CVPR 2026 Papers — Page 10
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 4071 papers
DSO: Direct Steering Optimization for Bias Mitigation
Lucas Monteiro Paes (Apple), Nicholas Apostoloff (Apple)
OptimizationTransformerLarge Language ModelReinforcement LearningVision Language ModelMultimodality
🎯 What it does: Propose a Direct Activation Optimization (DSO) method that learns linear transformations via reinforcement learning during inference to modulate model internal activations, thereby reducing gender-occupation bias in vision-language models (VLMs) and large language models (LLMs).
DTG-Restore: Training-Free Diffusion Refinement for Generative Video Super-Resolution
Hidir Yesiltepe (Virginia Tech), Jinrong Xie (Adobe)
Super ResolutionTransformerDiffusion modelRectified FlowVideo
🎯 What it does: Proposed a training-free diffusion correction framework called DTG-Restore, which can perform super-resolution reconstruction and structural correction on AI-generated or damaged videos without additional training of the model.
Dual Ascent Diffusion for Inverse Problems
Minseo Kim (Stanford University), Gordon Wetzstein (Stanford University)
RestorationOptimizationDiffusion modelImage
🎯 What it does: Proposed the DDiff dual ascending-descending framework, which leverages pre-trained diffusion models to achieve data matching, denoising, and dual variable updates in MAP inverse problems;
Dual Band Thermal Videography: Separating Time-Varying Reflection and Emission Near Ambient Conditions
Sriram Narayanan (Carnegie Mellon University), Srinivasa Narasimhan
RestorationOptimizationVideoPhysics Related
🎯 What it does: Separate reflection and emission under near-environment conditions using dual-band thermal imaging to recover object temperature and reflection field.
Dual Graph Regularized Deep Unfolding Network for Guided Depth Map Super-resolution
Zhiwei Zhong (City University of Hong Kong), Shiqi Wang (City University of Hong Kong)
Super ResolutionOptimizationConvolutional Neural NetworkImage
🎯 What it does: Proposed a depth unfolding network called LapNet based on dual graph Laplacian prior for depth map super-resolution under color guidance.
Dual-Agent Reinforcement Learning for Adaptive and Cost-Aware Visual-Inertial Odometry
Feiyang Pan (Southeast University), Guangbin Dou (Southeast University)
OptimizationReinforcement LearningSimultaneous Localization and MappingMultimodality
🎯 What it does: Introduce a dual-agent reinforcement learning strategy in visual inertial odometry, intelligently deciding when to activate the costly visual frontend and how to adaptively fuse IMU predictions with visual measurements, thereby reducing the frequency of VIBA calls.
Dual-branch Distilled Transformer for Efficient Asymmetric UAV Tracking
Hongtao Yang (Guangxi Normal University), Shuxiang Song (Guangxi Normal University)
Object TrackingKnowledge DistillationRepresentation LearningTransformerVideo
🎯 What it does: Propose a teacher-guided dual-branch distillation strategy to enhance feature representation and localization accuracy in lightweight drone trackers.
Dual-Estimator: Decoupling Global and Local Semantic Shift for Drift Compensation in Class-Incremental Learning
Fankang Xu (Nanjing University of Science and Technology), Zechao Li (Nanjing University of Science and Technology)
ClassificationKnowledge DistillationRepresentation LearningMixture of ExpertsImageBenchmark
🎯 What it does: Propose Dual-Estimator (Dual-E) by estimating local and global semantic drift to achieve drift compensation in sample-free class incremental learning.
Dual-Granularity Memory for Efficient Video Generation
Hongjun Wang (Inclusion AI), Tao Lin (Inclusion AI)
GenerationComputational EfficiencyKnowledge DistillationVideo
🎯 What it does: Propose a dual-grained memory framework (Context Memory and LCaM) to enhance the cyclic video generator, addressing temporal discontinuity caused by block isolation.
Dual-level Adaptation for Multi-Object Tracking: Building Test-Time Calibration from Experience and Intuition
Wen Guo (Shandong Technology and Business University), Junyu Gao (Chinese Academy of Sciences)
Object TrackingTransformerVideoBenchmark
🎯 What it does: Propose a test-time calibration framework (TCEI) based on experience and intuition, enhancing identity prediction performance in multi-object tracking through two-level caches of short-term transient memory and long-term historical experience.
Dual-level Adapter Boosting Prompt-free Curvilinear Structure Segmentation
Kai Zhu (Wuhan University of Science and Technology), Jun Cheng (Institute for Infocomm Research)
SegmentationTransformerImageBiomedical Data
🎯 What it does: Proposed a prompt-free curled structure segmentation framework (SACM) based on SAM, achieving fine-grained local adaptation and global cross-domain feature fusion through a two-layer adapter.
Dual-Level Confidence based Implicit Self-Refinement for Medical Visual Question Answering
Meihong Pan (Westlake University), Yefeng Zheng (Westlake University)
TransformerLarge Language ModelContrastive LearningMultimodalityBiomedical DataBenchmark
🎯 What it does: Proposes a dual-layer confidence self-optimization framework called DuCoR, which evaluates and weights pseudo-labels by leveraging the loss distribution of pseudo-labels and feature prototype similarity, thereby achieving adaptive distribution alignment in medical visual question answering.
Dual-Level Hypergraph Generation for Addressing Feature Scarcity in Whole-Slide Image Classification
Shuilian Yao (Dalian University Of Technology), Xin Fan (Dalian University Of Technology)
ClassificationGraph Neural NetworkDiffusion modelAuto EncoderImageBiomedical Data
🎯 What it does: This paper proposes a dual-layer hypergraph generation framework called Dual-HGNet to address the issue of insufficient minority class features in whole-slide image classification.
