Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileWorks243 from public data
- 4,387
- Action recognition with trajectory-pooled deep-convolutional descriptors1,205
This paper presents a new video representation, called trajectory-pooled deep-convolutional descriptor (TDD), which shares the merits of both hand-crafted features and deep-learned features, and achieves superior performance to the state of the art on these datasets.
- A review of convolutional neural networks in computer vision1,170
An elementary understanding of CNN components and their functions, including input layers, convolution layers, pooling layers, activation functions, batch normalization, dropout, fully connected layers, and output layers are presented.
- Scaffold-GS: Structured 3D Gaussians for View-Adaptive Rendering1,011
This work introduces Scaffold-GS, which uses an-chor points to distribute local 3D Gaussians, and predicts their attributes on-the-fly based on viewing direction and distance within the view frustum, and effectively reduces redun-dant Gaussians while delivering high-quality rendering.
- Temporal Action Detection with Structured Segment Networks992
This paper addresses an important and challenging task, namely detecting the temporal intervals of actions in untrimmed videos by presenting a framework called structured segment network (SSN), built on temporal proposals of actions.
- Temporal Segment Networks for Action Recognition in Videos989
The proposed TSN framework, called temporal segment network (TSN), aims to model long-range temporal structure with a new segment-based sampling and aggregation scheme and won the video classification track at the ActivityNet challenge 2016 among 24 teams.
- 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.
- VideoMAE V2: Scaling Video Masked Autoencoders with Dual Masking871
This paper shows that video masked autoencoder (VideoMAE) is a scalable and general self-supervised pre-trainer for building video foundation models, and presents a dual masking strategy for efficient pre-training, which enables the efficientPre-training of billion-level models in video.
- Bag of visual words and fusion methods for action recognition: Comprehensive study and good practice677
A comprehensive study of all steps in BoVW and different fusion methods is provided, and a simple yet effective representation is proposed, called hybrid supervector, by exploring the complementarity of different BoVW frameworks with improved dense trajectories.
- InternVideo: General Video Foundation Models via Generative and Discriminative Learning584
Without bells and whistles, InternVideo achieves state-of-the-art performance on 39 video datasets from extensive tasks including video action recognition/detection, video-language alignment, and open-world video applications.
- TEA: Temporal Excitation and Aggregation for Action Recognition575
This paper proposes a Temporal Excitation and Aggregation block, including a motion excitation module and a multiple temporal aggregation module, specifically designed to capture both short- and long-range temporal evolution, and achieves impressive results at low FLOPs on several action recognition benchmarks.
- InternVid: A Large-scale Video-Text Dataset for Multimodal Understanding and Generation557
A scalable approach to autonomously build a high-quality video-text dataset with large language models (LLM), thereby showcasing its efficacy in learning video-language representation at scale and introducing ViCLIP, a video- Text representation learning model based on ViT-L.
- UntrimmedNets for Weakly Supervised Action Recognition and Detection546
This paper presents a new weakly supervised architecture, called UntrimmedNet, which is able to directly learn action recognition models from untrimmed videos without the requirement of temporal annotations of action instances.
- 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.
- Real-Time Action Recognition with Enhanced Motion Vector CNNs444
This paper accelerates the deep two-stream architecture by replacing optical flow with motion vector which can be obtained directly from compressed videos without extra calculation, and introduces three strategies for this, initialization transfer, supervision transfer and their combination.
- InternVideo2: Scaling Foundation Models for Multimodal Video Understanding438
This work introduces InternVideo2, a new family of video foundation models (ViFM) that achieve the state-of-the-art results in video recognition, video-text tasks, and video-centric dialogue, highlighting its ability to reason and comprehend longer contexts.
- Appearance-and-Relation Networks for Video Classification403
This paper presents a new architecture, termed as Appearance-and-Relation Network (ARTNet), to learn video representation in an end-to-end manner, constructed by stacking multiple generic building blocks, called SMART, whose goal is to simultaneously model appearance and relation from RGB input in a separate and explicit manner.
- PyMAF: 3D Human Pose and Shape Regression with Pyramidal Mesh Alignment Feedback Loop402
A Pyramidal Mesh Alignment Feedback (PyMAF) loop to leverage a feature pyramid and rectify the predicted parameters explicitly based on the mesh-image alignment status in the authors' deep regressor is proposed.
- TAM: Temporal Adaptive Module for Video Recognition388
A new temporal adaptive module (TAM) to generate video-specific temporal kernels based on its own feature map that outperforms other temporal modeling methods consistently, and achieves the state-of-the-art performance under the similar complexity.
- 340
- 311
- 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.
- Unmasked Teacher: Towards Training-Efficient Video Foundation Models288
This paper proposes a training-efficient method for temporal-sensitive VFMs that integrates the benefits of existing methods, and masks out most of the low-semantics video tokens, but selectively align the unmasked tokens with IFM, which serves as the UnMasked Teacher (UMT).
