CVPR 2023 Papers — Page 7
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2353 papers
Dynamic Aggregated Network for Gait Recognition
Kang Ma (Beijing Institute of Technology), Yongzhen Huang (Beijing Normal University)
RecognitionConvolutional Neural NetworkContrastive LearningImageVideo
🎯 What it does: This paper proposes a Dynamic Aggregation Network (DANet) that adaptively aggregates local motion patterns through Local Convolutional Mixture Blocks (LCMB) and a Global Motion Pattern Aggregator (GMPA) to construct robust global gait features.
Dynamic Coarse-To-Fine Learning for Oriented Tiny Object Detection
Chang Xu (Wuhan University), Gui-Song Xia (Wuhan University)
Object DetectionImage
🎯 What it does: This paper proposes a Dynamic Coarse-Fine Learning (DCFL) framework specifically designed to address the detection of inclined small targets with extreme shapes and scales.
Dynamic Conceptional Contrastive Learning for Generalized Category Discovery
Nan Pu (University of Trento), Nicu Sebe (University of Trento)
ClassificationRepresentation LearningTransformerContrastive LearningImage
🎯 What it does: The study proposes a dynamic concept contrastive learning framework (DCCL) for generalized category discovery tasks, jointly learning representations and concept generation.
Dynamic Focus-Aware Positional Queries for Semantic Segmentation
Haoyu He (Monash University), Bohan Zhuang (Monash University)
Object DetectionSegmentationTransformerImage
🎯 What it does: A dynamic focus-aware position query (DFPQ) and high-resolution cross-attention (HRCA) have been designed and implemented to enhance the localization accuracy and detail recovery of DETR-style semantic segmentation models.
Dynamic Generative Targeted Attacks With Pattern Injection
Weiwei Feng (University of Science and Technology of China), Yongdong Zhang (University of Science and Technology of China)
Adversarial AttackConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: A dynamic generative target attack model is designed, utilizing pattern injection to achieve transferable targeted adversarial examples.
Dynamic Graph Enhanced Contrastive Learning for Chest X-Ray Report Generation
Mingjie Li (University of Technology Sydney), Xiaojun Chang (University of Technology Sydney)
GenerationTransformerContrastive LearningImageTextMultimodalityElectronic Health Records
🎯 What it does: For the task of generating chest X-ray reports, a dynamic knowledge graph structure is proposed, which enhances visual features through graph encoding and graph-visual cross-attention, and trains the model using image-text contrastive learning and matching loss.
Dynamic Graph Learning With Content-Guided Spatial-Frequency Relation Reasoning for Deepfake Detection
Yuan Wang (Institute of Automation, Chinese Academy of Sciences), Silong Peng (Institute of Automation, Chinese Academy of Sciences)
ClassificationAnomaly DetectionConvolutional Neural NetworkGraph Neural NetworkImageVideo
🎯 What it does: A dynamic graph learning framework SFDG based on spatial-frequency relationships is proposed for deepfake detection.
Dynamic Inference With Grounding Based Vision and Language Models
Burak Uzkent (Amazon Prime Video), Mohamed Omar (Amazon Prime Video)
RecognitionObject DetectionSegmentationComputational EfficiencyKnowledge DistillationTransformerReinforcement LearningVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes a dynamic reasoning framework for location-based vision-language tasks, capable of dynamically cropping word tokens from input image-text pairs, integrating visual and textual tokens, and conditionally skipping across multi-head self-attention and feedforward network layers in visual, textual, and multimodal Transformers.
Dynamic Neural Network for Multi-Task Learning Searching Across Diverse Network Topologies
Wonhyeok Choi (DGIST), Sunghoon Im (DGIST)
Object DetectionSegmentationNeural Architecture SearchImage
🎯 What it does: A framework is proposed for searching task-adaptive structures and sharing patterns for multi-task learning within a single network.
Dynamically Instance-Guided Adaptation: A Backward-Free Approach for Test-Time Domain Adaptive Semantic Segmentation
Wei Wang (Western University), Nicu Sebe (University of Trento)
SegmentationDomain AdaptationImage
🎯 What it does: The paper proposes a test-time domain adaptive semantic segmentation method (DIGA) that achieves online real-time adaptation without backpropagation.
DynamicDet: A Unified Dynamic Architecture for Object Detection
Zhihao Lin (Peking University), Xiaojie Chu (Peking University)
Object DetectionConvolutional Neural NetworkImage
🎯 What it does: A dynamic reasoning unified object detection framework called DynamicDet is proposed, allowing a single model to switch between different speed-accuracy points.
DynamicStereo: Consistent Dynamic Depth From Stereo Videos
Nikita Karaev (Meta AI), Christian Rupprecht (Visual Geometry Group University of Oxford)
Depth EstimationTransformerVideo
🎯 What it does: A dynamic stereo matching network based on Transformer, DynamicStereo, is proposed to achieve temporally consistent depth estimation in video sequences.
DyNCA: Real-Time Dynamic Texture Synthesis Using Neural Cellular Automata
Ehsan Pajouheshgar (École Polytechnique Fédérale de Lausanne), Sabine Süsstrunk (École Polytechnique Fédérale de Lausanne)
GenerationData SynthesisOptical FlowImageVideo
🎯 What it does: The DyNCA model is proposed, utilizing neural cellular automata to achieve real-time, controllable dynamic texture synthesis, capable of generating infinitely long and arbitrarily sized video textures.