Dual-Prototype-Guided Multi-task Learning for Unsupervised Anomaly Detection and Classification
Qianhao Luo (Dongguan University of Technology), Weiling Li (CISDI Group Co., Ltd)
ClassificationAnomaly DetectionTransformerContrastive LearningImageBiomedical Data
🎯 What it does: Propose an end-to-end multi-task learning framework PG-SFD, jointly training unsupervised anomaly detection and weakly supervised anomaly classification to address local visual ambiguity and cross-task feature conflicts.
Duala: Dual-Level Alignment of Subjects and Stimuli for Cross-Subject fMRI Decoding
Shumeng Li (Nanjing University), Yinghuan Shi (Nanjing University)
RetrievalDomain AdaptationTransformerSupervised Fine-TuningVision Language ModelDiffusion modelBiomedical DataMagnetic Resonance Imaging
🎯 What it does: Proposes the Duala two-layer alignment framework, which fine-tunes cross-subject visual decoding for new subjects using only one hour of fMRI data.
DualMirage: Hunting Stealthy Multimodal LLM Agents via CAPTCHAs with Contour and Adversarial Illusions
Bei Chen (Shanghai Jiao Tong University), Jianhua Li (Shanghai Jiao Tong University)
Anomaly DetectionSafty and PrivacyAdversarial AttackLarge Language ModelVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: Designed a CAPTCHA framework called DualMirage that detects and blocks multi-modal large language model agents by simultaneously utilizing psychological contour illusions and machine learning adversarial illusions.
DualPrim: Compact 3D Reconstruction with Positive and Negative Primitives
Xiaoxu Meng (Independent Researcher), Lin Gao (Institute of Computing Technology, Chinese Academy of Sciences)
GenerationData SynthesisNeural Radiance FieldImageMesh
🎯 What it does: Propose DualPrim, a differentiable 3D reconstruction framework based on positive and negative dual primitives (positive and negative super tetrahedra), which can directly learn and generate compact, structured meshes from multi-view images;
DualReg: Dual-Space Filtering and Reinforcement for Rigid Registration
Jiayi Li (University Of Science And Technology Of China), Juyong Zhang (University Of Chinese Academy Of Sciences)
Pose EstimationOptimizationPoint Cloud
🎯 What it does: Propose a dual-space optimization framework called DualReg, which denoises the corresponding points obtained from feature matching through an efficient 1-point RANSAC filtering and 3-point RANSAC refinement. Then, it constructs geometric proxy points based on the filtered corresponding points and simultaneously optimizes the rigid transformation in both the feature space and geometric space.
DualSplat: Robust 3D Gaussian Splatting via Pseudo-Mask Bootstrapping from Reconstruction Failures
Xu Wang (Beihang University), Yisong Chen (Peking University)
RestorationSegmentationDepth EstimationGaussian SplattingImage
🎯 What it does: Propose a two-stage 3D Gaussian Splatting (DualSplat) framework for disturbance-free real-time rendering in multi-view images with transient dynamic objects.
DUET-VLM: Dual stage Unified Efficient Token reduction for VLM Training and Inference
Aditya Kumar Singh (Advanced Micro Devices Inc), Emad Barsoum (Advanced Micro Devices Inc)
CompressionComputational EfficiencyTransformerVision Language ModelImageVideoMultimodality
🎯 What it does: Proposes a two-stage visual token compression framework named DUET-VLM, achieving significant reduction in visual tokens while preserving visual semantics.
DuetMerging: Synergizing Dynamic and Static Strategies for Mitigating Task Interference in Model Merging
Yan Li (Pengcheng Laboratory), Dongmei Jiang (Pengcheng Laboratory)
ClassificationRepresentation LearningTransformerMixture of ExpertsImageBenchmark
🎯 What it does: Proposed the DuetMerging framework, which constructs a shared expert pool using dynamic Tucker decomposition and corrects residuals at the static level through neuron-guided sparsification to achieve multi-task model merging
DuetSVG: Unified Multimodal SVG Generation with Internal Visual Guidance
Peiying Zhang (City University of Hong Kong), Difan Liu (Adobe Research)
Image TranslationGenerationTransformerSupervised Fine-TuningVision Language ModelMultimodality
🎯 What it does: Designed and implemented a unified multi-modal generation framework called DuetSVG, capable of simultaneously generating image tokens and SVG code, supporting tasks such as text-to-SVG, image-to-SVG, and SVG editing.
DUO-VSR: Dual-Stream Distillation for One-Step Video Super-Resolution
Zhengyao Lv (University of Hong Kong), Kwan-Yee K. Wong (University of Hong Kong)
Super ResolutionKnowledge DistillationTransformerDiffusion modelGenerative Adversarial NetworkVideo
🎯 What it does: Proposes DUO-VSR, a three-stage single-step video super-resolution method that achieves high-quality generation by leveraging evolution-guided distillation, dual-stream distillation, and preference-guided refinement.
DuoGen: Towards Autonomous Interleaved Multimodal Generation
Min Shi (Georgia Tech), Humphrey Shi (Georgia Tech)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringVision Language ModelDiffusion modelImageVideoTextMultimodality
🎯 What it does: Proposes DuoGen, a conversational generation framework capable of automatically generating and embedding multimodal content (text and images) during user interactions;
DuoMo: Dual Motion Diffusion for World-Space Human Reconstruction
Yufu Wang (Meta Reality Labs), Michael Zollhöfer
GenerationTransformerDiffusion modelVideoMesh
🎯 What it does: Propose DuoMo, a two-stage diffusion model that first estimates motion in camera coordinates and then lifts and refines it in world coordinates, directly generating mesh vertices rather than SMPL parameters.
DVAR: Dynamic Visual Autoregressive Modeling for Image Super-Resolution
Yu Zheng (Nanjing University of Science and Technology), Jian Yang (Nanjing University of Science and Technology)
Super ResolutionTransformerAuto EncoderImage
🎯 What it does: This paper proposes DVAR, a dynamic visual autoregressive image super-resolution model that can uniformly generate images at different target sizes.
