CVPR 2024 Papers — Page 8
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2716 papers
Edge-Aware 3D Instance Segmentation Network with Intelligent Semantic Prior
Wonseok Roh (Korea University), Sangpil Kim (Korea University)
Object DetectionSegmentationTransformerPoint Cloud
🎯 What it does: This paper proposes a Transformer-based 3D instance segmentation framework called EASE, which achieves precise segmentation of instances in complex scenes by introducing a semantic guidance network and an edge prediction module.
Edit One for All: Interactive Batch Image Editing
Thao Nguyen (University of Wisconsin Madison), Yong Jae Lee (University of Wisconsin Madison)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: Proposes an interactive batch image editing method that utilizes StyleGAN to transfer the editing results of a single image to multiple images, achieving a unified final state.
Editable Scene Simulation for Autonomous Driving via Collaborative LLM-Agents
Yuxi Wei (Shanghai Jiao Tong University), Yanfeng Wang (Shanghai Jiao Tong University)
Autonomous DrivingLarge Language ModelNeural Radiance FieldImage
🎯 What it does: The ChatSim system is proposed, which can edit and render editable high-fidelity 3D driving scenes through natural language commands and supports the import of external digital assets.
EditGuard: Versatile Image Watermarking for Tamper Localization and Copyright Protection
Xuanyu Zhang (Peking University), Jian Zhang (Peking University)
Image TranslationSafty and PrivacyTransformerFlow-based ModelImage
🎯 What it does: Proposes an active digital watermarking framework called EditGuard that can simultaneously achieve image copyright protection and localized forgery detection.
Effective Video Mirror Detection with Inconsistent Motion Cues
Alex Warren (Swansea University), Rynson W.H. Lau (City University of Hong Kong)
SegmentationAnomaly DetectionTransformerOptical FlowVideo
🎯 What it does: A deep learning framework based on video motion inconsistency detection for mirrors, called MG-VMD, is proposed, and a more challenging mirror video dataset, MMD, is constructed.
Efficient 3D Implicit Head Avatar with Mesh-anchored Hash Table Blendshapes
Ziqian Bai (Google), Yinda Zhang
GenerationPose EstimationComputational EfficiencyConvolutional Neural NetworkNeural Radiance FieldVideoMesh
🎯 What it does: A new 3D head digital human model is proposed, utilizing mesh-anchored hash table Blendshapes for real-time rendering and fine-grained controllability.
Efficient and Effective Weakly-Supervised Action Segmentation via Action-Transition-Aware Boundary Alignment
Angchi Xu (Sun Yat-sen University), Wei-Shi Zheng (Sun Yat-sen University)
SegmentationTransformerContrastive LearningVideo
🎯 What it does: A boundary alignment framework based on action transfer, ATBA, is proposed for weakly supervised action segmentation tasks.
Efficient Dataset Distillation via Minimax Diffusion
Jianyang Gu (Zhejiang University), Yiran Chen (Duke University)
ClassificationData SynthesisKnowledge DistillationDiffusion modelImage
🎯 What it does: Utilize diffusion models to distill large-scale datasets, generating smaller, representative datasets rich in original data information.
Efficient Deformable ConvNets: Rethinking Dynamic and Sparse Operator for Vision Applications
Yuwen Xiong (University of Toronto), Jifeng Dai (Tsinghua University)
Object DetectionSegmentationGenerationConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: This paper proposes Deformable Convolution v4 (DCNv4) as an efficient and deformable sparse convolution operation, replacing the previous DCNv3, forming faster and stronger visual backbone networks such as FlashInternImage.
Efficient Detection of Long Consistent Cycles and its Application to Distributed Synchronization
Shaohan Li (University of Minnesota), Gilad Lerman (University of Minnesota)
OptimizationComputational EfficiencyGraph
🎯 What it does: This paper proposes an efficient long-period consistency detection method called LongSync, which robustly solves the group synchronization problem in global structured light sequencing.
Efficient Hyperparameter Optimization with Adaptive Fidelity Identification
Jiantong Jiang (University of Western Australia), Ajmal Mian (University of Western Australia)
OptimizationHyperparameter SearchTabular
🎯 What it does: FastBO is proposed, an adaptive multi-fidelity Bayesian optimization method that determines suitable low-order precision for each configuration to build surrogate models.
Efficient LoFTR: Semi-Dense Local Feature Matching with Sparse-Like Speed
Yifan Wang (Zhejiang University), Xiaowei Zhou (Zhejiang University)
Pose EstimationComputational EfficiencyTransformerImage
🎯 What it does: We propose an efficient detector-free matcher called Efficient LoFTR based on LoFTR, which significantly improves inference speed while maintaining or enhancing matching accuracy.
Efficient Meshflow and Optical Flow Estimation from Event Cameras
Xinglong Luo (University of Electronic Science and Technology of China), Shuaicheng Liu (University of Electronic Science and Technology of China)
Computational EfficiencyConvolutional Neural NetworkOptical FlowMesh
🎯 What it does: Proposed the Meshflow and optical flow estimation method for event cameras, and released a high-resolution HREM dataset.
Efficient Model Stealing Defense with Noise Transition Matrix
Dong-Dong Wu (Southeast University), Min-Ling Zhang (Southeast University)
OptimizationAdversarial AttackConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A lightweight model stealing defense method called EMMA is proposed based on the Noise Transfer Matrix (NTM), which injects noise into the predicted posterior through linear mapping and explicitly adjusts the interaction between the defense and the attacker's model through a two-layer optimization, thereby weakening the attacker's cloning effect.
