CVPR 2026 Papers — Page 11
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 4071 papers
EG-3DVG: Expression and Geometry Aware Grounding Decoder for 3D Visual Grounding
GwangWook Park (Chungnam National University), Yeong Jun Koh (Chungnam National University)
Object DetectionSegmentationTransformerVision Language ModelContrastive LearningTextPoint Cloud
🎯 What it does: Proposes a unified 3D visual grounding framework named EG-3DVG, capable of simultaneously predicting 3D bounding boxes and semantic masks of target objects.
Ego-1K - A Large-Scale Multiview Video Dataset for Egocentric Vision
Jae Yong Lee (Meta Reality Labs), Jason Wither (Meta Reality Labs)
GenerationDepth EstimationGaussian SplattingVideoMultimodalityBenchmark
🎯 What it does: This paper proposes and releases the Ego-1K dataset, which contains 956 synchronized first-person multi-camera videos from 12+4 perspectives, focusing on hand-object interactions; it provides both original and distortion-corrected versions, accompanied by multi-modal data including hardware synchronization, online calibration, and IMU; additionally, a 4D novel view synthesis baseline based on stereo depth priors is built and evaluated on this dataset.
Ego-Grounding for Personalized Question-Answering in Egocentric Videos
Junbin Xiao (University of Science and Technology of China), Angela Yao (National University of Singapore)
TransformerLarge Language ModelPrompt EngineeringVision Language ModelVideoMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Proposed the MyEgo dataset and evaluated the self-localization capability of multimodal large language models in personalized perspective video question answering.
Ego-InBetween: Generating Object State Transitions in Ego-Centric Videos
Mengmeng Ge (Advanced Micro Devices, Inc.), Emad Barsoum (Advanced Micro Devices, Inc.)
GenerationData SynthesisSupervised Fine-TuningVision Language ModelVision-Language-Action ModelDiffusion modelVideo
🎯 What it does: Propose the Egocentric Instructed Visual State Transition (EIVST) task, generating a sequence of continuous intermediate frames from the initial state to the target state, and achieve first-person perspective action instruction-driven video generation.
Ego: Embedding-Guided Personalization of Vision-Language Models
Soroush Seifi (Toyota Motor Europe), Rahaf Aljundi (Toyota Motor Europe)
Recommendation SystemComputational EfficiencyRepresentation LearningPrompt EngineeringVision Language ModelImageVideoText
🎯 What it does: Propose a training-free, module-free personalization method called Ego, which leverages the internal attention mechanism of large vision-language models to extract visual memories;
Ego2Web: A Web Agent Benchmark Grounded in Egocentric Videos
Shoubin Yu (Google DeepMind), Boqing Gong (Google DeepMind)
TransformerLarge Language ModelVision-Language-Action ModelVideoTextMultimodalityBenchmark
🎯 What it does: Propose the Ego2Web benchmark, integrating first-person video perception with Web agent tasks, and design the Ego2WebJudge automated evaluation framework
EgoAVU: Egocentric Audio-Visual Understanding
Ashish Seth (Meta), Zhipeng Cai (Meta)
Data SynthesisRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningVision-Language-Action ModelVideoTextMultimodalityBenchmarkAudio
🎯 What it does: Built the EgoAVU data engine to automatically generate audio-visual-language data for egocentric videos, creating a large-scale training set EgoAVU‑Instruct (3M QA) and a validation set EgoAVU‑Bench (3K QA), and used these data to train and evaluate multimodal large language models (MLLMs).
Egocentric Visibility-Aware Human Pose Estimation
Peng Dai (ByteDance), Yang Zhang (ByteDance)
Pose EstimationTransformerAuto EncoderVideo
🎯 What it does: Proposed the Eva-3M real-world perspective visibility annotated dataset and extended the EMHI dataset based on it; designed the EvaPose visibility-aware 3D pose estimation framework;
EgoControl: Controllable Egocentric Video Generation via 3D Full-Body Poses
Enrico Pallotta (University of Bonn), Juergen Gall (University of Bonn)
GenerationTransformerDiffusion modelVideo
🎯 What it does: A first-person video generation model based on 3D full-body pose control;
EgoEdit: Dataset, Real-Time Streaming Model, and Benchmark for Egocentric Video Editing
Runjia Li (Snap Research), Willi Menapace (Snap Research)
GenerationTransformerLarge Language ModelVision Language ModelDiffusion modelRectified FlowVideoTextBenchmark
🎯 What it does: This paper proposes the EgoEdit ecosystem, which includes a manually constructed EgoEditData training set, the real-time inference EgoEdit-RT model, and a specialized benchmark for evaluating first-person video editing called EgoEditBench.
EgoFlow: Gradient-Guided Flow Matching for Egocentric 6DoF Object Motion Generation
Abhishek Saroha (Technische Universitaet Muenchen), Xi Wang (ETH Zuerich)
GenerationTransformerFlow-based ModelVideo
🎯 What it does: Propose EgoFlow, a flow-matching-based generative framework for generating 6DoF trajectories of objects from first-person videos, achieving physical plausibility during inference through gradient guidance;
EgoMind: Activating Spatial Cognition through Linguistic Reasoning in MLLMs
Zhenghao Chen (Beihang University), Di Huang (Beihang University)
TransformerSupervised Fine-TuningReinforcement LearningVision Language ModelMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Proposes the EgoMind framework, which utilizes Chain-of-Thought (CoT) to achieve multi-frame spatial cognition through pure 2D language reasoning.
EgoPoseFormer v2: Accurate Egocentric Human Motion Estimation for AR/VR
Zhenyu Li (Meta), Chenhongyi Yang (KAUST)
Pose EstimationTransformerImage
🎯 What it does: This paper proposes an end-to-end Transformer architecture, EgoPoseFormer v2, for estimating full-body 3D motion from multi-view images captured by head-mounted cameras, and introduces an automated annotation system to leverage massive unlabeled data to improve model performance.
