Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileWorks267 from public data
- BiSeNet: Bilateral Segmentation Network for Real-Time Semantic Segmentation2,851
A novel Bilateral Segmentation Network (BiSeNet) is proposed that makes a right balance between the speed and segmentation performance on Cityscapes, CamVid, and COCO-Stuff datasets.
- BiSeNet V2: Bilateral Network with Guided Aggregation for Real-Time Semantic Segmentation1,760
This work proposes an efficient and effective architecture with a good trade-off between speed and accuracy, termed Bilateral Segmentation Network (BiSeNet V2), and designs a guided aggregation layer to enhance mutual connections and fuse both types of feature representation.
- Learning a Discriminative Feature Network for Semantic Segmentation871
This work proposes a Discriminative Feature Network (DFN), which contains two sub-networks: Smooth Network and Border Network, which is specially design to handle the intra-class inconsistency problem and to make the bilateral features of boundary distinguishable with deep semantic boundary supervision.
- Lite-HRNet: A Lightweight High-Resolution Network503
An efficient high-resolution network, Lite-HRNet, is presented, which demonstrates superior results on human pose estimation over popular lightweight networks and can be easily applied to semantic segmentation task in the same lightweight manner.
- Domain Adaptation for Image Dehazing477
This work proposes a domain adaptation paradigm, which consists of an image translation module and two image dehazing modules, and applies a bidirectional translation network to bridge the gap between the synthetic and real domains by translating images from one domain to another.
- Context Prior for Scene Segmentation313
This work develops a Context Prior with the supervision of the Affinity Loss, an effective Context Prior Network that can selectively capture the intra-class and inter-class contextual dependencies, leading to robust feature representation.
- 260
- Temporal Context Aggregation Network for Temporal Action Proposal Refinement224
A Local-Global Temporal Encoder (LGTE) and Temporal Boundary Regressor (TBR) are designed to combine these two regression granularities in an end-to-end fashion, which achieves the precise boundaries and reliable confidence of proposals through progressive refinement.
- 183
- OadTR: Online Action Detection with Transformers174
The experimental results show that OadTR achieves higher training and inference speeds than current RNN based approaches, and significantly outperforms the state-of-the-art methods in terms of both mAP and mcAP.
- Re-ID Driven Localization Refinement for Person Search160
A differentiable ROI transform layer is developed to effectively transform the bounding boxes from the original images so that the box coordinates can be supervised by the re-ID training other than the original detection task.
- Learning a Discriminative Prior for Blind Image Deblurring160
This work forms the image prior as a binary classifier which can be achieved by a deep convolutional neural network (CNN) and is able to distinguish whether an input image is clear or not.
- 152
- Hybrid Relation Guided Set Matching for Few-shot Action Recognition138
A novel Hybrid Relation guided Set Matching (HyRSM) approach that incorporates two key components: hybrid relation module and set matching metric that can be highly informative and flexible to predict query categories under the few-shot settings is proposed.
- UniAnimate: taming unified video diffusion models for consistent human image animation124
This work proposes an alternative temporal modeling architecture based on a state space model to replace the original computation-consuming temporal Transformer and achieves superior synthesis results over existing state-of-the-art counterparts in both quantitative and qualitative evaluations.
- Blind Image Deblurring via Deep Discriminative Priors118
This work forms the image prior as a binary classifier using a deep convolutional neural network to handle image dehazing and develops an efficient numerical approach based on the half-quadratic splitting method and gradient descent algorithm to optimize the proposed model.
- CLIP-guided Prototype Modulating for Few-shot Action Recognition106
This work presents a CLIP-guided prototype modulating framework called CLIP-FSAR, which consists of two key components: a video-text contrastive objective and a prototype modulation that bridges the task discrepancy between CLIP and the few-shot video task by contrasting videos and corresponding class text descriptions.
- VideoLCM: Video Latent Consistency Model100
The VideoLCM framework is presented, which leverages the concept of consistency models from image generation to efficiently synthesize videos with minimal steps while maintaining high quality and computational efficiency.
- MoLo: Motion-Augmented Long-Short Contrastive Learning for Few-Shot Action Recognition100
A Motion-augmented Long-short Contrastive Learning (MoLo) method that contains two crucial components, including a long-short contrastive objective and a motion autodecoder, that can simultaneously learn long-range temporal context and motion cues for comprehensive few-shot matching.
- PLIP: Language-Image Pre-training for Person Representation Learning96
A novel language-image pre-training framework for person representation learning, termed PLIP, which significantly improves existing methods on all these tasks, but also shows great ability in the zero-shot and domain generalization settings.
- Structure-aware human pose estimation with graph convolutional networks94
This paper proposes a novel model, termed Pose Graph Convolutional Network (PGCN), to exploit the structural relationships between body key points to exploit these important relationships for pose estimation, and shows that this method consistently outperforms competing state-of-the-art methods.