Efficient Multi-scale Network with Learnable Discrete Wavelet Transform for Blind Motion Deblurring
Xin Gao (China University of Mining and Technology Beijing), Huaping Liu (University of Science and Technology of China)
RestorationConvolutional Neural NetworkImage
🎯 What it does: A multi-scale network MLWNet based on Single Input Multiple Output (SIMO) is proposed, introducing a learnable discrete wavelet transform (LWN) module for blind motion deblurring.
Efficient Multitask Dense Predictor via Binarization
Yuzhang Shang (Illinois Institute of Technology), Yan Yan (Illinois Institute of Technology)
SegmentationDepth EstimationCompressionComputational EfficiencyKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: Compression and acceleration of multi-task dense prediction through binary networks
Efficient Privacy-Preserving Visual Localization Using 3D Ray Clouds
Heejoon Moon (Hanyang University), Je Hyeong Hong (Hanyang University)
Pose EstimationSafty and PrivacyComputational EfficiencySimultaneous Localization and MappingImagePoint Cloud
🎯 What it does: This paper proposes a privacy-preserving visual localization framework based on 3D ray clouds, achieving real-time localization under a single frame image.
Efficient Scene Recovery Using Luminous Flux Prior
Zhongyu Li (University of Science and Technology of China), Lei Zhang (University of Science and Technology of China)
RestorationImage
🎯 What it does: A scene recovery method called LFP is proposed, which does not require learning and estimates transmittance using the rate of change of luminous flux, achieving image recovery under various adverse weather conditions (fog, dust storms, underwater, etc.).
Efficient Solution of Point-Line Absolute Pose
Petr Hruby (ETH Zurich), Marc Pollefeys (ETH Zurich)
Pose EstimationOptimizationComputational EfficiencyPoint Cloud
🎯 What it does: An algebraic optimal solver is proposed for the two minimum absolute pose estimation problems corresponding to point-line pairs (P2P1L and P1P2L), significantly reducing the polynomial degree and achieving fast solutions.
Efficient Stitchable Task Adaptation
Haoyu He (Monash University), Bohan Zhuang (Monash University)
ClassificationRecognitionDomain AdaptationComputational EfficiencyKnowledge DistillationSupervised Fine-TuningImage
🎯 What it does: The ESTA framework is proposed, which efficiently generates various resource-constrained fine-tuning models in task adaptation scenarios using model stitching.
Efficient Test-Time Adaptation of Vision-Language Models
Adilbek Karmanov (Mohamed bin Zayed University of Artificial Intelligence), Eric Xing (Carnegie Mellon University)
Domain AdaptationComputational EfficiencyTransformerVision Language ModelContrastive LearningImage
🎯 What it does: This paper proposes a training-free, dynamic caching adapter (TDA) that achieves instant adaptation of visual-language models by constructing positive and negative caches using CLIP features and pseudo-labels during testing.
Efficient Vision-Language Pre-training by Cluster Masking
Zihao Wei (University of Michigan), Andrew Owens (University of Michigan)
ClassificationRetrievalRepresentation LearningTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: This paper proposes a clustering-based image patch masking strategy for visual-language contrastive pre-training, which enhances representation quality while maintaining efficiency.
EfficientDreamer: High-Fidelity and Robust 3D Creation via Orthogonal-view Diffusion Priors
Zhipeng Hu (NetEase Fuxi AI Lab), Xin Yu (University of Queensland)
GenerationData SynthesisDiffusion modelScore-based ModelPoint CloudMesh
🎯 What it does: This paper presents EfficientDreamer, a comprehensive method for high-fidelity and stable 3D content generation using orthogonal perspective diffusion priors.
Efficiently Assemble Normalization Layers and Regularization for Federated Domain Generalization
Khiem Le (University of Notre Dame), Kok-Seng Wong (VinUniversity)
Domain AdaptationFederated LearningConvolutional Neural NetworkImageBiomedical Data
🎯 What it does: In federated domain generalization, a personalized normalization method gPerXAN is proposed by explicitly assembling Instance Normalization and Batch Normalization, with the addition of regularization guidance, while maintaining data privacy throughout.
EfficientSAM: Leveraged Masked Image Pretraining for Efficient Segment Anything
Yunyang Xiong (Meta AI Research), Vikas Chandra (Meta AI Research)
SegmentationTransformerAuto EncoderImage
🎯 What it does: A lightweight Segment Anything (EfficientSAM) framework based on the SAM model is proposed, and an efficient ViT encoder is obtained through SAM-guided occluded image pre-training (SAMI);
EFHQ: Multi-purpose ExtremePose-Face-HQ dataset
Trung Tuan Dao, Anh Tran
GenerationPose EstimationGenerative Adversarial NetworkImageVideoBenchmark
🎯 What it does: This study constructs a high-quality facial image dataset named EFHQ for extreme poses and provides multi-task subsets and cross-pose validation benchmarks based on it.
EFormer: Enhanced Transformer towards Semantic-Contour Features of Foreground for Portraits Matting
Zitao Wang (Xidian University), Peipei Zhao (Xidian University)
SegmentationTransformerImageVideo
🎯 What it does: Proposes EFormer, which utilizes cross-resolution self-attention and cross-attention to enhance the capture of low-frequency semantic and high-frequency contour features of human foregrounds, achieving trimap-free portrait matting.