- 278
- TEINet: Towards an Efficient Architecture for Video Recognition275
The proposed TEINet can achieve a good recognition accuracy on these datasets but still preserve a high efficiency, and is able to capture temporal structure flexibly and effectively, but also efficient for model inference.
- 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.
- SparseBEV: High-Performance Sparse 3D Object Detection from Multi-Camera Videos250
SparseBEV is proposed, a fully sparse 3D object detector that outperforms the dense counterparts and contains three key designs, which are scale-adaptive self attention to aggregate features with adaptive receptive field in BEV space, adaptive spatio-temporal sampling to generate sampling locations under the guidance of queries, and adaptive mixing to decode the sampled features with dynamic weights from the queries.
- 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.
- MeMOTR: Long-Term Memory-Augmented Transformer for Multi-Object Tracking207
This paper proposes MeMOTR, a long-term memory-augmented Transformer for multi-object tracking that is able to make the same object's track embedding more stable and distinguishable by leveraging longterm memory injection with a customized memory-attention layer.
- 205
- UniFormerV2: Spatiotemporal Learning by Arming Image ViTs with Video UniFormer204
This work proposes a generic paradigm to build a powerful family of video networks, by arming the pretrained ViTs with efficient UniFormer designs, which contains brand-new local and global relation aggregators, which allow for preferable accuracy-computation balance by seamlessly integrating advantages from both ViTs and UniFormer.
- Recovering 3D Human Mesh From Monocular Images: A Survey190
This is the first survey that focuses on the task of monocular 3D human mesh recovery and starts with the introduction of body models and then elaborate recovery frameworks and training objectives by providing in-depth analyses of their strengths and weaknesses.
- 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.
- Video Mamba Suite: State Space Model as a Versatile Alternative for Video Understanding185
The Video Mamba Suite is presented, a suite of Mamba-based model families together with a unified evaluation protocol covering four roles in video understanding: temporal modeling, cross-modal interaction, temporal adapters, and spatiotemporal modeling, and the Decomposed Bidirectional Mamba (DBM) block is proposed.
- 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.
- 180
- 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.
- 178
- Knowledge Guided Disambiguation for Large-Scale Scene Classification With Multi-Resolution CNNs159
A multi-resolution CNN architecture that captures visual content and structure at multiple levels is proposed and two knowledge guided disambiguation techniques to deal with the problem of label ambiguity are designed.
- 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.
- AdaMixer: A Fast-Converging Query-Based Object Detector151
This work proposes a fast-converging query-based detector, named AdaMixer, by improving the adaptability ofquery-based decoding processes in two aspects, and sheds light on a simple, accurate, and fast converging architecture for query- based object detectors.
- SketchyCOCO: Image Generation From Freehand Scene Sketches148
This work introduces the first method for automatic image generation from scene-level freehand sketches that allows for controllable image generation by specifying the synthesis goal via free hand sketches and builds a large-scale composite dataset called SketchyCOCO to support and evaluate the solution.
- StNet: Local and Global Spatial-Temporal Modeling for Action Recognition146
A novel spatial temporal network (StNet) architecture for both local and global spatial-temporal modeling in videos that outperforms several state-of-the-art approaches in action recognition and can strike a satisfying trade-off between recognition accuracy and model complexity is explored.
- A Comparative Study of Encoding, Pooling and Normalization Methods for Action Recognition145
The results show the new encoding methods can significantly improve the recognition accuracy compared with classical VQ and among them, Fisher kernel encoding and sparse encoding have the best performance.
- Thin-Slicing Network: A Deep Structured Model for Pose Estimation in Videos141
The proposed end-to-end architecture is evaluated on two widely used benchmarks for video-based pose estimation (Penn Action and JHMDB datasets) and outperforms several state-of-the-art methods.
- 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.
- Video Action Detection with Relational Dynamic-Poselets138
A relational model for action detection, which first decomposes human action into temporal “key poses” and then further into spatial “action parts”, which not only allows to localize the action in a video stream, but also enables a detailed pose estimation of an actor.
- VideoLLM: Modeling Video Sequence with Large Language Models134
This work proposes a novel framework called VideoLLM that leverages the sequence reasoning capabilities of pre-trained LLMs from natural language processing (NLP) for video sequence understanding and demonstrates that the understanding and Reasoning capabilities of LLMs can be effectively transferred to video understanding tasks.
- Task-specific Inconsistency Alignment for Domain Adaptive Object Detection131
This paper proposes Task-specific Inconsistency Alignment (TIA), a new alignment mechanism in separate task spaces, improving the performance of the detector on both subtasks, and adds a set of auxiliary predictors for both classification and localization branches, and exploits their behavioral inconsistencies as finer-grained domain-specific measures.
- EgoExoLearn: A Dataset for Bridging Asynchronous Ego- and Exo-centric View of Procedural Activities in Real World130
EgoExoLearn is introduced, a large-scale dataset that emulates the human demonstration following process, in which individuals record egocentric videos as they execute tasks guided by exocentric-view demonstration videos, to serve as an important resource for bridging the actions across views.
