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
- MixFormer: End-to-End Tracking with Iterative Mixed Attention911
This paper proposes a compact tracking framework, termed as MixFormer, built upon transformers, to utilize the flexibility of attention operations, and proposes a Mixed Attention Module (MAM) for simultaneous feature extraction and target information integration.
- 618
- Residual Feature Distillation Network for Lightweight Image Super-Resolution610
The feature distillation connection (FDC) is proposed that is functionally equivalent to the channel splitting operation while being more lightweight and flexible and can rethink the information multi-distillation network (IMDN) and propose a lightweight and accurate SISR model called residual feature distilling network (RFDN).
- 537
- TDN: Temporal Difference Networks for Efficient Action Recognition506
A new video architecture, termed as Temporal Difference Network (TDN), with a focus on capturing multi-scale temporal information for efficient action recognition, which provides a simple and principled temporal modeling framework and could be instantiated with the existing CNNs at a small extra computational cost.
- Learning Actor Relation Graphs for Group Activity Recognition306
This paper proposes to build a flexible and efficient Actor Relation Graph (ARG) to simultaneously capture the appearance and position relation between actors, and performs extensive experiments on two standard group activity recognition datasets.
- SportsMOT: A Large Multi-Object Tracking Dataset in Multiple Sports Scenes257
A new multi-object tracking framework, termed as MixSort, introducing a MixFormer-like structure as an auxiliary association model to prevailing tracking-by-detection trackers is proposed, which achieves state-of-the-art performance on SportsMOT and MOT17.
- Relaxed Transformer Decoders for Direct Action Proposal Generation230
A simple and efficient framework (RTD-Net) for direct action proposal generation, by re-purposing a Transformer-alike architecture, which is more efficient than previous proposal generation methods, without non-maximum suppression post-processing.
- Robust Object Modeling for Visual Tracking203
A robust object modeling framework for visual tracking (ROMTrack) is proposed, which simultaneously models the inherent template and the hybrid template features, thus resulting in a more robust object modeling framework.
- Negative Sample Matters: A Renaissance of Metric Learning for Temporal Grounding190
A Mutual Matching Network (MMN) is presented, to directly model the similarity between language queries and video moments in a joint embedding space and suggests that metric learning is still a promising method for temporal grounding via capturing the essential cross-modal correlation in aointing space.
- Extracting Motion and Appearance via Inter-Frame Attention for Efficient Video Frame Interpolation180
This paper proposes a new module to explicitly extract motion and appearance information via a unified operation and rethink the information process in inter-frame attention and reuse its attention map for both appearance feature enhancement and motion information extraction.
- Boundary-Aware Cascade Networks for Temporal Action Segmentation178
A new boundary-aware cascade network is presented by introducing a new cascading paradigm, called Stage Cascade, and a general and principled smoothing operation is designed, termed as local barrier pooling, to aggregate local predictions by leveraging semantic boundary information.
- MultiSports: A Multi-Person Video Dataset of Spatio-Temporally Localized Sports Actions157
This paper aims to present a new multi-person dataset of spatio-temporal localized sports actions, coined as MultiSports, which it hopes can serve as a standard benchmark for spatio/temporal action detection in the future.
- Actions as Moving Points138
This paper presents a conceptually simple, computationally efficient, and more precise action tubelet detection framework, termed as MovingCenter Detector (MOC-detector), by treating an action instance as a trajectory of moving points.
- 136
- Fast and Memory-Efficient Network Towards Efficient Image Super-Resolution134
A novel sequential attention branch is proposed, where every pixel is assigned an important factor according to local and global contexts, to enhance high-frequency details in a fast and memory-efficient network.
- Depth-aware salient object detection using anisotropic center-surround difference133
The results compared with several state-of-the-art 2D and depth-aware methods show that the proposed saliency-based object segmentation method has the most satisfactory overall performance.
- 130
- The Ninth Visual Object Tracking VOT2021 Challenge Results127
The Visual Object Tracking challenge VOT2021 is the ninth annual tracker benchmarking activity organized by the VOT initiative; results of 71 trackers are presented; many are state-of-the-art trackers published at major computer vision conferences or in journals in recent years.
- Self Supervision to Distillation for Long-Tailed Visual Recognition127
It is shown that soft label can serve as a powerful solution to incorporate label correlation into a multi-stage training scheme for long-tailed recognition, as well as a new distillation label generation module guided by self-supervision.
- LIP: Local Importance-Based Pooling118
A unified framework over the common downsampling layers of convolutional neural networks from a view of local aggregation based on importance is presented, and a simple, general, and effective pooling operation based on local importance modeling, termed as Local Importance-based Pooling (LIP).
- 111
- AIM 2020 Challenge on Efficient Super-Resolution: Methods and Results101
The AIM 2020 challenge on efficient single image super-resolution was to super-resolve an input image with a magnification factor x4 based on a set of prior examples of low and corresponding high resolution images with focus on the proposed solutions and results.
- Multi-modal deep feature learning for RGB-D object detection98
By introducing the modality-correlated feature representation, the proposed multi-modal RGB-D object detection approach is substantially superior to the state-of-the-art competitors.
- NTIRE 2022 Challenge on Efficient Super-Resolution: Methods and Results94
The NTIRE 2022 challenge was to super-resolve an input image with a magnification factor of ×4 based on pairs of low and corresponding high resolution images and the aim was to design a network for single image super-resolution that achieved improvement of efficiency measured according to several metrics.
- MGSampler: An Explainable Sampling Strategy for Video Action Recognition94
This paper presents a simple, sparse, and explainable frame sampler, termed as Motion-Guided Sampler (MGSampler), which yields a new principled and holistic sampling scheme, that could be incorporated into any existing video architecture.
