CVPR 2025 Papers — Page 8
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2871 papers
DualTalk: Dual-Speaker Interaction for 3D Talking Head Conversations
Ziqiao Peng (Renmin University of China), Zhaoxin Fan (Beijing Advanced Innovation Center for Future Blockchain and Privacy Computing)
GenerationData SynthesisRecurrent Neural NetworkTransformerVideoMultimodalityAudio
🎯 What it does: The DualTalk framework is proposed for multi-turn 3D talking head generation for dual speakers, supporting speaker/listener role switching to achieve coherent dialogue animation.
DUNE: Distilling a Universal Encoder from Heterogeneous 2D and 3D Teachers
Mert Bülent Sarıyıldız (NAVER LABS Europe), Yannis Kalantidis (NAVER LABS Europe)
SegmentationPose EstimationDepth EstimationKnowledge DistillationTransformerImage
🎯 What it does: This paper proposes a heterogeneous teacher distillation (co-distillation) method, training a ViT-Base encoder DUNE, which achieves excellent performance on 2D, 3D, and 3D human perception tasks simultaneously.
DV-Matcher: Deformation-based Non-rigid Point Cloud Matching Guided by Pre-trained Visual Features
Zhangquan Chen (Tsinghua Shenzhen International Graduate School), Ruqi Huang (Pengcheng Laboratory)
OptimizationRepresentation LearningContrastive LearningPoint Cloud
🎯 What it does: A novel unsupervised learning framework called DV-Matcher has been developed for dense correspondence of non-rigid point clouds.
DVHGNN: Multi-Scale Dilated Vision HGNN for Efficient Vision Recognition
Caoshuo Li (Xiamen University), Taisong Jin (Xiamen University)
RecognitionObject DetectionSegmentationGraph Neural NetworkImage
🎯 What it does: This paper proposes the DVHGNN architecture, which combines multi-scale hypergraphs with dynamic hypergraph convolution to efficiently capture high-order associations in visual features.
DViN: Dynamic Visual Routing Network for Weakly Supervised Referring Expression Comprehension
Xiaofu Chen (Mohammed Bin Zayed University of Artificial Intelligence), Yiyi Zhou (Xiamen University)
RecognitionObject DetectionSegmentationContrastive LearningImageMultimodality
🎯 What it does: This paper proposes a weakly supervised framework for referring expression understanding, DViN, which enhances fine-grained visual description capabilities through a dynamically routable visual encoder combination.
DyCoke: Dynamic Compression of Tokens for Fast Video Large Language Models
Keda Tao (Westlake University), Huan Wang (Westlake University)
CompressionComputational EfficiencyTransformerLarge Language ModelVision Language ModelVideo
🎯 What it does: This paper presents DyCoke, a training-free, pluggable dynamic visual token compression method designed to accelerate the inference of large video language models.
DyCON: Dynamic Uncertainty-aware Consistency and Contrastive Learning for Semi-supervised Medical Image Segmentation
Maregu Assefa (Khalifa University of Science and Technology), Naoufel Werghi (Khalifa University of Science and Technology)
SegmentationContrastive LearningImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A semi-supervised medical image segmentation framework named DyCON is proposed, which can improve segmentation accuracy with a small amount of labeled data.
DyFo: A Training-Free Dynamic Focus Visual Search for Enhancing LMMs in Fine-Grained Visual Understanding
Geng Li (Peking University), Yuxin Peng (Peking University)
RecognitionObject DetectionTransformerLarge Language ModelVision Language ModelImageMultimodality
🎯 What it does: This paper proposes DyFo, a training-free dynamic attention visual search method that enhances the performance of large multimodal models in fine-grained visual understanding through bidirectional interaction between LMM and visual experts, as well as MCTS simulating human visual search.
DyMO: Training-Free Diffusion Model Alignment with Dynamic Multi-Objective Scheduling
Xin Xie (University of New South Wales), Dong Gong (University of New South Wales)
GenerationData SynthesisTransformerLarge Language ModelDiffusion modelImageText
🎯 What it does: A training-free text-to-image diffusion model alignment method called DyMO is proposed, which can achieve consistency between text and human preferences during inference through multi-objective dynamic scheduling.
Dyn-HaMR: Recovering 4D Interacting Hand Motion from a Dynamic Camera
Zhengdi Yu (Imperial College London), Tolga Birdal (Imperial College London)
Pose EstimationOptimizationSimultaneous Localization and MappingVideo
🎯 What it does: This paper proposes Dyn-HaMR, which can recover the 4D global motion trajectory of both hands from monocular videos captured by a moving camera.
Dynamic Camera Poses and Where to Find Them
Chris Rockwell (NVIDIA), Chen-Hsuan Lin (NVIDIA)
Pose EstimationRobotic IntelligenceTransformerVision Language ModelSimultaneous Localization and MappingOptical FlowVideoBenchmark
🎯 What it does: This paper presents the DynPose-100K dataset, which collects and annotates 100,000 internet videos containing dynamic scenes with estimable camera poses, aiming to advance applications in real video generation, 4D scene reconstruction, and robotic learning.
Dynamic Content Prediction with Motion-aware Priors for Blind Face Video Restoration
Lianxin Xie (South China University of Technology), Hau San Wong (City University of Hong Kong)
RestorationConvolutional Neural NetworkGenerative Adversarial NetworkVideo
🎯 What it does: A blind face video restoration framework named DCP-MP is proposed, which utilizes a motion-aware semantic dictionary to predict and optimize high-quality content of video frames, and enhances the structural information of the generator through a structural feature correction module, ultimately achieving high-fidelity and temporally consistent facial video restoration.