DynIBaR: Neural Dynamic Image-Based Rendering
Zhengqi Li (Google Research), Noah Snavely (Google Research)
GenerationData SynthesisNeural Radiance FieldOptical FlowVideo
🎯 What it does: A new perspective synthesis method for dynamic scenes based on a volumetric image rendering framework is proposed, capable of handling long-duration, complex camera and object motion in monocular videos.
E2PN: Efficient SE(3)-Equivariant Point Network
Minghan Zhu (University of Michigan), Huei Peng (University of Michigan)
ClassificationPose EstimationRetrievalComputational EfficiencyConvolutional Neural NetworkPoint Cloud
🎯 What it does: A three-dimensional point cloud equivariant convolutional network named E2PN is proposed, utilizing the quotient space S²×R³ to achieve SE(3) equivariant feature learning.
EC2: Emergent Communication for Embodied Control
Yao Mu (Hong Kong University), Chuang Gan (Massachusetts Institute of Technology IBM Watson AI Lab)
Representation LearningRobotic IntelligenceTransformerLarge Language ModelReinforcement LearningVideoTextMultimodality
🎯 What it does: Proposes the EC2 framework, which combines unsupervised emergent communication with GPT-style language models for pre-training multimodal representations applicable to few-shot robotic control.
ECON: Explicit Clothed Humans Optimized via Normal Integration
Yuliang Xiu (Max Planck Institute for Intelligent Systems), Michael J. Black (Max Planck Institute for Intelligent Systems)
GenerationPose EstimationImageMesh
🎯 What it does: This study proposes a full-body 3D human reconstruction method called ECON based on a single RGB image, capable of generating high-detail, complete 3D body shapes even in the presence of loose clothing and complex poses. The method first predicts front and back normal maps, utilizes the SMPL-X prior to construct a 2.5D surface, and then performs shape completion to ultimately obtain a complete mesh.
EcoTTA: Memory-Efficient Continual Test-Time Adaptation via Self-Distilled Regularization
Junha Song (Qualcomm AI Research), Sungha Choi (Qualcomm AI Research)
SegmentationDomain AdaptationKnowledge DistillationImage
🎯 What it does: This paper proposes a memory-efficient continuous testing adaptive framework (EcoTTA), which achieves online adaptation to the target domain by attaching a lightweight meta-network to a pre-trained model and only updating these meta-networks during testing.
EDA: Explicit Text-Decoupling and Dense Alignment for 3D Visual Grounding
Yanmin Wu (Peking University), Jian Zhang (Peking University)
RecognitionObject DetectionTransformerContrastive LearningTextPoint Cloud
🎯 What it does: A 3D visual localization method (EDA) is proposed, which achieves more accurate target localization by decoupling natural language sentences into various semantic components and densely aligning each component with point cloud objects.
Edge-Aware Regional Message Passing Controller for Image Forgery Localization
Dong Li (University of Science and Technology of China), Zheng-Jun Zha (University of Science and Technology of China)
SegmentationGraph Neural NetworkImage
🎯 What it does: A region message passing controller based on edge perception (ERMPC) is proposed, which dynamically constructs graphs through learnable edge information, significantly suppressing the feature coupling between forged and real areas, thereby improving the accuracy of image forgery localization.
EDGE: Editable Dance Generation From Music
Jonathan Tseng (Stanford University), Karen Liu
GenerationData SynthesisPose EstimationTransformerDiffusion modelVideoMultimodalityAudio
🎯 What it does: An editable dance generation model called EDGE is proposed, which utilizes a diffusion model and the music feature extractor Jukebox to generate dances that are synchronized with musical beats, physically feasible, and freely editable.
Edges to Shapes to Concepts: Adversarial Augmentation for Robust Vision
Aditay Tripathi (Indian Institute of Science), Pradeep Shenoy (Google Research India)
ClassificationObject DetectionSegmentationTransformerGenerative Adversarial NetworkImage
🎯 What it does: A lightweight adversarial data augmentation method called ELEAS is proposed, which encourages the model to learn shape features by mixing edge maps with randomly shuffled block images.
EDICT: Exact Diffusion Inversion via Coupled Transformations
Bram Wallace (Salesforce Research), Nikhil Naik (Salesforce Research)
RestorationGenerationDiffusion modelImage
🎯 What it does: A precise diffusion inverse method named EDICT is proposed, utilizing coupled transformations to achieve complete reconstruction and editing of real images.
EditableNeRF: Editing Topologically Varying Neural Radiance Fields by Key Points
Chengwei Zheng (Tsinghua University), Feng Xu (Tsinghua University)
GenerationData SynthesisDepth EstimationNeural Radiance FieldOptical FlowVideo
🎯 What it does: This paper proposes an editable dynamic scene NeRF model called EditableNeRF, which can automatically reconstruct from a single camera video without human supervision and supports three-dimensional editing with vertex changes.
EFEM: Equivariant Neural Field Expectation Maximization for 3D Object Segmentation Without Scene Supervision
Jiahui Lei (University of Pennsylvania), Kostas Daniilidis (University of Pennsylvania)
SegmentationPose EstimationPoint CloudMesh
🎯 What it does: An unsupervised 3D instance segmentation method EFEM is proposed, which segments in the scene point cloud using the shape prior of a single object, and can simultaneously estimate pose and reconstruction.