- Context-Aware RCNN: A Baseline for Action Detection in Videos87
This work empirically finds the recognition accuracy is highly correlated with the bounding box size of an actor, and thus higher resolution of actors contributes to better performance, and develops a surpringly effective baseline (Context-Aware RCNN) that achieves new state-of-the-art results on two challenging action detection benchmarks of AVA and JHMDB.
- Target Adaptive Context Aggregation for Video Scene Graph Generation86
This paper designs an efficient method for frame-level VidSGG, termed as Target Adaptive Context Aggregation Network (TRACE), with a focus on capturing spatio-temporal context information for relation recognition.
- 81
- 73
- 69
- Anchor-based Plain Net for Mobile Image Super-Resolution63
The authors' model can outperform 8-bit quantized FSRCNN by nearly 2dB in terms of PSNR, while satisfying realistic needs at the same time, and adopt quantization-aware training strategy to further boost the performance.
- LinK: Linear Kernel for LiDAR-based 3D Perception62
This work proposes a new method, called LinK, to achieve a wider-range perception receptive field in a convolution-like manner with two core designs, to replace the static kernel matrix with a linear kernel generator, which adaptively provides weights only for non-empty voxels.
- 61
- Fully convolutional online tracking60
This work presents the first fully convolutional online tracking framework (FCOT), with a focus on enabling online learning for both classification and regression branches, and introduces an online regression model generator (RMG) based on the carefully designed anchor-free box regression branch.
- Automatic Wheat Leaf Rust Detection and Grading Diagnosis via Embedded Image Processing System60
A wheat leaf diseases detection system based on embedded image recognition technology that can be used as agricultural robot to inspect in the field, realizing the intelligentization in detection, identification, diagnosis and classification of crop disease.
- Translate-to-Recognize Networks for RGB-D Scene Recognition58
A unified framework to integrate the tasks of cross-modal translation and modality-specific recognition, termed as Translate-to-Recognize Network TRecgNet, which achieves superior performance to the existing state-of-the-art methods, especially for recognition solely based on a single modality.
- 53
- Video Visual Relation Detection via Multi-modal Feature Fusion51
This paper proposes a novel video visual relation detection method with multi-model feature fusion that outperforms other methods by a large margin and earned the first place inVisual relation detection task of Video Relation Understanding Challenge (VRU), ACMMM 2019.
- 50
- CoMo: Learning Continuous Latent Motion from Internet Videos for Scalable Robot Learning48
The proposed CoMo aims to learn more precise continuous latent motion from internet-scale videos by employing an early temporal difference mechanism to increase the shortcut learning difficulty and explicitly enhance motion cues, and proposes a temporal contrastive learning (Tcl) scheme.
- 48
- STMixer: A One-Stage Sparse Action Detector45
A query-based adaptive feature sampling module is presented, which endows the proposed one-stage sparse action detector with the flexibility of mining a set of discriminative features from the entire spatiotemporal domain and a dual-branch feature mixing module, which allows the STMixer to dynamically attend to and mix video features along the spatial and the temporal dimension respectively for better feature decoding.
- Dual DETRs for Multi-Label Temporal Action Detection39
This work proposes a new Dual-level query-based TAD framework, namely DualDETR, to detect actions from both instance-level and boundary-level, and introduces a two-branch decoding structure that builds distinctive decoding processes for different lev-els, facilitating explicit capture of temporal cues and se-mantics at each level.
- Efficient Video Action Detection with Token Dropout and Context Refinement37
This work proposes an end-to-end framework for efficient video action detection (EVAD) based on vanilla ViTs that reduces the overall GFLOPs by 43% and improves real-time inference speed by 40% with no performance degradation.
- EventBuilder36
The proposed EventBuilder system that enables people to automatically generate a summary for a given event in real-time by visualizing different social media such as Wikipedia and Flickr and efficiently summarizes knowledge structures based on the metadata of photos/videos and Wikipedia articles.
- 35
- MTNet: Learning Modality-aware Representation with Transformer for RGBT Tracking34
This paper proposes a modality-aware tracker based on transformer, termed MTNet, which achieves satisfactory results compared with the state-of-the-art competitors on three RGBT benchmarks while reaching real-time speed.
- 34
- Exploring overall opinions for document level sentiment classification with structural SVM32
This work proposes a novel method for document sentiment classification which adequately explores the effect of overall opinion sentences and uses a structural SVM to encode the overall opinion Sentence encoding for this task.
- AnomalyR1: A GRPO-based End-to-end MLLM for Industrial Anomaly Detection31
This study is the first to deliver an end-to-end VLM-based IAD solution that demonstrates the transformative potential of ROAM-enhanced GRPO, positioning the framework as a forward-looking cornerstone for next-generation intelligent anomaly detection systems in industrial applications with limited defective data.
- New Algorithms for Barrier Coverage with Mobile Sensors30
An O(n2.5 log n) time algorithm for moving n sensors to the perimeter of the given circle such that the new positions of sensors form a regular n-gon and the maximum of the distances travelled by mobile sensors is minimized.
- Transferring Foundation Models for Generalizable Robotic Manipulation29
A novel paradigm that effectively leverages language-reasoning segmentation mask generated by internet-scale foundation models, to condition robot manipulation tasks, and develops a two-stream 2D policy model based on imitation learning, which processes raw images and object masks to predict robot actions with a local-global perception manner.
- Cross-Domain Gated Learning for Domain Generalization29
The proposed CDG training strategy can excellently enforce the network to exploit the intrinsic features of objects from the multi-domain data, and achieve a new state-of-the-art domain generalization performance on these benchmarks.
- 29
- Crowd Counting via Multi-layer Regression29
By using multi-layer regression, MRNet achieves significant improvement in crowd counting accuracy, and outperforms the state-of-the-art methods.