Dynamic Derivation and Elimination: Audio Visual Segmentation with Enhanced Audio Semantics
Chen Liu (University of Queensland), Xin Yu (University of Queensland)
SegmentationTransformerVideoMultimodalityAudio
🎯 What it does: This paper proposes a dynamic derivation and elimination audio-video segmentation framework named DDESeg, which can dynamically derive multiple independent audio source semantics from mixed audio and enhance discrimination, and then filter out audio semantics that do not match the scene through a visually guided dynamic elimination module, achieving more accurate audio-video alignment and segmentation.
Dynamic Group Normalization: Spatio-Temporal Adaptation to Evolving Data Statistics
Yair Smadar (Ariel University), Assaf Hoogi (Ariel University)
ClassificationObject DetectionSegmentationImage
🎯 What it does: Dynamic Group Normalization (DGN) is proposed, which achieves adaptive normalization by dynamically adjusting channel groups during training to better match the statistical characteristics of the data.
Dynamic Integration of Task-Specific Adapters for Class Incremental Learning
Jiashuo Li (Xi'an Jiaotong University), Yihong Gong (Xi'an Jiaotong University)
ClassificationComputational EfficiencyKnowledge DistillationTransformerVision Language ModelImage
🎯 What it does: Proposes a Dynamic Task-specific Adapter (DIA) framework to achieve example-free category incremental learning, addressing task interference and feature drift issues.
Dynamic Motion Blending for Versatile Motion Editing
Nan Jiang (Peking University), Siyuan Huang (State Key Laboratory of General Artificial Intelligence)
GenerationData SynthesisPose EstimationRetrievalDiffusion modelVideoText
🎯 What it does: This paper proposes MotionReFit, an end-to-end text-driven human motion editing framework that enables semantic and stylistic editing of body parts and time segments.
Dynamic Neural Surfaces for Elastic 4D Shape Representation and Analysis
Awais Nizamani (Murdoch University), Anuj Srivastava (Florida State University)
OptimizationMesh
🎯 What it does: This paper proposes a Dynamic Spherical Neural Surface (D-SNS) that achieves continuous space-time representation of 4D surfaces based on spherical parameterization, allowing for direct execution of spatiotemporal registration, geometric path computation, and mean surface estimation on this representation.
Dynamic Pseudo Labeling via Gradient Cutting for High-Low Entropy Exploration
Jae Hyeon Park (Dongguk University), Sung In Cho (Dongguk University)
ClassificationSegmentationContrastive LearningImage
🎯 What it does: This paper proposes a new Dynamic Pseudo Labeling method that incorporates high-entropy and low-entropy samples into training through Gradient Cutting technology, thereby making fuller use of unlabeled data within a consistency training framework.
Dynamic Stereotype Theory Induced Micro-expression Recognition with Oriented Deformation
Bohao Zhang (East China Normal University), Gaoqi He (East China Normal University)
RecognitionRecurrent Neural NetworkContrastive LearningVideo
🎯 What it does: This paper proposes SODA4MER, a micro-expression recognition method that does not require vertex annotations. It utilizes a self-supervised deformation estimator, muscle group priors, contrastive learning, and dynamic stiffness theory to achieve local deformation capture and biphasic temporal modeling.
Dynamic Updates for Language Adaptation in Visual-Language Tracking
Xiaohai Li (Guangxi Normal University), Shuxiang Song (Guangxi Normal University)
Object TrackingDomain AdaptationTransformerLarge Language ModelVision Language ModelImageVideoMultimodality
🎯 What it does: The DUTrack framework is proposed to enhance tracking robustness by dynamically updating multimodal reference information in visual-language tracking.
DynamicScaler: Seamless and Scalable Video Generation for Panoramic Scenes
Jinxiu Liu (South China University of Technology), Ming-Hsuan Yang (Google DeepMind)
GenerationData SynthesisDiffusion modelVideoText
🎯 What it does: Proposes the DynamicScaler framework, achieving training-free, scalable dynamic image generation for any aspect ratio and 360° panoramas;
DynaMoDe-NeRF: Motion-aware Deblurring Neural Radiance Field for Dynamic Scenes
Ashish Kumar (Indian Institute of Technology Madras), Rajagopalan A. N. (Indian Institute of Technology Madras)
RestorationNeural Radiance FieldVideo
🎯 What it does: This paper studies a method for removing motion blur from dynamic scenes and rendering new viewpoints using multi-view videos.
DynFocus: Dynamic Cooperative Network Empowers LLMs with Video Understanding
Yudong Han (Beijing Institute of Technology), Ming Yang (Huawei)
CompressionComputational EfficiencyTransformerLarge Language ModelVision Language ModelVideoText
🎯 What it does: A low-cost and efficient video encoding framework is constructed for long videos, enabling large language models (LLMs) to significantly reduce the number of tokens for video frames while maintaining visual details, thus achieving efficient video understanding.
DynPose: Largely Improving the Efficiency of Human Pose Estimation by a Simple Dynamic Framework
Yalong Xu (Nanjing University of Science and Technology), Nannan Wang (Xidian University)
Pose EstimationComputational EfficiencyKnowledge DistillationConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes DynPose, a dynamic framework based on a lightweight router that dynamically selects lightweight or heavyweight networks for human pose estimation based on pose difficulty, significantly improving inference speed.