Effective Ambiguity Attack Against Passport-Based DNN Intellectual Property Protection Schemes Through Fully Connected Layer Substitution
Yiming Chen (University of Macau), Jiantao Zhou (University of Macau)
Adversarial AttackImage
🎯 What it does: A type of ambiguity attack targeting passport-based deep learning model IP protection is designed, generating counterfeit passports that maintain the original model's performance using a minimal amount of training data.
Efficient and Explicit Modelling of Image Hierarchies for Image Restoration
Yawei Li (ETH Zurich), Luc Van Gool (ETH Zurich)
RestorationSuper ResolutionTransformerImage
🎯 What it does: A GRL network based on anchor stripe self-attention is proposed, achieving efficient explicit modeling of global, regional, and local features of images.
Efficient Frequency Domain-Based Transformers for High-Quality Image Deblurring
Lingshun Kong (Nanjing University of Science and Technology), Jinshan Pan (China Electronics Technology Group Corporation)
RestorationComputational EfficiencyTransformerImage
🎯 What it does: An efficient Transformer model utilizing frequency domain characteristics is proposed for high-quality image deblurring.
Efficient Hierarchical Entropy Model for Learned Point Cloud Compression
Rui Song (Peking University), Ge Li (Peking University)
CompressionPoint Cloud
🎯 What it does: An efficient Hierarchical Attention Entropy Model (EHEM) is proposed for point cloud compression, balancing compression performance and decoding speed.
Efficient Loss Function by Minimizing the Detrimental Effect of Floating-Point Errors on Gradient-Based Attacks
Yunrui Yu (University of Macau), Cheng-Zhong Xu (University of Macau)
OptimizationAdversarial AttackImage
🎯 What it does: A new loss function MIFPE is proposed, which can reduce gradient distortion caused by floating-point errors in gradient attacks, thereby more accurately assessing model robustness.
Efficient Map Sparsification Based on 2D and 3D Discretized Grids
Xiaoyu Zhang (Chinese University of Hong Kong), Yun-Hui Liu (Chinese University of Hong Kong)
CompressionOptimizationComputational EfficiencySimultaneous Localization and MappingPoint Cloud
🎯 What it does: Sparse compression of large-scale visual SLAM maps, retaining only a subset that ensures positioning performance.
Efficient Mask Correction for Click-Based Interactive Image Segmentation
Fei Du (Alibaba Group), Fan Wang (Alibaba Group)
SegmentationConvolutional Neural NetworkImage
🎯 What it does: An efficient click-based interactive image segmentation method is proposed, which quickly corrects the mask after each click through click-guided self-attention and correlation modules.
Efficient Movie Scene Detection Using State-Space Transformers
Md Mohaiminul Islam (University of North Carolina Chapel Hill), Gedas Bertasius (University of North Carolina Chapel Hill)
Object DetectionComputational EfficiencyTransformerContrastive LearningVideo
🎯 What it does: A movie scene detection model for long videos, TranS4mer, is proposed, which can efficiently capture short-term and long-term dependencies, improving the accuracy of scene boundary judgment.
Efficient Multimodal Fusion via Interactive Prompting
Yaowei Li (University of Technology Sydney), Yi Yang (Zhejiang University)
ClassificationComputational EfficiencyTransformerPrompt EngineeringMultimodality
🎯 What it does: This paper proposes an efficient multimodal fusion method called PMF, which utilizes a frozen single-modal pre-trained Transformer and adds interactive prompts in the deeper layers to achieve cross-modal feature fusion.
Efficient On-Device Training via Gradient Filtering
Yuedong Yang (University of Texas at Austin), Radu Marculescu (University of Texas at Austin)
ClassificationSegmentationComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: A Gradient Filtering method is proposed, which constructs a gradient mapping with fewer unique elements by block averaging the gradient map during backpropagation, significantly reducing the computational load and memory usage of backpropagation in convolutional layers.
Efficient RGB-T Tracking via Cross-Modality Distillation
Tianlu Zhang (Xidian University), Jungong Han (University of Sheffield)
Object TrackingComputational EfficiencyKnowledge DistillationConvolutional Neural NetworkImageMultimodality
🎯 What it does: A cross-modal distillation framework is proposed to transfer knowledge from a dual-stream RGB-T tracker to a single-stream lightweight model, achieving efficient RGB-T tracking.
Efficient Robust Principal Component Analysis via Block Krylov Iteration and CUR Decomposition
Shun Fang (Wuhan University of Science and Technology), Shoulie Xie (Institute for Infocomm Research A*STAR)
RestorationAnomaly DetectionOptimizationComputational EfficiencyImageVideo
🎯 What it does: This paper proposes an efficient robust principal component analysis (eRPCA) algorithm that can simultaneously recover low-rank matrices and sparse errors without prior knowledge of the rank, significantly reducing computational complexity.
Efficient Scale-Invariant Generator With Column-Row Entangled Pixel Synthesis
Thuan Hoang Nguyen (VinAI Research), Anh Tran (VinAI Research)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: A new arbitrary scale image generation network called CREPS is proposed, which can generate scale-consistent and detail-coherent images at different resolutions.
Efficient Second-Order Plane Adjustment
Lipu Zhou (Meituan)
Pose EstimationOptimizationComputational EfficiencySimultaneous Localization and MappingPoint Cloud
🎯 What it does: An efficient plane adjustment algorithm using the Newton method is proposed, which can quickly solve large-scale joint optimization problems of planes and attitudes.