Ego-Exo4D: Understanding Skilled Human Activity from First- and Third-Person Perspectives
Kristen Grauman (Meta), Michael Wray (University of Bristol)
RecognitionPose EstimationVideoMultimodalityPoint CloudBenchmarkAudio
🎯 What it does: A large-scale, multi-view, multi-modal first-person and third-person video dataset called Ego‑Exo4D has been created, along with corresponding benchmark tasks such as cross-view and proficiency assessment.
Egocentric Whole-Body Motion Capture with FisheyeViT and Diffusion-Based Motion Refinement
Jian Wang (Max Planck Institute for Informatics), Christian Theobalt (Max Planck Institute for Informatics)
Pose EstimationTransformerDiffusion modelVideo
🎯 What it does: The paper proposes a full-body motion capture framework based on a monocular fisheye camera, jointly estimating the 3D poses of the body and hands, and refining the estimates using a diffusion model.
EgoExoLearn: A Dataset for Bridging Asynchronous Ego- and Exo-centric View of Procedural Activities in Real World
Yifei Huang (OpenGVLab), Yu Qiao (OpenGVLab)
Domain AdaptationRepresentation LearningData-Centric LearningTransformerContrastive LearningVideoMultimodality
🎯 What it does: This work proposes the EgoExoLearn dataset, which collects 747 synchronized and asynchronous first-person and third-person videos, encompassing 120 hours of daily and laboratory tasks, and provides eye-tracking, fine-grained multimodal annotations, and four cross-view benchmarks (association, prediction, planning, skill assessment).
EgoGen: An Egocentric Synthetic Data Generator
Gen Li (ETH Zurich), Siyu Tang (ETH Zurich)
GenerationData SynthesisReinforcement LearningImageMultimodality
🎯 What it does: This work presents EgoGen, a scalable synthetic data generation system for first-person perspective, which utilizes virtual human motion synthesis and multimodal rendering to generate egocentric data with precise annotations.
EgoThink: Evaluating First-Person Perspective Thinking Capability of Vision-Language Models
Sijie Cheng (Tsinghua University), Yang Liu (Tsinghua University)
Large Language ModelVision Language ModelImageMultimodalityBenchmark
🎯 What it does: Construct the EgoThink benchmark to evaluate the multi-dimensional thinking abilities of Vision-Language models in terms of objects, activities, localization, reasoning, prediction, and planning from a first-person perspective.
EGTR: Extracting Graph from Transformer for Scene Graph Generation
Jinbae Im (NAVER Cloud AI), Seunghyun Park (NAVER)
Object DetectionGenerationTransformerGraph
🎯 What it does: This paper proposes a lightweight one-stage scene graph generation model, EGTR, which directly constructs an inter-object relationship graph using the weights and query/key vectors from the multi-head self-attention in the DETR decoder, and predicts predicates with a shallow relationship classification head; it also introduces adaptive smoothing and connectivity prediction as auxiliary training.
ElasticDiffusion: Training-free Arbitrary Size Image Generation through Global-Local Content Separation
Moayed Haji-Ali (Rice University), Vicente Ordonez (Rice University)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: Proposes ElasticDiffusion, a training-free decoding method that allows pre-trained diffusion models to generate high-quality images at any size.
Elite360D: Towards Efficient 360 Depth Estimation via Semantic- and Distance-Aware Bi-Projection Fusion
Hao Ai (Hong Kong University of Science and Technology), Lin Wang (Hong Kong University of Science and Technology)
Depth EstimationConvolutional Neural NetworkPoint Cloud
🎯 What it does: A new framework named Elite360D is proposed for efficient 360-degree depth estimation, combining a dual-projection fusion method that incorporates semantic and distance awareness.
EMAGE: Towards Unified Holistic Co-Speech Gesture Generation via Expressive Masked Audio Gesture Modeling
Haiyang Liu (University of Tokyo), Michael J. Black (Max Planck Institute for Intelligent Systems)
GenerationPose EstimationTransformerMultimodalityAudio
🎯 What it does: This paper proposes the EMAGE framework, which achieves audio-based full-body coordinated speaking posture generation and enhances generation quality through masked posture reconstruction; it also releases a unified BEAT2 dataset in SMPL-X/FLAME format.
Embodied Multi-Modal Agent trained by an LLM from a Parallel TextWorld
Yijun Yang (Southern University of Science and Technology), Yuhui Shi (Southern University of Science and Technology)
Robotic IntelligenceReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningVision Language ModelTextMultimodalityBenchmark
🎯 What it does: A multimodal embodied agent named EMMA has been constructed, utilizing LLM to train VLM in executing embodied tasks in the visual world based on expert behavior in a parallel text world.
EmbodiedScan: A Holistic Multi-Modal 3D Perception Suite Towards Embodied AI
Tai Wang (Shanghai AI Laboratory), Jiangmiao Pang (Shanghai AI Laboratory)
Object DetectionSegmentationPose EstimationRobotic IntelligenceTransformerVision Language ModelSimultaneous Localization and MappingTextMultimodalityPoint CloudBenchmark
🎯 What it does: Proposes the EmbodiedScan dataset and benchmark, providing 5k perspective-centered RGB-D scans, 1M language prompts, 160k 3D boxes, 80 classes of semantic occupancy, and presents an Embodied Perceptron baseline model capable of handling arbitrary viewpoints.