DVGT: Driving Visual Geometry Transformer
Sicheng Zuo (Tsinghua University), Jiwen Lu (Tsinghua University)
Autonomous DrivingTransformerSimultaneous Localization and MappingImagePoint Cloud
🎯 What it does: Proposed and implemented a vision geometry Transformer (DVGT) that operates under different camera configurations without camera priors, capable of directly generating global dense 3D point maps and vehicle poses from multi-view unposed image sequences.
DyaDiT: A Multi-Modal Diffusion Transformer for Socially Favorable Dyadic Gesture Generation
Yichen Peng, Kris Kitani
GenerationTransformerDiffusion modelAuto EncoderMultimodalityAudio
🎯 What it does: Propose DyaDiT, a multimodal diffusion Transformer that generates gestures aligned with conversational contexts based on dual-person audio and social context.
DyFCLT: Dynamic Frequency-Decoupled Cross-Modal Learning Transformer for Multimodal Tiny Object Detection
Chaolang Li (Sun Yat-sen University), Zhuoran Zheng (Qilu University of Technology)
Object DetectionTransformerMultimodality
🎯 What it does: This paper proposes a dynamic frequency decoupling cross-modal learning transformer (DyFCLT) for multi-modal small target detection.
Dynamic Black-hole Emission Tomography with Physics-informed Neural Fields
Berthy T. Feng (Caltech), Katherine L. Bouman (Caltech)
Neural Radiance FieldImageComputed TomographyPhysics Related
🎯 What it does: Propose a physics-informed dynamic black hole emitter field (PI-DEF) method, using differentiable neural rendering to reconstruct four-dimensional (time + 3D) emissivity and velocity fields under sparse EHT observations.
Dynamic Exposure Burst Image Restoration
Woohyeok Kim (POSTECH), Sunghyun Cho
RestorationConvolutional Neural NetworkOptical FlowImageVideo
🎯 What it does: Propose a dynamic exposure burst image recovery (DEBIR) pipeline that achieves high-quality image reconstruction in low-light conditions by predicting the optimal exposure time for each frame, combined with BAENet and burst image recovery networks.
Dynamic Important Example Mining for Reinforcement Finetuning
Haoru Tan (University of Hong Kong), Xiaojuan Qi (University of Hong Kong)
OptimizationReinforcement LearningMultimodalityBenchmark
🎯 What it does: This study proposes the Dynamic Importance Sample Mining (DIEM) framework, integrating gradient alignment importance estimation and constrained reweighting into every optimization step of Reinforcement Learning Fine-tuning (RFT) to achieve adaptive and self-organized data usage;
Dynamic Label Noise Suppression with Optimal Teacher Pool for Facial Expression Recognition
Yuzhuang Yang (Xidian University), Qigong Sun (SenseTime Research)
RecognitionKnowledge DistillationConvolutional Neural NetworkImageBenchmark
🎯 What it does: Propose a dynamic label noise suppression framework OTP-NS based on an optimal teacher pool (OTP) to address performance degradation in Facial Expression Recognition (FER) caused by noisy labels;
Dynamic Logits Adjustment and Exploration for Test-Time Adaptation in Vision Language Models
Haoyan Wu (University Of Electronic Science And Technology Of China), Wen Li (University Of Electronic Science And Technology Of China)
ClassificationDomain AdaptationTransformerVision Language ModelImageText
🎯 What it does: This paper proposes DLAE, a test-time adaptation framework for visual-language models, which enhances the zero-shot/ few-shot performance of pre-trained models such as CLIP under distribution drift by utilizing Dynamic Logits Adjustment (DLA) and Consistency-Guided Exploratory Cache (CEC).
Dynamic Magic: Unleashing Restricted Knowledge for Lifelong Person Re-Identification
Jinjia Peng (Hebei University), Huibing Wang (Dalian Maritime University)
RecognitionDomain AdaptationMeta LearningTransformerSupervised Fine-TuningMixture of ExpertsContrastive LearningImage
🎯 What it does: Propose a dynamically scalable lifelong person re-identification framework named VIA, which achieves continuous learning and knowledge retention on large pre-trained models by leveraging Unseen-domain Adapter, Domain-aware Dispatch, and Holistic Domain Controller.
Dynamic Momentum Recalibration in Online Gradient Learning
Zhipeng Yao (Shenyang University of Chemical Technology), Dazhou Li (Shenyang University of Chemical Technology)
OptimizationImageStochastic Differential Equation
🎯 What it does: Proposed a novel optimizer SGDF based on optimal linear filtering, dynamically recalibrates momentum, and real-time minimizes the mean squared error of gradient estimates.
Dynamic Stream Network for Combinatorial Explosion Problem in Deformable Medical Image Registration
Shaochen Bi (Hong Kong University of Science and Technology), Hao Chen (Hong Kong University of Science and Technology)
TransformerOptical FlowBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: Proposes a novel dynamic flow network DySNet to address the combinatorial explosion problem in deformable medical image registration.
Dynamic Token Reweighting for Robust Vision-Language Models
Tanqiu Jiang (Stony Brook University), Ting Wang (Stony Brook University)
Safty and PrivacyComputational EfficiencyAdversarial AttackTransformerVision Language ModelMultimodality
🎯 What it does: Propose DTR, an inference-time defense method that resists multimodal jailbreak attacks by dynamically reweighting visual tokens in the KV cache
Dynamic Visual SLAM using a General 3D Prior
Xingguang Zhong (University of Bonn), Cyrill Stachniss (University of Bonn)
Pose EstimationDepth EstimationSimultaneous Localization and MappingImageVideo
🎯 What it does: Propose a monocular visual SLAM system that integrates the depth prior of a feedforward reconstruction model and moving object segmentation, achieving robust camera pose estimation, scale-consistent dense depth reconstruction, and real-time moving object segmentation in dynamic scenes.