EgoProx: Evaluating MLLMs on Egocentric 3D Proximity Reasoning Across a Cognitive Hierarchy
Jinzhao Li (Tsinghua University), Miao Liu (Tsinghua University)
TransformerSupervised Fine-TuningAgentic AIPrompt EngineeringVision Language ModelVideoMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Propose the EgoProx benchmark to evaluate the 3D approximate reasoning ability of multimodal large language models from a first-person perspective, and build an agent-based data generation engine;
EgoRoC: Towards Egocentric Robotic Control via Task-Agnostic Visual Alignment
Wei Feng (Tianjin University), Mingyan Li (Tianjin University)
Robotic IntelligenceTransformerLarge Language ModelVision-Language-Action ModelDiffusion modelContrastive LearningImageMultimodality
🎯 What it does: Proposed a pluggable Egocentric Alignment Head (EgoRoC), which first performs perspective alignment in Vision-Language-Action systems and then executes actions, reducing training redundancy and enhancing robustness across tasks and environments.
EgoSound: Benchmarking Sound Understanding in Egocentric Videos
Bingwen Zhu (Fudan University), Xiangyang Xue (Fudan University)
Large Language ModelVision Language ModelVideoMultimodalityBenchmarkAudio
🎯 What it does: Proposed the EgoSound benchmark for systematic evaluation of multimodal large language models (MLLMs) in understanding and reasoning about sounds from a first-person perspective.
EgoX: Egocentric Video Generation from a Single Exocentric Video
Taewoong Kang (KAIST AI), Jaegul Choo (KAIST AI)
GenerationData SynthesisSupervised Fine-TuningDiffusion modelVideoPoint Cloud
🎯 What it does: Propose a framework named EgoX that can generate corresponding first-person perspective videos (egocentric) using only a single external camera video (exocentric), achieving perspective conversion and scene consistency.
EgoXtreme: A Dataset for Robust Object Pose Estimation in Egocentric Views under Extreme Conditions
Taegyoon Yoon (Seoul National University), Hyung-Sin Kim (Seoul National University)
RestorationPose EstimationSimultaneous Localization and MappingVideoBenchmark
🎯 What it does: Introduce and release the EgoXtreme dataset, which collects 775.5 minutes of panoramic RGB videos from 15 participants using Aria glasses, covering three scenarios: industrial maintenance, sports, and emergency rescue, providing 6D pose annotations for 13 objects; meanwhile, benchmark the performance of existing RGB-only zero-shot pose estimation models (FoundPose, GigaPose, PicoPose) on this dataset.
EI-Part:Explode for Completion and Implode for Refinement
Wanhu Sun (Chinese University of Hong Kong), Xiaoguang Han (Chinese University of Hong Kong)
GenerationTransformerDiffusion modelFlow-based ModelAuto EncoderMesh
🎯 What it does: Proposed the EI-Part framework, which employs an Explode-Implode strategy to achieve 3D part-level generation, ensuring structural coherence, geometric合理性, fine details, and high efficiency.
Elastic Weight Consolidation Done Right for Continual Learning
Xuan Liu (Sun Yat-sen University), Xiaobin Chang (Sun Yat-sen University)
ClassificationConvolutional Neural NetworkTransformerImageMultimodality
🎯 What it does: Propose an operation called Logits Reversal (LR) to address the gradient vanishing issue in traditional Elastic Weight Consolidation (EWC) for weight importance estimation and the redundant constraints problem in Memory Aware Synapses (MAS), leading to the improved EWC-DR method.
Elastic3D: Controllable Stereo Video Conversion with Guided Latent Decoding
Nando Metzger (ETH Zurich), Federico Tombari (Google)
GenerationDiffusion modelAuto EncoderVideo
🎯 What it does: Propose Elastic3D, an end-to-end, warp-free monocular video-to-stereo video system that generates the right view in one go using a conditional latent diffusion model and achieves high-quality, controllable stereo videos through guided VAE decoding.
ElasticFormer: Detecting Objects in HRW Shots via Elastic Computing Vision Transformer
Wenxi Li (East China Normal University), Yuchen Guo (Tsinghua University)
Object DetectionTransformerImage
🎯 What it does: Propose ElasticFormer, a sparse visual Transformer that elastically allocates computational resources for object detection in high-resolution wide-angle images.
Electromagnetic Inverse Scattering from a Single Transmitter
Yizhe Cheng (Peking University), Yizhou Wang (Peking University)
RestorationImageMeshPhysics Related
🎯 What it does: Propose a fully end-to-end data-driven framework that uses a multi-layer perceptron (MLP) to directly predict the relative permittivity from the scattering field of one or multiple transmitters, achieving high-quality reconstruction for single-transmitter inverse scattering problems.
ELiC: Efficient LiDAR Geometry Compression via Cross-Bit-depth Feature Propagation and Bag-of-Encoders
Junsik Kim (Electronics and Telecommunications Research Institute), Soowoong Kim (Electronics and Telecommunications Research Institute)
CompressionConvolutional Neural NetworkMixture of ExpertsPoint Cloud
🎯 What it does: Propose ELiC, a real-time LiDAR geometry compression framework capable of achieving throughput above 10 FPS under 12-bit sampling.
Eliciting Complex Spatial Reasoning in MLLMs through Wide-Baseline Matching
Hao Zhong (State Key Laboratory of CAD & CG, Zhejiang University), Chunhua Shen (State Key Laboratory of CAD & CG, Zhejiang University)
Data SynthesisTransformerReinforcement LearningVision Language ModelImageVideoPoint CloudBenchmark
🎯 What it does: Proposes a wide-baseline matching task and evaluation benchmark ReasonMatch-Bench for multimodal large language models, constructs an scalable data generation pipeline and a Dynamic Correspondence Reinforcement Learning (DCRL) framework based on reinforcement learning to enhance the model's cross-perspective spatial reasoning capabilities.