Embracing Unimodal Aleatoric Uncertainty for Robust Multimodal Fusion
Zixian Gao (University of Electronic Science and Technology of China), Heng Tao Shen (Kyushu Institute of Technology)
ClassificationRecognitionContrastive LearningMultimodality
🎯 What it does: A framework for robust multimodal fusion using aleatoric uncertainty is proposed, named EAU.
EMCAD: Efficient Multi-scale Convolutional Attention Decoding for Medical Image Segmentation
Md Mostafijur Rahman (University of Texas at Austin), Radu Marculescu (University of Texas at Austin)
SegmentationConvolutional Neural NetworkTransformerImageBiomedical Data
🎯 What it does: This paper proposes an efficient multi-scale convolutional attention decoder, EMCAD, for medical image segmentation.
Emergent Open-Vocabulary Semantic Segmentation from Off-the-shelf Vision-Language Models
Jiayun Luo (Nanyang Technological University), Boyang Li (Nanyang Technological University)
SegmentationHyperparameter SearchTransformerVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: This paper proposes a zero-shot, unsupervised open-source vocabulary semantic segmentation method called PnP-OVSS, which directly utilizes the cross-modal attention of a pre-trained vision-language model, GradCAM, and Salience Dropout to generate pixel-level segmentation.
EmoGen: Emotional Image Content Generation with Text-to-Image Diffusion Models
Jingyuan Yang (Shenzhen University), Hui Huang (Shenzhen University)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: This paper proposes the Emotion Image Content Generation (EICG) task, which aims to generate images that clearly express emotions while also possessing diverse semantics.
EMOPortraits: Emotion-enhanced Multimodal One-shot Head Avatars
Nikita Drobyshev (Imperial College London), Maja Pantic (Imperial College London)
GenerationData SynthesisGenerative Adversarial NetworkContrastive LearningImageVideoMultimodalityAudio
🎯 What it does: The EMOPortraits model is proposed, achieving single-shot neural head portrait animation that supports strong, asymmetric expressions and voice-driven animation.
Emotional Speech-driven 3D Body Animation via Disentangled Latent Diffusion
Kiran Chhatre (KTH Royal Institute of Technology), Timo Bolkart (Max Planck Institute for Intelligent Systems)
GenerationPose EstimationTransformerDiffusion modelMultimodalityAudio
🎯 What it does: A system called AMUSE based on latent diffusion models is proposed, which can directly generate 3D human poses with emotional expressions from speech and supports controllable editing of emotions and styles by integrating emotional and style latent variables from different speech inputs.
EmoVIT: Revolutionizing Emotion Insights with Visual Instruction Tuning
Hongxia Xie (Jilin University), Wen-Huang Cheng (National Taiwan University)
RecognitionTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringVision Language ModelImageMultimodality
🎯 What it does: Developed the EmoVIT framework, which utilizes emotion visual instruction data generated by GPT-4 for instruction tuning in visual emotion recognition;
Empowering Resampling Operation for Ultra-High-Definition Image Enhancement with Model-Aware Guidance
Wei Yu (University of Science and Technology of China), Feng Zhao (University of Science and Technology of China)
RestorationSuper ResolutionConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes a 'Model-Aware Resampling' method (LMAR), which predicts compensation convolution kernels specific to each UHD image and downsampling scale, embedding the compensation information into the downsampled image to achieve a synergistic optimization of resampling and enhancement models.
Emu Edit: Precise Image Editing via Recognition and Generation Tasks
Shelly Sheynin (Meta), Yaniv Taigman (Meta)
Image TranslationGenerationTransformerLarge Language ModelDiffusion modelImageTextBenchmark
🎯 What it does: This paper presents Emu Edit, a multi-task image editing model that supports natural language instructions and can accurately perform various editing operations such as region editing, global editing, and text editing.
En3D: An Enhanced Generative Model for Sculpting 3D Humans from 2D Synthetic Data
Yifang Men (Alibaba Group), Xuansong Xie (Peking University)
GenerationData SynthesisDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: We propose En3D, a complete pipeline for zero-shot generation of high-quality 3D human models directly from synthetic 2D images, which includes three main modules: 3D generation, geometric sculpting, and explicit texturing.
End-to-End Spatio-Temporal Action Localisation with Video Transformers
Alexey A. Gritsenko (Google), Anurag Arnab (Google)
RecognitionObject DetectionTransformerVideo
🎯 What it does: This paper proposes a completely Transformer-based end-to-end spatiotemporal action localization model called STAR, which directly maps videos to a series of action categories and corresponding bounding box tubelets, without the need for additional human detectors or memory banks.
End-to-End Temporal Action Detection with 1B Parameters Across 1000 Frames
Shuming Liu (King Abdullah University of Science and Technology), Bernard Ghanem (King Abdullah University of Science and Technology)
RecognitionTransformerSupervised Fine-TuningVideo
🎯 What it does: Proposed the AdaTAD framework, achieving end-to-end temporal action detection with 1 billion parameters and 1536 frames;
Endow SAM with Keen Eyes: Temporal-spatial Prompt Learning for Video Camouflaged Object Detection
Wenjun Hui (Beijing Jiaotong University), Yao Zhao (Beijing Jiaotong University)
Object DetectionConvolutional Neural NetworkPrompt EngineeringVideo
🎯 What it does: This paper proposes TSP-SAM, an end-to-end spatiotemporal prompt learning framework for covert object detection in videos.