Dynamic-eDiTor: Training-Free Text-Driven 4D Scene Editing with Multimodal Diffusion Transformer
Dong In Lee (Purdue University), Karthik Ramani (Purdue University)
GenerationTransformerDiffusion modelGaussian SplattingOptical FlowVideoTextMultimodality
🎯 What it does: Proposed a no-training, text-based 4D scene editing framework called Dynamic-eDiTor, which can perform multi-view, time-consistent edits on pre-trained 4D Gaussian Splatting (4DGS) without retraining the model.
Dynamic-Static Decomposition for Novel View Synthesis of Dynamic Scenes with Spiking Neurons
Lingyun Dai (Zhejiang University), Gang Pan (Zhejiang University)
GenerationData SynthesisSpiking Neural NetworkGaussian SplattingVideo
🎯 What it does: This paper proposes a dynamic scene novel view synthesis framework based on dynamic-static decomposition, utilizing a 4D fine-grained mask field and spiking neurons to achieve precise segmentation of dynamic and static Gaussian primitives, thus enabling high-quality real-time rendering.
DynamicGTR: Leveraging Graph Topology Representation Preferences to Boost VLM Capabilities on Graph QAs
Yanbin Wei (Southern University of Science and Technology), Yu Zhang (Southern University of Science and Technology)
Representation LearningTransformerVision Language ModelGraphChain-of-Thought
🎯 What it does: This paper proposes the DynamicGTR framework, which dynamically selects the most suitable graph topology representation (GTR) for Vision-Language models (VLM) in graph question answering, enhancing the accuracy and conciseness of zero-shot graph QA.
Dynamics-Aware Preference Optimization for Vision-Language Models
Jusheng Zhang (Sun Yat-sen University), Keze Wang (Sun Yat-sen University)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningVision Language ModelImageTextMultimodality
🎯 What it does: Proposed and implemented a two-stage vision-language model preference alignment method called CW-DPO, which first smooths the loss surface through constrained SFT, and then adaptively adjusts the gradient of DPO using cooling weights based on the average token log-prob, thereby alleviating the 'compression effect' caused by easy negative samples and achieving a more stable optimization process.
Dynamics: Language-Based Representation for Inferring Rigid-Body Dynamics From Videos
Chia-Hsiang Kao (Cornell University), Ning Zhou (Amazon)
OptimizationExplainability and InterpretabilityRepresentation LearningTransformerVision Language ModelOptical FlowVideoText
🎯 What it does: Propose a method that converts rigid body dynamics information into interpretable language descriptions (YAML) and directly predicts scene configurations from monocular videos using a vision-language model (VLM).
DynamicsBoost: Dynamic Plausible Video Generation via Annotation-Free Continuation Preference Optimization
Jiaxing Li (Nanyang Technological University), Hao-Xiang Guo (Skywork AI)
GenerationTransformerSupervised Fine-TuningPrompt EngineeringDiffusion modelFlow-based ModelVideo
🎯 What it does: Generate unannotated preference pairs through video continuation and post-train pre-trained video generation models using asymmetric DPO.
DynamicTree: Interactive Real Tree Animation via Sparse Voxel Spectrum
Yaokun Li (Sun Yat-sen University), Tianfan Xue (Chinese University Of Hong Kong Mmlab)
GenerationData SynthesisConvolutional Neural NetworkDiffusion modelGaussian SplattingVideoMesh
🎯 What it does: Leverage sparse voxel spectrum representation bound with 3D Gaussian Splatting to achieve high-quality, long-term, real-time interactive animation of real trees;
DynamicVGGT: Learning Dynamic Point Maps for 4D Scene Reconstruction in Autonomous Driving
Zhuolin He (Fudan University), Xiangyang Xue (Fudan University)
Pose EstimationDepth EstimationAutonomous DrivingTransformerGaussian SplattingPoint Cloud
🎯 What it does: Proposes a unified feed-forward framework called DynamicVGGT for 4D dynamic point cloud reconstruction in autonomous driving scenarios, providing functionalities such as camera pose estimation, depth prediction, and novel view synthesis.
DynBridge: Bridging Imagination and Control through Interaction Dynamics for Robot Manipulation
Alex Wang (State Grid Corporation of China), Mengmeng Wang (Zhejiang University of Technology)
Robotic IntelligenceTransformerVision-Language-Action ModelWorld ModelVideoBenchmark
🎯 What it does: Proposes an end-to-end framework called DynBridge, which tightly integrates the robot's 'imagination' of future visual outcomes with control decisions through a shared interaction dynamics representation to achieve more consistent robot manipulation.
DynFusion: Rethinking Condition Fusion for Adaptive Multi-Conditional Text-to-Image Generation
Zheng Fang (University Of Warwick), Hongkai Wen (University Of Warwick)
GenerationTransformerDiffusion modelFlow-based ModelAuto EncoderMultimodality
🎯 What it does: Propose the DynFusion framework to achieve dynamic fusion of multiple conditions (depth maps, edges, reference images, etc.) during text-to-image generation, enabling controllable and flexible image synthesis.
E-3DPSM: A State Machine for Event-based Egocentric 3D Human Pose Estimation
Mayur Deshmukh (MPI for Informatics), Vladislav Golyanik (MPI for Informatics)
Pose EstimationTransformerTime Series
🎯 What it does: Proposed an event-driven continuous pose state machine called E-3DPSM for real-time 3D human pose estimation using monocular head-mounted cameras.