- Holmes-VAU: Towards Long-term Video Anomaly Understanding at Any Granularity91
HIVAU-70k is introduced, a large-scale benchmark for hierarchical video anomaly understanding across any granularity and a semi-automated annotation engine that efficiently scales high-quality annotations by combining manual video segmentation with recursive free-text annotation using large language models (LLMs).
- Holmes-VAD: Towards Unbiased and Explainable Video Anomaly Detection via Multi-modal LLM89
Extensive experimental results validate the generality and interpretability of the proposed Holmes-VAD, establishing it as a novel interpretable technique for real-world video anomaly analysis.
- FGN: Fully Guided Network for Few-Shot Instance Segmentation84
A Fully Guided Network for few-shot instance segmentation is presented, which introduces different guidance mechanisms into the various key components in Mask R-CNN in order to make full use of the guidance effect from the support set and adapt better to the inter-class generalization.
- UFineBench: Towards Text-based Person Retrieval with Ultra-fine Granularity80
CFAM, a more efficient al-gorithm especially designed for text-based person retrieval with ultra fine-grained texts is proposed, which achieves fine granularity mining by adopting a shared cross-modal granularity de-coder and hard negative match mechanism.
- Self-Supervised Learning for Semi-Supervised Temporal Action Proposal80
This paper designs an effective Self-supervised Semi- supervised Temporal Action Proposal framework, which significantly outperforms state-of-the-art semi- Supervised methods and even matches fully-super supervised methods.
- Multi-Centroid Representation Network for Domain Adaptive Person Re-ID79
This paper presents a novel Multi-Centroid Memory (MCM) to adaptively capture different identity information within the cluster and proposes Second-Order Nearest Interpolation (SONI) to obtain abundant and informative negative samples.
- Probabilistic class structure regularized sparse representation graph for semi-supervised hyperspectral image classification75
By incorporating the class structure information into the SR model, PCSSR can learn a discriminative graph from the data and outperforms state of the art methods on Hyperion and AVIRIS hyperspectral data.
- MAR: M asked Autoencoders for Efficient A ction R ecognition72
This work proposes Masked Action Recognition (MAR), which reduces redundant computation by discarding a proportion of patches and operating only on a portion of the videos, and can reduce the computational cost of ViT by 53%.
- A Recipe for Scaling up Text-to-Video Generation with Text-free Videos70
A novel text-to-video generation framework, termed TF-T2V, which can directly learn with text-free videos and validate the effectiveness and generalizability of the ideology on both native text-to-video generation and compositional video synthesis paradigms.
- Hierarchical Spatio-temporal Decoupling for Text-to- Video Generation70
HiGen is proposed, a diffusion model-based method that improves performance by decoupling the spatial and temporal factors of videos from two perspectives, i.e., structure level and content level.
- Adversarial Semantic Data Augmentation for Human Pose Estimation69
Semantic Data Augmentation (SDA), a method that augments images by pasting segmented body parts with various semantic granularity, is proposed, which exploits a generative network to dynamiclly predict tailored pasting configuration.
- 69
- 69
- Style Transformer for Image Inversion and Editing67
A transformer-based image inversion and editing model for pretrained StyleGAN which is not only with less distortions, but also of high quality and flexibility for editing, and investigates the reference-and label-based attribute editing through a pretrained latent classifier.
- GTNet: Generative Transfer Network for Zero-Shot Object Detection65
An IoU-Aware Generative Adversarial Network (IoUGAN) is designed as the feature synthesizer, which can be easily integrated into GTNet, and performs favorably against the state-of-the-art ZSD approaches.
- 63
- Do Not Disturb Me: Person Re-identification Under the Interference of Other Pedestrians63
A novel deep network termed Pedestrian-Interference Suppression Network (PISNet) is presented, which leverages a Query-Guided Attention Block (QGAB) to enhance the feature of the target in the gallery, under the guidance of the query.
- Equidistance constrained metric learning for person re-identification61
An algorithm for learning a Mahalanobis distance for person re-identification that outperforms the state-of-the-art methods on CUHK01, CUHK03, Market1501 and DukeMTMC-reID datasets, and achieves very competitive performance on the widely used VIPeR dataset.
- 58
- 57
- Decoupled and Memory-Reinforced Networks: Towards Effective Feature Learning for One-Step Person Search56
To reconcile the conflicts of multiple objectives, the standard tightly coupled pipelines are simplified and a deeply decoupled multi-task learning framework is established and a memory-reinforced mechanism is built to boost the identification feature learning.
- 53
- Open-Vocabulary Semantic Segmentation with Image Embedding Balancing52
A novel framework for open-vocabulary semantic segmentation called EBSeg is proposed, incorpo-rating an Adaptively Balanced Decoder (AdaB Decoder) and a Semantic Structure Consistency loss (SSC Loss) to learn a consistent semantic structure from CLIP.
- Graph coloring based surveillance video synopsis50
This work proposes a novel approach to deal with tube rearrangement by formulating it as a graph coloring problem, which can generate more compact video synopsis with less collision artifacts than the existing methods.