Efficient Semantic Segmentation by Altering Resolutions for Compressed Videos
Yubin Hu (Tsinghua University), Yong-Jin Liu (Tsinghua University)
SegmentationCompressionComputational EfficiencyConvolutional Neural NetworkOptical FlowVideo
🎯 What it does: A resolution-changing framework for compressed video semantic segmentation, AR-Seg, is proposed, which processes high-resolution keyframes and low-resolution non-keyframes in parallel and improves accuracy through cross-resolution feature fusion.
Efficient Verification of Neural Networks Against LVM-Based Specifications
Harleen Hanspal (Imperial College London), Alessio Lomuscio (Safe Intelligence)
Pose EstimationComputational EfficiencyRobotic IntelligenceFlow-based ModelAuto EncoderImage
🎯 What it does: A reversible encoding head is proposed to be embedded in the network to be verified, defining specifications in the latent space, thereby efficiently verifying the robustness of neural networks against nonlinear transformations such as pose, viewpoint, and occlusion.
Efficient View Synthesis and 3D-Based Multi-Frame Denoising With Multiplane Feature Representations
Thomas Tanay (Huawei Noah's Ark Lab), Matteo Maggioni (Huawei Noah's Ark Lab)
RestorationData SynthesisConvolutional Neural NetworkAuto EncoderImage
🎯 What it does: A three-dimensional multi-frame denoising and view synthesis method based on Multi-Plane Features (MPF) is proposed, which transfers Multi-Plane Images (MPI) to feature space, utilizing a learnable encoder-renderer to achieve cross-depth consistency, significantly improving denoising and synthesis quality.
EfficientSCI: Densely Connected Network With Space-Time Factorization for Large-Scale Video Snapshot Compressive Imaging
Lishun Wang (Chengdu Institute of Computer Application Chinese Academy of Sciences, University of Chinese Academy of Sciences), Xin Yuan (Westlake University, Zhejiang University)
RestorationCompressionConvolutional Neural NetworkTransformerVideo
🎯 What it does: This paper proposes an end-to-end network called EfficientSCI for reconstructing high-quality HD video from single-shot video Snapshot Compressive Imaging (SCI).
EfficientViT: Memory Efficient Vision Transformer With Cascaded Group Attention
Xinyu Liu (Chinese University of Hong Kong), Yixuan Yuan (Chinese University of Hong Kong)
Object DetectionComputational EfficiencyTransformerImage
🎯 What it does: This study investigates the inference speed bottleneck of Vision Transformers and proposes two key modules: Memory-Efficient Sandwich Block and Cascaded Group Attention, constructing the EfficientViT framework.
Ego-Body Pose Estimation via Ego-Head Pose Estimation
Jiaman Li (Stanford University), Jiajun Wu (Stanford University)
GenerationPose EstimationTransformerDiffusion modelSimultaneous Localization and MappingOptical FlowVideo
🎯 What it does: Predicting full-body 3D motion from head-mounted camera video involves first estimating head pose, and then generating full-body pose using a conditional diffusion model based on the head pose.
Egocentric Audio-Visual Object Localization
Chao Huang (University of Rochester), Chenliang Xu (University of Rochester)
RecognitionObject DetectionConvolutional Neural NetworkContrastive LearningVideoMultimodalityAudio
🎯 What it does: This study investigates the fine-grained association between sound and vision in first-person perspective videos and proposes a self-supervised framework for locating sound-emitting objects.
Egocentric Auditory Attention Localization in Conversations
Fiona Ryan (Georgia Institute of Technology), Vamsi Krishna Ithapu (Meta Reality Labs Research)
RecognitionObject DetectionTransformerVideoMultimodalityAudio
🎯 What it does: This study investigates an algorithm for locating the listener's focus of attention using subject perspective video and multi-channel audio.
Egocentric Video Task Translation
Zihui Xue (University of Texas at Austin), Lorenzo Torresani (Meta AI)
RecognitionOptimizationTransformerVideoMultimodality
🎯 What it does: Proposes the EgoTask Translation (EgoT2) framework, which integrates features from multiple tasks through a task translator to uniformly enhance the performance of egocentric video multi-tasking.
Elastic Aggregation for Federated Optimization
Dengsheng Chen (Meituan), Enhua Wu (University of Macau)
OptimizationFederated LearningImageText
🎯 What it does: This paper proposes a flexible aggregation algorithm to alleviate the client drift problem in federated learning.
EMT-NAS:Transferring Architectural Knowledge Between Tasks From Different Datasets
Peng Liao (East China University of Science and Technology), Wenli Du (East China University of Science and Technology)
ClassificationNeural Architecture SearchImageBiomedical Data
🎯 What it does: This paper proposes an Evolutionary Multi-Task Neural Architecture Search (EMT-NAS), which significantly improves the accuracy of various tasks and reduces search time by sharing network architecture knowledge between classification tasks on different datasets and training weights separately.
End-to-End 3D Dense Captioning With Vote2Cap-DETR
Sijin Chen (Fudan University), Tao Chen (Fudan University)
Object DetectionGenerationTransformerMultimodalityPoint Cloud
🎯 What it does: An end-to-end single-stage 3D fine-grained description framework called Vote2Cap-DETR is proposed, capable of simultaneously performing object detection and natural language description.