- 29
- 28
- Spatiotemporal Predictive Pre-training for Robotic Motor Control27
A simple yet effective visual pre-training framework termed as STP for robotic motor control is presented, which jointly performs spatial and temporal prediction with different masking ratios and a dual-decoder design by utilizing large-scale egocentric video data.
- 27
- 27
- Tra-MoE: Learning Trajectory Prediction Model from Multiple Domains for Adaptive Policy Conditioning26
This paper proposes a sparsely-gated MoE (Top-1 gating strategy) architecture for trajectory model, coined as Tra-MoE, and introduces an adaptive policy conditioning technique by learning 2D mask representations for predicted trajectories, which is explicitly aligned with image observations to guide action prediction more flexibly.
- 26
- Image retargeting using multi-map constrained region warping26
This paper proposes a novel image retargeting approach based on region warping, which emphasizes the image parts with important content while reducing the visual distortion over the whole image.
- Lightweight Super-Resolution Head for Human Pose Estimation24
This work proposes the SR head, which predicts heatmaps with a spatial resolution higher than the input feature maps by super-resolution, to effectively reduce the quantization error and the dependence on further post-processing in heatmap prediction as a Super-Resolution (SR) task.
- Temporal Perceiver: A General Architecture for Arbitrary Boundary Detection24
A general architecture with Transformer is presented, offering a unified solution to the detection of arbitrary generic boundaries, ranging from shot-level, event- level, to scene-level GBDs, and results show that the class-agnostic Perceiver achieves comparable detection accuracy but better generalization ability compared to dataset-specific counterparts.
- NTIRE 2025 Challenge on Image Super-Resolution ($\times 4$): Methods and Results23
The NTIRE 2025 image super-resolution ($\times 4$) challenge, one of the associated competitions of the 10th NTIRE Workshop at CVPR 2025, aims to recover high-resolution images from low-resolution counterparts generated through bicubic downsampling with a $\times 4$ scaling factor.
- HyPSAM: Hybrid Prompt-Driven Segment Anything Model for RGB-Thermal Salient Object Detection23
A novel hybrid prompt-driven segment anything model (HyPSAM), which leverages the zero-shot generalization capabilities of the segment anything model (SAM) for RGB-T SOD, and a plug-and-play refinement network (P2RNet) which serves as a general optimization strategy to guide SAM in refining saliency maps by using hybrid prompts.
- Interactive RGB-D Image Segmentation Using Hierarchical Graph Cut and Geodesic Distance23
This paper utilizes Euclidean distance on RGB space and geodesic distance on 3D space to measure how likely a pixel belongs to foreground or background in color and depth respectively, and integrates the color cue and depth cue into a unified Graph Cut framework to obtain the optimal segmentation result.
- The development and evaluation of Chinese digital science and technology museum23
The feedbacks indicate that CDSTM has positive impacts on mitigating traditional science and technology museum challenges, and has the potential to promote science education and science literacy in public as well.
- SportsHHI: A Dataset for Human-Human Interaction Detection in Sports Videos22
A new video visual relation detection task: video human-human interaction detection, and a dataset named SportsHHI is proposed, which contains 34 high-level interaction classes from basketball and volleyball sports and is aimed at stimulating research on human interaction understanding in videos.
- Joint Modeling of Feature, Correspondence, and a Compressed Memory for Video Object Segmentation22
A unified VOS framework, coined as Joint-Former, for joint modeling the three elements of feature, correspondence, and a compressed memory, which achieves a new state-of-art performance on DAVIS 2017 val/test-dev and YouTube-VOS 2018/2019 val benchmarks, out-performing existing works by a large margin.
- Optimal Point Movement for Covering Circular Regions21
A by-product of the techniques is an algorithm for dynamically maintaining the maximum matching of a circular convex bipartite graph that can handle each vertex insertion or deletion on the graph in O(log2n) time.
- Topic Modeling in Semantic Space with Keywords19
This paper proposes a novel method called TDCS (Topic Distilling with Compressive Sensing) for explicit and accurate modeling the topic implied by several keywords, which is efficient for subsequent text mining tasks.
- 19
- Cross-Modal Pyramid Translation for RGB-D Scene Recognition18
This paper presents a cross-modal pyramid translation strategy to perform multi-scale image generation with a carefully designed layer-wise perceptual supervision, and devise a feature selection module to adaptively enhance the discriminative information during the translation procedure.
- Dynamically Visual Disambiguation of Keyword-based Image Search18
An adaptive multi-model framework that resolves polysemy by visual disambiguation by adapting to the dynamic changes in the search results and employing the proposed saliency-guided deep multi-instance learning network to remove outliers and learn classification models for visual disAmbiguation.
- 18
- Fast approximate geodesic paths on triangle mesh17
A new algorithm to compute a geodesic path over a triangle mesh is presented, based on Novotni’s propagating wavefront method, which can handle both convex and non-convex surfaces or even with boundaries.
- Earth Science Digital Museum (ESDM): Toward a new paradigm for museums17
Practical demonstrations show that ESDM is suitable for formal and informal Earth Science education, including classroom education, online education and life-long learning.
- Human Object Interaction Detection via Multi-level Conditioned Network16
A novel multi-level conditioned network that fuses extra spatial-semantic knowledge with visual features to enhance the reasoning capablity of CNN is proposed and is superior to the state-of-the-arts.
- 16
- 16
- 15
- Asymmetric Masked Distillation for Pre-Training Small Foundation Models15
A new asymmetric masked distillation framework for pre-training relatively small models with au-toencoding, where the teacher model is enabled to see more context information with a lower masking ratio, while the student model is still equipped with a high masking ratio.