- Disentangling Spatial and Temporal Learning for Efficient Image-to-Video Transfer Learning48
DiST is presented, which disentangles the learning of spatial and temporal aspects of videos by using a dual-encoder structure, where a pre-trained foundation model acts as the spatial encoder, and a lightweight network is introduced as the temporal encoder.
- 48
- UniAnimate-DiT: Human Image Animation with Large-Scale Video Diffusion Transformer46
To preserve the robust generative capabilities of the original Wan2.1 model, Low-Rank Adaptation (LoRA) technique is implemented to fine-tune a minimal set of parameters, significantly reducing training memory overhead.
- 45
- Representative Graph Neural Network44
This paper presents a Representative Graph (RepGraph) layer to dynamically sample a few representative features, which dramatically reduces redundancy and can improve the performance on the COCO dataset compared to the non-local operation.
- 42
- Tunnel Try-on: Excavating Spatial-temporal Tunnels for High-quality Virtual Try-on in Videos41
Tunnel Try-on is a diffusion-based framework named ''Tunnel Try-on'' that excavates a ''focus tunnel'' in the input video that gives close-up shots around the clothing regions and develops an environment encoder to extract the context information outside the tunnels.
- 40
- 40
- Hard sample mining makes person re-identification more efficient and accurate39
An Adaptive Hard Sample Mining algorithm is proposed for training a robust person re-identification model that has good generalization performance that can be combined with various existing models readily and alleviate the under-fitting and over-fitting problem simultaneously.
- Towards General Low-Light Raw Noise Synthesis and Modeling38
This work synthesizes the signal-dependent and signal-independent noise in a physics-and learning-based manner, respectively, and presents an effective multi-scale discriminator termed Fourier transformer discriminator (FTD) to distinguish the noise distribution accurately.
- 36
- 34
- 34
- 33
- Weakly Supervised Person Search with Region Siamese Networks33
This paper presents a weakly supervised setting where only bounding box annotations are available, and provides an effective baseline model termed Region Siamese Networks (R-SiamNets), which surpasses several fully supervised methods, such as OIM and MGTS, by a clear margin.
- Relevant region prediction for crowd counting33
Relevant Region Prediction (RRP) for crowd counting is proposed, which consists of the Count Map and the Region Relation-Aware Module (RRAM), which obviously outperforms other existing state-of-the-art methods.
- 33
- 32
- WorldLens: Full-Spectrum Evaluations of Driving World Models in Real World29
This work introduces WorldLens, a full-spectrum benchmark evaluating how well a model builds, understands, and behaves within its generated world, and develops WorldLens-Agent, an evaluation model distilled from these annotations to enable scalable, explainable scoring.
- 28
- Text detection approach based on confidence map and context information28
An approach based on the confidence map and context information is proposed to robustly detect texts in natural scenes to improve the recall rate and achieve the competitive performances by comparing with the other state-of-the-art methods.
- 26
- Pyramid-Based Multi-structure Local Binary Pattern for Texture Classification26
The drawback of conventional LBP operators in describing some textures that has the same small structures but differential large structures is investigated and a multi-structure local binary pattern operator is achieved by executing the LBP method on different layers of image pyramid.
- 25
- On selection and combination of weak learners in AdaBoost25
A new AdaBoost algorithm is proposed to make improvement in the two aspects of selecting the most discriminative weak learners and combining the weak learners optimally by tuning the coefficients using kernel-based perceptron.
- CTR-Driven Advertising Image Generation with Multimodal Large Language Models24
A novel reward model is proposed to fine-tune pre-trained MLLMs through Reinforcement Learning (RL), which can jointly utilize multimodal features and accurately reflect user click preferences, enhancing the overall relevance and effectiveness of the advertising images.
- 24
- Learning from Untrimmed Videos: Self-Supervised Video Representation Learning with Hierarchical Consistency24
A hierarchical consistency learning framework HiCo is presented, and it is shown that not only HiCo can generate stronger representations on untrimmed videos, it also improves the representation quality when applied to trimmed videos, in contrast to standard contrastive learning that fails to learn appropriate representations from untrimed videos.
- 24
- Pose-guided spatiotemporal alignment for video-based person Re-identification24
A pose-guided spatiotemporal alignment method to align video sequences for video-based person re-id tasks and shows that the proposed method performs favorably against state-of-the-art methods, even deep learning-based approaches.
- 24
- 24
- 23
- Joint image deblurring and matching with feature-based sparse representation prior23
This approach utilizes two-directional two-dimensional (2D)2PCA to extract feature vectors from images and obtains a sparse representation prior in a robust feature space rather than the original pixel space, thus mitigating the influence of image blur.
- 23
- 23
- Hierarchical Feature Embedding for Visual Tracking22
A hierarchical feature embedding model is proposed which separately learns the instance and category information, and progressively embeds them, and develops the instance-aware and category-aware modules that collaborate from different semantic levels to produce discriminative and robust feature embeddings.