End-to-End Vectorized HD-Map Construction With Piecewise Bezier Curve
Limeng Qiao (MEGVII Technology), Chi Zhang (MEGVII Technology)
Object DetectionAutonomous DrivingTransformerImagePoint Cloud
🎯 What it does: An end-to-end high-precision HD map construction network BeMapNet has been designed and implemented, utilizing a unified piecewise Bézier curve for vectorized representation, completely eliminating the post-processing step.
End-to-End Video Matting With Trimap Propagation
Wei-Lun Huang (National Taiwan University), Ming-Sui Lee (National Taiwan University)
Object DetectionSegmentationConvolutional Neural NetworkVideo
🎯 What it does: An end-to-end video matting model FTP-VM is proposed, integrating trimap propagation and video matting functionalities.
Endpoints Weight Fusion for Class Incremental Semantic Segmentation
Jia-Wen Xiao (Nankai University), Ming-Ming Cheng (Nankai University)
SegmentationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: A strategy called Endpoints Weight Fusion (EWF) is proposed, which does not require further training and is used for class-incremental semantic segmentation, combined with knowledge distillation to enhance the model's memory of old knowledge.
Energy-Efficient Adaptive 3D Sensing
Brevin Tilmon (University of Florida), Jian Wang (Snap Inc.)
Depth EstimationOptimizationComputational EfficiencyPoint Cloud
🎯 What it does: An energy-efficient adaptive 3D perception system is proposed and implemented, utilizing a camera to generate attention maps that project structured light only in the regions of interest, significantly reducing power consumption and enhancing eye safety while maintaining the same maximum measurement distance.
Enhanced Multimodal Representation Learning With Cross-Modal KD
Mengxi Chen (Shanghai Jiao Tong University), Ya Zhang (Shanghai Jiao Tong University)
RecognitionRetrievalKnowledge DistillationRepresentation LearningGenerative Adversarial NetworkContrastive LearningVideoMultimodalityAudio
🎯 What it does: A cross-modal knowledge distillation method called AMID is proposed, which enhances the representation capability of the target modality by maximizing the mutual information between the teacher-student and teacher-auxiliary models while minimizing the conditional entropy.
Enhanced Stable View Synthesis
Nishant Jain (Indian Institute of Technology Roorkee), Luc Van Gool (ETH Zurich)
Data SynthesisDepth EstimationOptimizationConvolutional Neural NetworkGraph Neural NetworkImageBenchmark
🎯 What it does: This paper addresses the problem of viewpoint synthesis based on freely moving cameras in outdoor scenes by integrating complementary information from multi-view stereo (MVS) and monocular depth prediction, jointly refining camera poses, and employing a graph neural network to achieve multi-rotation averaging, ultimately resulting in more detailed and realistic new viewpoint rendering.
Enhanced Training of Query-Based Object Detection via Selective Query Recollection
Fangyi Chen (Carnegie Mellon University), Marios Savvides (Carnegie Mellon University)
Object DetectionTransformerImage
🎯 What it does: This paper proposes a Selective Query Recollection (SQR) training strategy to enhance the final stage performance of query-based object detectors.
Enhancing Deformable Local Features by Jointly Learning To Detect and Describe Keypoints
Guilherme Potje (Federal University of Minas Gerais), Erickson R. Nascimento (Microsoft)
Object DetectionRetrievalConvolutional Neural NetworkReinforcement LearningImage
🎯 What it does: Designed and trained an end-to-end keypoint detection and description network DALF for robust local feature matching in non-rigid deformation scenarios.
Enhancing Multiple Reliability Measures via Nuisance-Extended Information Bottleneck
Jongheon Jeong (Korea Advanced Institute of Science and Technology), Jinwoo Shin (Korea Advanced Institute of Science and Technology)
Object DetectionRepresentation LearningAdversarial AttackConvolutional Neural NetworkTransformerAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: A novel information bottleneck framework (NIB) is proposed, and an autoencoder version (AENIB) is implemented to enhance the model's performance on various robustness metrics by learning representations of negligible noise.
Enhancing the Self-Universality for Transferable Targeted Attacks
Zhipeng Wei (Fudan University), Yu-Gang Jiang (Fudan University)
Adversarial AttackConvolutional Neural NetworkImage
🎯 What it does: A transfer-based targeted attack method is proposed based on Self-Universality (SU), which generates perturbations that are insensitive to different local areas, more targeted, and do not require additional auxiliary networks by maximizing the feature similarity between the global image and the local image obtained from random cropping.
Enlarging Instance-Specific and Class-Specific Information for Open-Set Action Recognition
Jun Cen (Hong Kong University of Science and Technology), Qifeng Chen (Hong Kong University of Science and Technology)
RecognitionContrastive LearningVideo
🎯 What it does: In the open set action recognition (OSAR) task, a Prototypical Similarity Learning (PSL) framework is proposed, which enhances instance specificity (IS) and class specificity (CS) information by preserving instance differences in the feature space and introducing video shuffling;
Ensemble-Based Blackbox Attacks on Dense Prediction
Zikui Cai (University of California Riverside), M. Salman Asif (University of California Riverside)
Object DetectionSegmentationAdversarial AttackImage
🎯 What it does: This paper studies a black-box attack method based on multi-model ensemble, which can generate targeted adversarial samples for object detection and semantic segmentation models.