E-comIQ-ZH: A Human-Aligned Dataset and Benchmark for Fine-Grained Evaluation of E-commerce Posters with Chain-of-Thought
Meiqi Sun (Taobao & Tmall Group, Alibaba Group), Junxiong Zhu (Taobao & Tmall Group, Alibaba Group)
Explainability and InterpretabilityData-Centric LearningLarge Language ModelSupervised Fine-TuningVision Language ModelContrastive LearningMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Proposed a multi-dimensional quality assessment framework for Chinese e-commerce posters, constructed the E-comIQ-18k dataset and the evaluation model E-comIQ-M based on Qwen-2.5-VL, and released the automated evaluation benchmark E-comIQ-Bench;
E-RayZer: Self-supervised 3D Reconstruction as Spatial Visual Pre-training
Qitao Zhao (Carnegie Mellon University), Hanwen Jiang (Adobe Research)
GenerationPose EstimationDepth EstimationRepresentation LearningTransformerGaussian SplattingVideo
🎯 What it does: Built a self-supervised 3D Gaussian splatting model called E-RayZer, capable of learning camera poses and scene 3D geometry from unlabeled videos alone, serving as a spatial vision pre-training framework.
E$^2$-SCI: Elastic Edge-Cloud Speculative Decoding via Credit Inertia
Senyao Li (Huazhong University of Science and Technology), Ruixuan Li (Huazhong University of Science and Technology)
Computational EfficiencyTransformerTextBenchmark
🎯 What it does: This paper proposes an E-SCI 2 framework for edge-cloud collaborative inference, which utilizes credit inertia to adaptively adjust the validation thresholds of the draft model and the target model, and achieves asynchronous pipeline through Progressive Lookahead Concurrency (PLC) to reduce inference latency and improve throughput.
E2EGS: Event-to-Edge Gaussian Splatting for Pose-Free 3D Reconstruction
Yunsoo Kim (KAIST), Hyun Myung (KAIST)
GenerationGaussian SplattingSimultaneous Localization and Mapping
🎯 What it does: Propose E2EGS, a pose-agnostic 3D reconstruction framework that uses only event streams.
E3AD: An Emotion-Aware Vision-Language-Action Model for Human-Centric End-to-End Autonomous Driving
Yihong Tang (McGill University), Chengzhong Xu (University of Macau)
Autonomous DrivingTransformerLarge Language ModelSupervised Fine-TuningVision-Language-Action ModelTextMultimodalityChain-of-Thought
🎯 What it does: Developed an emotion-aware visual-language-action (E3AD) framework for open-domain end-to-end autonomous driving, capable of understanding passenger natural language instructions, inferring emotions, and generating safe executable trajectories with verbal feedback.
EagleNet: Energy-Aware Fine-Grained Relationship Learning Network for Text-Video Retrieval
Yuhan Chen (Sun Yat-sen University), Xiaochun Cao (Sun Yat-sen University)
RetrievalGraph Neural NetworkVision Language ModelContrastive LearningVideoTextMultimodality
🎯 What it does: This paper proposes EagleNet, an energy-aware fine-grained relationship learning network that generates context-aware text embeddings to improve text-video retrieval performance.
EagleVision: A Dual-Stage Framework with BEV-grounding-based Chain-of-Thought for Spatial Intelligence
Jiaxu Wan (Beihang University), Yifan Yang (Beihang University)
Reinforcement LearningVision Language ModelSimultaneous Localization and MappingVideoTextChain-of-Thought
🎯 What it does: Proposed a two-stage framework called EagleVision, achieving video spatial reasoning through macro perception (semantic and perspective-diverse keyframe selection) and micro verification (spatial Chain-of-Thought based on BEV active perspective queries).
EarlyTom: Early Token Compression Completes Fast Video Understanding
Hesong Wang (Zhejiang University), Huan Wang (Alibaba Cloud Computing)
Computational EfficiencyTransformerLarge Language ModelVision Language ModelVideo
🎯 What it does: Proposed a training-free EarlyTom framework that performs early frame merging and spatial token selection within the visual encoder of video LLMs to compress visual tokens;
Easy2Hard: From Partially to Fully Unmatched Modalities as Negative Samples in Contrastive Learning
Zhicheng Yang (Southern Illinois University), Xiaopeng Jiang (Southern Illinois University)
RetrievalTransformerContrastive LearningMultimodalityBiomedical DataElectronic Health RecordsElectrocardiogram
🎯 What it does: Propose the Easy2Hard framework for multi-modal contrastive learning with M>2, which first fine-grainedly divides negative samples into two categories: partial mismatch (easy) and complete mismatch (hard). It gradually shifts the training focus from easy to hard negative samples through sigmoid curriculum scheduling, achieving more refined contrastive learning.
Easy3E: Feed-Forward 3D Asset Editing via Rectified Voxel Flow
Shimin Hu (University of Science and Technology of China), Juyong Zhang (University of Science and Technology of China)
GenerationData SynthesisVision Language ModelDiffusion modelFlow-based ModelRectified FlowMesh
🎯 What it does: Propose a Feed-Forward 3D asset editing framework named Easy3E that can accomplish geometry deformation and texture refinement from a single editing view and a rough 3D mask.
EasyOmnimatte: Taming Pretrained Inpainting Diffusion Models for End-to-End Video Layered Decompositio
Yihan Hu (GVC Lab, Great Bay University), Xiaodong Cun (GVC Lab, Great Bay University)
SegmentationGenerationTransformerSupervised Fine-TuningMixture of ExpertsDiffusion modelVideo
🎯 What it does: Proposed a unified end-to-end video Omnimatte method that directly generates foreground layers, alpha mattes, and background layers using a fine-tuned pre-trained video inpainting diffusion model.
EasyV2V: A High-quality Instruction-based Video Editing Framework
Jinjie Mai (KAUST), Ashkan Mirzaei (KAUST)
GenerationData SynthesisTransformerLarge Language ModelAuto EncoderVideoTextMultimodality
🎯 What it does: Proposes the EasyV2V framework, achieving unified data, architecture, and control for multimodal instruction-based video editing.