DynRefer: Delving into Region-level Multimodal Tasks via Dynamic Resolution
Yuzhong Zhao (University of Chinese Academy of Sciences), Fang Wan (University of Chinese Academy of Sciences)
RecognitionObject DetectionTransformerLarge Language ModelVision Language ModelImageMultimodality
🎯 What it does: We propose DynRefer, a method for multi-task region-level visual language reasoning achieved through dynamic resolution (simulating human retinal focus and saccade).
DynScene: Scalable Generation of Dynamic Robotic Manipulation Scenes for Embodied AI
Sangmin Lee (Soongsil University), Heewon Kim (Soongsil University)
GenerationRobotic IntelligenceDiffusion modelTextBenchmark
🎯 What it does: Proposes the DynScene framework, which can generate dynamic robot operation scenes from textual instructions;
EAP-GS: Efficient Augmentation of Pointcloud for 3D Gaussian Splatting in Few-shot Scene Reconstruction
Dongrui Dai (Tsinghua University), Yuxiang Xing (Tsinghua University)
RestorationData SynthesisComputational EfficiencyGaussian SplattingSimultaneous Localization and MappingPoint Cloud
🎯 What it does: A point cloud augmentation technique based on attention (APA) is proposed, which improves the initialization of 3D Gaussian splatting under conditions of limited views, thereby enhancing the quality of scene reconstruction.
Early-Bird Diffusion: Investigating and Leveraging Timestep-Aware Early-Bird Tickets in Diffusion Models for Efficient Training
Lexington Whalen (Georgia Institute of Technology), Yingyan Lin
GenerationComputational EfficiencyDiffusion modelImage
🎯 What it does: This study proposes an efficient diffusion model training framework, EB-Diff-Train, which utilizes early 'early bird' sparse subnetworks. It achieves a training speed increase of over 10 times without the need to fully train a dense model, while maintaining generation quality.
EarthDial: Turning Multi-sensory Earth Observations to Interactive Dialogues
Sagar Soni (IBM Research), Salman Khan (Mohamed bin Zayed University of AI)
ClassificationObject DetectionTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageMultimodality
🎯 What it does: This paper presents EarthDial, a conversational visual-language model specifically designed for Earth observation (EO) data, capable of processing multispectral, multi-temporal, and multi-resolution remote sensing images to perform tasks such as classification, detection, annotation, question answering, and visual reasoning.
EASEMVC:Efficient Dual Selection Mechanism for Deep Multi-View Clustering
Baili Xiao (National University of Defense Technology), Xinwang Liu (National University of Defense Technology)
OptimizationRepresentation LearningContrastive LearningImage
🎯 What it does: The EASEMVC framework is proposed to address the representation degradation problem caused by inconsistent view pairing and low-quality views in multi-view clustering. It designs a dual selection mechanism for view selection and sample weighting, achieving end-to-end training by combining contrastive learning and clustering loss.
Easy-editable Image Vectorization with Multi-layer Multi-scale Distributed Visual Feature Embedding
Ye Chen (Shanghai Jiao Tong University), Bingbing Ni (USC-SJTU Institute of Cultural and Creative Industry)
Image TranslationGenerationCompressionImage
🎯 What it does: This paper proposes a vectorized image representation based on multi-layer multi-scale proxy geometric nodes, which can embed texture features within a single shape layer, achieving an editable representation of high-quality natural images.
EasyCraft: A Robust and Efficient Framework for Automatic Avatar Crafting
Suzhen Wang (Netease Fuxi AI Lab), Xin Yu (University of Queensland)
Image TranslationGenerationTransformerDiffusion modelImageText
🎯 What it does: The EasyCraft framework is proposed to achieve automated avatar creation based on images or text, and it can be seamlessly transferred between different game engines.
EasyHOI: Unleashing the Power of Large Models for Reconstructing Hand-Object Interactions in the Wild
Yumeng Liu (University of Hong Kong), Wenping Wang (Texas A&M University)
Pose EstimationOptimizationDiffusion modelImage
🎯 What it does: Reconstructing 3D models of hand-object interactions from single-view images using large pre-trained models and prior-guided optimization.
EBS-EKF: Accurate and High Frequency Event-based Star Tracking
Albert W. Reed (Kitware), Scott McCloskey (Kitware)
Object TrackingPose EstimationSimultaneous Localization and MappingVideo
🎯 What it does: A star map tracking algorithm based on event cameras, EBS-EKF, is proposed, which can estimate camera pose in real-time under low light conditions.
ECBench: Can Multi-modal Foundation Models Understand the Egocentric World? A Holistic Embodied Cognition Benchmark
Ronghao Dang (Alibaba DAMO Academy), Lidong Bing (Alibaba DAMO Academy)
Robotic IntelligenceTransformerLarge Language ModelVision Language ModelVideoTextMultimodalityBenchmark
🎯 What it does: This paper presents ECBench, a comprehensive human perspective question-answering benchmark that covers static, dynamic scenes, and hallucinations, aimed at systematically evaluating the capabilities of large visual language models in embodied cognition tasks.
EchoMatch: Partial-to-Partial Shape Matching via Correspondence Reflection
Yizheng Xie (Technical University of Munich), Daniel Cremers (Technical University of Munich)
SegmentationRetrievalDiffusion modelContrastive LearningImagePoint Cloud
🎯 What it does: Proposes the EchoMatch framework, which implements part-to-part shape matching and predicts overlapping areas using corresponding reflection methods.