EqMotion: Equivariant Multi-Agent Motion Prediction With Invariant Interaction Reasoning
Chenxin Xu (Shanghai Jiao Tong University), Yanfeng Wang (Shanghai Jiao Tong University)
Pose EstimationAutonomous DrivingOptimizationGraph Neural NetworkGraphTime SeriesSequentialPhysics Related
🎯 What it does: This study presents EqMotion, a multi-agent motion prediction model that maintains the invariance of Euclidean geometric transformations and remains invariant in interactive reasoning.
Equiangular Basis Vectors
Yang Shen (Nanjing University of Science and Technology), Xiu-Shen Wei (Nanjing University of Science and Technology)
ClassificationObject DetectionSegmentationImage
🎯 What it does: Proposes Equiangular Basis Vectors (EBVs), using fixed equiangular vectors to replace traditional fully connected + softmax classifiers;
Equivalent Transformation and Dual Stream Network Construction for Mobile Image Super-Resolution
Jiahao Chao (East China Normal University), Lydia Dehbi (Chengdu Institute of Computer Applications of Chinese Academy of Sciences)
RestorationSuper ResolutionConvolutional Neural NetworkImage
🎯 What it does: An equivalent transformation and dual-stream network structure is proposed for real-time image super-resolution on mobile devices.
ERM-KTP: Knowledge-Level Machine Unlearning via Knowledge Transfer
Shen Lin (Xidian University), Willy Susilo (University of Wollongong)
ClassificationExplainability and InterpretabilityComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a knowledge-level machine forgetting method called ERM-KTP for achieving class-level interpretable forgetting.
ERNIE-ViLG 2.0: Improving Text-to-Image Diffusion Model With Knowledge-Enhanced Mixture-of-Denoising-Experts
Zhida Feng (Baidu Inc), Haifeng Wang (Baidu Inc)
GenerationData SynthesisTransformerMixture of ExpertsDiffusion modelImageText
🎯 What it does: A Chinese text-to-image generation system based on diffusion models, ERNIE-ViLG 2.0, is proposed, which enhances image quality and text consistency through knowledge enhancement and multi-expert denoising techniques.
ESLAM: Efficient Dense SLAM System Based on Hybrid Representation of Signed Distance Fields
Mohammad Mahdi Johari (Idiap Research Institute), François Fleuret (University of Geneva)
OptimizationComputational EfficiencyRobotic IntelligenceNeural Radiance FieldSimultaneous Localization and MappingPoint Cloud
🎯 What it does: A real-time RGB-D SLAM system based on TSDF and multi-scale axis-aligned feature planes is proposed, capable of achieving high-precision geometric and appearance reconstruction without pre-training.
EVA: Exploring the Limits of Masked Visual Representation Learning at Scale
Yuxin Fang (Huazhong University of Science and Technology), Yue Cao (Beijing Academy of Artificial Intelligence)
Object DetectionSegmentationRepresentation LearningTransformerContrastive LearningImageMultimodality
🎯 What it does: A 1B parameter ViT model called EVA was constructed and pre-trained, using a mask feature reconstruction MIM task on publicly available image data, and applied to various visual downstream tasks and cross-modal CLIP models.
Evading DeepFake Detectors via Adversarial Statistical Consistency
Yang Hou (Kyushu University), Jianjun Zhao (Kyushu University)
Adversarial AttackGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes two natural degradation-based adversarial attack methods, StatAttack and MStatAttack, to reduce the statistical differences between deepfake images and real images in both spatial and frequency domains, thereby bypassing various DeepFake detectors.
Evading Forensic Classifiers With Attribute-Conditioned Adversarial Faces
Fahad Shamshad (Mohamed bin Zayed University of AI), Karthik Nandakumar (Mohamed bin Zayed University of AI)
GenerationData SynthesisAdversarial AttackMeta LearningGenerative Adversarial NetworkImage
🎯 What it does: By utilizing the hierarchical latent space of StyleGAN2 and CLIP text guidance, adversarial fake faces with specified attributes are generated, capable of misleading forensic classifiers.
EVAL: Explainable Video Anomaly Localization
Ashish Singh (University of Massachusetts), Erik G. Learned-Miller (University of Massachusetts)
Anomaly DetectionExplainability and InterpretabilityConvolutional Neural NetworkVideo
🎯 What it does: An interpretable single-scene video anomaly localization framework is proposed, which achieves anomaly detection by learning high-level object and motion attributes.
Event-Based Blurry Frame Interpolation Under Blind Exposure
Wenming Weng (University of Science and Technology of China), Zhiwei Xiong (University of Science and Technology of China)
RestorationVideo
🎯 What it does: This paper proposes a blind exposure blur frame interpolation method based on event cameras, which can recover high frame rate clear videos from low frame rate blurry videos without knowing the exposure time.
Event-Based Frame Interpolation With Ad-Hoc Deblurring
Lei Sun (Zhejiang University), Luc Van Gool (KU Leuven)
RestorationRecurrent Neural NetworkImageVideo
🎯 What it does: A unified event-driven frame interpolation framework called REFID is proposed, which can simultaneously achieve event information-assisted deblurring and frame interpolation without a separate deblurring step.
Event-Based Shape From Polarization
Manasi Muglikar (University of Zurich), Davide Scaramuzza (University of Zurich)
Data SynthesisDepth EstimationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a Shape from Polarization (SfP) system utilizing a rotating polarizer and an event camera, which includes both physical modeling and deep learning estimation methods.