Enhance Image Classification via Inter-Class Image Mixup with Diffusion Model
Zhicai Wang (University of Science and Technology of China), Qi Tian (Huawei Inc.)
ClassificationGenerationSupervised Fine-TuningDiffusion modelImage
🎯 What it does: This paper proposes a cross-class image mixing (Diff-Mix) data augmentation method based on diffusion models, utilizing the fine-tuned Stable Diffusion to interpolate images from different categories, thereby enhancing background diversity while maintaining the authenticity of foreground objects, to improve domain-specific image classification tasks.
Enhanced Motion-Text Alignment for Image-to-Video Transfer Learning
Wei Zhang (University of Science and Technology of China), Jieping Ye (Alibaba Cloud)
ClassificationRecognitionTransformerLarge Language ModelSupervised Fine-TuningContrastive LearningVideoText
🎯 What it does: A video recognition framework MoTED based on CLIP is proposed, achieving symmetrical alignment of vision and text through the automatic generation of motion-related descriptions.
Enhancing 3D Fidelity of Text-to-3D using Cross-View Correspondences
Seungwook Kim, Peng Wang
GenerationData SynthesisOptimizationDiffusion modelScore-based ModelNeural Radiance FieldText
🎯 What it does: A method named CorrespondentDream is proposed, which enhances the geometric accuracy of NeRF by obtaining cross-view correspondences through unsupervised extraction of features from a multi-view diffusion model during the zero-shot text-to-3D generation process.
Enhancing 3D Object Detection with 2D Detection-Guided Query Anchors
Haoxuanye Ji (Zhengzhou University), Erkang Cheng
Object DetectionAutonomous DrivingTransformerPoint Cloud
🎯 What it does: A method is proposed to infer 3D query anchors from 2D detection boxes to enhance the performance of multi-camera 3D object detection.
Enhancing Intrinsic Features for Debiasing via Investigating Class-Discerning Common Attributes in Bias-Contrastive Pair
Jeonghoon Park (Korea Advanced Institute of Science and Technology), Jaegul Choo (Korea Advanced Institute of Science and Technology)
ClassificationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a method to achieve debiasing in image classification by explicitly guiding the model to learn the intrinsic features of the target class through 'bias-contrastive pairs' in the absence of biased labels.
Enhancing Multimodal Cooperation via Sample-level Modality Valuation
Yake Wei (Renmin University of China), Di Hu (Renmin University of China)
ClassificationRecognitionData-Centric LearningConvolutional Neural NetworkOptical FlowVideoMultimodalityAudio
🎯 What it does: This paper proposes a sample-level multimodal contribution evaluation metric based on Shapley values, and implements a gain resampling strategy for low-contribution modalities based on this evaluation, significantly enhancing the collaborative effect of multimodal learning models.
Enhancing Post-training Quantization Calibration through Contrastive Learning
Yuzhang Shang (Illinois Institute of Technology), Yan Yan (Illinois Institute of Technology)
ClassificationObject DetectionOptimizationConvolutional Neural NetworkTransformerContrastive LearningImage
🎯 What it does: A framework (CL-Calib) is proposed to calibrate activation quantization parameters using contrastive learning (CL) during post-training quantization (PTQ), reducing quantization noise by maximizing the mutual information between full-precision activations and quantized activations.
Enhancing Quality of Compressed Images by Mitigating Enhancement Bias Towards Compression Domain
Qunliang Xing (Beihang University), Ying Chen (Beihang University)
RestorationSuper ResolutionCompressionGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a technique to alleviate the bias towards the compressed domain when enhancing the quality of compressed images; by conditioning the discriminator of the generative adversarial network and introducing domain divergence regularization, the enhanced images are made closer to the original domain, thereby improving perceptual quality and the authenticity of the compressed images.
Enhancing the Power of OOD Detection via Sample-Aware Model Selection
Feng Xue (Shanghai Jiao Tong University), Falong Tan (Hunan University)
Anomaly DetectionImage
🎯 What it does: This paper proposes a sample-aware model selection method ZODE based on a model zoo to enhance the performance of out-of-distribution (OOD) detection without prior knowledge.
Enhancing Video Super-Resolution via Implicit Resampling-based Alignment
Kai Xu (National University of Singapore), Angela Yao (National University of Singapore)
RestorationSuper ResolutionOptical FlowVideo
🎯 What it does: This paper studies the alignment and resampling problem in video super-resolution, proposing an implicit resampling alignment module and integrating it into existing VSR frameworks to enhance performance.
Enhancing Vision-Language Pre-training with Rich Supervisions
Yuan Gao (Stanford University), Stefano Soatto (AWS AI Labs)
Image TranslationObject DetectionSegmentationTransformerVision Language ModelImageMultimodality
🎯 What it does: A visual-language pre-training framework S4 based on webpage screenshots is proposed, utilizing HTML structure and visual information to generate ten supervised tasks.
Enhancing Visual Continual Learning with Language-Guided Supervision
Bolin Ni (Chinese Academy of Sciences), Shiming Xiang (Chinese Academy of Sciences)
ClassificationRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImage
🎯 What it does: LingoCL is proposed, a paradigm for guiding visual continual learning through semantic targets generated by pre-trained language models.
Enhancing Visual Document Understanding with Contrastive Learning in Large Visual-Language Models
Xin Li (Tencent YouTu Lab), Xing Sun (Tencent YouTu Lab)
Representation LearningTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: A pre-training framework based on Document Object Contrast Learning (DoCo) is proposed to enhance the visual representation ability of large visual-language models in text-rich scenarios.