EchoMimicV2: Towards Striking, Simplified, and Semi-Body Human Animation
Rang Meng (Ant Group), Chenguang Ma (Ant Group)
GenerationData SynthesisPose EstimationDiffusion modelVideoBenchmarkAudio
🎯 What it does: Proposes the EchoMimicV2 method, which generates high-quality upper-body animation videos using audio, reference images, and gesture sequences;
EchoONE: Segmenting Multiple Echocardiography Planes in One Model
Jiongtong Hu (Shenzhen University), Dong Ni (Shenzhen People's Hospital)
SegmentationConvolutional Neural NetworkImageBiomedical DataUltrasound
🎯 What it does: A unified model EchoONE has been designed and implemented, capable of performing structural segmentation of multi-plane echocardiograms within a single model, utilizing the SAM framework, PC-Mask semantic mask prompts, and LFFA local feature fusion.
EchoTraffic: Enhancing Traffic Anomaly Understanding with Audio-Visual Insights
Zhenghao Xing (Chinese University of Hong Kong), Pheng-Ann Heng (Chinese University of Hong Kong)
Anomaly DetectionAutonomous DrivingTransformerLarge Language ModelVision Language ModelVideoMultimodalityAudio
🎯 What it does: This paper presents EchoTraffic, a multimodal large language model that integrates audio and visual information for understanding traffic anomalies.
EchoWorld: Learning Motion-Aware World Models for Echocardiography Probe Guidance
Yang Yue (Tsinghua University), Gao Huang (Tsinghua University)
SegmentationPose EstimationTransformerContrastive LearningWorld ModelImageBiomedical DataUltrasound
🎯 What it does: The EchoWorld framework is proposed, achieving precise guidance of ultrasound probes through world model pre-training and motion-aware attention mechanisms.
ECVC: Exploiting Non-Local Correlations in Multiple Frames for Contextual Video Compression
Wei Jiang (Bytedance), Li Zhang (Bytedance)
CompressionOptical FlowVideo
🎯 What it does: A context video compression method based on multi-frame non-local correlation (ECVC) is proposed, which combines local and non-local contexts to enhance inter-prediction and compression performance.
EDCFlow: Exploring Temporally Dense Difference Maps for Event-based Optical Flow Estimation
Daikun Liu (Southeast University), Changyin Sun (Southeast University)
Autonomous DrivingComputational EfficiencyConvolutional Neural NetworkOptical FlowVideo
🎯 What it does: A lightweight event camera optical flow network EDCFlow is designed, which combines time-domain dense differential features with low-resolution associations to achieve event optical flow estimation at higher resolutions.
EDEN: Enhanced Diffusion for High-quality Large-motion Video Frame Interpolation
Zihao Zhang (Fudan University), Zuxuan Wu (Fudan University)
RestorationGenerationData SynthesisTransformerDiffusion modelVideo
🎯 What it does: This paper proposes a diffusion model named EDEN for high-quality large motion video frame interpolation.
Edge-SD-SR: Low Latency and Parameter Efficient On-device Super-Resolution with Stable Diffusion via Bidirectional Conditioning
Isma Hadji (Samsung AI Center), Georgios Tzimiropoulos (Queen Mary University of London)
RestorationSuper ResolutionDiffusion modelAuto EncoderImage
🎯 What it does: This paper presents Edge-SD-SR, a low-latency, parameter-efficient Stable Diffusion super-resolution model designed for edge devices.
EdgeDiff: Edge-aware Diffusion Network for Building Reconstruction from Point Clouds
Yujun Liu (Shenzhen University), Guorong Cai (Jimei University)
RestorationGenerationTransformerDiffusion modelPoint Cloud
🎯 What it does: This paper proposes EdgeDiff, a noise-to-wireframe paradigm based on a conditional diffusion model for reconstructing 3D wireframe models of buildings from point clouds.
EdgeMovingNet: Edge-preserving Point Cloud Reconstruction via Joint Geometry Features
Xinran Yang (Nanjing University), Yanwen Guo (Nanjing University)
Point Cloud
🎯 What it does: Proposes EdgeMovingNet, which jointly predicts the direction, distance, and normal of points to edges, generating accurate edge points for point cloud reconstruction.
EdgeTAM: On-Device Track Anything Model
Chong Zhou (Meta AI), Bilge Soran (Feeling AI)
Object TrackingSegmentationComputational EfficiencyKnowledge DistillationTransformerImageVideo
🎯 What it does: Proposes EdgeTAM, a method for efficiently implementing SAM 2 video object tracking and segmentation on edge devices.
Edit Away and My Face Will not Stay: Personal Biometric Defense against Malicious Generative Editing
Hanhui Wang (University of Southern California), Zhengzhong Tu (Texas A&M University)
Safty and PrivacyAdversarial AttackDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes FACELOCK, a privacy protection method that eliminates facial biometric information after editing with diffusion models by generating adversarial perturbations.
EditAR: Unified Conditional Generation with Autoregressive Models
Jiteng Mu (University of California San Diego), Xiaolong Wang (NVIDIA)
Image TranslationSegmentationGenerationKnowledge DistillationTransformerImage
🎯 What it does: A unified autoregressive model called EditAR is proposed, capable of simultaneously completing various conditional image generation tasks, including image editing, depth-to-image, edge-to-image, and segmentation-to-image.
EditSplat: Multi-View Fusion and Attention-Guided Optimization for View-Consistent 3D Scene Editing with 3D Gaussian Splatting
Dong In Lee (Korea University), Sangpil Kim (Korea University)
GenerationOptimizationDiffusion modelGaussian SplattingPoint Cloud
🎯 What it does: This paper proposes a text-driven 3D scene editing framework called EditSplat based on a 3D Gaussian Splatting model;
EDM: Equirectangular Projection-Oriented Dense Kernelized Feature Matching
Dongki Jung (NAVER LABS), Suyong Yeon (NAVER LABS)
Pose EstimationSimultaneous Localization and MappingImage
🎯 What it does: The first learning-based dense feature matching method for panoramic images, EDM, is proposed, utilizing a spherical camera model for matching and pose estimation.