- 130
- MoFAP: A Multi-level Representation for Action Recognition128
This paper proposes a multi-level video representation by stacking the activations of motion features, atoms, and phrases (MoFAP), and achieves the state-of-the-art performance on four challenging datasets.
- 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.
- 127
- 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).
- 116
- InternGPT: Solving Vision-Centric Tasks by Interacting with ChatGPT Beyond Language113
By incorporating pointing instructions, the proposed iGPT significantly improves the efficiency of communication between users and chatbots, as well as the accuracy of chatbots in vision-centric tasks, especially in complicated visual scenarios where the number of objects is greater than 2.
- 111
- 110
- DDT: Decoupled Diffusion Transformer108
A new decoupled architecture with a dedicated condition encoder for semantic extraction alongside a specialized velocity decoder is proposed, which enhances inference speed by enabling the sharing self-condition between adjacent denoising steps.
- 108
- Latent Hierarchical Model of Temporal Structure for Complex Activity Classification106
A latent hierarchical model (LHM) to describe the decomposition of complex activity into sub-activities in a hierarchical way and achieves the state-of-the-art performance on the Hollywood2 dataset and the Olympic Sports dataset.
- Two-Stream SR-CNNs for Action Recognition in Videos105
A new deep architecture is proposed by incorporating human/object detection results into the framework, called two-stream semantic region based CNNs (SR-CNNs), which not only shares great modeling capacity with the original two-stream CNNs, but also exhibits the ability of leveraging semantic cues for action understanding.
- Actionness Estimation Using Hybrid Fully Convolutional Networks104
A new deep architecture for actionness estimation is presented, called hybrid fully convolutional network (HFCN), which is composed of appearance FCN (A-FCN) and motionFCN (M-FCNs), which leverage the strong capacity of deep models to estimate actionness maps from the perspectives of static appearance and dynamic motion.
- 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.
- DCAN: Improving Temporal Action Detection via Dual Context Aggregation91
The Multi-Path Temporal Context Aggregation (MTCA) is design to achieve smooth context aggregation on boundary level and precise evaluation of boundaries and Coarse-to-fine Matching (CFM) is designed to aggregate context on the proposal level and refine the matching map from coarse to fine.
- Locally Supervised Deep Hybrid Model for Scene Recognition90
This paper proposes a novel locally supervised deep hybrid model (LS-DHM) that effectively enhances and explores the convolutional features for scene recognition, and proposes an efficient Fisher Convolutional vector (FCV) that successfully rescues the orderless mid-level semantic information of scene image.
- 90
- Fully Sparse 3D Occupancy Prediction89
A novel fully sparse occupancy network, termed SparseOcc, which initially reconstructs a sparse 3D representation from camera-only inputs and subsequently predicts semantic/instance occupancy from the 3D sparse representation by sparse queries, and a thoughtful ray-based evaluation metric, namely RayIoU, to solve the inconsistency penalty along the depth axis raised in traditional voxel-level mIoU criteria.
- FineAction: A Fine-Grained Video Dataset for Temporal Action Localization88
This work develops a novel large-scale and fine-grained video dataset, coined as FineAction, which takes distinct characteristics of fine action classes with rich diversity, dense annotations of multiple instances, and co-occurring actions of different classes, which introduces new opportunities and challenges for temporal action localization.
- Dynamic Sampling Networks for Efficient Action Recognition in Videos87
The experimental results demonstrate that DSN is able to greatly improve the inference efficiency by only using less than half of the clips, which can still obtain a slightly better or comparable recognition accuracy to the state-of-the-art approaches.
- 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.
- Memory-and-Anticipation Transformer for Online Action Understanding84
This paper rethink the temporal dependence of event evolution and proposes a novel memory-anticipation-based paradigm to model an entire temporal structure, including the past, present, and future, which significantly outperforms all existing methods.
- PixNerd: Pixel Neural Field Diffusion82
This work proposes to model the patch-wise decoding with neural field and present a single-scale, single-stage, efficient, end-to-end solution, coined as pixel neural field diffusion~(PixelNerd).
- Structured Sparse R-CNN for Direct Scene Graph Generation80
A simple, sparse, and unified framework, termed as Structured Sparse R-CNN, with a set of learnable triplet queries and a structured triplet detector which could be jointly optimized from the training set in an end-to-end manner.
- Dynamic MDETR: A Dynamic Multimodal Transformer Decoder for Visual Grounding75
A new dynamic multimodal transformer decoder that outperforms TransVG (ResNet-50) but only brings marginal extra computational cost relative to TransVG, and achieves competitive trade-offs between computation and accuracy.
- Progressive Visual Prompt Learning with Contrastive Feature Re-formation73
This paper proposes a novel structure called Progressive Visual Prompt (ProVP), and introduces a contrastive feature re-formation technique for visual prompt learning that significantly stabilizes the training process and enhances both the adaptation and generalization capabilities of visual prompt learning in V-L models.