- Completed local similarity pattern for color image recognition22
Experimental results show that CLSP outperforms state-of-the-art color image feature extraction methods and has the merits of robustness and compactness of coding based feature extraction method.
- Lookup Table meets Local Laplacian Filter: Pyramid Reconstruction Network for Tone Mapping21
A novel strategy that integrates global and local operators by utilizing closed-form Laplacian pyramid decomposition and reconstruction is explored, which achieves simultaneous global tone manipulation and local edge detail preservation in an end-to-end manner.
- 21
- Multi-structure local binary patterns for texture classification21
A multi-structure local binary pattern (Ms-LBP) operator is achieved by executing the extended LBP operator on different layers of an image pyramid by changing the shape of sampling region and getting an extended L BP operator.
- 20
- LEDTD: Local edge direction and texture descriptor for face recognition19
LEDTD differs from the existing Kirsch based methods in a manner that it not only considers image edge direction information but also extracts image texture feature by encoding the edge response directions of center and its neighborhood pixels by employing local XOR binary coding strategy.
- Attribute-specific Control Units in StyleGAN for Fine-grained Image Manipulation18
This work proposes a simple yet effective method to detect attribute-specific control units, which consist of multiple channels of feature maps and modulation styles and collaboratively manipulate the modulation style channels and feature maps in control units rather than individual ones to obtain the semantic and spatial disentangled controls.
- CCSD: cross-camera self-distillation for unsupervised person re-identification17
This paper proposes a cross-camera self-distillation (CCSD) framework for unsupervised person Re-ID to alleviate the effect of camera variation, and proposes a camera-aware cluster refinement mechanism, which first splits each cluster into multiple clusters according to the camera views, and then refines them into more compact clusters.
- 17
- Multiscale face recognition in cluttered backgrounds based on visual attention16
This study proposes the attention developmental network to recognize multiscale faces without using face detectors and can attain at least 13% of accuracy improvement over bionic neural networks and ResNet-based recognition networks on the same model scale with less training epochs.
- 16
- Cross-video Identity Correlating for Person Re-identification Pre-training15
A Cross-video Identity-cOrrelating pre-traiNing (CION) framework, which seeks the identity correlation from cross-video images by modeling it as a progressive multi-level denoising problem, and contributes a model zoo named ReIDZoo to meet diverse research and application needs in this field.
- 15
- 15
- 14
- Multi-level Temporal Pyramid Network for Action Detection14
This paper proposes a Multi-Level Temporal Pyramid Network (MLTPN), which can learn rich and discriminative features for different action instances with different durations and designs a simple yet effective feature fusion module to fuse the multi-level multi-scale features.
- 14
- Motion-blur kernel size estimation via learning a convolutional neural network14
Quantitative and qualitative results of experiments show that the method can not only estimate the motion blur kernel size accurately but also help to yield better pictures with estimated kernel sizes as input parameters.
- 13
- 13
- High-Resolution Photo Enhancement in Real-Time: A Laplacian Pyramid Network12
A pyramid network called LLF-LUT++ is introduced, which integrates global and local operators through closed-form Laplacian pyramid decomposition and reconstruction, which enables fast processing of high-resolution images while also achieving excellent performance.
- 12
- Log-Gabor Weber descriptor for face recognition12
A local image descriptor for face recognition, called Log-Gabor Weber descriptor (LGWD), which applies the Weber local binary pattern operator to each transform response, and yields a better performance than state-of-the art methods.
- 12
- Learning to Tell Apart: Weakly Supervised Video Anomaly Detection via Disentangled Semantic Alignment11
This work proposes a novel Disentangled Semantic Alignment Network (DSANet) to explicitly separate abnormal and normal features from coarse-grained and fine-grained aspects, enhancing the distinguishability of video anomaly detection results.
- Query-centric distance modulator for few-shot classification11
A QCDM is proposed to generate query-related weights for each query sample to address the issue of similarity between samples.
- Color and Luminance Separated Enhancement for Low-Light Images with Brightness Guidance11
A straightforward yet effective strategy for Retinex decomposition is introduced, dividing images into colormaps and graymaps as new estimations for reflectance and illumination maps, and addressing the dual challenge of perturbation removal and brightness adjustment in illumination maps by incorporating brightness guidance.
- 11
- CSENet: Cascade semantic erasing network for weakly-supervised semantic segmentation11
A Cascade Semantic Erasing Network (CSENet) is proposed to expand seeds effectively and reasonably and forces the network to exploit relevant feature response for non-discriminative object districts.
- Orientation Adaptive YOLOv3 for Object Detection in Remote Sensing Images11
This paper modified YOLOv3 based on the oriented bounding box (OBB) for object detection in remote images to solve the problems above and can obtain bounding boxes more suitable for large aspect ratio objects.
- 11
- Adaptive Prototype Replay for Class Incremental Semantic Segmentation10
The Adaptive prototype replay (Adapter) for CISS is proposed, which comprises an adaptive deviation compensation (ADC) strategy and an uncertainty-aware constraint (UAC) loss that dynamically updates the stored prototypes based on the estimated representation shift distance to match the updated representation of old class.