EEE-Bench: A Comprehensive Multimodal Electrical And Electronics Engineering Benchmark
Ming Li (University of Tokyo), Konstantinos Psounis (University of Southern California)
Large Language ModelPrompt EngineeringImageMultimodalityBenchmarkChain-of-Thought
🎯 What it does: A multimodal benchmark EEE-Bench has been constructed in the field of electrical and electronic engineering, containing 2860 questions covering 10 subfields, including technical images such as circuit diagrams and waveforms.
Effective Cloud Removal for Remote Sensing Images by an Improved Mean-Reverting Denoising Model with Elucidated Design Space
Yi Liu (Tongji University), Yichao Zhang (Fudan University)
RestorationTransformerDiffusion modelImageMultimodalityStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: A cloud removal method based on the Mean Regression Diffusion Model (EMRDM) is proposed, constructing a modular framework.
Effective SAM Combination for Open-Vocabulary Semantic Segmentation
Minhyeok Lee (Yonsei University), Sangyoun Lee (Yonsei University)
SegmentationTransformerVision Language ModelImage
🎯 What it does: A single-stage open vocabulary semantic segmentation model ESC-Net is proposed, which integrates CLIP visual-language alignment and SAM decoder to achieve pixel-level classification without region proposals.
Efficient ANN-Guided Distillation: Aligning Rate-based Features of Spiking Neural Networks through Hybrid Block-wise Replacement
Shu Yang (Zhejiang University), Erping Li (Zhejiang University)
Computational EfficiencyKnowledge DistillationSpiking Neural NetworkImage
🎯 What it does: This paper proposes an ANN-guided SNN distillation framework based on rate representation, which gradually aligns the intermediate feature space of ANN and SNN during SNN training using a block-wise replacement strategy, ultimately achieving efficient low-timestep inference.
Efficient Data Driven Mixture-of-Expert Extraction from Trained Networks
Uranik Berisha (Robert Bosch GmbH), Alexandru Paul Condurache (University of Lubeck)
Computational EfficiencyKnowledge DistillationTransformerMixture of ExpertsImage
🎯 What it does: On the existing Vision Transformer (ViT), a Mixture-of-Experts (MoE) variant is formed by extracting expert subnetworks after training with clustered activation patterns.
Efficient Decoupled Feature 3D Gaussian Splatting via Hierarchical Compression
Zhenqi Dai (Northwestern Polytechnical University), Yanning Zhang (Northwestern Polytechnical University)
SegmentationCompressionComputational EfficiencyAuto EncoderGaussian SplattingPoint Cloud
🎯 What it does: This paper proposes DF-3DGS, which decouples color and semantic fields, significantly reducing the number of 3D Gaussians and storage required for the semantic field through hierarchical compression, while improving semantic segmentation performance.
Efficient Depth Estimation for Unstable Stereo Camera Systems on AR Glasses
Yongfan Liu (University of California), Hyoukjun Kwon (University of California)
Depth EstimationImage
🎯 What it does: Achieve low-latency, real-time stereo depth estimation on AR glasses, proposing two models: MultiHeadDepth (optimized cost volume) and HomoDepth (eliminating preprocessing).
Efficient Diffusion as Low Light Enhancer
Guanzhou Lan (Northwestern Polytechnical University), Bin Zhao (Northwestern Polytechnical University)
RestorationGenerationKnowledge DistillationDiffusion modelImage
🎯 What it does: This paper proposes a low-light image enhancement diffusion model distillation framework named ReDDiT, aimed at significantly reducing the number of sampling steps while maintaining high-quality output.
Efficient Dynamic Scene Editing via 4D Gaussian-based Static-Dynamic Separation
Joohyun Kwon (DGIST), Junmo Kim (KAIST)
Image TranslationComputational EfficiencyKnowledge DistillationScore-based ModelGaussian SplattingVideo
🎯 What it does: We propose Instruct-4DGS, a 4D dynamic scene instruction editing method that only edits static 3D Gaussians and refines the temporal aspect through score distillation.
Efficient Event-Based Object Detection: A Hybrid Neural Network with Spatial and Temporal Attention
Soikat Hasan Ahmed (Forschungszentrum Julich), Emre Neftci (Forschungszentrum Julich)
Object DetectionAutonomous DrivingComputational EfficiencyConvolutional Neural NetworkRecurrent Neural NetworkSpiking Neural NetworkVideo
🎯 What it does: A hybrid SNN-ANN attention bridge model is proposed for target detection in event cameras, combining high temporal SNN layers with low temporal ANN layers (optionally DWConvLSTM) for efficient detection.
Efficient Fine-Tuning and Concept Suppression for Pruned Diffusion Models
Reza Shirkavand (University of Maryland), Heng Huang (University of Maryland)
GenerationOptimizationKnowledge DistillationDiffusion modelImage
🎯 What it does: A two-layer optimization framework is proposed for efficient fine-tuning on pruned diffusion models while simultaneously suppressing undesirable concepts.
Efficient Long Video Tokenization via Coordinate-based Patch Reconstruction
Huiwon Jang (KAIST), Younggyo Seo (UC Berkeley)
GenerationCompressionTransformerVideo
🎯 What it does: A video tokenizer named CoordTok has been developed, capable of learning to map long videos to low-dimensional discrete/continuous tokens through coordinate mapping, achieving efficient compression of 128-frame videos.