Ensemble Diversity Facilitates Adversarial Transferability
Bowen Tang (University of Electronic Science and Technology of China), Heng Tao Shen (University of Electronic Science and Technology of China)
Adversarial AttackConvolutional Neural NetworkTransformerReinforcement LearningImage
🎯 What it does: Proposes the SMER (Stochastic Mini-batch Ensemble Reweighting with Reinforcement Learning) method, which gradually generates transferable adversarial examples by leveraging the diversity of multiple models.
Entangled View-Epipolar Information Aggregation for Generalizable Neural Radiance Fields
Zhiyuan Min (Zhejiang University), Yi Yang (Zhejiang University)
GenerationData SynthesisConvolutional Neural NetworkTransformerNeural Radiance FieldImage
🎯 What it does: This paper proposes a generalized NeRF model, EVE-NeRF, that can directly generate new views in unseen scenes.
Entity-NeRF: Detecting and Removing Moving Entities in Urban Scenes
Takashi Otonari (Tokyo Institute of Technology), Kiyoharu Aizawa (University of Tokyo)
RestorationObject DetectionSegmentationConvolutional Neural NetworkNeural Radiance FieldImage
🎯 What it does: Proposed the Entity-NeRF method for detecting and removing moving objects of various scales in urban scenes to construct a static NeRF;
EpiDiff: Enhancing Multi-View Synthesis via Localized Epipolar-Constrained Diffusion
Zehuan Huang, Lu Sheng
GenerationData SynthesisDiffusion modelImage
🎯 What it does: A diffusion model called EpiDiff is proposed, which is based on local phase constraints to generate multi-view consistent and high-quality images from a single view.
Epistemic Uncertainty Quantification For Pre-Trained Neural Networks
Hanjing Wang (Rensselaer Polytechnic Institute), Qiang Ji (Rensselaer Polytechnic Institute)
ClassificationAnomaly DetectionConvolutional Neural NetworkImage
🎯 What it does: A gradient-based method for quantifying the representational uncertainty of pre-trained models is proposed, which does not require additional data or modifications to the model.
Equivariant Multi-Modality Image Fusion
Zixiang Zhao (Xi'an Jiaotong University), Luc Van Gool (ETH Zurich)
Object DetectionSegmentationConvolutional Neural NetworkTransformerImageMultimodalityMagnetic Resonance Imaging
🎯 What it does: This paper proposes a self-supervised multimodal image fusion framework EMMA based on the invariance prior, addressing the challenge of lacking real fused image labels.
Equivariant Plug-and-Play Image Reconstruction
Matthieu Terris (University Paris-Saclay), Julian Tachella
RestorationSuper ResolutionImageMagnetic Resonance Imaging
🎯 What it does: By applying random equivariant transformations (rotation, flipping, translation) to the trained denoiser during the inference phase and performing inverse transformations on the output, an equivariant denoiser was achieved, significantly improving reconstruction stability and image quality under implicit prior frameworks such as PnP, RED, and ULA.
ERMVP: Communication-Efficient and Collaboration-Robust Multi-Vehicle Perception in Challenging Environments
Jingyu Zhang (Fudan University), Liang Song (Fudan University)
Object DetectionAutonomous DrivingSimultaneous Localization and MappingPoint Cloud
🎯 What it does: This paper proposes the ERMVP framework, which enables multi-vehicle collaborative 3D object detection in environments with limited communication and significant localization errors.
Error Detection in Egocentric Procedural Task Videos
Shih-Po Lee (Northeastern University), Ehsan Elhamifar (Stony Brook University)
SegmentationAnomaly DetectionGraph Neural NetworkContrastive LearningVideo
🎯 What it does: This paper proposes a new framework called EgoPED for detecting first-person perspective execution errors using only normal video training, and publicly releases the EgoPER dataset containing various types of errors.
ES3: Evolving Self-Supervised Learning of Robust Audio-Visual Speech Representations
Yuanhang Zhang (University of Chinese Academy of Sciences), Xilin Chen (University of Chinese Academy of Sciences)
RecognitionRepresentation LearningTransformerContrastive LearningVideoMultimodalityAudio
🎯 What it does: This paper proposes a self-supervised learning framework ES3, which gradually learns shared, unique, and collaborative information between audio and video using an 'evolution' strategy to obtain robust multimodal speech representations.
ESCAPE: Encoding Super-keypoints for Category-Agnostic Pose Estimation
Khoi Duc Nguyen (University of Wisconsin Madison), Gim Hee Lee (National University of Singapore)
Pose EstimationImage
🎯 What it does: Proposes the ESCAPE framework, which utilizes super keypoint priors to achieve category-agnostic pose estimation, avoiding the use of keypoint identifiers.
EscherNet: A Generative Model for Scalable View Synthesis
Xin Kong (Imperial College London), Andrew J. Davison (Imperial College London)
GenerationData SynthesisTransformerDiffusion modelGaussian SplattingImage
🎯 What it does: Designed and implemented EscherNet, a scalable view synthesis framework based on a multi-view conditional diffusion model;
ESR-NeRF: Emissive Source Reconstruction Using LDR Multi-view Images
Jinseo Jeong (Seoul National University), Gunhee Kim (Seoul National University)
RestorationGenerationNeural Radiance FieldImage
🎯 What it does: This paper proposes ESR-NeRF, which can reconstruct light sources in a scene from low dynamic range multi-view images and achieve controllable scene editing.