Echoes of Ownership: Adversarial-Guided Dual Injection for Copyright Protection in MLLMs
Chengwei Xia (Lanzhou University), Yi Yang (Zhejiang University)
Adversarial AttackSupervised Fine-TuningVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: Designed and implemented a trigger image method based on adversarial dual injection for tracking copyright ownership of multimodal large language models in black-box environments.
Echoes Over Time: Unlocking Length Generalization in Video-to-Audio Generation Models
Christian Simon (Sony Group Corporation), Yuki Mitsufuji (Sony Group Corporation)
GenerationFlow-based ModelVideoMultimodalityAudio
🎯 What it does: Train short video clips (about 8 seconds) and generate audio lasting over 5 minutes during inference; propose the MMHNet framework, integrating multi-modal hierarchical routing and non-causal Mamba-2 to achieve length generalization;
EchoFoley: Event-Centric Hierarchical Control for Video Grounded Creative Sound Generation
Bingxuan Li (University of Illinois Urbana-Champaign), Yulei Niu (ByteDance Intelligent Creation)
GenerationTransformerLarge Language ModelAgentic AIVision Language ModelDiffusion modelVideoTextMultimodalityChain-of-ThoughtAudio
🎯 What it does: Proposes the EchoFoley task and EchoVidia framework to address event-level fine-grained control and hierarchical instruction parsing in video sound effect generation.
EchoPOSE: 6D Pose Estimation of Sparse Echocardiograms for Left-Ventricular 3D Shape Reconstruction
Lucas Iijima (Imperial College London), Choon Hwai Yap (Imperial College London)
Pose EstimationTransformerBiomedical DataUltrasound
🎯 What it does: Propose a 6D pose estimation based on the deep network EchoPOSE, combining graphical harmonic deformation to recover the 3D shape of the left ventricle from sparse 2D ultrasound slices.
EchoVDiff: Cardiac-Cycle Echocardiography Video Generation from Arbitrary Single Frame
Jiansong Zhang (Shenzhen University), Linlin Shen (Shenzhen University)
GenerationTransformerDiffusion modelAuto EncoderVideoBiomedical DataUltrasound
🎯 What it does: Develop a phase-aware diffusion model (EchoVDiff) capable of generating a complete cardiac cycle video from any single-frame cardiac ultrasound image.
EcoAlign: An Economically Rational Framework for Efficient LVLM Alignment
Ruoxi Cheng (Beijing Electronic Science and Technology Institute), Hongyi Zhang (Nanyang Technological University)
OptimizationComputational EfficiencyGraph Neural NetworkVision Language ModelMultimodalityChain-of-Thought
🎯 What it does: Designed an economic rationality alignment framework called EcoAlign, which balances safety and utility through multi-modal graph search under limited computational budgets.
EcoSplat: Efficiency-controllable Feed-forward 3D Gaussian Splatting from Multi-view Images
Minh-Quan Viet Bui (KAIST), Munchurl Kim (KAIST)
GenerationTransformerGaussian SplattingImage
🎯 What it does: Proposes EcoSplat, a method capable of generating 3D Gaussian splatting representations through a feed-forward approach from multi-view inputs, with explicit control over the number of retained Gaussians, achieving high-quality, adjustable view synthesis.
Edge-Focused Super-Resolution for Omnidirectional Images with Spherical Geometric Augmentation
Shaolin Wang (Southwest University), Jianfeng Li (Southwest University)
Super ResolutionConvolutional Neural NetworkImage
🎯 What it does: Proposes an edge-core super-resolution network (Edge-Focused Block) combined with spherical geometry rotation-translation data augmentation for high-magnification (8×, 16×) panoramic image super-resolution.
Edge-RecViT: Efficient Vision Transformer via Semantic-Refined Dynamic Recursion
YiZhou Li (Xi'an Jiaotong-Liverpool University), Xianyi Zhao (Xi'an Jiaotong-Liverpool University)
ClassificationComputational EfficiencyTransformerImage
🎯 What it does: Propose Edge-RecViT, an efficient visual Transformer that significantly reduces FLOPs and parameter count through edge-aware token adapters and recursive parameter sharing.
Edges Compete for Trust: Group Relative Edge Optimization for Building Reconstruction from Point Clouds
Yujun Liu (Shenzhen University), Qingquan Li (Shenzhen University)
OptimizationTransformerReinforcement LearningPoint Cloud
🎯 What it does: Proposed a Group Relative Edge Optimization (GREO) training strategy that converts geometric matching rewards into group relative rewards, applying dense supervision to all edge candidates.
EDGS: Eliminating Densification for Efficient Convergence of 3DGS
Dmytro Kotovenko (LMU Munich), Björn Ommer (LMU Munich)
GenerationOptimizationComputational EfficiencyGaussian SplattingImage
🎯 What it does: In 3D Gaussian Splatting (3DGS), a novel initial configuration based on dense image correspondence triangulation is proposed, directly generating a large number of high-quality Gaussians in one go, completely eliminating the subsequent densification step.
Edit-As-Act: Goal-Regressive Planning for Open-Vocabulary 3D Indoor Scene Editing
Seongrae Noh (Korea University), HyeongYeop Kang (Korea University)
GenerationLarge Language ModelVision-Language-Action ModelTextMesh
🎯 What it does: This paper proposes Edit-As-Act, a 3D indoor scene editing framework based on goal inverse planning, which can perform minimal physically feasible editing operations while preserving the original scene according to natural language instructions.
Edit-aware RAW reconstruction
Abhijith Punnappurath (Samsung Electronics), Michael S. Brown (Samsung Electronics)
RestorationDiffusion modelImage
🎯 What it does: Propose a pluggable edit-aware loss function to directly optimize the quality of the final sRGB rendering during RAW reconstruction.