Efficient Motion-Aware Video MLLM
Zijia Zhao (Institute of Automation Chinese Academy of Sciences), Jing Liu (Baichuan Inc.)
CompressionComputational EfficiencyTransformerLarge Language ModelVision Language ModelVideoTextMultimodality
🎯 What it does: This paper proposes EMA, an efficient motion-aware video multimodal large language model that utilizes the structure of compressed video to reduce visual input redundancy while preserving motion information.
Efficient Personalization of Quantized Diffusion Model without Backpropagation
Hoigi Seo (Seoul National University), Se Young Chun (Seoul National University)
GenerationOptimizationComputational EfficiencyDiffusion modelImage
🎯 What it does: This paper proposes an efficient method for personalized quantization of diffusion models.
Efficient Test-time Adaptive Object Detection via Sensitivity-Guided Pruning
Kunyu Wang (University of Science and Technology of China), Zheng-Jun Zha (University of Science and Technology of China)
Object DetectionDomain AdaptationAutonomous DrivingComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: A sensitivity-guided channel pruning adaptive object detection framework for continuous testing is proposed, aiming to enhance adaptation efficiency and computational resource utilization by suppressing and pruning channels that negatively impact performance in the target domain.
Efficient Transfer Learning for Video-language Foundation Models
Haoxing Chen (Ant Group), Zhangxuan Gu (Ant Group)
ClassificationRecognitionTransformerLarge Language ModelContrastive LearningVideoMultimodality
🎯 What it does: A multi-modal spatio-temporal adapter (MSTA) is proposed for efficient transfer learning of video-language foundational models, achieving parameter-efficient adaptation for video action recognition tasks.
Efficient Video Face Enhancement with Enhanced Spatial-Temporal Consistency
Yutong Wang (Beijing Institute of Technology), Dixin Luo (Beijing Institute of Technology)
RestorationTransformerGenerative Adversarial NetworkVideo
🎯 What it does: A two-stage video face enhancement framework based on 3D-VQGAN is proposed, which can simultaneously achieve blind video face restoration and de-ghosting.
Efficient Video Super-Resolution for Real-time Rendering with Decoupled G-buffer Guidance
Mingjun Zheng (Nanjing University of Science and Technology), Jinshan Pan (Nanjing University of Science and Technology)
RestorationSuper ResolutionConvolutional Neural NetworkVideo
🎯 What it does: Designed and implemented an asynchronous UNet network RDG for real-time rendering, utilizing decoupled G-buffer guidance to achieve detail-rich and temporally stable video super-resolution;
Efficient Visual State Space Model for Image Deblurring
Lingshun Kong (Nanjing University of Science and Technology), Jinshan Pan (Nanjing University of Science and Technology)
RestorationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes an Efficient Visual State Space Model (EVSSM) for image deblurring, combining geometric transformations with SSM and frequency domain discriminative FFN to achieve an end-to-end deblurring network.
EfficientLLaVA: Generalizable Auto-Pruning for Large Vision-language Models
Yinan Liang (Tsinghua University), Jiwen Lu (Tsinghua University)
OptimizationComputational EfficiencyTransformerVision Language ModelMultimodality
🎯 What it does: This paper proposes EfficientLLaVA, an automatic pruning method for large visual-language models (LVLM), which achieves efficient inference with a minimal number of proxy samples without significantly losing task performance.
EfficientViM: Efficient Vision Mamba with Hidden State Mixer based State Space Duality
Sanghyeok Lee (Korea University), Hyunwoo J. Kim (KAIST)
ClassificationObject DetectionSegmentationConvolutional Neural NetworkImage
🎯 What it does: A lightweight visual network called EfficientViM is proposed, which efficiently captures global context using Hidden State Mixing with State Space Duality (HSM-SSD) and enhances representation capability through multi-stage hidden state fusion.
EffiDec3D: An Optimized Decoder for High-Performance and Efficient 3D Medical Image Segmentation
Md Mostafijur Rahman (University of Texas), Radu Marculescu (University of Texas)
SegmentationComputational EfficiencyConvolutional Neural NetworkBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper proposes an efficient 3D decoder, EffiDec3D, aimed at significantly reducing the parameter count and computational requirements of 3D medical image segmentation models.
Effortless Active Labeling for Long-Term Test-Time Adaptation
Guowei Wang (South China University of Technology), Changxing Ding (South China University of Technology)
Domain AdaptationImage
🎯 What it does: An adaptive method called EATTA for long-term testing is proposed, which requires labeling at most one sample per batch, utilizes feature perturbation to identify boundary samples, and achieves a balance between supervised and unsupervised objectives through dynamic weighting of gradient norms.
Ego4o: Egocentric Human Motion Capture and Understanding from Multi-Modal Input
Jian Wang (Max Planck Institute for Informatics), Christian Theobalt (Max Planck Institute for Informatics)
GenerationPose EstimationTransformerLarge Language ModelMultimodality
🎯 What it does: The Ego4o framework is proposed, utilizing IMU, first-person images, and optional textual descriptions for 3D motion capture and generating motion descriptions.
EgoLife: Towards Egocentric Life Assistant
Jingkang Yang (Nanyang Technological University), Ziwei Liu (Nanyang Technological University)
TransformerLarge Language ModelSupervised Fine-TuningVideoTextMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: Created the EgoLife dataset, a 300-hour long-term first-person multimodal daily life dataset, and designed the EgoLifeQA evaluation task and EgoButler system based on this dataset, achieving long-context life assistant functionality.