- 10
- 10
- HTSTL: Head-and-Tail Search Network With Scale-Transfer Layer for Traffic Sign Text Detection10
An effective detector for traffic sign text is designed, whose pipeline only consists of a preprocessing module to tackle with some complex situations, a Fully Convolutional Network (FCN) in which a Scale-transfer layer is proposed to speed up the network and a simple post-processing step.
- A discriminant sparse representation graph-based semi-supervised learning for hyperspectral image classification10
This paper presents a discriminant sparse representation (DSR) graph for semi-supervised learning (SSL) to address the classification of hyperspectral image with a paucity of labeled samples.
- Face recognition with Riesz binary pattern10
Experimental results on three databases demonstrate that the proposed RBP descriptor is more discriminant in extracting image information and can provide a higher classification rate compared to some state-of-the-art image representation methods.
- Action recognition through discovering distinctive action parts10
This paper presents a method of automatically discovering distinctive midlevel action parts from video for recognition of human actions by learning and selecting a collection of discriminative and representative action part detectors directly from video data using novel coverage-entropy curves (CE curves).
- 10
- 9
- 9
- Parallel Reasoning Network for Human-Object Interaction Detection9
A new transformer-based method named Parallel Reasoning Network (PR-Net), which constructs two independent predictors for instance-level localization and relation-level understanding, which has achieved competitive results on HICO-DET and V-COCO benchmarks.
- Online Unsupervised Learning Classification of Pedestrian and Vehicle for Video Surveillance9
Experimental results illustrate that the unsupervised learning algorithm is adapted to polymorphism of the pedestrians and diversity of the vehicles with high classification accuracy.
- GlanceVAD: Exploring Glance Supervision for Label-efficient Video Anomaly Detection8
This paper presents a novel labeling paradigm, termed "glance annotation", to achieve a better balance between anomaly detection accuracy and annotation cost, and proposes a customized GlanceVAD method, that leverages gaussian kernels as the basic unit to compose the temporal anomaly distribution.
- Few-Shot Learning for Annotation-Efficient Nucleus Instance Segmentation8
This work adopts the meta-learning based FSL paradigm, and extends the basic definition of few-shot instance segmentation (FSIS) to generalized few-shot instance segmentation (GFSIS), introducing a structural guidance mechanism into the GFSIS network to cope with the intrinsic challenges of nucleus segmentation.
- 8
- 8
- 8
- Camera Style and Identity Disentangling Network for Person Re-identification.8
A Camstyle-Identity Disentangling (CID) network for person ReID that disentangle the ID feature and camstyle feature in the latent space and is able to eliminate the camstyle features in the backbone during the training process.
- 8
- 7
- 7
- Semi-supervised Discriminant Analysis and Sparse Representation-based self-training for Face Recognition7
A new semi-supervised self-training approach is proposed which is used to automatically augment the manually labeled training set with new unlabeled data and outperforms other face recognition methods under 3 different configurations, namely transductive, semi- supervised and single training image.
- Textured image segmentation based on modulation models7
This work provides a feature vector that strengthens useful information and has fewer dimensionalities simultaneously in textured image segmentation based on amplitude-modulation frequency- modulation models.
- 6
- 6
- Jointly detecting and multiple people tracking by semantic and scene information6
A new method for online multiple people tracking is proposed, which combines the detection process and the single object tracking process, and establishes the interactions between them.
- 6
- 6
- Spatial multi-scale gradient orientation consistency for place instance and Scene category recognition6
This paper presents a novel image feature called Gradient Orientation Consistency (GOC) and an image representation based on the spatial multi-scale GOC histogram for place instance and scene category recognition and shows that it outperforms the state of the art.
- 6
- Biologically Inspired Scene Context for Object Detection Using a Single Instance6
The proposed object detection method is robust to large intra-class variations, occlusions, a minor variety of poses, low-revolution conditions, background clutter etc., and there is no off-line training.
- 5
- Is Your Driving World Model an All-Around Player?5
WorldLens is introduced, a unified benchmark that measures world-model fidelity across the full spectrum, from pixel quality and 4D geometry to closed-loop driving and human perceptual alignment, through five complementary aspects and 24 standardized dimensions and forms a unified ecosystem for assessing generated worlds not merely by visual appeal, but by physical and behavioral fidelity.
- 5
- Continual Gaussian Mixture Distribution Modeling for Class Incremental Semantic Segmentation5
This work proposes a Continual Gaussian Mixture Distribution (CoGaMiD) modeling method, which uses the means and covariance matrices of the Gaussian Mixture Models (GMMs) to model the complete feature distributions of learned classes and introduces a Dynamic Adjustment strategy that utilizes the features of previous classes within incoming data streams to update the stored GMMs.
- 5
- 5
- Superpixel-Based Temporally Aligned Representation for Video-Based Person Re-Identification5
This work proposes an effective super pixel-based temporally aligned representation for video-based person re-identification, which represents a video sequence only using one walking cycle, and proposes a superpixel-based representation to improve spatial alignment.