Eliminate Distance Differences Induced by Backdoor Attacks: Layer-Selective Training and Clipping to Mask Backdoor Models
Xuzeng Li (Beijing Jiaotong University), Dong In Kim (Beijing Jiaotong University)
Federated LearningSafty and PrivacyAdversarial AttackImage
🎯 What it does: Proposes a framework called LaySelFL that achieves stealthy backdoor attacks in federated learning through layer selection and pruning.
ELITE: Efficient Gaussian Head Avatar from a Monocular Video via Learned Initialization and Test-time Generative Adaptation
Kim Youwang (POSTECH), Tae-Hyun Oh (KAIST)
GenerationConvolutional Neural NetworkDiffusion modelNeural Radiance FieldGaussian SplattingVideoMesh
🎯 What it does: This paper proposes ELITE, which utilizes the Mesh2Gaussian Prior Model to achieve rapid, high-quality, high-fidelity, and highly animated head Gaussian avatar initialization from monocular videos. During testing, it accomplishes detail completion and identity preservation through single-step diffusion enhancement and generation adaptation, solving the challenges of missing views and expressions in monocular videos.
Elucidating the Design Space of Arbitrary-Noise-Based Diffusion Models
Xingyu Qiu (Harbin Institute of Technology), Shuo Li (Harbin Institute of Technology)
RestorationDiffusion modelImageBiomedical DataMagnetic Resonance ImagingComputed TomographyStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: Propose the EDA framework, which extends the EDM general design space to arbitrary noise patterns, supports various noise covariances, and achieves image restoration without additional computational overhead while maintaining modularity;
Elucidating the SNR-t Bias of Diffusion Probabilistic Models
Meng Yu (Lanzhou University), Kun Zhan (AMAP Alibaba Group)
GenerationExplainability and InterpretabilityComputational EfficiencyDiffusion modelImage
🎯 What it does: The paper proposes a differential correction method targeting the SNR-t bias that occurs in diffusion models during the inference phase, utilizing wavelet domain frequency correction to enhance generation quality.
ELV-Halluc: Benchmarking Semantic Aggregation Hallucinations in Video Understanding
Hao Lu (SenseTime Research), Lewei Lu (SenseTime Research)
TransformerLarge Language ModelReinforcement LearningVision Language ModelVideoTextMultimodalityBenchmark
🎯 What it does: Proposed the ELV-Halluc benchmark to systematically evaluate semantic aggregation hallucination (SAH) in event-level videos, and designed adversarial question-answer pairs and the SAH Ratio metric based on this benchmark.
ELVIS: Enhance Low-Light for Video Instance Segmentation in the Dark
Joanne Lin (University of Bristol), Nantheera Anantrasirichai
SegmentationData SynthesisTransformerVideo
🎯 What it does: Designed an end-to-end low-light video instance segmentation framework ELVIS, including an unsupervised low-light synthesis pipeline, an unsupervised degradation estimation network VDP-Net, and an enhanced decoder, to improve the performance of existing VIS models on dark-light videos.
EMAD: Evidence-Centric Grounded Multimodal Diagnosis for Alzheimer's Disease
Qiuhui Chen (East China University of Science and Technology), Yi Hong (Shanghai Jiao Tong University)
ClassificationExplainability and InterpretabilityKnowledge DistillationData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningVision Language ModelMultimodalityTabularBiomedical DataMagnetic Resonance ImagingAlzheimer's Disease
🎯 What it does: Developed EMAD, an end-to-end multimodal vision-language framework capable of generating Alzheimer's disease diagnostic reports with three-layer localization (sentence-evidence-anatomy) based on 3D structural MRI and clinical data.
EmbodiedSplat: Online Feed-Forward Semantic 3DGS for Open-Vocabulary 3D Scene Understanding
Seungjun Lee (National University of Singapore), Gim Hee Lee (National University of Singapore)
SegmentationRepresentation LearningRobotic IntelligenceConvolutional Neural NetworkRecurrent Neural NetworkVision Language ModelGaussian SplattingTextPoint Cloud
🎯 What it does: Proposes EmbodiedSplat, an online, full-scenario 3D Gaussian Splatting semantic understanding framework for embodied tasks such as robotics, which can construct panoramic scenes and perform open-vocabulary semantic segmentation at near real-time speed (5-6 FPS) while cameras capture images in real-time;
EmbodMocap: In-the-Wild 4D Human-Scene Reconstruction for Embodied Agents
Wenjia Wang (University of Hong Kong), Taku Komura (University of Hong Kong)
Pose EstimationDepth EstimationOptimizationRobotic IntelligencePrompt EngineeringNeural Radiance FieldSimultaneous Localization and MappingImageVideo
🎯 What it does: Proposed a low-cost, portable 4D human-scene reconstruction system called EmbodMocap based on two mobile iPhones, capable of simultaneously capturing precise human motion and complete scene geometry in real environments;
Emergent Extreme-View Geometry in 3D Foundation Models
Yiwen Zhang (Cornell University), Hadar Averbuch-Elor (Cornell University)
Pose EstimationDepth EstimationTransformerSupervised Fine-TuningImageBenchmark
🎯 What it does: Investigated the geometric reasoning ability of 3D foundation models under extreme, non-overlapping perspectives, discovering that their shared backbone has formed an 'internal 3D language,' and proposed a lightweight alignment scheme that only updates a few biases to improve pose estimation, while contributing the MegaUnScene unseen scene benchmark.
Emergent Outlier View Rejection in Visual Geometry Grounded Transformers
Jisang Han (KAIST AI), Chen Feng (New York University)
Pose EstimationDepth EstimationAnomaly DetectionTransformerImage
🎯 What it does: Without additional training, perspective filtering of the input image set is achieved by leveraging the attention mechanism and feature similarity within VGGT, thereby discarding noisy perspectives in unsupervised feed-forward 3D reconstruction.
EMGauss: Continuous Slice-to-3D Reconstruction via Dynamic Gaussian Modeling in Volume Electron Microscopy
Yumeng He (Shanghai Jiao Tong University), Xiaokang Yang (Shanghai Jiao Tong University)
Gaussian SplattingBiomedical Data
🎯 What it does: Reconstruct 2D slices into a continuous 3D volume via dynamic Gaussian splatting, achieving anisotropy compensation and arbitrary depth slice synthesis; during training, use teacher-student pseudo-label self-supervision to enhance reconstruction quality of sparse slices.