Edit2Perceive: Image Editing Diffusion Models Are Strong Dense Perceivers
Yiqing Shi (Peking University), Mike Zheng Shou (National University Of Singapore)
SegmentationDepth EstimationTransformerDiffusion modelFlow-based ModelImageOrdinary Differential Equation
🎯 What it does: This work proposes Edit2Perceive, a unified diffusion Transformer framework that transforms image-to-image (I2I) diffusion models into dense perception models for monocular depth estimation, surface normal estimation, and interactive matting.
EditCtrl: Disentangled Local and Global Control for Real-Time Generative Video Editing
Yehonathan Litman (Meta Reality Labs), Caleb Leak (Meta Reality Labs)
GenerationTransformerDiffusion modelAuto EncoderVideo
🎯 What it does: Proposed the EditCtrl framework, which achieves efficient real-time video editing by leveraging local context and global embeddings, performing computations only in the editing region.
EditMGT: Unleashing Potentials of Masked Generative Transformers in Image Editing
Wei Chow (ByteDance), Songhua Liu (Shanghai Jiao Tong University)
Image TranslationTransformerLarge Language ModelVision Language ModelImage
🎯 What it does: Proposed EditMGT, an image editing framework based on Masked Generative Transformers (MGT), which can complete high-resolution (≥1024) image editing within 2 seconds and significantly reduce editing leakage issues.
Editprint: General Digital Image Forensics via Editing Fingerprint with Self-Augmentation Training
Haiwei Wu (University of Electronic Science and Technology of China), Jiantao Zhou (University of Macau)
Anomaly DetectionRepresentation LearningConvolutional Neural NetworkVision Language ModelContrastive LearningImage
🎯 What it does: Designed and trained a self-supervised general image forensics feature called Editprint, which leverages an online self-incremental editing pool to simulate massive camera internal and external processing chains, thereby learning features capable of distinguishing different forensics tasks (e.g., SID, SNP, CSI).
EduDiag: A Benchmark for Educational Diagnostic Reasoning with Error Tracing and Correction on Large Multimodal Models
Jiali Chen (South China University of Technology), Yi Cai (Key Laboratory of Big Data and Intelligent Robot Ministry of Education)
TransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningTextMultimodalityBenchmark
🎯 What it does: Created the EduDiag benchmark to evaluate the ability of large multimodal models in error tracking and correction.
EE-RL: Vision Language Guided Reinforcement Learning with Explorer and Expert model for End-to-End Autonomous Driving
Xiaolong Li (Chang'an University), Xiangmo Zhao (Chang'an University)
Autonomous DrivingComputational EfficiencyReinforcement LearningVision Language ModelImageVideoMultimodalityBenchmark
🎯 What it does: Propose the EE-RL framework, which combines RL explorers with VLM experts, and designs dual replay buffers and StateHash mechanisms to achieve efficient learning
EEGiT: Teaching Vision Transformers to Understand the EEG signal
Jiahao Zhou (Xidian University), Cheng Deng (Beijing Institute for General Artificial Intelligence)
ClassificationRetrievalRepresentation LearningTransformerContrastive LearningBiomedical Data
🎯 What it does: Propose the EEGiT framework, which converts EEG signals into structured EEG patches and utilizes a pre-trained Vision Transformer (ViT) to extract semantic features for visual stimulus decoding.
EffectErase: Joint Video Object Removal and Insertion for High-Quality Effect Erasing
Yang Fu (Fudan University), Henghui Ding (Fudan University)
Image HarmonizationRestorationTransformerDiffusion modelAuto EncoderVideo
🎯 What it does: Propose the EffectErase framework, jointly learning video object removal and insertion, and constructing a large-scale hybrid dataset VOR;
EffectMaker: Unifying Reasoning and Generation for Customized Visual Effect Creation
Shiyuan Yang (Tencent Hunyuan), Jing Liao (City University Of Hong Kong)
GenerationData SynthesisTransformerLarge Language ModelVision Language ModelDiffusion modelImageVideoTextMultimodality
🎯 What it does: Proposes a unified reasoning-generation framework called EffectMaker, enabling customized transfer of visual effects (VFX) using reference videos and target images;
Efficiency Follows Global-Local Decoupling
Zhenyu Yang (Nanjing University of Science and Technology), Fumin Shen (University of Electronic Science and Technology of China)
ClassificationObject DetectionSegmentationConvolutional Neural NetworkImage
🎯 What it does: Proposed the ConvNeur two-branch network, where one branch is responsible for local detail extraction and the other for global context aggregation, achieving efficient global-local decoupling through gated fusion;
Efficient All-Pairs Correlation Volume Sampling for Optical Flow Estimation
Karlis Martins Briedis (DisneyResearch Studios), Studios 0000-0003-1473-1878
OptimizationComputational EfficiencyOptical FlowVideo
🎯 What it does: An efficient sampling algorithm based on block sparse matrix multiplication is proposed to address the fully connected correlation volume sampling in methods like RAFT, significantly reducing memory usage and runtime.
Efficient and High-Fidelity Omni Modality Retrieval
Chuong Huynh (University of Maryland), Abhinav Shrivastava (University of Maryland)
RetrievalLarge Language ModelTextMultimodalityAudio
🎯 What it does: Proposed OmniRet, the first retrieval model capable of handling composite queries across text, visual, and audio modalities.
Efficient and Training-Free Single-Image Diffusion Models
Haojun Qiu (University of Toronto), David B. Lindell (University of Toronto)
GenerationTransformerDiffusion modelScore-based ModelAuto EncoderImage
🎯 What it does: Propose a training-free single-image diffusion model that constructs a closed-form denoiser using all patches from the input image, and generates new images with internal structures similar to the original through multi-scale coarse-to-fine sampling.