Event-Based Video Frame Interpolation With Cross-Modal Asymmetric Bidirectional Motion Fields
Taewoo Kim (Korea Advanced Institute of Science and Technology), Kuk-Jin Yoon (Korea Advanced Institute of Science and Technology)
Image TranslationRestorationTransformerOptical FlowImageVideoMultimodality
🎯 What it does: A cross-modal asymmetric bidirectional motion field estimation framework based on event cameras (EIF-BiOFNet) is proposed, and a large-scale high frame rate event-image dataset ERF-X170FPS is constructed for video frame interpolation.
Event-Guided Person Re-Identification via Sparse-Dense Complementary Learning
Chengzhi Cao (University of Science and Technology of China), Zheng-Jun Zha (University of Science and Technology of China)
RecognitionRetrievalSpiking Neural NetworkVideo
🎯 What it does: A Sparse-Dense Complementary Learning Network (SDCL) based on event cameras is proposed, which enhances video person re-identification performance by guiding video frame feature extraction through event streams.
EventNeRF: Neural Radiance Fields From a Single Colour Event Camera
Viktor Rudnev (Max Planck Institute for Informatics), Vladislav Golyanik (Max Planck Institute for Informatics)
RestorationGenerationData SynthesisNeural Radiance FieldImageVideo
🎯 What it does: Using a single-channel color event camera's event stream to train NeRF, achieving high-quality, dense, and view-consistent RGB rendering directly from events.
Evolved Part Masking for Self-Supervised Learning
Zhanzhou Feng (National Key Laboratory for Multimedia Information Processing, Peking University), Shiliang Zhang (Peng Cheng Laboratory)
ClassificationObject DetectionSegmentationTransformerContrastive LearningImage
🎯 What it does: An adaptive evolved part masking method is proposed, which dynamically generates masking patterns using the attention of visual models during training to enhance the pre-training effect of Masked Image Modeling.
EvShutter: Transforming Events for Unconstrained Rolling Shutter Correction
Julius Erbach (Huawei Technologies), Yuanyou Li (Huawei Technologies)
RestorationOptical FlowImage
🎯 What it does: A method called EvShutter is proposed to correct rolling shutter (RS) artifacts using a single RGB image and event information.
Exact-NeRF: An Exploration of a Precise Volumetric Parameterization for Neural Radiance Fields
Brian K. S. Isaac-Medina (Durham University), Toby P. Breckon (Durham University)
GenerationData SynthesisNeural Radiance FieldPoint Cloud
🎯 What it does: Proposes Exact-NeRF, which uses pyramid-based precise volume parameterization to achieve accurate positional encoding for NeRF, replacing the traditional Gaussian approximation.
EXCALIBUR: Encouraging and Evaluating Embodied Exploration
Hao Zhu (Carnegie Mellon University), Luca Weihs (Allen Institute for Artificial Intelligence)
Robotic IntelligenceRecurrent Neural NetworkLarge Language ModelReinforcement LearningTextBenchmark
🎯 What it does: This paper proposes the EXCALIBUR benchmark, allowing embodied agents to engage in long-term open exploration within virtual houses and testing their understanding of physical and visual properties through natural language question answering; it also provides a VR interface for human comparative experiments.
Executing Your Commands via Motion Diffusion in Latent Space
Xin Chen (Tencent), Gang Yu (Tencent)
GenerationData SynthesisPose EstimationTransformerDiffusion modelAuto EncoderVideoTextSequential
🎯 What it does: In this paper, the authors propose a conditional human motion generation method based on a latent space diffusion model (Motion Latent-Diffusion, MLD). This method first compresses the original motion sequences into low-dimensional latent variables using a Transformer-VAE, and then trains a conditional diffusion model in this latent variable space to achieve motion synthesis based on action categories or text descriptions.
Exemplar-FreeSOLO: Enhancing Unsupervised Instance Segmentation With Exemplars
Taoseef Ishtiak (Carleton University), Yuhong Guo (Carleton University)
Object DetectionSegmentationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: An unsupervised instance segmentation framework called Exemplar-FreeSOLO is proposed, which enhances the model's discriminative power by extracting example objects from unlabeled images as upstream information.
EXIF As Language: Learning Cross-Modal Associations Between Images and Camera Metadata
Chenhao Zheng (University of Michigan), Andrew Owens (University of Michigan)
ClassificationAnomaly DetectionRepresentation LearningTransformerContrastive LearningImageTextMultimodality
🎯 What it does: This paper learns a joint embedding representation that captures camera attributes by converting the camera's EXIF metadata into text and performing contrastive learning with image patches.
Explaining Image Classifiers With Multiscale Directional Image Representation
Stefan Kolek (Ludwig-Maximilians-Universität München), Ron Levie (Technion-Israel Institute of Technology)
Explainability and InterpretabilityConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A mask interpretation method based on wavelet and shearlet transforms, called ShearletX, is proposed for interpreting the decisions of image classifiers.
Explicit Boundary Guided Semi-Push-Pull Contrastive Learning for Supervised Anomaly Detection
Xincheng Yao (Shanghai Jiao Tong University), Chongyang Zhang (Shanghai Jiao Tong University)
Anomaly DetectionFlow-based ModelContrastive LearningImageBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: In semi-supervised anomaly detection, the authors propose the BGAD model, which trains an anomaly detector by guiding semi-pull-and-push contrastive learning with a small number of known anomaly samples through explicit boundary separation.