- MGMAE: Motion Guided Masking for Video Masked Autoencoding73
A motion guided masking strategy is introduced that explicitly incorporates motion information to build temporal consistent masking volume and sample a set of temporal consistent cubes from videos to relieve the information leakage issue in time and encourage the MGMAE to learn more useful structure information.
- Object-Scene Convolutional Neural Networks for event recognition in images71
This paper designs a new architecture, called Object-Scene Convolutional Neural Network (OS-CNN), which is decomposed into object net and scene net, which extract useful information for event understanding from the perspective of objects and scene context, respectively, and investigates different network architectures for OS-CNN design.
- Boosting VLAD with Supervised Dictionary Learning and High-Order Statistics69
A high-order VLAD (H-VLAD) for visual recognition is proposed, which leverages two kinds of high- order statistics in the VL AD-like framework, namely diagonal covariance and skewness and allows for efficient computation.
- 67
- Single Image Highlight Removal with a Sparse and Low-Rank Reflection Model64
A sparse and low-rank reflection model for specular highlight detection and removal using a single input image that is competitive with previous methods, especially in some challenging scenarios featuring natural illumination, hue-saturation ambiguity and strong noises.
- 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.
- InternVideo-Ego4D: A Pack of Champion Solutions to Ego4D Challenges62
In these five tasks, the performance of InternVideo-Ego4D comprehensively surpasses the baseline methods and the champions of CVPR2022, demonstrating the powerful representation ability of Intern video as a video foundation model.
- A self-trained semisupervised SVM approach to the remote sensing land cover classification62
A novel semisupervised SVM model that uses self-training approach and the Gustafson-Kessel fuzzy clustering algorithm is proposed for the selection of unlabeled points to reduce the impact of ineffective labels for remote sensing land cover classification.
- OmniCorpus: A Unified Multimodal Corpus of 10 Billion-Level Images Interleaved with Text60
This paper introduces OmniCorpus, a 10 billion-scale image-text interleaved dataset that has 15 times larger scales while maintaining good data quality and is more flexible, easily degradable from an image-text interleaved format to pure text corpus and image-text pairs.
- 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.
- 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.
- Reduced-difference pyramid: a data structure for progressive image transmission53
This paper introduces a reduced-difference pyramid data structure where the number of nodes, corresponding to a set of decorrelated difference values, is exactly equal to thenumber of pixels.
- Transferring Deep Object and Scene Representations for Event Recognition in Still Images52
This paper empirically investigates the correlation of the concepts of object, scene, and event, and proposes an iterative selection method to identify a subset of object and scene classes deemed most relevant for representation transfer, and develops three transfer techniques.
- Sparse Global Matching for Video Frame Interpolation with Large Motion49
A new pipeline for VFI is introduced, which can effectively integrate global-level information to alleviate issues associated with large motion and demonstrates the state-of-the-art performance on these VFI subsets with large motion.
- 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
- PDPP: Projected Diffusion for Procedure Planning in Instructional Videos47
This paper model the whole intermediate action sequence distribution with a diffusion model (PDPP), and thus transform the planning problem to a sampling process from this distribution, and removes the expensive intermediate supervision.
- 46
- 46
- Action and Gesture Temporal Spotting with Super Vector Representation46
This paper proposes an action and gesture spotting system, which is mainly composed of three steps: temporal segmentation, clip classification, and post processing, and uses some post processing techniques to void the detection of false positives.
- LeviTor: 3D Trajectory Oriented Image-to-Video Synthesis45
This work proposes a pioneering method for 3D trajectory control in image-to-video synthesis by abstracting object masks into a few cluster points, accompanied by the depth information and the instance information, which are finally fed into a video diffusion model as the control signal.
- 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.
- 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.
- Task Preference Optimization: Improving Multimodal Large Language Models with Vision Task Alignment35
This work proposes Task Preference Optimization (TPO), a novel method that utilizes differentiable task preferences derived from typical fine-grained visual tasks that significantly enhances the MLLM’s multimodal capabilities and task-specific performance.
- 35
- AlphaBlock: Embodied Finetuning for Vision-Language Reasoning in Robot Manipulation34
A novel framework for learning high-level cognitive capabilities in robot manipulation tasks, such as making a smiley face using building blocks is proposed, and a closed-loop multi-modal embodied planning model that autoregressively generates plans by taking image observations as input is proposed.
- A novel adaptive density-based spatial clustering of application with noise based on bird swarm optimization algorithm34
A novel adaptive density-based spatial clustering of application with noise based on bird swarm optimization algorithm (BSA-DBSCAN) is proposed that can reasonably search the Eps parameter value and can obtain the higher accuracy of clustering.
- Adapting Short-Term Transformers for Action Detection in Untrimmed Videos33
This paper designs effective crosssnippet propagation modules to gradually exchange short-term video information among different snippets from two levels, and introduces a cross-snippet propagation strategy to enable multi-snippet temporal feature interaction inside the backbone.