- EXEMPLAR-BASED LINEAR DISCRIMINANT ANALYSIS FOR ROBUST OBJECT TRACKING5
This paper proposes a novel tracking framework based on exemplar detector rather than category detector, an ensemble of exemplar-based linear discriminant analysis (ELDA) detectors, which is quite specific and discriminative.
- Collaborative multicue fusion using the cross-diffusion process for salient object detection5
This paper presents a multicue fusion method using the cross-diffusion process (CDP) for salient object detection that allows for collaborative fusion, that is, the individual visual cues to be fused are able to interact and exchange information with each other during the fusion procedure, leading to more robust and effective fusion results.
- A low‐cost real‐time face tracking system for ITSs and SDASs5
Experimental results on three challenging video sequences show that the proposed face tracking system outperforms the state‐of‐the‐art face tracker and consumes only 5–13% CPU resources of a low‐spec onboard computer while processing in real time.
- 4
- 4
- EWST: an extreme weather scene text detector with dehazing and localization refinement4
Experimental results illustrate that the method is superior to some state-of-the-art methods in the hazing scene and the hybrid scene, and it is found that the general non-maximum suppression (NMS) method or the soft-NMS method are difficult to precisely match the ground truth, and the incomplete feature extraction will affect the final performance.
- 4
- Context-aware Proposal Network for Temporal Action Detection4
This technical report presents the first place winning solution for temporal action detection task in CVPR-2022 AcitivityNet Challenge, which builds on top of BMN, GTAD and PRN to aggregate contextual information by randomly masking some proposal features.
- 4
- Joint Image Deblurring and Matching with Blurred Invariant‐Based Sparse Representation Prior4
JDM-BISR is proposed, which obtains the sparse representation prior in the robust blurred invariant space rather than the original pixel space and thus can effectively improve the quality of image deblurring and the accuracy of image matching.
- Vehicle parts detection based on Faster - RCNN with location constraints of vehicle parts feature point4
A new approach to vehicle parts detection is proposed by combining Faster RCNN and three level cascaded convolutional neural network (DCNN), which can precisely predict feature point position, which is the center of vehicle part.
- Learning deep features with adaptive triplet loss for person reidentification4
This paper makes the attempt in jointly learning both the global full-body and local body-parts features of the input persons with a multichannel convolutional neural network (CNN) model, which is trained by an adaptive triplet loss function that serves to minimize the distance between the same person and maximize thedistance between different persons.
- Enhancement of ELDA Tracker Based on CNN Features and Adaptive Model Update4
This work improves the ELDA tracking algorithm by deep convolutional neural network (CNN) features and adaptive model update and builds the background model as a Gaussian mixture model (GMM) to adapt to complex scenes, which is initialized offline and updated online.
- 4
- Instance-based attention: where could humans look first when searching for an object instance4
An instance-based attention model is presented to predict where humans could look first when searching for an object instance, and its application in image synthesis is shown.
- FilmWorld: Agentic Novel-to-Film Generation through Dynamic Cinematic World Modeling3
FilmEval is introduced, a systematic evaluation framework that couples a difficulty-graded benchmark of 15 representative novels with an automated protocol of nine objective metrics spanning three dimensions: cinematic presentation, film consistency, and novel fidelity.
- Structural Pruning via Spatial-aware Information Redundancy for Semantic Segmentation3
A novel approach, denoted as Spatial-aware Information Redundancy Filter Pruning (SIRFP), which aims to reduce feature redundancy between channels by introducing a spatial-aware redundancy metric based on feature maps, thus endowing the pruning process with location sensitivity to better adapt to pruning segmentation networks.
- 3
- REPAIR: Rank Correlation and Noisy Pair Half-Replacing With Memory for Noisy Correspondence3
A generalized framework termed as Rank corrElation and noisy Pair hAlf-replacing wIth memoRy (REPAIR), which benefits from maintaining a memory bank for features of matched pairs, and calculates the distances between the features in the memory bank and those of the target pair for each respective modality to estimate the soft correspondence label of the target pair.
- Object-Aware Video Matting with Cross-Frame Guidance3
A trimap-free Object-Aware Video Matting (OAVM) framework, which can perceive different objects, enabling joint recognition of foreground objects and refinement of edge details, and designs a Sequential Foreground Merging augmentation strategy to diversify sequential scenarios and enhance capacity of the network for object discrimination.
- 3
- 3
- 3
- Partial Forward Blocking: A Novel Data Pruning Paradigm for Lossless Training Acceleration3
Partial Forward Blocking is proposed, a novel framework for lossless training acceleration that significantly reduces the computational overhead of deep-layer forward passes and back-propagation for pruned samples, while also eliminating the need for auxiliary backward computations and proxy model training.
- ParamCrop: Parametric Cubic Cropping for Video Contrastive Learning3
It is shown that adaptively controlling the disparity between two augmented views along the training process enhances the quality of the learned representations, and this adaptive and gradual increase in the disparity yielded by ParamCrop is beneficial to learning a strong and generalized representation for downstream tasks.