- Dual Graph Networks for Pose Estimation in Crowded Scenes15
It is argued that learning structural information based on human joints alone is not enough to address human body variations and could be prone to overfitting, and a pose refinement network is proposed, coined as dual graph network (DGN), to jointly learn its structural information of body joints and limbs by incorporating the cooperative constraints between two branches.
- 15
- How important is location information in saliency detection of natural images15
The experimental results show that location information plays a positive role inSaliency detection, and the proposed method can outperform most state-of-the-art saliency detection methods and handle natural images with different object positions and multiple salient objects.
- 15
- 15
- 14
- Efficient Test-Time Prompt Tuning for Vision-Language Models13
The results consistently show that Self-TPT not only significantly reduces inference costs but also achieves state-of-the-art performance, effectively balancing the efficiency-efficacy trade-off.
- RAVSS: Robust Audio-Visual Speech Separation in Multi-Speaker Scenarios with Missing Visual Cues13
This study proposes a simultaneous multi-speaker separation framework that can facilitate the concurrent separation of multiple speakers within a singular process, and introduces speaker-wise interactions to establish distinctions and correlations among speakers.
- Joint Learning for Relationship and Interaction Analysis in Video with Multimodal Feature Fusion13
This paper proposes a solution to the DVU task which applies joint learning of interaction and relationship prediction and multimodal feature fusion to handle the deep video understanding task.
- Visual Relation of Interest Detection13
The results show that IPNet achieves the best performance compared with the baselines on visual relation detection, scene graph generation and image captioning.
- 13
- Stereo GrabCut: Interactive and Consistent Object Extraction for Stereo Images13
This paper generates consistent segmentation by putting the correspondence relationship in a graph cut framework and leverage depth information, obtained by stereo matching, to give a pre-estimation of foreground and background.
- A bag-of-objects retrieval model for web image search13
This paper proposes a novel bag-of-objects retrieval model for image search reranking of object queries that can significantly outperform the existing reranking methods on object queries.
- 13
- 13
- Open-Vocabulary Spatio-Temporal Action Detection12
This paper builds two benchmarks based on the existing STAD datasets and proposes a simple but effective method based on pretrained video-language models (VLM) that achieves a promising performance on novel classes.
- 12
- Stereo GrabCut: Object Extraction for Stereo Images12
This paper generates consistent segmentation by putting the correspondence relationship in a graph cut framework and leverage depth information, obtained by stereo matching, to give a pre-estimation of foreground and background.
- Jointly modeling association and motion cues for robust infrared UAV tracking11
This work proposes a novel tracker for UAV object tracking, named MAMC, that employs a data augmentation strategy to enhance the training dataset and leverages a motion estimation algorithm with window jitter compensation to address the tracking instability due to background clutter and occlusion.
- 11
- 11
- User-aware topic modeling of online reviews11
A novel model called User Item Sentiment Topic (UIST) is proposed which incorporates users and items for topic modeling and produces topic–word, user–topic, and item–topic distributions simultaneously.
- A Tai Chi Training System Based on Fast Skeleton Matching Algorithm11
A Tai Chi training system based on Microsoft's Kinect, which automatically evaluates a user's performance and provides real-time feedback for the user to refine his current posture is introduced.
- AutoLUT: LUT-Based Image Super-Resolution with Automatic Sampling and Adaptive Residual Learning10
Automatic Sampling (AutoSample) is introduced, a flexible LUT sampling approach where sampling weights are automatically learned during training to adapt to pixel variations and expand the receptive field without added inference cost, enabling detailed information flow and improving the network’s ability to reconstruct fine details.
- Steganographer Detection via Multi-Scale Embedding Probability Estimation10
A novel content-adaptive steganographer detection method via embedding probability estimation that is superior to the state-of-the-art methods in both spatial and frequency domains is proposed.
- Learning frequency and memory-aware prompts for multi-modal object tracking9
Learning Frequency and Memory-Aware Prompts is presented, a dual-adapter framework that injects lightweight prompts into a frozen RGB tracker and preserves the efficiency of prompt learning while strengthening cross-modal interaction and temporal coherence.
- CAGNet: a context-aware graph neural network for detecting social relationships in videos9
A context-aware graph neural network (CAGNet) is proposed for VSRGG, which effectively generates social relationship graphs through message passing, capturing the context of the video.
- 9
- 9
- 8
- Efficient Single-image Super-resolution Using Dual path Connections with Multiple scale Learning8
By introducing dual path connections inspired by Dual path Networks into EMSRDPN, it uses residual connections and dense connections in an integrated way in most network layers and achieves better performance and comparable or even better parameter and inference efficiency over state-of-the-art methods.
- Multimodal Analysis for Deep Video Understanding with Video Language Transformer8
To answer the queries of DVUC, a joint learning framework is used to simultaneously predict multiple tasks with visual, text, audio and pose features and uses video language transformer which learns cross-modal information for matching videos with text choices.
- 8
- 8
- CycleACR: Cycle Modeling of Actor-Context Relations for Video Action Detection7
This paper proposes to select actor-related scene context, rather than directly laveraging raw video scenario, to improve relation modeling, and develops a Cycle Actor-Context Relation network (CycleACR), which introduces the core design of A2C-R for more effective relation modeling.
- MaxSR: Image Super-Resolution Using Improved MaxViT7
A single image super-resolution model based on recent hybrid vision transformer of MaxViT, named as MaxSR is presented, which instantiate proposed model for classical single imagesuper-resolution (MaxSR) and lightweight single imageSuperresolution (maxSR-light) and Experiments show that the MaxSR and MaxSR- light establish new state-of-the-art performance efficiently.