EMMA: Concept Erasure Benchmark with Comprehensive Semantic Metrics and Diverse Categories
Lu Wei (University of Osaka), Noa Garcia (University of Osaka)
GenerationPrompt EngineeringVision Language ModelDiffusion modelImageTextBenchmark
🎯 What it does: Proposed the EMMA benchmark for systematic evaluation of concept erasure methods in text-to-image models.
EMMA: Extracting Multiple physical parameters from Multimodal Data
Farhat Shaikh (Arizona State University), Sandeep Gupta (Arizona State University)
Convolutional Neural NetworkRecurrent Neural NetworkVideoMultimodalityTabularPhysics RelatedOrdinary Differential EquationAudio
🎯 What it does: EMMA is a physics-constrained multi-modal framework that directly recovers multiple identifiable dynamic parameters of a system from raw data, including explicit, implicit, and calibration-invariant dynamics, using video, audio, and charts.
EMO-R3: Reflective Reinforcement Learning for Emotional Reasoning in Multimodal Large Language Models
Yiyang Fang (Wuhan University), Mang Ye (Wuhan University)
Explainability and InterpretabilityTransformerLarge Language ModelReinforcement LearningVision Language ModelMultimodality
🎯 What it does: Proposed and implemented the EMO-R3 framework, leveraging structured emotional thinking and reflective rewards to enhance the emotional reasoning capabilities of multimodal large language models.
EmoDiffTalk: Emotion-aware Diffusion for Editable 3D Gaussian Talking Head
Chang Liu (Beijing Normal University), Shi-Sheng Huang (Beijing Normal University)
GenerationTransformerDiffusion modelGaussian SplattingTextPoint CloudAudio
🎯 What it does: Propose EmoDiffTalk, an editable emotion-driven 3D talking-head generation framework based on 3D Gaussian Splatting, supporting audio-driven and text-based emotional editing.
EmoStyle: Emotion-Driven Image Stylization
Jingyuan Yang (Shenzhen University), Hui Huang (Shenzhen University)
Image TranslationGenerationTransformerPrompt EngineeringVision Language ModelDiffusion modelAuto EncoderImageTextMultimodality
🎯 What it does: Proposed and implemented the Affective Image Stylization (AIS) framework EmoStyle, which performs emotion-driven artistic stylization on user-input images using emotional words as the sole prompt while maintaining content consistency.
EmoTaG: Emotion-Aware Talking Head Synthesis on Gaussian Splatting with Few-Shot Personalization
Haolan Xu (Michigan State University), Xiaoming Liu (University of North Carolina at Chapel Hill)
GenerationData SynthesisConvolutional Neural NetworkTransformerGaussian SplattingVideoMeshAudio
🎯 What it does: Propose EmoTaG, a few-shot emotion-driven 3D talking head synthesis framework based on FLAME-Gaussian structural prior, capable of rapidly personalizing and generating emotionally rich, synchronized 3D avatars from only 5 seconds of video.
EmoThinker: Advancing Visual-Acoustic Emotion Analysis via Structural Token Selection and Chain-of-Thought Reasoning
Qinfu Xu (Beijing Institute Of Technology), Tianyu Liu (Beijing Institute Of Technology)
RecognitionLarge Language ModelSupervised Fine-TuningVision Language ModelImageMultimodalityBenchmarkChain-of-ThoughtAudio
🎯 What it does: Propose the EmoThinker framework, achieving multimodal sentiment analysis through structured token selection and audio evidence extraction, and construct the CoET dataset to enable explicit reasoning.
EMR-Diff: Edge-aware Multimodal Residual Diffusion Model for Hyperspectral Image Super-resolution
Tao Zhang (Hangzhou Dianzi University), Chenggang Yan (Hangzhou Dianzi University)
RestorationSuper ResolutionConvolutional Neural NetworkDiffusion modelImageMultimodality
🎯 What it does: Proposed the Edge-aware Multimodal Residual Diffusion Model (EMR-Diff) to fuse low-resolution HSI with high-resolution MSI, generating high-resolution HSI.
Enabling Supervised Learning of Generative Signatures for Generalized AI-Generated Images Detection
Jianwei Fei (University of Florence), Alessandro Piva (University of Florence)
Anomaly DetectionConvolutional Neural NetworkImage
🎯 What it does: Propose a supervised learning framework based on simulating generated traces, which can extract a generic signature (GenSign) from real images for AI-generated image detection.
ENC-Bench: A Benchmark for Evaluating Multimodal Large Language Models in Electronic Navigational Chart Understanding
Ao Cheng (National University of Defense Technology), Qingyong Hu (Intelligent Game and Decision Lab)
TransformerLarge Language ModelVision Language ModelImageMultimodalityBenchmark
🎯 What it does: Propose the ENC-Bench benchmark to evaluate the capability of multimodal large language models in understanding electronic nautical charts (ENC).
End-to-End Hyper-Relational Information Extraction for Engineering Diagrams via Dynamically Tokenized Relation Transformer
Tianyou Bai (Institute of Automation, Chinese Academy of Sciences), Cheng-Lin Liu (Institute of Automation, Chinese Academy of Sciences)
Object DetectionTransformerContrastive LearningImageGraph
🎯 What it does: Proposed an end-to-end engineering diagram information extraction framework called DTRT, which can simultaneously detect symbols, line segments, text, and directly generate hyper-relational knowledge graphs.
End-to-End Language-Action Model for Humanoid Whole Body Control
Yuxuan Wang (Peking University), Zongqing Lu (Peking University)
Robotic IntelligenceTransformerReinforcement LearningVision-Language-Action ModelFlow-based ModelTextMultimodality
🎯 What it does: Built an end-to-end language-action framework called SENTINEL, directly mapping natural language instructions and robot body perception to low-level control actions, achieving whole-body control without intermediate motion representations.