EgoLM: Multi-Modal Language Model of Egocentric Motions
Fangzhou Hong (Meta Reality Labs Research), Lingni Ma (Meta Reality Labs Research)
Object TrackingGenerationPose EstimationTransformerLarge Language ModelSupervised Fine-TuningVideoTextMultimodality
🎯 What it does: This paper presents EgoLM, a multimodal large language model that integrates wearable sparse motion sensors with first-person video for tasks such as full-body motion tracking, motion narration, and text-to-motion generation.
EgoPressure: A Dataset for Hand Pressure and Pose Estimation in Egocentric Vision
Yiming Zhao (ETH Zürich), Christian Holz (ETH Zürich)
Pose EstimationTransformerImageMeshBenchmark
🎯 What it does: Collected and released the EgoPressure dataset, which includes egocentric perspective hand pressure and 3D pose annotations, and provides benchmark models and evaluation results.
EgoTextVQA: Towards Egocentric Scene-Text Aware Video Question Answering
Sheng Zhou (University of Science and Technology of China), Angela Yao (National University of Singapore)
Autonomous DrivingTransformerLarge Language ModelVision Language ModelVideoTextMultimodality
🎯 What it does: The EgoTextVQA dataset has been constructed, containing 1.5K perspective videos and 7K real scene text-related QA pairs, focusing on outdoor driving and indoor housekeeping tasks.
EIDT-V: Exploiting Intersections in Diffusion Trajectories for Model-Agnostic, Zero-Shot, Training-Free Text-to-Video Generation
Diljeet Jagpal (University of Bath), Vinay P. Namboodiri (University of Bath)
GenerationData SynthesisLarge Language ModelDiffusion modelVideoText
🎯 What it does: A training-free, model-agnostic text-to-video generation method is proposed, achieving temporally consistent video generation through diffusion trajectory crossing and grid prompt switching.
EigenGS Representation: From Eigenspace to Gaussian Image Space
Lo-Wei Tai (National Tsing Hua University), Tyng-Luh Liu (Academia Sinica)
RestorationDomain AdaptationRepresentation LearningGaussian SplattingImage
🎯 What it does: Proposes the EigenGS method, which utilizes PCA's eigenvectors to learn a 2D Gaussian mixture model for rapid initialization and reconstruction of images;
Electromyography-Informed Facial Expression Reconstruction for Physiological-Based Synthesis and Analysis
Tim Büchner (Friedrich Schiller University Jena), Joachim Denzler (Friedrich Schiller University Jena)
Image TranslationGenerationData SynthesisGenerative Adversarial NetworkVideoMultimodality
🎯 What it does: This paper proposes a framework for electrode-occluded facial expression reconstruction called EIFER, based on 3DMM and unsupervised CycleGAN, which can recover facial geometry and appearance under the condition of synchronously collected facial expressions and electromyographic (sEMG) signals, achieving bidirectional mapping from sEMG to expression and from expression to sEMG.
Embodied Scene Understanding for Vision Language Models via MetaVQA
Weizhen Wang (University of California), Bolei Zhou (University of California)
Autonomous DrivingMeta LearningTransformerSupervised Fine-TuningPrompt EngineeringVision Language ModelMultimodalityBenchmark
🎯 What it does: A MetaVQA benchmark is proposed to evaluate and enhance the capabilities of Vision Language Models (VLM) in embedded scene understanding, and it is fine-tuned to serve as a self-driving planner evaluated in closed-loop simulations.
Embracing Collaboration Over Competition: Condensing Multiple Prompts for Visual In-Context Learning
Jinpeng Wang (Tsinghua University), Shu-Tao Xia (Artificial Intelligence Research Center)
Object DetectionSegmentationPrompt EngineeringImage
🎯 What it does: This paper proposes a multi-prompt compression method in visual context learning, which integrates multiple candidate prompts into a single fine-grained prompt through a lightweight Condencer module, thereby enhancing the performance of visual models in tasks such as foreground segmentation, single object detection, and image coloring.
EmoDubber: Towards High Quality and Emotion Controllable Movie Dubbing
Gaoxiang Cong (Institute of Computing Technology, Chinese Academy of Sciences), Qingming Huang (Institute of Computing Technology, Chinese Academy of Sciences)
GenerationData SynthesisFlow-based ModelContrastive LearningVideoAudio
🎯 What it does: A controllable emotional movie dubbing model called EmoDubber is proposed, addressing three major challenges: audio-video synchronization, clear pronunciation, and controllable emotions.
EMOE: Modality-Specific Enhanced Dynamic Emotion Experts
Yiyang Fang (Wuhan University), Mang Ye (Wuhan University)
RecognitionKnowledge DistillationTransformerMixture of ExpertsMultimodality
🎯 What it does: A dynamic modality weight allocation and unimodal distillation multimodal emotion recognition framework EMOE based on Mixture of Experts is proposed.
EmoEdit: Evoking Emotions through Image Manipulation
Jingyuan Yang (Shenzhen University), Hui Huang (Hebrew University of Jerusalem)
GenerationData SynthesisTransformerLarge Language ModelDiffusion modelImage
🎯 What it does: The EmoEdit framework is proposed, utilizing emotional words as prompts to edit the content of user-provided images to evoke target emotions; a large-scale AIM dataset, EmoEditSet, is constructed, along with the design of an Emotion Adapter and instruction loss.
EmotiveTalk: Expressive Talking Head Generation through Audio Information Decoupling and Emotional Video Diffusion
Haotian Wang (University of Science and Technology of China), Qingfeng Liu (iFLYTEK)
GenerationData SynthesisTransformerDiffusion modelVideoMultimodalityAudio
🎯 What it does: Proposes the EmotiveTalk framework, which utilizes vision-guided audio information decoupling and a diffusion model to generate controllable emotional talking head videos from a single portrait and audio.