Estimating Extreme 3D Image Rotations using Cascaded Attention
Shay Dekel (Bar Ilan University), Martin Cadik (Brno University of Technology)
Pose EstimationKnowledge DistillationTransformerImage
🎯 What it does: An end-to-end network based on Transformer is proposed to estimate extreme 3D rotations (only estimating pitch and yaw, assuming roll is 0) when the overlap between image pairs is minimal or nonexistent.
Estimating Noisy Class Posterior with Part-level Labels for Noisy Label Learning
Rui Zhao (Xi'an Jiaotong University), Bo Dong (Xi'an Jiaotong University)
ClassificationSupervised Fine-TuningImage
🎯 What it does: This paper proposes to enhance the posterior estimation of noisy classes by incorporating instance-based clipping multi-label part supervision in noisy label learning.
eTraM: Event-based Traffic Monitoring Dataset
Aayush Atul Verma (Arizona State University), Yezhou Yang (Arizona State University)
Object DetectionAutonomous DrivingVideo
🎯 What it does: This paper presents eTraM, a static event camera traffic monitoring dataset with 10 hours of data and 2 million annotations, and systematically evaluates event-based object detection models.
EvalCrafter: Benchmarking and Evaluating Large Video Generation Models
Yaofang Liu (Tencent AI Lab), Ying Shan (Tencent AI Lab)
GenerationLarge Language ModelPrompt EngineeringVideoTextBenchmark
🎯 What it does: Constructed the EvalCrafter framework to conduct multi-dimensional evaluations of large-scale text-to-video models, generating over 700 diverse prompts and designing multiple objective metrics.
Evaluating Transferability in Retrieval Tasks: An Approach Using MMD and Kernel Methods
Mengyu Dai (Microsoft), Joshua Correa (Salesforce)
RetrievalDomain AdaptationImageTextMultimodality
🎯 What it does: A RetMMD metric based on MMD and kernel methods is proposed to evaluate the transferability of pre-trained models in retrieval tasks, without the need for fine-tuning;
EVCap: Retrieval-Augmented Image Captioning with External Visual-Name Memory for Open-World Comprehension
Jiaxuan Li (University of Tokyo), Hideki Nakayama (National Institute of Informatics)
GenerationRetrievalTransformerLarge Language ModelVision Language ModelImageTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes a retrieval-enhanced image captioning model called EVCAP, which utilizes external visual-name memory to retrieve object names and feeds them as prompts to a frozen large language model to generate image captions.
EvDiG: Event-guided Direct and Global Components Separation
Xinyu Zhou (Peking University), Boxin Shi (Peking University)
RestorationConvolutional Neural NetworkImageVideoBenchmark
🎯 What it does: A method for direct and global illumination separation based on event cameras, called EvDiG, is proposed. It utilizes event capture to quickly detect shadow changes, combines RGB images for initial coarse separation, and then employs a two-stage network, EvSepNet and ImColorNet, to denoise and restore colors, achieving high-quality separation at video frame rates.
Event Stream-based Visual Object Tracking: A High-Resolution Benchmark Dataset and A Novel Baseline
Xiao Wang (Anhui University), Jin Tang (Anhui University)
Object TrackingKnowledge DistillationTransformerMultimodalityBenchmark
🎯 What it does: A visual target tracking framework based on hierarchical knowledge distillation for event cameras is designed, achieving high-speed and low-latency tracking using only event signals, and a high-resolution event tracking dataset, EventVOT, is proposed.
Event-assisted Low-Light Video Object Segmentation
Hebei Li (University of Science and Technology of China), Xiaoyan Sun (Institute of Artificial Intelligence, Hefei Comprehensive National Science Center)
SegmentationTransformerVideo
🎯 What it does: To address the problem of video object segmentation (VOS) in low-light environments, an event camera-assisted end-to-end framework is proposed.
Event-based Structure-from-Orbit
Ethan Elms (University of Adelaide), Tat-Jun Chin (Stanford University)
Object TrackingPose EstimationDepth EstimationOptimizationSimultaneous Localization and MappingPoint CloudBenchmark
🎯 What it does: This paper proposes an event camera-based structure-from-orbit (eSfO) method for reconstructing the sparse 3D structure of rotating objects and estimating rotational trajectory parameters when observed by a static event camera.
Event-based Visible and Infrared Fusion via Multi-task Collaboration
Mengyue Geng (Peking University), Yonghong Tian (Peking University)
Image TranslationRestorationObject DetectionConvolutional Neural NetworkRecurrent Neural NetworkTransformerContrastive LearningImageVideoMultimodality
🎯 What it does: To address the issue of visible and infrared image fusion under extreme lighting and high-speed motion scenarios, this paper proposes an event camera-based fusion framework called EVIF, which achieves visible texture reconstruction, event-guided infrared image deblurring, and collaborative fusion of all three components.
EventDance: Unsupervised Source-free Cross-modal Adaptation for Event-based Object Recognition
Xu Zheng (Hong Kong University of Science and Technology), Lin Wang (Hong Kong University of Science and Technology)
RecognitionObject DetectionDomain AdaptationOptical FlowImageVideo
🎯 What it does: An unsupervised, source-free cross-modal (image → event) adaptation framework called EventDance is proposed, addressing the problem of how to utilize a pre-trained source image model for object recognition in event cameras when there is no source-labeled image data.