Endless World: Real-Time 3D-Aware Long Video Generation
Ke Zhang (Johns Hopkins University), Vishal M. Patel (Johns Hopkins University)
GenerationTransformerDiffusion modelVideo
🎯 What it does: Proposed a real-time, infinite-length, 3D-consistent video generation framework called Endless World
Energy Waveify and Redistribution for Test-Time Adaptation: A Control System Perspective
Zhenbin Wang (Sichuan University), Wei Huang (Sichuan University)
Domain AdaptationComputational EfficiencyImage
🎯 What it does: Propose a test-time energy adaptive framework APT without sampling or source data, treating energy as complex waves and redistributing energy using wave equations.
Energy-GS: Image Energy-guided Pose Alignment Gaussian Splatting with redesigned pose gradient flow
Yu Gao (Beijing Institute of Technology), Yi Yang (Beijing Institute of Technology)
Pose EstimationOptimizationGaussian SplattingImage
🎯 What it does: Achieved joint optimization of 3D Gaussian Splatting and camera pose under the condition of using only RGB images; by redesigning the camera gradient flow and introducing an image-based energy progressive alignment strategy, stable coarse-to-fine pose correction and scene reconstruction were realized.
EnergyAction: Unimanual to Bimanual Composition with Energy-Based Models
Mingchen Song (Harbin Institute Of Technology (Shenzhen)), Weili Guan (Harbin Institute Of Technology (Shenzhen))
OptimizationRobotic IntelligenceReinforcement LearningDiffusion modelFlow-based ModelImageText
🎯 What it does: Integrate a pre-trained single-arm manipulation strategy into dual-arm tasks using an energy-based model (EBM) to achieve dual-arm collaborative manipulation.
Enhance-then-Balance Modality Collaboration for Robust Multimodal Sentiment Analysis
Kang He (Wuhan University), Donghong Ji (Wuhan University)
ClassificationKnowledge DistillationTransformerContrastive LearningMultimodality
🎯 What it does: This paper proposes a two-stage enhanced-balanced multimodal collaboration framework (EBMC), which first performs semantic decoupling of modalities and weak modality compensation, and then achieves robustness and balance in multimodal sentiment analysis through energy-guided modality coordination and instance-aware trust distillation.
Enhancing Accuracy of Uncertainty Estimation in Appearance-based Gaze Tracking with Probabilistic Evaluation and Calibration
Qiaojie Zheng (Colorado School of Mines), Xiaoli Zhang (Colorado School of Mines)
Pose EstimationImage
🎯 What it does: Propose a post-hoc calibration method to correct error uncertainty estimation in visual pupil tracking caused by domain shift;
Enhancing Continual Learning of Vision-Language Models via Dynamic Prefix Weighting
Hyeonseo Jang (Yonsei University), Kibok Lee (Yonsei University)
Computational EfficiencyRepresentation LearningMeta LearningTransformerVision Language ModelMultimodalityBenchmark
🎯 What it does: For the continual learning scenario of visual-language models (VLM), a dynamic prefix weighting (DPW) framework is proposed, which can adaptively assign weights to prefixes and adapters on each input token, thereby achieving efficient task adaptation while preserving pre-trained knowledge.
Enhancing Descriptive Captions with Visual Attributes for Multimodal Perception
Yanpeng Sun, Jingdong Wang
ClassificationRecognitionObject DetectionDepth EstimationLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: Designed Cap-Workflow, an end-to-end pipeline that automatically generates image descriptions rich in fine-grained attributes and relationships by leveraging multiple visual experts and large language models.
Enhancing Hands in 3D Whole-Body Pose Estimation with Conditional Hands Modulator
Gyeongsik Moon (Korea University)
Pose EstimationTransformerImageMesh
🎯 What it does: Propose the Hand4Whole++ framework, which integrates pre-trained full-body and hand pose estimators. It improves wrist orientation and full-body consistency by applying a lightweight CHAM module to conditionally modulate full-body features without retraining the full-body model, while directly fusing hand details and hand shapes from the hand estimator into the SMPL-X mesh via differentiable rigid alignment.
Enhancing Mixture-of-Experts Specialization via Cluster-Aware Upcycling
Sanghyeok Chu (Seoul National University), Bohyung Han (Seoul National University)
ClassificationRetrievalKnowledge DistillationTransformerMixture of ExpertsVision Language ModelImageMultimodality
🎯 What it does: Proposes a cluster-aware hybrid expert initialization strategy (Cluster-aware Upcycling), which initializes experts and routers by partitioning the activation space of dense models into semantic clusters, and introduces expert ensemble self-distillation (EESD) to enhance training stability.
Enhancing Out-of-Distribution Detection with Extended Logit Normalization
Yifan Ding (Linköping University), Gabriel Eilertsen (Linköping University)
Anomaly DetectionConvolutional Neural NetworkImage
🎯 What it does: Proposed a new training objective ELogitNorm to address the feature collapse problem caused by LogitNorm, and improved OOD detection and ID confidence calibration.
Enhancing Part-Level Point Grounding for Any Open-Source MLLMs
Jin-Cheng Jhang (National Tsing Hua University), Cheng-Hao Kuo (Amazon)
Object DetectionTransformerSupervised Fine-TuningVision Language ModelMultimodality
🎯 What it does: Provide accurate part-level point localization capability for any open-source multimodal large language models (MLLMs) while preserving their original pre-trained functions
Enhancing Spatial Understanding in Image Generation via Reward Modeling
Zhenyu Tang (Peking University), Daquan Zhou (Peking University)
GenerationReinforcement Learning from Human FeedbackReinforcement LearningDiffusion modelImageTextMultimodalityStochastic Differential Equation
🎯 What it does: Construct an 80K adversarial spatial preference dataset and train a specialized reward model called SpatialScore. Use this reward model to perform online reinforcement learning on FLUX.1-dev, significantly enhancing the spatial relationship understanding of text-to-image generation models.