Explicit Visual Prompting for Low-Level Structure Segmentations
Weihuang Liu (University of Macau), Xiaodong Cun (Tencent AI Lab)
SegmentationTransformerPrompt EngineeringImage
🎯 What it does: This paper proposes a unified low-level structure segmentation framework—Explicit Visual Prompting (EVP), which achieves segmentation for four types of tasks: forgery, shadows, defocus blur, and camouflaged targets, by applying adjustable parameter prompts to image embeddings and high-frequency components on a frozen Vision Transformer backbone.
Exploiting Completeness and Uncertainty of Pseudo Labels for Weakly Supervised Video Anomaly Detection
Chen Zhang (Institute of Information Engineering), Ming-Hsuan Yang (University of California)
Anomaly DetectionVideo
🎯 What it does: This paper proposes a two-stage weakly supervised video anomaly detection framework that enhances the quality of pseudo-labels by leveraging the completeness and uncertainty of pseudo-labels, thereby improving anomaly detection performance.
Exploiting Unlabelled Photos for Stronger Fine-Grained SBIR
Aneeshan Sain (University of Surrey), Yi-Zhe Song (University of Surrey)
RetrievalKnowledge DistillationTransformerContrastive LearningImage
🎯 What it does: A strong FG-SBIR baseline based on PVT is proposed, significantly improving retrieval accuracy through internal triplet loss, EMA training stabilization, and knowledge distillation from unlabeled photos.
Exploring and Exploiting Uncertainty for Incomplete Multi-View Classification
Mengyao Xie (Tianjin University), Qinghua Hu (Tianjin University)
ClassificationMultimodality
🎯 What it does: A framework for incomplete multi-view classification based on uncertainty modeling, UIMC, is proposed, which achieves robust classification through multiple sampling and evidence reasoning.
Exploring and Utilizing Pattern Imbalance
Shibin Mei (Shanghai Jiao Tong University), Bingbing Ni (Shanghai Jiao Tong University)
Domain AdaptationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes identifying pattern imbalance in datasets and enhancing domain generalization performance through dynamic distribution training.
Exploring Data Geometry for Continual Learning
Zhi Gao (Beijing Institute of Technology), Yuwei Wu (Beijing Institute of Technology)
ClassificationData-Centric LearningImage
🎯 What it does: This paper explores and utilizes the geometric structure of non-Euclidean mixed curvature spaces in continual learning, dynamically expanding the underlying space and maintaining the geometric relationships of old data through angle regularization and neighbor robustness loss, thereby alleviating catastrophic forgetting.
Exploring Discontinuity for Video Frame Interpolation
Sangjin Lee (Yonsei University), Sangyoun Lee (Korea Institute of Science and Technology)
Image TranslationRestorationConvolutional Neural NetworkOptical FlowVideoTextBenchmark
🎯 What it does: In the video frame interpolation task, this paper proposes a data augmentation strategy called Figure-Text Mixing (FTM) and a continuous/non-continuous motion separation module D-map, further enhancing the model's robustness through the supervised loss L_D.
Exploring Incompatible Knowledge Transfer in Few-Shot Image Generation
Yunqing Zhao, Ngai-Man Cheung
GenerationDomain AdaptationGenerative Adversarial NetworkImage
🎯 What it does: This paper addresses the task of generating images from very few samples and explores and resolves the issue of 'incompatible knowledge transfer' that arises during the transfer process.
Exploring Intra-Class Variation Factors With Learnable Cluster Prompts for Semi-Supervised Image Synthesis
Yunfei Zhang (South China University of Technology), Hau San Wong (City University of Hong Kong)
GenerationData SynthesisPrompt EngineeringGenerative Adversarial NetworkImage
🎯 What it does: A learnable cluster prompt generation network (LCP-GAN) is proposed to model intra-class variation in semi-supervised conditional image synthesis through soft clustering and CLIP-guided learnable prompts.
Exploring Motion Ambiguity and Alignment for High-Quality Video Frame Interpolation
Kun Zhou (Chinese University of Hong Kong Shenzhen), Jiangbo Lu (SmartMore Corporation)
RestorationGenerationConvolutional Neural NetworkOptical FlowVideo
🎯 What it does: A texture consistency loss and a guided cross-scale pyramid alignment framework are proposed for high-quality video frame interpolation and extrapolation.
Exploring Structured Semantic Prior for Multi Label Recognition With Incomplete Labels
Zixuan Ding (Xidian University), Jungong Han (University of Sheffield)
ClassificationRecognitionGraph Neural NetworkContrastive LearningImage
🎯 What it does: For the task of multi-label recognition with missing labels, a Structured Semantic Prior Reasoning Network (SCPNet) and its self-supervised learning strategy are proposed to leverage the semantic relationships between labels in the CLIP model to enhance recognition performance.
Exploring the Effect of Primitives for Compositional Generalization in Vision-and-Language
Chuanhao Li (Beijing Institute of Technology), Yuwei Wu (Beijing Institute of Technology)
RecognitionSegmentationTransformerVision Language ModelContrastive LearningVideoMultimodality
🎯 What it does: This paper proposes a self-supervised learning framework by analyzing the semantic and labeling effects of raw units such as words, image regions, and video frames on visual and language tasks. It utilizes masked generation to create both invariant and variant samples, training the model to learn semantic invariance and variance, thereby enhancing combinatorial generalization ability.