- OCSampler: Compressing Videos to One Clip with Single-step Sampling33
OCSampler designs a new paradigm of learning instance-specific video condensation policies to select frames only in a single step, and extends the proposed method with a frame number budget, enabling the framework to produce correct predictions in high confidence with as few frames as possible.
- Cross-Architecture Self-supervised Video Representation Learning31
This paper introduces a temporal self-supervised learning module able to predict an Edit distance explicitly between two video sequences in the temporal order, which enables the model to learn a rich temporal representation that compensates strongly to the video-level representation learned by the CACL.
- 30
- 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
- 29
- Progressive Attention on Multi-Level Dense Difference Maps for Generic Event Boundary Detection28
An effective and end-to-end learnable framework (DDM-Net) to tackle the diversity and complicated semantics of event boundaries, and demonstrates the effectiveness of richer motion representation and more sophisticated aggregation, in handling the diversity of GEBD.
- Learning Optical Flow and Scene Flow With Bidirectional Camera-LiDAR Fusion26
This paper proposes a novel end-to-end framework, which consists of 2D and 3D branches with multiple bidirectional fusion connections between them in specific layers, and applies a point-based 3D branch to extract the LiDAR features, as it preserves the geometric structure of point clouds.
- Modeling Fine-Grained Hand-Object Dynamics for Egocentric Video Representation Learning25
This work proposes EgoVideo, a model with a new lightweight motion adapter to capture fine-grained hand-object motion information and introduces HOD, a novel pipeline employing a hand-object detector and a large language model to generate high-quality narrations with detailed descriptions of hand-object dynamics.
- BIVDiff: A Training-Free Framework for General-Purpose Video Synthesis via Bridging Image and Video Diffusion Models25
A training-free general-purpose video synthesis framework, coined as BIVDiff, via bridging specific image diffusion models and general text-to-video foundation diffusion models that enables flexible image model selection for different purposes with strong task generalization and high efficiency.
- 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.
- Better Exploiting OS-CNNs for Better Event Recognition in Images24
This paper addresses the problem of cultural event recognition in still images and focuses on applying deep learning methods on this problem by utilizing the successful architecture of Object-Scene Convolutional Neural Networks (OS-CNNs) to perform event recognition.
- Vinci: A Real-time Smart Assistant Based on Egocentric Vision-language Model for Portable Devices22
Vinci is a vision-language system designed to provide real-time, comprehensive AI assistance on portable devices that integrates an egocentric vision foundation model with a large language model (LLM), enabling advanced functionalities such as scene understanding, temporal grounding, video summarization, and future planning.
- Logit Normalization for Long-Tail Object Detection22
This paper proposes Logit Normalization (LogN), a simple technique to self-calibrate the classification logits of detectors in a similar way to Batch Normalization (BN), which can serve as a strong baseline for long-tail object detection and is expected to inspire future research in this field.
- 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.
- Knowledge Integration Networks for Action Recognition22
This work proposes a two-level knowledge encoding mechanism which contains a Cross Branch Integration module for encoding the auxiliary knowledge into medium-level convolutional features, and an Action Knowledge Graph (AKG) for effectively fusing high-level context information.
- Vinci: A Real-time Embodied Smart Assistant based on Egocentric Vision-Language Model21
Vinci, a real-time embodied smart assistant built upon an egocentric vision-language model, is introduced, a real-time embodied smart assistant built upon an egocentric vision-language model designed for deployment on portable devices such as smartphones and wearable cameras.
- Structured Triplet Learning with POS-Tag Guided Attention for Visual Question Answering20
This paper explores mechanisms of incorporating part-ofspeech tag guided attention, convolutional n-grams, triplet attention interactions between the image, question and candidate answer, and structured learning for triplets based on image-question pairs and achieves the state-of-the-art performance.
- Modeling spatial layout for scene image understanding via a novel multiscale sum-product network20
A new deep architecture MSPN is proposed for image segmentation, which utilizes multiscale unary potentials as the inputs and models the spatial layouts of image content in a hierarchical manner and characterizes scene spatial layouts in a fine-to-coarse manner to enforce the consistency in labeling.
- Hybrid Super Vector with Improved Dense Trajectories for Action Recognition20
This work employs two advanced super vector methods, namely Fisher Vector and soft Vector of Locally Aggregated Descriptors to encode them separately, and concatenates them into a Hybrid Super Vector, and a linear SVM classifier is used to predict labels.
- 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.
- Finding Action Tubes with a Sparse-to-Dense Framework18
This paper proposes for the first time, an efficient framework that generates action tube proposals from video streams with a single forward pass in a sparse-to-dense manner and introduces a new dynamic feature sampling module (DTS) to effectively approximate the tube output while keeping the system tractable.
- 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.
- Acoustic Emission from Breaking a Bamboo Chopstick18
The acoustic emission from breaking a bamboo chopstick or a bundle of spaghetti is found to exhibit similar behavior as the famous seismic laws of Gutenberg and Richter, Omori, and Båth, and this model is deterministic and relies on the existence of a structured cross section, either fibrous or layered.