EventEgo3D: 3D Human Motion Capture from Egocentric Event Streams
Christen Millerdurai (Max Planck Institute for Informatics), Vladislav Golyanik (Max Planck Institute for Informatics)
Pose EstimationVideo
🎯 What it does: This paper proposes a 3D human pose capture method implemented with a real-time monocular fisheye event camera named EventEgo3D, along with a lightweight head-mounted device and synthetic and real event datasets.
EventPS: Real-Time Photometric Stereo Using an Event Camera
Bohan Yu (Peking University), Boxin Shi (Peking University)
Depth EstimationOptimizationComputational EfficiencyConvolutional Neural NetworkSupervised Fine-TuningImageVideo
🎯 What it does: This paper proposes EventPS, which utilizes event cameras to achieve real-time photometric stereo, estimating the surface normals of objects through event signals.
Evidential Active Recognition: Intelligent and Prudent Open-World Embodied Perception
Lei Fan (Northwestern University), Ying Wu (Northwestern University)
RecognitionObject DetectionRobotic IntelligenceConvolutional Neural NetworkReinforcement LearningImage
🎯 What it does: A proactive recognition framework based on evidence theory has been constructed. The model outputs uncertainty for each perspective and performs Dempster–Shafer combination across multiple perspectives, ultimately obtaining the credibility of the target category to guide the robot's movement strategy.
EVS-assisted Joint Deblurring Rolling-Shutter Correction and Video Frame Interpolation through Sensor Inverse Modeling
Rui Jiang (OMNIVISION), Andreas Suess
RestorationVideo
🎯 What it does: A joint method based on the fusion of event sensors (EVS) and CMOS image sensors (CIS) is proposed to achieve deblurring, rolling shutter correction, and high frame rate video interpolation.
Exact Fusion via Feature Distribution Matching for Few-shot Image Generation
Yingbo Zhou (East China Normal University), Mingsong Chen (East China Normal University)
GenerationData SynthesisAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: This paper studies a few-shot image generation framework called F2DGAN that achieves precise fusion through feature distribution matching.
ExACT: Language-guided Conceptual Reasoning and Uncertainty Estimation for Event-based Action Recognition and More
Jiazhou Zhou (Hong Kong University of Science and Technology), Lin Wang (Hong Kong University of Science and Technology)
RecognitionVision Language ModelContrastive LearningVideoText
🎯 What it does: Proposes the ExACT framework, which enhances event camera action recognition through language guidance, combining Adaptive Fine-grained Event Representation (AFE) and Concept Reasoning and Uncertainty Estimation (CRUE) modules.
ExMap: Leveraging Explainability Heatmaps for Unsupervised Group Robustness to Spurious Correlations
Rwiddhi Chakraborty (UiT The Arctic University of Norway), Michael C. Kampffmeyer (UiT The Arctic University of Norway)
ClassificationExplainability and InterpretabilityImage
🎯 What it does: This paper proposes ExMap, a two-stage unsupervised method that first uses Layer-wise Relevance Propagation (LRP) to obtain interpretable heatmaps, then clusters to generate pseudo group labels, which are subsequently used in existing group robust learning strategies to enhance the model's robustness against spurious features.
Expandable Subspace Ensemble for Pre-Trained Model-Based Class-Incremental Learning
Da-Wei Zhou (Nanjing University), De-Chuan Zhan (Nanjing University)
ClassificationRecognitionTransformerSupervised Fine-TuningImage
🎯 What it does: This paper proposes a scalable subspace ensemble method (EASE) based on pre-trained models, which constructs dedicated subspaces for each new task by inserting lightweight adapters on a frozen pre-trained model in class-incremental learning. It utilizes a semantic-guided prototype completion technique to reconstruct old class prototypes under zero-sample conditions, ultimately achieving incremental learning without forgetting and without examples.
Explaining CLIP's Performance Disparities on Data from Blind/Low Vision Users
Daniela Massiceti (Microsoft Research), Cecily Morrison (Microsoft Research)
ClassificationObject DetectionGenerationTransformerVision Language ModelContrastive LearningImage
🎯 What it does: Evaluate and quantify the performance differences of CLIP on images taken by blind/low vision (BLV) users, and explore the reasons and possible mitigation measures.
Explaining the Implicit Neural Canvas: Connecting Pixels to Neurons by Tracing their Contributions
Namitha Padmanabhan (University of Maryland), Abhinav Shrivastava (University of Maryland)
SegmentationAutonomous DrivingExplainability and InterpretabilityConvolutional Neural NetworkOptical FlowVideo
🎯 What it does: The XINC framework is proposed, which explains the internal working mechanism of implicit neural representations (INR) using contribution maps, revealing the characteristics of INR in aspects such as color, edges, distributed representations, and motion-driven contribution changes.
Exploiting Diffusion Prior for Generalizable Dense Prediction
Hsin-Ying Lee (University of California), Ming-Hsuan Yang (University of California)
SegmentationGenerationDepth EstimationDiffusion modelImage
🎯 What it does: Using a pre-trained text-to-image diffusion model as a prior, a deterministic diffusion mapping is proposed to transform the random generation process into a deterministic mapping suitable for dense prediction. Based on this, the model is fine-tuned to complete predictions for five tasks: depth, normals, semantic segmentation, and intrinsic image decomposition.