- 7
- ADNet: An Asymmetric Dual-Stream Network for RGB-T Salient Object Detection7
This paper proposes a novel asymmetric dual-stream network (ADNet) that leverages an asymmetric backbone to extract four stages of RGB features and four stages of thermal features and introduces the Channel-Spatial Interaction module.
- APP-Net: Auxiliary-point-based Push and Pull Operations for Efficient Point Cloud Classification7
This work proposes a new local aggregator of linear complexity for point cloud classification, coined as APP, and introduces an auxiliary container as an anchor to exchange features between the source point and the aggregating center.
- 7
- Hierarchical Visual Relationship Detection7
A novel VRD task named hierarchical visual relationship detection (HVRD), which encourages predictions with abstract yet compatible relationship triplets when the confidence level of the specific image content is relatively low and can handle the inevitable ambiguity of groundtruth annotation in VRD.
- Video Summary Quality Evaluation Based on 4C Assessment and User Interaction7
This chapter proposes a novel full-reference evaluation framework to assess the quality of the video summary according to various user requirements and shows a good performance in automatic video summary evaluation.
- 7
- 6
- 6
- DPL: Decoupled Prompt Learning for Vision-Language Models6
This paper theoretically investigates the collaborative process between prompts and instances by reformulating the original self-attention into four separate sub-processes, and introduces language-conditioned textual prompting based on decoupled attention to naturally preserve the generalization of text input.
- 6
- How Important is Location in Saliency Detection6
Experimental results show that location basedSaliency detection can obtain much better performance than random selection, even better than most state-of-the-art saliency detection methods.
- 6
- 6
- Semiconducting bilinear deep learning for incomplete image recognition6
Owing to exploiting the embedding information according to the reliable features rather than any completion of missing features, the proposed semiconducting bilinear deep belief networks has demonstrated outstanding recognition ability on two standard datasets and one constructed dataset, comparing with both incomplete image recognition techniques and existing deep learning models.
- 6
- RGB-D Video Object Segmentation via Enhanced Multi-store Feature Memory5
A novel RGB-D VOS method via multi-store feature memory for robust segmentation via multi-store feature memory is proposed, which adaptively combines features from both modalities and effectively utilizes the Segmentation Anything Model to refine the segmentation mask.
- Spatial-Temporal Human-Object Interaction Detection5
The experimental results demonstrate that the proposed novel method for ST-HOID outperforms the baselines generated by the state-of-the-art methods of image human-object interaction detection, video visual relation detection and video human- object interaction recognition.
- Hyperspectral image fusion based on L-NMF5
According to the experiments on the simulated and real data, the proposed fusion method obtains balanced fused results, and it is more suitable for the HSI fusion task.
- 5
- 5
- 5
- 5
- 5
- 5
- 4
- Deep Video Understanding with Video-Language Model4
A solution called Dual Branches Video Modeling (DBVM) is proposed, which combines knowledge graph (KG) and VLMs, leveraging their strengths and addressing shortcomings, in multimodal question answering for long-form videos.
- Heterogeneous Learning for Scene Graph Generation4
A plug-and-play Heterogeneous Learning Branch is proposed, which enhances the independent representation capability of relation features in SGG methods and significantly improves the performance of all these methods in the common evaluation criteria for SGG task.
- Memory Recursive Network for Single Image Super-Resolution4
The proposed MRNet achieves state-of-the-art SISR performance while using much fewer parameters and introduces a ShuffleConv module to fuse the features from different recursive stages, which is much more effective than using plain convolutional combinations.
- Object Trajectory Proposal via Hierarchical Volume Grouping4
This paper proposes a novel object trajectory proposal method using hierarchical volume grouping, which represents a given video with hierarchical volumes by mapping hierarchical regions with optical flow and ranks the object candidates using a multi-modal fusion scoring mechanism.
- The impact of “Cloud Computing and Big Data” on the future trends of seismic exploration4
The methodology and basic features of “Cloud Computing and Big Data” are introduced, the relationship between cloud computing, big data and Internet of Things, Internet + as well as other key technologies are analyzed and the influence of ‘cloud computing and big data’ on the evolution of seismic technology is discussed.
- Say Cheese: Personal Photography Layout Recommendation Using 3D Aesthetics Estimation4
This paper focuses on how to aid people to master daily life photography using a computational layout recommendation method using a 3D layout aesthetic estimation model to rank the proposed photos.
- Parallel surface reconstruction on GPU4
A novel GPU-based parallel surface reconstruction algorithm that can run more than 10 times faster compared to the CPU-based implementations of this algorithm.
- 4
- Appraisal expression extraction based on parse tree structure4
A phrase tree structure to present the appraisal expression pattern is proposed, as well as a method based on approximate convolution tree kernels to calculate the similarity between these structures, and a similarity-based pattern matching method to extract the appraisal expressions.
- On Volume Distribution Features Based 3D Model Retrieval4
The method first finds three principal directions for a model by employing the principal component analysis method, and rotates the model to align it in a reference frame, and three sets of planes are used to slice the model along to the directions respectively.
- Towards Practical Real-Time Low-Latency Music Source Separation3
This paper introduces a lightweight real-time low-latency model called Real-Time Single-Path TFC-TDF UNET (RT-STT), which is based on the Dual-Path TFC-TDF UNET (DTTNet), and demonstrates the superiority of single-path modeling over dual-path modeling in real-time models.
- 3
- 3
- Hybrid Improvements in Multimodal Analysis for Deep Video Understanding3
This paper improved the joint learning method which was previously proposed in many aspects, including few shot learning, optical flow feature, entity recognition, and video description matching, and verified the effectiveness of these measures through experiments.
- Instance of Interest Detection3
A Cross-influential Network for IOID is proposed, which integrates both visual saliency and semantic context for image semantic description and outperforms the state-of-the-art baselines on this dataset.