Enhancing the Security of Visual Speaker Authentication Based on Dynamic Lip-Print Analysis
Yi He (Shanghai Jiao Tong University), Shilin Wang (Shanghai Jiao Tong University)
RecognitionConvolutional Neural NetworkVideo
🎯 What it does: Propose a visual speaker authentication framework based on dynamic lip features to enhance security against Replay and DeepFake attacks
Enhancing Unregistered Hyperspectral Image Super-Resolution via Unmixing-based Abundance Fusion Learning
Yingkai Zhang (Beijing Institute of Technology), Ying Fu (Hangzhou Dianzi University)
Super ResolutionConvolutional Neural NetworkImage
🎯 What it does: Propose a super-resolution framework for unregistered hyperspectral images based on spectral decomposition, utilizing spectral unmixing of low-resolution HSI and unregistered high-resolution RGB reference images to achieve the fusion of spatial-spectral features.
Enhancing Video Vision Language Model with Hippocampal Sensing
Xu Cao (PediaMed AI)
TransformerReinforcement LearningVision Language ModelContrastive LearningVideoTextChain-of-ThoughtAudio
🎯 What it does: Designed and trained a video vision-language model named HippoVLM, leveraging cross-modal prediction and reinforcement learning to achieve active information selection and joint reasoning.
Enhancing Visual Representation with Textual Semantics: Textual Semantics-Powered Prototypes for Heterogeneous Federated Learning
Xinghao Wu (Beihang University), Wei Chen (Beihang University)
Federated LearningSafty and PrivacyRepresentation LearningTransformerLarge Language ModelPrompt EngineeringContrastive LearningMultimodality
🎯 What it does: This paper proposes the FedTSP method, which generates fine-grained class descriptions using large language models and constructs semantically rich global prototypes through pre-trained text models. It combines learnable prompts to achieve cross-modal alignment, thereby improving prototype quality and model performance in heterogeneous federated learning.
Envision, Attend, Then Respond: Counterfactual Hallucination Mitigation in Large Vision-Language Models
Yuxuan Liang (Fudan University), Xiangyang Xue (Fudan University)
Anomaly DetectionVision Language ModelDiffusion modelContrastive LearningImageText
🎯 What it does: Propose the Envision-Attend-Respond (EnAR) framework, which utilizes diffusion models to generate visual impressions and guides LVLM to focus on counterfactual elements in images, thereby suppressing counterfactual hallucinations in vision-language models without additional training.
Envisioning the Future, One Step at a Time
Stefan Andreas Baumann (LMU Munich), Björn Ommer (LMU Munich)
GenerationTransformerDiffusion modelFlow-based ModelVideoBenchmark
🎯 What it does: This paper proposes an autoregressive diffusion model based on a single image, capable of rapidly generating diverse and physically consistent future scenes through step-by-step reasoning of sparse point trajectories.
EpiAgent: An Agent-Centric System for Ancient Inscription Restoration
Shipeng Zhu (Southeast University), Hui Xue (Southeast University)
RestorationLarge Language ModelAgentic AIVision Language ModelDiffusion modelImageTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: Designed and implemented EpiAgent, an agent-centric epitaph restoration system that employs an LLM-planned Observe-Conceive-Execute-Reevaluate cycle, achieving multi-modal analysis, experience-driven tool selection, composable specialized restoration tools, and multi-perspective evaluation to recover damaged inscriptions from both visual and textual perspectives.
Erasing Thousands of Concepts: Towards Scalable and Practical Concept Erasure for Text-to-Image Diffusion Models
Hoigi Seo (Seoul National University), Se Young Chun (Seoul National University)
GenerationMixture of ExpertsDiffusion modelImageText
🎯 What it does: This paper proposes an expandable concept elimination framework, ETC, which can eliminate over 2000 concepts in a single pass within text-to-image diffusion models while achieving precise removal while maintaining image quality.
EReCu: Pseudo-label Evolution Fusion and Refinement with Multi-Cue Learning for Unsupervised Camouflage Detection
Shuo Jiang (Hangzhou Dianzi University), Gang Pan (Zhejiang University)
Object DetectionTransformerImageBenchmark
🎯 What it does: Proposes an unsupervised camouflaged object detection framework named EReCu based on teacher-student self-evolving pseudo-label fusion and multi-clue local perception.
eRetinexGS: Retinex Modeling for Low-Light Scene Enhancement via Event Streams and 3D Gaussian Splatting
Haojie Yan (State Key Lab of Brain-Machine Intelligence Zhejiang University), Gang Pan (State Key Lab of Brain-Machine Intelligence Zhejiang University)
RestorationGaussian SplattingImageMultimodality
🎯 What it does: Fusing event camera and low-light frames, using 3D Gaussian Splatting (3DGS) to achieve Retinex decomposition and enhancement in low-light scenes, resulting in color-consistent and detail-rich normally illuminated views.
ERMoE: Eigen-Reparameterized Mixture-of-Experts for Stable Routing and Interpretable Specialization
Anzhe Cheng (University of Southern California), Paul Bogdan (University of Southern California)
ClassificationRetrievalExplainability and InterpretabilityTransformerMixture of ExpertsImageBiomedical DataMagnetic Resonance ImagingAlzheimer's Disease
🎯 What it does: Propose ERMoE by reparameterizing expert weights on an orthogonal feature basis and using Eigenbasis Score for routing based on feature-basis vector alignment, eliminating auxiliary load balancing loss to enhance MoE's stability and interpretability.
ESAM++: Efficient Online 3D Perception on the Edge
Qin Liu (Stanford University), Andrea Colaco (Google)
SegmentationComputational EfficiencyConvolutional Neural NetworkTransformerPoint Cloud
🎯 What it does: Proposes ESAM++, a lightweight online 3D scene perception framework for edge devices, replacing the previously time-consuming 3D sparse UNet with a 3D sparse feature pyramid network (SFPN) to achieve multi-scale point cloud feature extraction.