- 3
- Joint image restoration and matching method based on distance-weighted sparse representation prior3
This paper proposes a joint blurred image restoration and matching method based on distance-weighted sparse representation (JRM-DSR), which utilizes the sparse representation prior to exploit the correlation between Restoration and matching.
- Robust and Efficient Ellipse Fitting Using Tangent Chord Distance3
A novel geometric distance, called Tangent Chord Distance (TCD), is introduced, to formulate the ellipse fitting problem, under the least squares framework, and can thereby significantly reduce the computational load without sacrificing the performance.
- 3
- 3
- Re-ranking Person Re-identification with Adaptive Hard Sample Mining3
An adaptive hard sample mining method is used to re-train the selected samples in order to distinguish similar persons, which is applied for re-ranking the re-ID results, and proposes a coarse-fine tuning mechanism, which adaptively inflicts different punishment on the negative samples with the ranking information.
- Local Fractional Order Derivative Vector Quantization Pattern for Face Recognition3
A novel local image descriptor, local fractional order derivative vector quantization pattern (fVQP), based on image local directional fractionalorder derivative feature vector and vector quantification method for face recognition, which has the advantages of better image recognition performance and robust to noise.
- Object localization with mid-level part detectors3
A method for object detection combining the effectiveness of a set of mid-level parts learned weak-supervised from object bounding box annotations that is competitive with the state of art in localizing the object.
- 3
- 3
- 2
- Learning Unpaired Image Dehazing with Physics-based Rehazy Generation2
This paper proposes a novel training strategy for unpaired image dehazing, termed Rehazy, to improve both dehazing performance and training stability and introduces a dual-branch framework for dehazing network training, which progressively restores clean scenes from coarse to fine.
- 2
- 2
- 2
- Adaptive Semantic Consistency for Cross-domain Few-shot Classification2
A simple plug-and-play Adaptive Semantic Consistency (ASC) framework, which improves cross-domain robustness by preserving source transfer capability during the finetuning stage and enables explicit transfer of source domain knowledge to prevent the model from overfitting the target domain.
- 2
- 2
- 2
- A Cuboid Bi-Level Log Operator for Action Classification2
This work proposes a 3-D cuboid bi-level Laplacian-of-Gaussian (CBLoG) operator with high speed and invariant space–time scale for detecting abrupt changes of signals in videos and outperforms the state of the art tested on the KTH data set and achieves impressive performance on the UCF sports data set.
- Data Association Based Multi-target Tracking Using a Joint Formulation2
This work uses a Hough Forest (HF) based learning framework to effectively learn a discriminative codebook of features among tracklets by utilizing appearance and motion cues stored in the leaf nodes to directly model the joint formulation of pairs of tracklets for association in the CRF framework.
- 2
- 2
- 1
- 1
- ClothHMR: 3D Mesh Recovery of Humans in Diverse Clothing from Single Image1
ClothHMR can accurately recover 3D meshes of humans wearing diverse clothing, precisely estimating their body shapes and poses and significantly outperforms existing state-of-the-art methods across benchmark datasets and in-the-wild images.
- 1
- Instance-Aware Robust Consistency Regularization for Semi-Supervised Nuclei Instance Segmentation1
The proposed Instance-Aware Robust Consistency Regularization Network (IRCR-Net) significantly enhances semi-supervised nuclei instance segmentation performance across multiple public datasets compared to existing approaches, even surpassing fully supervised methods in some scenarios.
- L-Man: A Large Multi-modal Model Unifying Human-centric Tasks1
A large multi-modal model which utilizes LLMs to unify various human-centric tasks shows significant superiority on human-centric tasks compared with existing large multi-modal models, and also achieves even better results on downstream datasets compared with respective task-specific models.
- Taming Consistency Distillation for Accelerated Human Image Animation1
This paper proposes the DanceLCM approach complemented by several enhancements to improve visual quality and motion continuity at low-step regime, and achieves results comparable to state-of-the-art video diffusion models with a mere 2-4 inference steps.
- 1
- 1
- 1
- 1
- 1
- 1
- Blind deblurring of Gaussian blurred images by blurred edge image1
The edges of the blurred images to be incorporated with the blurred image as the input of the DeepEdgeGAN to provide a strong prior constraint for the restoration, which is beneficial to solve the problem that gradients of the restored images with GANs methods tend to be smooth and not clear enough.
- Co-occurrence relationship encoding via channel merging for vehicle part recognition1
A deep neural network is constructed for vehicle part recognition that can use the co-occurrence relationship and recognize two vehicle part simultaneously simultaneously and performs favorably against the state-of-the-art vehicle recognition algorithms.
- Blurred Template Matching Based on Cascaded Network1
This paper proposes a blurred template matching method based on a cascaded network, which combines a coarse matching network and a fine matching network, and significantly outperforms the state-of-art on the accuracy, speed and robustness.