- End-to-end dense video grounding via parallel regression17
This paper tackles a new problem of dense video grounding, by simultaneously localizing multiple moments with a paragraph as input and devise a robust proposal-level attention loss to guide the training of PRVG, which is invariant to moment duration and contributes to model convergence.
- Does Video-Text Pretraining Help Open-Vocabulary Online Action Detection?16
This research establishes a robust base-line for zero-shot transfer in online action detection, enabling scalable solutions for open-world temporal understanding and proposes OV-OAD, a zero-shot online action detector that leverages vision-language models and learns solely from text supervision.
- 16
- Multiclass object detection by combining local appearances and context16
A novel approach for multiclass object detection by combining local appearances and contextual constraints is presented, which achieves state-of-the-art performance in multiclass objects detection.
- 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
- StageInteractor: Query-based Object Detector with Cross-stage Interaction14
A new query-based object detector with cross-stage interaction with cross-stage interaction, coined as StageInter-actor, and a cross-stage label assigner is designed to improve the one-to-one label assignment.
- 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.
- 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
- 12
- VideoEval: Comprehensive Benchmark Suite for Low-Cost Evaluation of Video Foundation Model11
A comprehensive benchmark suite is built to address the task adaptability of VFMs under few-shot conditions and assessing their representation power by directly applying to downstream tasks, and a large-scale study on 20 popular open-source vision foundation models is conducted.
- 11
- 11
- Data-efficient Event Camera Pre-training via Disentangled Masked Modeling10
This paper designs a semantic-uniform masking method to address the learning imbalance caused by the varying reconstruction difficulties of different regions in non-uniform data when using random masking, and decomposes the traditional hybrid masked modeling process into two branches.
- 10
- Learning Discriminative Feature Representation for Open Set Action Recognition10
This approach involves augmenting the input features with their reconstruction obtained from a reconstruction-based model in unsupervised training on known classes to learn the open-set classifier, forcing it to associate low correspondence both when the feature is from unknown classes as well as when the input feature and its reconstruction variant are inconsistent with each other.
- Adaptive image denoising for speckle noise images based on fuzzy logic10
An image denoising model based on fuzzy logic, which can eliminate speckle noise in the image well, improve the recognition of the image, and facilitate the acquisition of image information by doctors is proposed.
- Designing an Artificial Intelligence Platform to Assist Undergraduate in Art and Design to Develop a Personal Learning Plans10
A learning information service platform for undergraduate that will create new learning mode and opportunities to constructing students’ knowledge structure and hope to provide a new framework for the further innovation in higher education model.
- VideoTG-R1: Boosting Video Temporal Grounding via Curriculum Reinforcement Learning on Reflected Boundary Annotations9
A Boundary Reflection Agent that utilizes MLLMs to predict query-relevant timestamps outside the annotated intervals, allowing us to identify and filter out partially annotated samples, thereby reducing ambiguity and enabling data-efficient training.
- An Egocentric Vision-Language Model based Portable Real-time Smart Assistant9
Vinci is a vision-language system designed to provide real-time, comprehensive AI assistance on portable devices that integrates an egocentric vision foundation model with a large language model (LLM), enabling advanced functionalities such as scene understanding, temporal grounding, video summarization, and future planning.
- 9
- RefineTAD: Learning Proposal-free Refinement for Temporal Action Detection9
This work proposes a learnable, proposal-free refinement method for fine boundary localization, named RefineTAD, and proposes an offset focusing strategy to progressively refine the predicted results of TAD models in a coarse-to-fine manner with the authors' multi-scale offsets.
- 9
- Block Transform Image Coding By Multistage Vector Quantization With Optimal Bit Allocation9
A multistage vector quantization technique applied to the transform coefficients, where the effective number of bits assigned to each coefficient is proportional to the coefficient variance, which is well suited for progressive image transmission.
- 8
- CaReBench: A Fine-Grained Benchmark for Video Captioning and Retrieval8
Surprisingly, the experiment re-sults show that VLLE performs comparably to CLIP-based models on traditional benchmarks and has a stronger capability of fine-grained representation with lower spatial-temporal bias.
- Approximation algorithms for the dynamic k-level facility location problems8
A combinatorial primal-dual approximation algorithm is presented for the dynamic k-level facility location problem which finds a solution within 6 times the optimum when considering time factor.
- V4D: 4D Covolutional Neural Networks for Video-level Representations Learning8
This paper proposes Video-level 4D Convolutional Neural Networks, namely V4D, to model the evolution of long-range spatio-temporal representation with 4D convolutions, as well as preserving 3D spatio -temporal representations with residual connections.
- Weakly Supervised Temporal Sentence Grounding via Positive Sample Mining7
Positive Sample Mining (PSM) is proposed, a novel framework that mines positive samples from the training set to provide more discriminative supervision and introduces a PSM-guided contrastive loss to ensure that the anchor proposal is closer to similar samples and further from dissimilar ones.