EthoCLIP: Ontology-Enhanced Video-Language Pretraining for Animal Behavior Understanding
Yinuo Jing (Beijing University of Posts and Telecommunications), Zhanyu Ma (Beijing University of Posts and Telecommunications)
ClassificationGraph Neural NetworkTransformerLarge Language ModelVision Language ModelContrastive LearningVideoText
🎯 What it does: This paper first unifies five animal behavior datasets under the Neuro Behavior Ontology (NBO) standard, constructing the AnimalBand dataset containing 74,671 videos; subsequently, it proposes EthoCLIP, a model that integrates ontology semantics and graph attention within a visual-language contrastive learning framework.
Eulerian Gaussian Splatting using Hashed Probability Pyramids
Mia Gaia Polansky (Harvard University), Dor Verbin (Google DeepMind)
GenerationGaussian SplattingImage
🎯 What it does: Proposed a probabilistic distribution-based Eulerian Gaussian splatting method (EGS), which dynamically creates and deletes primitives during training without manual rules by sampling Gaussian primitives using learnable distributions for scene geometry.
EV-CGNet: Co-visible Focused 3D-guided 2D Event Keypoint Detection Network
Yuan Gao (University of Science and Technology of China), Tianzhu Zhang (University of Science and Technology of China)
Pose EstimationTransformerContrastive LearningBenchmark
🎯 What it does: Proposes EV-CGNet, an event keypoint detection network that utilizes 3D-guided 2D feature prototype learning and co-visible region focused detection/descriptor learning.
EVA: Efficient Reinforcement Learning for End-to-End Video Agent
Yaolun Zhang (SenseTime Research), Lewei Lu (SenseTime Research)
Computational EfficiencyRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningAgentic AIVision-Language-Action ModelVideoMultimodality
🎯 What it does: Proposed an end-to-end video agent EVA based on reinforcement learning, adopting a loop reasoning framework that first plans and then perceives.
Evaluating Generative Models via One-Dimensional Code Distributions
Zexi Jia (WeChat AI, Tencent Inc.), Jie Zhou (WeChat AI, Tencent Inc.)
GenerationTransformerImage
🎯 What it does: Propose a generative model evaluation framework based on discrete visual codes, and introduce two unsupervised metrics: distribution consistency metric CHD and no-reference quality assessment CMMS.
EVATok: Adaptive Length Video Tokenization for Efficient Visual Autoregressive Generation
Tianwei Xiong (University Of Hong Kong), Xihui Liu (University Of Hong Kong)
GenerationTransformerReinforcement LearningVideo
🎯 What it does: Proposes EVATok, a content-adaptive video tokenization framework that achieves efficient variable-length video tokenization through four stages (proxy tokenizer, proxy reward search, router training, final adaptive tokenizer) for autoregressive video generation.
Event Stream Filtering via Probability Flux Estimation
Jinze Chen (University of Science and Technology of China), Zheng-Jun Zha (University of Science and Technology of China)
RestorationTime SeriesSequentialBenchmark
🎯 What it does: Proposed an event filtering framework called EDFilter based on probabilistic flux, which can real-time denoise and reconstruct continuous radiation changes.
Event Structural Valley: A Unified Theoretical and Practical Framework for Event Camera Autofocus
Xijie Xiang (Peking University), Yonghong Tian (Peking University)
Time SeriesSequential
🎯 What it does: This paper studies the self-focusing problem of event cameras in dynamic environments and proposes a self-focusing method based on the dual-peak-valley structure theory of event rate curves.
Event-Based Motion Deblurring Using Task-Oriented 3D Gaussian Event Representations
Shengdong Xue (Beijing University of Technology), Yongjian Deng (Nankai University)
RestorationComputational EfficiencyConvolutional Neural NetworkGaussian Splatting
🎯 What it does: Proposed a learnable 3D Gaussian event representation module combined with a two-stage attention fusion network to achieve event-driven motion deblurring
Event-based Motion Deblurring with Unpaired Data
Hoonhee Cho, Kuk-Jin Yoon
RestorationConvolutional Neural NetworkGaussian SplattingImageMultimodality
🎯 What it does: To address the event stream and blurry images captured by event cameras, this paper designs a learnable 3D Gaussian event representation module and constructs a two-stage fusion network to achieve motion deblurring.
Event-based Visual Deformation Measurement
Yuliang Wu (University of Science and Technology of China), Zheng-Jun Zha (University of Science and Technology of China)
OptimizationComputational EfficiencyOptical FlowImageVideoMultimodalityBenchmark
🎯 What it does: This paper proposes an event-frame fusion visual deformation measurement (Event-based Deformation Measurement, EVDM) framework, leveraging the high temporal density motion cues provided by event cameras and the high spatial resolution of traditional frame images to achieve dense deformation field regression for large-scale and rapid deformations.
Event-Illumination Collaborative Low-light Image Enhancement with a High-resolution Real-world Dataset
Senyan Xu (University of Science and Technology of China), Zheng-Jun Zha (University of Science and Technology of China)
RestorationConvolutional Neural NetworkTransformerImageBenchmark
🎯 What it does: Proposed an event-illumination collaborative low-light image enhancement framework EIC-LIE, which utilizes event and image information jointly to improve the quality of low-light images.
Event6D: Event-based Novel Object 6D Pose Tracking
Jae-Young Kang (KAIST), Kuk-Jin Yoon (KAIST)
Object TrackingPose EstimationConvolutional Neural NetworkRecurrent Neural NetworkOptical FlowImageSequential
🎯 What it does: This paper proposes EventTrack6D, a method for 6D object pose tracking using event cameras, capable of tracking novel objects at extremely high frame rates without requiring per-object specific training.
EventDrive: Event Cameras for Vision-Language Driving Intelligence
Dongyue Lu (National University of Singapore), Wei Tsang Ooi (National University of Singapore)
Data SynthesisAutonomous DrivingLarge Language ModelMixture of ExpertsVision Language ModelMultimodalityBenchmark
🎯 What it does: Proposes EventDrive—a full-stack driving dataset and benchmark integrating event cameras, RGB frames, and language supervision, covering four levels: perception, understanding, prediction, and planning; as well as EventDrive-VLM, an end-to-end driving inference framework that incorporates asynchronous event encoding, dynamic time-scale selection, and multi-expert gating into large vision-language models.
