- Robust Visual Object Tracking via Sparse Representation and Reconstruction–
This paper proposes a new approach for robust visual object tracking via sparse representation and reconstruction, where two main contributions are devoted in terms of object representation and location respectively.
- –
- –
- –
- Real-time pedestrian learning-tracking with information fusion–
The results of the experiments show good robustness and performance when facing extreme cases such as the object rotation, long-period occlusion, and high similarity between object and background.
- Visual object tracking via sparse reconstruction–
With the coefficient vector of the sparse representation, the tracked object is reconstructed in an instantaneous sample set, which improves the tracking adaptation to background variation, object shape change, and partial occlusion.
- Shaking video stabilization with content completion–
A new stabilization algorithm to counterbalance the shaking motion in a video based on classical Kandade-Lucas- Tomasi (KLT) method is presented in this paper and experiments demonstrate the effectiveness of the proposed algorithm.
- Text detection in natural scene images with feature combination–
Experimental results on a large dataset show that the combination of features in different detection stages is competent for text detection task.
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.