- 3
- Adaptive saliency cuts3
This paper proposes a novel saliency cuts method named adaptiveSaliency cuts, which takes advantage of all the input cues in a unified framework and adjusts its components adaptively, and is the first adaptive saliency cut method for different input cues.
- Object Proposal via Depth Connectivity Constrained Grouping3
This paper proposes a novel object proposal method on RGB-D images with the constraint of depth connectivity, which can improve the key techniques in grouping based object proposal effectively, including segment generation, hypothesis expansion and candidate ranking.
- Smart Thumbnail: Automatic Image Cropping by Mining Canonical Query Objects3
The experimental results demonstrate that this query-dependent thumbnailing approach can outperform down-sampling and saliency-based methods on both object localization accuracy and general thumbnail quality.
- Image annotation by modeling Supporting Region Graph3
A light-weight classification model, Approximated Supporting Region Graph (ASRG), in order to handle more relevant regions efficiently, with which a large number of supporting regions are selected and their features are utilized to represent the contextual information in the training and prediction for each image region.
- Feature Selection for Neural Network-based Chinese Text Categorization3
The superiority of the proposed feature selection method for NN based Chinese text categorization is proved after comparing the categorization performance of different feature selection methods and different feature dims in the experiments.
- Text Information Retrieval Based on Latent Semantic Analysis3
The ideas, characters and implementation of text information retrieval based on Latent semantic analysis are introduced.
- 2
- GLAD: Generative Language-Assisted Visual Tracking for Low-Semantic Templates2
A pioneering Generative Language-AssisteD tracking model, GLAD, which utilizes diffusion models for the generative multi-modal fusion of text description and template image to bolster compatibility between language and image and enhance template image semantic information is introduced.
- VMonarch: Efficient Video Diffusion Transformers with Structured Attention2
This work proposes VMonarch, a novel attention mechanism for Video DiTs that enables efficient computation over the dynamic sparse patterns with structured Monarch matrices, and introduces a recomputation strategy to mitigate artifacts arising from instabilities during alternating minimization of Monarch matrices.
- RASA: Replace Anyone, Say Anything -- A Training-Free Framework for Audio-Driven and Universal Portrait Video Editing2
This paper introduces a training-free universal portrait video editing framework that provides a versatile and adaptable editing strategy that supports portrait appearance editing conditioned on the changed first reference frame, as well as lip editing conditioned on varied speech, or a combination of both.
- Text-Guided Nonverbal Enhancement Based on Modality-Invariant and -Specific Representations for Video Speaking Style Recognition2
A text-guided nonverbal enhancement method, TNvE, which is composed of two core modules: a text-guided nonverbal representation selection module employs cross-modal attention based on modality-invariant representations, and a modality-invariant and -specific representation decoupling module incorporates modality-specific representations and decouples them from modality-invariant representations, enabling a more comprehensive understanding of multimodal data.
- 2
- RGB-D Tracking via Hierarchical Modality Aggregation and Distribution Network2
A novel network, denoted as HMAD (Hierarchical Modality Aggregation and Distribution), which leverages the distinct feature representation strengths of RGB and depth modalities, giving prominence to a hierarchical approach for feature distribution and fusion, thereby enhancing the robustness of RGB-D tracking.
- 2
- 2
- Belief Map Enhancement Network for Accurate Human Pose Estimation2
The Belief Map Enhancement Network (En-hanceNet) is proposed for more accurate human pose estimation by enlarging the belief maps by using the efficient sub-pixel operations, which not only increases the belief map resolution but also corrects some wrong predictions at the same time.
- 2
- 2
- 2
- A Saliency-Guided Method for Automatic Photo Refocusing2
This paper presents a simple but effective method to perform image refocusing automatically and rapidly, leveraging current saliency detection methods to locate the region of interest for a given photo rapidly.
- Pre-stack Kirchhoff Time Migration on Hadoop and Spark2
Two parallel algorithms of the plre-stack Kirchhoff time migration based on Hadoop and Sark respectively are proposed and the results show that both of implementations are efficient and scalable and the PKTM on Spark exhibits better performance than the one onHadoop.
- Flat3D: Browsing Stereo Images on a Conventional Screen2
Flat3D for animating stereoscopy only through a conventional screen (2D) based on the motion parallax is proposed and an adaptive fixation acquisition approach combining color and depth cues, as well as employing a probability-based view synthesis to generate the view sequences is developed.
- CLSH: Cluster-based Locality-Sensitive Hashing2
A novel cluster-based locality-sensitive hashing approach, which extends the conventional LSH framework and aims at indexing and searching large scale high-dimensional datasets and can cope with large scale feature dataset.
- 2
- Approximation algorithms for cutting a convex polyhedron out of a sphere2
Three approximation solutions to the problem of cutting P out of Q by a sequence of plane cuts by improving upon the previous O(n^3) time O(log^2n)-factor approximation solution.
- Research on Saliency Map Based Flower Image Classification Algorithm2
This work proposed to use the saliency map to assist the feature extraction process, which could improve the robustness of the system and also discuss the method of combining different features using SVM classifier.
- CUDA based Point Cloud Denoising Algorithm2
A Compute Unified Device Architecture (CUDA)-based improved bilateral filtering point cloud denoising algorithm that uses Gaussian-weighted method of calculating normal vector and effectively improves the accuracy of normal vector calculation.
- Unsupervised saliency detection based on 2D Gabor and Curvelets transforms2
A novel saliency map model called Gabor & Curvelets based Saliency Map (GCSMP) relying on 2D Gabor and Curvelet transforms is proposed, which takes advantage of Garbor transforms's spatial localization and curvelet transform's edge and directional information.