- FreeRet: MLLMs as Training-Free Retrievers7
This work presents FreeRet, a plug-and-play framework that turns any MLLM into a two-stage retriever, and demonstrates that pretrained MLLMs, when carefully harnessed, can serve as strong retrieval engines without training.
- 7
- 7
- 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
- Approximation algorithm for a generalized Roman domination problem in unit ball graphs6
This paper presents the first constant approximation algorithm with a guaranteed performance ratio at most 6(k+2) in unit ball graphs, where k is a positive integer.
- Stochastic Layer-Wise Shuffle for Improving Vision Mamba Training5
A stochastic layer-wise shuffle regularization is proposed, which empowers successfully scaling non-hierarchical Vision Mamba to a large size (about 300M) in a supervised setting and can outperform the supervised ViTs of similar size.
- VLG: General Video Recognition with Web Textual Knowledge5
A unified visual-linguistic framework (VLG) to solve the problem of GVR with an effective two-stage training paradigm is presented, and extensive results show that the VLG obtains the state-of-the-art performance under four settings.
- 4
- 4
- 4
- Image Generation from Freehand Scene Sketches4
The first method for automatic image generation from scene-level freehand sketches is introduced with an attribute vector bridged generative adversarial network called edgeGAN which supports high visual-quality image content generation without using freehand sketches as training data.
- 4
- 4
- 4
- Considering the Spatial Layout Information of Bag of Features (BoF) Framework for Image Classification4
The proposed SDP method uses an extension model of Gauss mixture model to estimate the spatial layout distributions of the visual vocabulary and outperforms the traditional spatial pooling methods, and is competitive with the state-of-the-art classification accuracy on several challenging image datasets.
- 4
- Study on Application of Scale Invariant Feature Transform Algorithm on Automated Geometric Correction of Remote Sensing Images4
A two-step method is proposed, which firstly conducts coarse match on feature points, and then further conducts fine correction on the coarsely matched feature points by using the least squares technique, which is an effective automatic matching method for remote sensing images.
- An Improved Elman Neural Network with Profit Factors and Its Applications4
An improved model is proposed by introducing the direction and time profit factors to Elman neural network and applied to the fields of the finial investing and atmospheric environment.
- Arbitrary Generative Video Interpolation3
The Timestamp-aware Rotary Position Embedding (TaRoPE) is proposed, which modulates positions in temporal RoPE to align generated frames with target normalized timestamps and enables fine-grained control over frame timestamps, addressing the inflexibility of fixed-position paradigms in prior work.
- 3
- Harvest Video Foundation Models via Efficient Post-Pretraining3
This paper proposes an efficient framework to harvest video foundation models from image ones by randomly dropping input video patches and masking out input text during the post-pretraining procedure, which is extremely efficient and can be trained in less than one day on 8 GPUs.
- 3
- 2
- Evolaris: A Roadmap to Self-evolving Software Intelligence Management2
Evolaris is a self-evolving software intelligence system built on a multi-agent framework that could continuously improve the precision, timeliness, and scalability of software threat analysis, and offers a sustainable foundation for proactive decision-making and strengthens the broader ecosystem of security threat understanding.
- An Approximation Algorithm for a Variant of Dominating Set Problem2
A distributed approximation algorithm for the total dominating set problem in general graphs is given based on LP relaxation techniques and obtains a fractional total dominate set that is, at most, k(1+Δ1k)Δ 1k times the size of the optimal solution to this problem.
- 2
- A CU-level adaptive decision method for CNN-based in-loop filtering2
A Coding-unit (CU) level Adaptive Decision approach (CAD) which employs an early decision for each CU, based on their coding parameters, which achieves comparable performance with that of the RDO scheme while effectively reduces the encoding time complexity.
- 2
- 2
- A Novel Approach for Robust Surveillance Video Content Abstraction2
A novel framework to efficiently and robustly convert a surveillance video clip into one abstraction image containing the integrated contour of interested objects is proposed, which has the following novelties: an improved w-SIFT algorithm for Y-axis frames offset calculation, a trapezoid-based compensation algorithm for X-axis perspective distortion correction, and an incremental video content integration approach.
- 1
- SORCE: Small Object Retrieval in Complex Environments1
Experimental results show that the multi-embedding approach through MLLM and ReP significantly outperforms existing T2IR methods on SORCE-1K, a new benchmark consisting of images with complex environments and textual queries describing less conspicuous small objects with minimal contextual cues from other salient objects.
- 1
- 1
- 1
- 1
- Is Self-Supervised Pretraining Good for Extrapolation in Molecular Property Prediction?1
It is empirically revealed that while models were unable to accurately extrapolate absolute property values, self-supervised pretraining enables them to learn relative tendencies of unobserved property values and improve extrapolation performance.
- 1
- Exploring State Change Capture of Heterogeneous Backbones @ Ego4D Hands and Objects Challenge 20221
This technical report describes the method using heterogeneous backbones for the Ego4D Object State Change Classification and PNR Temporal Localization Challenge, which achieves an accuracy of 0.796 on OSCC while achieving an absolute temporal localization error of0.516 on PNR.