EventGait: Towards Robust Gait Recognition with Event Streams
Senyan Xu (University Of Science And Technology Of China), Xueyang Fu (University Of Electronic Science And Technology Of China)
RecognitionConvolutional Neural NetworkSpiking Neural NetworkMixture of ExpertsTime Series
🎯 What it does: Propose a dual-stream network called EventGait for non-invasive gait-based identity recognition using event cameras;
EventHub: Data Factory for Generalizable Event-Based Stereo Networks without Active Sensors
Luca Bartolomei, Guillermo Gallego
GenerationDepth EstimationDomain AdaptationKnowledge DistillationData-Centric LearningTransformerNeural Radiance FieldImageMultimodality
🎯 What it does: Built a data factory named EventHub, which generates proxy depth labels and proxy events through neural rendering and cross-modal distillation using RGB images, enabling the training of an event stereo matching network without LiDAR supervision.
Every Error has Its Magnitude: Asymmetric Mistake Severity Training for Multiclass Multiple Instance Learning
Sungrae Hong (Korea Advanced Institute of Science and Technology), Mun Yong Yi (Korea Advanced Institute of Science and Technology)
ClassificationImageBiomedical Data
🎯 What it does: Proposed an error severity training strategy for multi-class multi-instance learning aimed at improving classification errors in whole slide image (WSI) diagnosis, particularly clinically significant errors.
Evidential Deep Partial Label Learning to Quantify Disambiguation Uncertainty
Jinfu Fan (Qingdao University), Linqing Huang (Qingdao University)
ClassificationExplainability and InterpretabilityImage
🎯 What it does: Propose the ED-PLL framework, which quantifies uncertainty through evidence learning in partial label learning, achieving candidate label disambiguation and reliability prediction.
Evidential Neural Radiance Fields
Ruxiao Duan (Yale University), Alex Wong (Yale University)
GenerationNeural Radiance FieldImageBenchmark
🎯 What it does: Propose Evidential Neural Radiance Fields (Evidential NeRF), enabling simultaneous estimation of model uncertainty and data uncertainty with a single forward pass;
Evidential Transformation Network: Turning Pretrained Models into Evidential Models for Post-hoc Uncertainty Estimation
Yongchan Chun (Korea University), Heuiseok Lim (Korea University)
Explainability and InterpretabilityComputational EfficiencyImageTextBenchmark
🎯 What it does: Propose a lightweight post-processing module called Evidential Transformation Network (ETN), which converts any pre-trained model into a reliable evidence deep learning model by learning sample-dependent affine transformations in the logit space.
EVLF: Early Vision-Language Fusion for Generative Dataset Distillation
Wenqi Cai (University of Toyama), Chao Zhang (University of Toyama)
Data SynthesisKnowledge DistillationTransformerDiffusion modelContrastive LearningImageText
🎯 What it does: Propose an early vision-language fusion (EVLF) method, using lightweight cross-attention between the output of the image encoder and the decoder of the diffusion model to integrate semantic information with visual features early, addressing the problem of visual detail loss caused by traditional late fusion.
Evo-1: Lightweight Vision-Language-Action Model with Preserved Semantic Alignment
Tao Lin (Shanghai Jiao Tong University), Bo Zhao (Shanghai Jiao Tong University)
Robotic IntelligenceTransformerSupervised Fine-TuningVision-Language-Action ModelDiffusion modelFlow-based ModelTextMultimodalityBenchmark
🎯 What it does: Designed and implemented a lightweight Vision-Language-Action model, Evo-1, which maintains VLM semantic alignment through two-stage training, combines cross-modal diffusion transformers and integration modules, achieving end-to-end multimodal perception and continuous action generation.
Evo-Retriever: LLM-Guided Curriculum Evolution with Viewpoint-Pathway Collaboration for Multimodal Document Retrieval
Weiqing Li (Alibaba Cloud Computing), Hao Henry Wang (Alibaba Cloud Computing)
RetrievalLarge Language ModelVision Language ModelContrastive LearningMultimodality
🎯 What it does: Propose a retrieval framework named Evo-Retriever, which addresses the issues of insufficient spatial awareness, textual ambiguity, and bottlenecks caused by static training schedules in complex visual document retrieval through co-evolution of the model and training plan.
EvObj: Learning Evolving Object-centric Representations for 3D Instance Segmentation without Scene Supervision
Jiahao Chen (Hong Kong Polytechnic University), Bo Yang (Hong Kong Polytechnic University)
SegmentationConvolutional Neural NetworkReinforcement LearningDiffusion modelAuto EncoderPoint Cloud
🎯 What it does: Proposes the EvObj framework to achieve unsupervised 3D instance segmentation, bridging the domain gap between synthetic and real point clouds through object candidate identification and evolution modules, as well as object completion modules.
EvoComp: Learning Visual Token Compression for Multimodal Large Language Models via Semantic-Guided Evolutionary Labeling
Jiafei Song (OPPO), Bailin Na (OPPO)
CompressionComputational EfficiencyTransformerVision Language ModelMultimodality
🎯 What it does: Proposes the EvoComp framework, which compresses visual tokens in multimodal large language models (MLLMs) using a lightweight encoder-only transformer, supervised by evolutionary label generation.
EvoGraph-R1: Self-Evolving Multimodal Knowledge Hypergraphs for Agentic Retrieval
Jiashi Lin (Northwestern Polytechnical University), Junjun He (Shanghai Artificial Intelligence Laboratory)
RetrievalGraph Neural NetworkTransformerLarge Language ModelReinforcement LearningAgentic AITextMultimodalityGraphBenchmarkRetrieval-Augmented Generation
🎯 What it does: This paper proposes the EvoGraph-R1 framework, treating a multimodal knowledge graph as a dynamic MDP environment, allowing agents to self-evolve the knowledge graph during reasoning to enhance retrieval and inference.