- 2
- Content-based Copy Detection Task2
This year, the authors participated in TRECVID 2010 content-based copy detection task and proposed a method using Non-orthogonal Binary Subspace in their accurate detection process.
- 2
- Content-Based Retrieval of Cultural Relic Images in Digital Museum2
A novel relevance feedback method is proposed which combines semantic annotation to visual feature implicitly and the user can unknowingly use professional semantic information to retrieve images.
- News Information Extraction for Web Resource2
This paper introduces an approach of semi-automatically extracting from Web resource and gives the system which extracts useful Olympic news information and experiment results of it.
- Skeleton Extraction Algorithm Using Reeb Graph Based on Facets2
A practical skeleton extraction framework based on facets using Reeb graph was proposed to extract the skeleton which contains enough information of the model, such as gesture and topological features, to be able to be used as descriptor of 3D model retrieval.
- Information Extraction Technology for Web Forums2
A specification f information extraction is defined, and a visualization tool is provided to help users generate rules together with the background engine, which automatically download and extracts semantic information units inside Web forums.
- VL-UniTrack: A Unified Framework with Visual-Language Prompts for UAV-Ground Visual Tracking1
VL-UniTrack is proposed, a fully unified framework enhanced by visual-language prompts that achieves state-of-the-art performance on the latest benchmark and breaks the barrier of feature isolation to facilitate sufficient cross-view interaction.
- SwiTrack: Tri-State Switch for Cross-Modal Object Tracking1
SwiTrack is proposed, a novel state-switching framework that redefines CMOT through the deployment of three specialized streams that incorporate dynamic template reconstruction to iteratively update template features and employ a similarity alignment loss to reinforce feature consistency.
- FSDM: An efficient video super-resolution method based on Frames-Shift Diffusion Model1
The proposed Frames-Shift Diffusion Model optimizes video super-resolution by simplifying the integration of temporal data, thus addressing key challenges in the field.
- 1
- 1
- 1
- Weakly supervised instance segmentation via peak mining and filtering1
The pseudo‐supervision can be used to train an instance segmentation network in a fully supervised manner and outperforms other counterparts by a margin of more than 6 %, 4%, and 3% with the mean average precision (mAP) at IoU threshold.
- IAA-VSR: An iterative alignment algorithm for video super-resolution1
In this paper, a unified Iterative Alignment Algorithm (IAA) is proposed for more accurate frame alignment in VSR, which greatly eases the alignment problem for all kinds of motions.
- Human-Centric Visual Relation Segmentation Using Mask R-CNN and VTransE1
A novel human-centric visual relation segmentation method based on Mask R-CNN model and VTransE model, which outperforms other methods in Person in Context Challenge.
- Personalized Recommendation of Photography Based on Deep Learning1
The main job is to propose an optimized feature representation method in the unlabeled data set, and to train by the deep learning convolutional neural network (CNN), and finally achieve the recommended purpose.
- Don’t Be Confused: Region Mapping Based Visual Place Recognition1
It is found that images of identified places can be effectively matched by remarkable regions like building facades under limited geometry and illumination changes.
- HeterStyle1
A novel video style transfer application, named HeterStyle, which can stylize different regions in the video with adaptive intensities and minimizes a heterogeneous style transfer loss function considering content, style and temporal consistency in a Convolutional Neural Networks based optimization framework.
- Adaptive video object proposals by a context-aware model1
The results compared with several state-of-the-arts show that the proposed adaptive context-aware model for video object proposals has the most satisfactory overall performance for multi-object proposals in videos.
- 1
- Depth Extraction from a Light Field Camera Using Weighted Median Filtering1
A novel method to extract an accurate depth map from several refocused images by uniformly changing the focal length and applying a weighted median filter for refinement, which suppresses depth noise and supplies a well recovery of object boundaries and fine structures.
- 1
- 1
- A Complementary Aggregation Approach for Local Stereo Matching Using Color and Correlation Cues1
A new soft segmentation approach for correlation-based aggregation is proposed, which enables the approach to achieve a better performance in both highly textured areas and depth discontinuities than existed methods.
- Research of fusing image classification into object localization1
The experimental results prove that fusing image classification into object localization can really improve the performance of object localization.
- 1
- Nanjing University in TRECVID 20091
This report describes the details of the system for copy detection task in TRECVID 2009, which submitted 4 runs,2 for video-only queries and 2 for video + audio queries.
- Algorithm of Geology Mesh Surface Reconstruction Based on Constraints1
A geology surface reconstruction algorithm based on geology constraints is introduced to aim at generating linear discontinuous horizon pieces in 3D geology modeling by applying the "marching cubes" algorithm.
- An Efficient Probabilistic Syntactic Analysis Algorithm for Chinese1
An efficient probabilistic syntactic analysis algorithm that recognizes all of the phrases which contain some function word, and analyzes the relationship of the words to form the syntactic tree, shows that it can gain a high precision.
- Semantic Drift Analysis for Web Forums1
The purpose is to analyze the semantic drift phenomenon, and further find the integrated semantic units in Web forum sites with multi-dimension document model.
- Word Sense Disambiguation Algorithm Based on Semantic Information1
This paper puts forward a new multi-strategy method of word sense disambiguation based on HowNet from the viewpoint of semantic information that works with information of part of speech, association instances, semantic relation and so on.
- Design and Implementation of Web-based Full-text Search Engine1
This paper introduces the design and implementation of a Web-based full-text search engine and gives the test result of the system.
- Overview of Information Retrieval Technology for Web1
Several traditional information retrieval models and latent semantic analysis technique are introduced and a brief description of the question-answering system is given, and possible applications of question-ANSWering system in the future are analyzed.