- Submission to Generic Event Boundary Detection Challenge@CVPR 2022: Local Context Modeling and Global Boundary Decoding Approach1
A local context modeling sub-network is proposed to perceive diverse patterns of generic event boundaries, and it generates powerful video representations and reliable boundary confidence, and global boundary decoding sub- network is exploited to decode event boundaries from a global view.
- A Wolf Pack Optimization Theory Based Improved Density Peaks Clustering Approach1
A Wolf Pack optimization theory based Density Peaks Clustering approach (WPA-DPC), which introduces d c parameter into the Wolf Pack Al algorithm to speed up the search and makes silhouette index in the search process as the fitness value, and the optimal location of the wolf group is the parameter value at the end.
- An Approximation Algorithm for the Dynamic k-level Facility Location Problem1
A combinatorial primal-dual approximation algorithm is presented for the dynamic k-level facility location problem which finds a solution within 6 times the optimum when considering time factor.
- A Local Density Shape Context Algorithm for Point Pattern Matching in Three Dimensional Space1
An effect of Local Density Shape Context algorithm (LDSC) is proposed, which outperforms the three compared classical methods in most cases and is robust to outliers and noise.
- 19.2: OLED Lifetime Improvement Based on Optimization of Luminance Compensation Considering Luminance Uniformity1
The experimental results indicate that the OLED lifetime can be extended by 28% same as previous method under the luminance compensation optimization and luminance uniformity improvement.
- 1
- Bundle of paper money recognition based on image entropy1
Experimental in different algorithms, including color training algorithm and the median filter and edge detection algorithm, demonstrates its real-time, effectiveness and robustness in the management of bank paper.
- Reflection on Teaching of College English Writing1
Research on college English writing is summarized, improvement measures are put forward, and future direction of research is predicted so as to provide reference for future research.
- 1
- –
- Approximation algorithms for the total dominating set problem–
A distributed algorithm for the minimum TDS problem via the LP relaxation techniques and outputs a fractional total dominating set of expected size O(kΔ2k)|TDSOPT.
- –
- –
- –
- Navigating the Pitfalls: From Digital Tools to Core Competencies in China’s Junior Secondary English Teaching–
This study constructs a theoretical model for technology-competency integration, and proposes three transformative pathways: the creation of digitally intelligent scenarios, the design of critical thinking tasks, and the implementation of precise feedback mechanisms.
- SA-GNAS: Seed Architecture Expansion for Efficient Large-scale Graph Neural Architecture Search–
The proposed SA-GNAS is a novel framework based on seed architecture expansion for efficient large-scale GNAS that outperforms human-designed state-of-the-art GNN architectures and existing graph NAS methods and can significantly reduce the search time.
- Bridging The Gaps Between Token Pruning and Full Pre-training via Masked Fine-tuning–
Inspired by MAE which performs masking and reconstruction self-supervised task, masked fine-tuning is devised to bridge the gaps between pre-trained base models used for initialization and token pruning based dynamic vision transformers, by masking image patches and predicting the image class label based on left unmasked patches.
- Cost Allocation and Strategyproof Mechanism for Cover-set Games–
This study designs a cost allocation method that can recover no less than a fraction of the total cost, and presents a strategyproof charging mechanism for agents who are selfish service providers with privately known valuations.
- NJU MCG - Sensetime Team Submission to Pre-training for Video Understanding Challenge Track II–
This paper follows the basic pipeline of temporal segment networks and uses the latest transformer-based architectures, e.g., Swin Transformer, DeiT, CLIP-ViT, to enhance the representation power and derive an ensemble of deep models to attain a high classification accuracy.
- Cross-modal Pretraining and Matching for Video Understanding–
A general cross-modal pair discrimination (CPD) framework to capture this correlation between a video and its associated text, and presents a Contrastive and Compatible Matching Network (C2M-Net), to directly model the relations between language queries and video moments in a joint embedding space.
- –
- –
- –
- –
- –
- –
- –
- –
- –
- Research and Development of Plant Information Management System–
The browsing and inquiring ways of plant information management system were introduced as well as the main functional modules.
- –
- –
- –
- Data Collation and Integration of Multisource Geological Basic Information–
The theories and main methods of multisource data integration for constructing geological spatial database are researched and the solutions to deal with the problems existing in integrated database are proposed.
- An Automatic Algorithm for Image Segmentation in Urine Sediment Examination–
The Mumford-Shah model based Level Set algorithm can achieve urine sediment examinations accurately with both fast speed and strong robustness to the noise.
- A Design of Video Collection System Based on DSK–
This video collection system based on TMS320C6711 DSK will serve as an application platform for future researches of image processing and decoding algorithm of video compression.
- –
- Editing and Innovation on News Program of Broadcasting–
To increace the radio ratings, the optimized program should be not only edited ond innovated continuously, but also distinguish for its characteristic from different media, only theses can broaden its market share.
- –
- –
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.