Efficient Encoder-Free Fourier-based 3D Large Multimodal Model
Guofeng Mei (Fondazione Bruno Kessler), Fabio Poiesi (Fondazione Bruno Kessler)
Computational EfficiencyRepresentation LearningGraph Neural NetworkTransformerLarge Language ModelVision Language ModelMultimodalityPoint Cloud
🎯 What it does: Proposed Fase3D, an encoder-free Fourier-based 3D large model that directly serializes point clouds and maps them to the LLM's word embedding space, achieving 3D scene understanding and language tasks.
Efficient Equivariant Transformer for Self-Driving Agent Modeling
Scott Xu (Waabi), Raquel Urtasun (Waabi)
Autonomous DrivingTransformerPoint Cloud
🎯 What it does: Propose the DriveGATr transformer model, which realizes SE(2)-equivariant traffic scene agent modeling through 2D projection geometric algebra multivector encoding.
Efficient Frame Selection for Long Video Understanding via Reinforcement Learning
Yaxuan Qin (Tencent), Yancheng He (Tencent)
Computational EfficiencyTransformerSupervised Fine-TuningReinforcement LearningVision Language ModelVideoBenchmark
🎯 What it does: Propose a lightweight, query-adaptive frame selector to help multimodal large language models (MLLMs) select keyframes when processing long videos, thereby improving inference performance.
Efficient Hybrid SE(3)-Equivariant Visuomotor Flow Policy via Spherical Harmonics for Robot Manipulation
Qinglun Zhang (University of Electronic Science and Technology of China), Shuaicheng Liu (University of Electronic Science and Technology of China)
Robotic IntelligenceTransformerFlow-based ModelRectified FlowImagePoint Cloud
🎯 What it does: Propose E3Flow, a SE(3)-equivariant visual motion policy integrating spherical harmonics and flow matching, achieving efficient reasoning and data efficiency for robotic manipulation tasks;
Efficient Real-Time Raw-to-Raw Denoising for Extreme Low-Light Ultra HD Video on Mobile Devices
Charantej Pochimireddy (Samsung R&D Institute), Raj Gadde (Samsung R&D Institute)
RestorationData SynthesisComputational EfficiencyKnowledge DistillationConvolutional Neural NetworkVideo
🎯 What it does: This paper proposes a complete system capable of real-time raw-to-raw denoising for 4K/8K ultra-high-definition videos under extremely low illumination (<1lx), compatible with existing ISP pipelines and achieving low-power, low-latency processing on mobile devices.
Efficient Training for Human Video Generation with Entropy-Guided Prioritized Progressive Learning
Changlin Li (Stanford University), Xiaojun Chang (University of Science and Technology of China)
GenerationDiffusion modelVideo
🎯 What it does: Propose an efficient training framework named Ent-Prog for pose-guided human video diffusion models;
Efficient Unrolled Networks for Large-Scale 3D Inverse Problems
Romain Vo (CNRS, ENS de Lyon, Laboratoire de Physique), Julián Tachella (CNRS, ENS de Lyon, Laboratoire de Physique)
Computational EfficiencyConvolutional Neural NetworkBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: Proposed an efficient deconvolution network for solving large-scale 3D inverse problems, achieving state-of-the-art performance in 3D X-ray cone-beam CT and 3D multi-coil accelerated MRI.
Efficient Video Object Segmentation and Tracking with Recurrent Dynamic Submodel
Weidong Tang (Xidian University), Wangbo Zhao (Harbin Institute of Technology)
Object TrackingSegmentationTransformerVideo
🎯 What it does: Proposes the RDS (Recurrent Dynamic Submodel) framework, achieving dynamic submodel selection and efficient fine-tuning for video object segmentation and tracking through a prediction-aware router and importance-aware LoRA adapter.
Efficient Weighted Sampling via Score-based Generative Models
Heasung Kim (University of Texas at Austin), Gustavo De Veciana (University of Texas at Austin)
GenerationData SynthesisScore-based ModelImageStochastic Differential Equation
🎯 What it does: Proposes an efficient framework called LAGS (Lightweight Approximation with uncertainty-adaptive Guidance Scheduling), which is training-free and can directly utilize existing pre-trained Score-based Generative Models (SGM) for weighted sampling.
Efficiently Reconstructing Dynamic Scenes One D4RT at a Time
Chuhan Zhang (Google DeepMind), Mehdi S. M. Sajjadi (Google DeepMind)
Object TrackingGenerationData SynthesisPose EstimationDepth EstimationTransformerVideoPoint CloudBenchmark
🎯 What it does: Dynamic 4D scene reconstruction and tracking of videos, proposing a unified interface decoder that can predict 3D positions at any spatial and temporal point.
EfficientMonoHair: Fast Strand-Level Reconstruction from Monocular Video via Multi-View Direction Fusion
Da Li (King Abdullah University of Science and Technology), Ivan Viola (King Abdullah University of Science and Technology)
GenerationOptimizationComputational EfficiencyTransformerNeural Radiance FieldVideo
🎯 What it does: Proposes EfficientMonoHair, a fast framework combining implicit neural networks and multi-view geometry fusion, achieving high-precision hair-level geometric reconstruction using monocular videos.
EfficientVPR: Toward Efficient Visual Place Recognition via Scene-Aware Prompt Tuning and Adaptive Feature Enhancement
Wenjing Tang (State Key Laboratory of AI Safety, Institute of Computing Technology, Chinese Academy of Sciences), Yongjun Xu (State Key Laboratory of AI Safety, Institute of Computing Technology, Chinese Academy of Sciences)
RecognitionRetrievalTransformerPrompt EngineeringContrastive LearningImage
🎯 What it does: Proposed a lightweight one-stage visual place recognition framework called EfficientVPR, which achieves significant improvements in retrieval speed and compressed feature dimension while maintaining high recognition accuracy through scene-aware visual prompt tuning of pre-trained models and instance-dependent key local feature enhancement.