- On Intra-page and Inter-page Semantic Analysis of Web Pages1
This paper explores the intra-page semantic structure of Web pages and extends the processing unit from the whole page to a finer granularity, i.e., semantic information blocks within pages, and proposes corresponding feature selection and quantification methods.
- –
- –
- –
- –
- –
- –
- Reproducibility Companion Paper: Human Object Interaction Detection via Multi-level Conditioned Network–
The file structure of the source code is explained and the details of the experiments settings are published, and a program for component analysis is provided to assist other researchers with experiments on alternative models that are not included in the authors' experiments.
- Tuning ANFIS Using a Simplified Sparrow Search Algorithm–
The results of simulations indicate that, training ANFIS by SSSA reduce the root of mean of squared error in a more effective pattern, and the proposed technique increases the way to solve nonlinear regression problems.
- Human-centric Spatio-Temporal Video Grounding via the Combination of Mutual Matching Network and TubeDETR–
This technical report represents the solution for the Human-centric Spatio-Temporal Video Grounding (HC-STVG) track of the 4th Person in Context workshop and challenge built on the basis of TubeDETR and Mutual Matching Network (MMN).
- Reproducibility Companion Paper–
The file structure of the source code is explained and the details of the ViROI dataset are published, which can be used to retrain the model with custom parameters and list the parameters that can be modified for custom training and inference.
- –
- Reproducibility Companion Paper: Instance of Interest Detection–
The file structure of the source code and the details of the IOID dataset are explained, which can be used to retrain the model with custom parameters and a program for component analysis is provided to help other researchers to do experiments with alternative models that are not included in the authors' experiments.
- Salient object detection via multi-scale neural network–
This paper proposes heterogeneous dilated block to effectively increases the receptive field of the network, while alleviating the gridding effect problem caused by dilated convolution, and replaces the traditional interpolation up-sampling layer with a fully learnable up-Sampling module to solve the blurry artifacts and improve the accuracy.
- Robust and Real-Time Visual Tracking Based on Complementary Learners–
This paper exploits the low dimensional complementary features and an adaptive online detector with the average peak-to-correlation energy to improve tracking accuracy and time efficiency and appropriately integrate several complementary features in the correlation filter based discriminative framework and combine with the global color histogram to further boost the overall performance.
- –
- –
- –
- –
- –
- Combination of region and contour models for interactive image segmentation–
A novel interactive image segmentation method, which makes a combination of the classical graph cut and active contours based models, which can produce high quality segmentations with little user interaction and achieve a surprisingly high efficiency.
- A Dynamic Extension and Data Migration Method Based on PVFS–
Results show that the dynamic data node extension method could improve the performance of PVFS and reduce the probability of hot point effectively and the subsequent data migration algorithm based on PVFS.
- A New Data Replication Scheme for PVFS2–
A new data replication scheme in PVFS2 is presented, where the backup operation is done on the servers, therefore the user experience is not affected while creating copies of files, and the read operation of PVfs2 is optimized.
- Fast Binocular Depth Inference via Bidirectional Motion Based Interpolation–
A novel bidirectional motion-based interpolation framework is proposed, which avoids frame-by-frame matching through making use of the motion estimation and the redundancy between frames and is competent for fast and accurate binocular video depth acquisition.
- –
- Research on Emergency Plan Digitalization Method of Environmental Disaster–
This approach improves the automatic generation of emergency plan and adjusts the plan according to the site condition; it renders emergency plan fine visualization and flexible adaption when disaster happens.
- –
- –
- Fast Calculating Simplification Error of Triangular Mesh Using CUDA–
A highly paralleled algorithm to calculate the simplification error of triangular meshes by using CUDA, which verify the advantages of GPU on high performance calculation compared to CPU.
- –
- A Geometric Spread Approach of 3-D Reconstruction–
This paper introduces a method of 3D surface reconstruction based on the 3D Delaunay triangulation of points, and uses the local area incremental algorithm to build the triangular mesh.
- A Precomputed Slicing Based Volume Rendering Technique–
To improve the rendering performance and solve the major bottoleneck of the mass-data volume rendering system a new slice based hardware acelerated volume rendering technique is given.
- Mesh Similarity Measuring Based on CUDA–
A fast algorithm which could measure similarity between two meshes interactively and develop parallel method to construct uniform grid for fast space indexing of triangles is presented.
- Multi-resolution Ratio Extended Octree Model Based on Mass Seismic Data–
Experimental results show that the model is an efficient organization style of mass seismic data, and it has well application value of the geological modeling.
- Study on Music Genre Classification Method Based on Improved Voting Mechanism–
A method called NNVote based on clip genre distribution feature and a method called RhythmNNVote which combines the high level rhythm feature and low level features are proposed which improves about 10% compared with MaxVote.
- The Application of Colasoft Network Protocol Analyzer to Teaching–
Based on some teaching cases, one kind of homemade software named as Colasoft Network Protocols Analyzer to teaching was introduced in this paper, which mainly involve Network Environment Building and Protocol Packet Analyzer.
- –
- Software Development Using Classification Trees Based on Metric–
This paper discusses data mining and its applications to software engineering data analysis.
- Design and Implementation of Chinese TTS Engine: SmartTalk–
This paper concentrates on the design and implementation of a Chinese Text-to-Speech ( TTS) engine named SmallTalk, which deals mainly with the design of Chinese speech library and Chinese word library.
- Design and Implementation of Word and Speech Libraries in Chinese Text-to-speech Conversion–
This paper describes the design and implementation of Chinese word library and speech library in Chinese text-to-speech conversion.
- –
- Design and Implementation of an Education-oriented Automatic Webpage Classification System–
A framework of education-oriented automatic Webpage classification system, focused on the extraction ofmetadata and selection and quantification of feature, based on decision tree, using C4.5 algorithm is proposed.
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.