EMOVA: Empowering Language Models to See, Hear and Speak with Vivid Emotions
Kai Chen (Hong Kong University of Science and Technology), Hang Xu (Huawei Noah's Ark Lab)
RecognitionGenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodalityAudio
🎯 What it does: An end-to-end multimodal large language model EMOVA has been developed, capable of simultaneously perceiving visual, textual, and auditory information, and supporting emotional voice conversations.
Emphasizing Discriminative Features for Dataset Distillation in Complex Scenarios
Kai Wang (National University of Singapore), Yang You
Knowledge DistillationConvolutional Neural NetworkImageBenchmark
🎯 What it does: The EDF (Emphasize Discriminative Features) method is proposed, which utilizes Grad-CAM activation mapping to dynamically enhance discriminative regions and employs Common Pattern Dropout to eliminate low-loss gradients, thereby improving dataset distillation; at the same time, a CompDD benchmark aimed at complex scenes is constructed.
Empowering Large Language Models with 3D Situation Awareness
Zhihao Yuan (Chinese University of Hong Kong Shenzhen), Zhen Li (Chinese University of Hong Kong Shenzhen)
RecognitionObject DetectionPose EstimationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageMultimodalityPoint Cloud
🎯 What it does: Proposes the View2Cap dataset and Situation Grounding module to enhance the situational awareness capability of LLMs in 3D perspectives.
Empowering LLMs to Understand and Generate Complex Vector Graphics
Ximing Xing (Beihang University), Qian Yu (Beihang University)
GenerationTransformerLarge Language ModelSupervised Fine-TuningImageText
🎯 What it does: Developed the LLM4SVG framework, enabling large language models to understand and generate SVG graphics.
Empowering Vector Graphics with Consistently Arbitrary Viewing and View-dependent Visibility
Yidi Li (University of Chinese Academy of Sciences), Haiyong Jiang (University of Chinese Academy of Sciences)
GenerationData SynthesisOptimizationDiffusion modelGaussian SplattingTextPoint Cloud
🎯 What it does: A text-based vector graphic generation framework called Dream3DVG is proposed, which allows for arbitrary viewpoints and consistent visibility dependent on the viewpoint.
Encapsulated Composition of Text-to-Image and Text-to-Video Models for High-Quality Video Synthesis
Tongtong Su (Zhejiang University), Dongming Lu (Zhejiang University)
GenerationData SynthesisDiffusion modelVideoTextStochastic Differential Equation
🎯 What it does: A training-independent encapsulated video synthesis framework EVS is proposed, which can simultaneously improve the image quality and motion consistency of text-to-video (T2V) generation.
End-to-End HOI Reconstruction Transformer with Graph-based Encoding
Zhenrong Wang (Shenzhen University), Dongjiang Li (JD Explore Academy)
Object DetectionPose EstimationGraph Neural NetworkTransformerImageMesh
🎯 What it does: An end-to-end HOI Reconstruction Transformer (HOI-TG) is proposed, capable of simultaneously reconstructing 3D human and object meshes from a single RGB image, and implicitly learning the interactions between the two through self-attention.
End-to-End Implicit Neural Representations for Classification
Alexander Gielisse (Delft University of Technology), Jan van Gemert (Delft University of Technology)
ClassificationMeta LearningTransformerImage
🎯 What it does: An end-to-end implicit neural representation (SIREN) classification framework is proposed, which simultaneously updates SIREN parameters and the Transformer classifier during the training loop.
Enduring, Efficient and Robust Trajectory Prediction Attack in Autonomous Driving via Optimization-Driven Multi-Frame Perturbation Framework
Yi Yu (Wuhan University), Zhuangzhuang Zhang (City University of Hong Kong)
Autonomous DrivingOptimizationAdversarial AttackPoint Cloud
🎯 What it does: A multi-frame attack framework based on LiDAR, OMP-Attack, is proposed, which can continuously induce trajectory prediction errors on multi-frame historical trajectories.
EnergyMoGen: Compositional Human Motion Generation with Energy-Based Diffusion Model in Latent Space
Jianrong Zhang (University of Technology Sydney), Yi Yang (Zhejiang University)
GenerationData SynthesisTransformerDiffusion modelAuto EncoderText
🎯 What it does: Proposes EnergyMoGen, a framework that combines latent diffusion models with energy models for text-driven multi-concept human action generation.
Enhanced Contrastive Learning with Multi-view Longitudinal Data for Chest X-ray Report Generation
Kang Liu (Xidian University), Qiguang Miao (Xidian University)
GenerationTransformerLarge Language ModelContrastive LearningImageTextMultimodalityElectronic Health Records
🎯 What it does: This study proposes a multi-view longitudinal contrastive learning framework (MLRG) that integrates current multi-view X-ray images with patients' historical longitudinal data. It utilizes the spatiotemporal information in reports for visual and textual pre-training, and introduces tokenized absence encoding during the report generation phase to handle missing prior knowledge, thereby generating more accurate and coherent chest X-ray reports.
Enhanced OoD Detection through Cross-Modal Alignment of Multi-Modal Representations
Jeonghyeon Kim (Seoul National University of Science and Technology), Sangheum Hwang (Seoul National University of Science and Technology)
ClassificationAnomaly DetectionSupervised Fine-TuningVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: This paper proposes a Cross-Modal Alignment (CMA) multimodal fine-tuning method for multimodal visual language models (such as CLIP) to enhance out-of-domain detection (OoDD) and ID classification performance.