- Densenet-Based Multi-scale Recurrent Network for Video Restoration with Gaussian Blur1
This work proposes an efficient approach to produce a significant amount of realistic training data and introduces a novel multi-scale recurrent network architecture to deblur frames taking temporal information into account and demonstrates the effectiveness of the proposed method.
- Graph-Based Scale-Aware Network for Human Parsing1
A graph-Based Scale-Aware Network for human parsing that embeds a Graph-Based Part Reasoning Layer into the backbone network to reason the relative relationship between human parts and adopts an edge supervision to further improve the performance.
- Center-Level Verification Model for Person Re-identification1
A novel center-level verification (CLEVER) model for the siamese network, which simply represents the distribution as a center and calculates the loss based on the center, which gets state-of-the-art results on CUHK03, CUHK01 and VIPeR datasets.
- Detection of vehicle parts based on Faster R-CNN and relative position information1
This paper presents a method which corrects original score with relative position information between two parts, and chooses the box with maximum comprehensive score as the final result.
- 1
- Mixture Models for Object Detection1
An approach based on mixture of multiple components and mid-level part models for object detection in natural scenes using multi-component models and part models to describe the global variation and local deformation respectively.
- 1
- Text Detection in Natural Scene Images Leveraging Context Information1
This paper focuses on building a confidence map model by integrating the candidate appearance and the relationships with its adjacent candidates by leveraging context information for text detection in natural scene images.
- A hierarchical feature graph matching method for recognition of complex human activities1
This paper focuses on the problem of recognition of complex human activities in an example-based video retrieval framework and proposes a new method based on hierarchical feature-graph matching that does not require any preprocessing and gives reasonable results with even a small number of query examples.
- 1
- Visual codebook construction for class-specific recognition1
A simple and effective method on the basis of a feedforward hierarchical model, a robust local descriptor is proposed and an a priori statistical scheme is applied to the class-specific feature-learning stage of visual feature codebook construction.
- –
- –
- –
- –
- –
- –
- –
- Selecting Samples on Graphs: A Unified Dataset Pruning Framework for Lossless Training Acceleration–
This work presents a unified graph-based DP framework that outperforms existing DP methods while substantially reducing training cost, reducing training time by over 40\% without sacrificing accuracy on ImageNet-1k with ResNet-50.
- –
- –
- –
- INTACT: Ego-Guided Typed Sparse Evidence Retrieval for Heterogeneous Collaborative Perception–
INTACT is proposed, an ego-guided typed sparse evidence retrieval framework for heterogeneous collaborative perception that lets the ego vehicle issue typed evidence queries that express suspected objects and evidence-deficient regions and changes the compatibility requirement from global feature-map interpretability to local, typed response comparability under ego-issued queries.
- Achieving Text-Based Person Retrieval With Any Granularity–
This paper introduces a new paradigm, Text-based Person Retrieval with Any Granularity, and proposes the Cross-modal Multi-grained Aligning and Matching (CMAM) framework, establishing a foundational benchmark and a robust baseline, paving the way for more practical person retrieval systems.
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- Scene text detection by leveraging multi-channel information and local context–
An approach based on multi-channel information and local context is proposed to detect texts in natural scenes and demonstrates that the proposed method have achieved the state-of-the-art performance.
- Selecting good regions to deblur via relative total variation–
This paper applies a metric named relative total variation (RTV) to discriminate the structure regions from smooth and texture and shows that the proposed method can help to estimate the kernel accurately.
- Global Feature Learning with Human Body Region Guided for Person Re-identification–
A multi-channel convolutional neural network with a novel embedding training strategy for training, which uses local features to guide learning more robust global features in person reidentification.
- Week texture objects pose estimation based on 3D model–
This paper proposes a 3D pose estimation method for week texture objects, by performing point matching of a test image to a matched rendering image of an object rather than its 3D model.
- –
- –
- –
- Hough Voting with Distinctive Mid-Level Parts for Object Detection–
An efficient method for object detection in natural scenes is presented via generalized Hough transform of distinctive midlevel parts, which are more meaningful than low-level patches such as lines or corners and would be able to cover the key structures of object.
- Fast Shape Re-ranking with Neighborhood Induced Similarity Measure–
This paper addresses the shape retrieval problem by casting it into the task of identifying "authority" nodes in an inferred similarity graph and also by re-ranking the shapes using the average similarity between a node and its neighboring nodes.
- –
- Object detection with geometric context of keypoints described as lifetime–
This paper proposed a novel method to represent local regions around keypoints, called lifetime, and showed that the approach has an excellent performance for object detection and localization.
- –
- –
Publication data from OpenAlex, with missing venues and authors filled in from Crossref; citation counts are the higher of OpenAlex and Semantic Scholar, last synced 2026-10-12. One-sentence summaries under some papers are written by Semantic Scholar’s model. Citation counts may be lower than on Google Scholar, which indexes more sources.
Report an error
Wrong papers, two people merged into one, or a profile that should not be here? Tell us and we will fix or hide it.