CVPR 2025 Papers — Page 9
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2871 papers
Enhanced then Progressive Fusion with View Graph for Multi-View Clustering
Zhibin Dong (National University of Defense Technology), En Zhu (National University of Defense Technology)
ClassificationRepresentation LearningGraph Neural NetworkContrastive LearningImage
🎯 What it does: A multi-view clustering framework EPFMVC is proposed, which integrates feature enhancement and progressive fusion to improve the clustering accuracy of multi-view data.
Enhanced Visual-Semantic Interaction with Tailored Prompts for Pedestrian Attribute Recognition
Junyi Wu (Fuzhou University), Qiang Wu (University of Technology Sydney)
RecognitionTransformerPrompt EngineeringVision Language ModelImageText
🎯 What it does: This paper proposes a cross-modal pedestrian attribute recognition framework EVSITP based on visual-language interaction, which enhances semantic representation using learnable prompts under image conditions and achieves bidirectional modal interaction.
Enhancing 3D Gaze Estimation in the Wild using Weak Supervision with Gaze Following Labels
Pierre Vuillecard (Idiap Research Institute), Jean-Marc Odobez (Idiap Research Institute)
RecognitionPose EstimationTransformerImageVideo
🎯 What it does: A weakly supervised 3D gaze estimation framework ST-WSGE is proposed, and a modality-agnostic Gaze Transformer (GaT) model is designed to achieve unified training and inference for images and videos.
Enhancing Adversarial Transferability with Checkpoints of a Single Model's Training
Shixin Li (Huazhong University of Science and Technology), Linchen Yu (Huazhong University of Science and Technology)
Adversarial AttackConvolutional Neural NetworkImage
🎯 What it does: This paper studies a black-box attack method that generates transferable adversarial examples using checkpoints from a single model training process.
Enhancing Creative Generation on Stable Diffusion-based Models
Jiyeon Han (Korea Advanced Institute of Science and Technology), Jaesik Choi (Korea Advanced Institute of Science and Technology)
GenerationDiffusion modelImage
🎯 What it does: A training-independent C3 method is proposed, which enhances creative generation by amplifying low-frequency features in the shallow blocks of the U-Net in Stable Diffusion.
Enhancing Dance-to-Music Generation via Negative Conditioning Latent Diffusion Model
Changchang Sun (University of Illinois Chicago), Yan Yan (University of Illinois Chicago)
GenerationDiffusion modelVideoAudio
🎯 What it does: A bidirectional conditional latent diffusion model (PN-Diffusion) is proposed for generating dance to music using positive and negative rhythm information.
Enhancing Dataset Distillation via Non-Critical Region Refinement
Minh-Tuan Tran (Monash University), Dinh Phung (Monash University)
Data SynthesisKnowledge DistillationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes a Non-Critical Region Refinement Data Distillation (NRR-DD) method, which defines critical and non-critical regions through CAM, updating only non-critical pixels while using Distance-Based Representation (DBR) for knowledge transfer to generate high-quality, low-storage-cost synthetic data.
Enhancing Diversity for Data-free Quantization
Kai Zhao (Aalborg University), Bin Yang (East China Normal University)
Data SynthesisOptimizationKnowledge DistillationConvolutional Neural NetworkTransformerFlow-based ModelGenerative Adversarial NetworkImage
🎯 What it does: A GAN-based unsupervised quantization framework is proposed, utilizing multi-layer feature mixing and normalized flow attention to generate richer synthetic calibration data, thereby enhancing the performance of quantized models.
Enhancing Facial Privacy Protection via Weakening Diffusion Purification
Ali Salar (University of Oulu), Guoying Zhao (University of Oulu)
GenerationSafty and PrivacyDiffusion modelImage
🎯 What it does: This paper proposes a method for adversarial modification in latent space using diffusion models, introducing learnable non-text embeddings and self-attention constraints to generate facial privacy protection images that effectively conceal the original identity while maintaining visual quality.
Enhancing Few-Shot Class-Incremental Learning via Training-Free Bi-Level Modality Calibration
Yiyang Chen (Nanjing University), Wenbin Li (Nanjing University)
ClassificationRecognitionTransformerLarge Language ModelVision Language ModelContrastive LearningImage
🎯 What it does: A training-free framework based on CLIP, called BiMC, is proposed to address the few-shot incremental learning problem.
Enhancing Online Continual Learning with Plug-and-Play State Space Model and Class-Conditional Mixture of Discretization
Sihao Liu (Harbin Institute of Technology), Bernard Ghanem (King Abdullah University of Science and Technology)
ClassificationOptimizationMixture of ExpertsContrastive LearningImage
🎯 What it does: This paper proposes a pluggable module S6MOD, which can significantly enhance the adaptability of online continual learning (OCL) and prevent catastrophic forgetting by adding branches based on the Selective State Space Model (S6) while keeping the original network structure unchanged.
Enhancing Privacy-Utility Trade-offs to Mitigate Memorization in Diffusion Models
Chen Chen (University of Sydney), Chang Xu (University of Sydney)
GenerationSafty and PrivacyTransformerLarge Language ModelPrompt EngineeringDiffusion modelImageText
🎯 What it does: This paper proposes a method that significantly reduces the model's risk of memorization of training data during the inference phase of diffusion models through two strategies: prompt re-anchoring (PR) and semantic prompt search (SS), while maintaining or enhancing the alignment between text and generated images.
Enhancing SAM with Efficient Prompting and Preference Optimization for Semi-supervised Medical Image Segmentation
Aishik Konwer (GE Healthcare), Taha Kass-Hout (GE Healthcare)
SegmentationLarge Language ModelPrompt EngineeringImageBiomedical DataComputed TomographyUltrasound
🎯 What it does: In the medical image segmentation task, a SAM enhancement framework combining unsupervised prompts and direct preference optimization is proposed, which can automatically generate fine segmentation results in low-labeling scenarios.
Enhancing Testing-Time Robustness for Trusted Multi-View Classification in the Wild
Wei Liu (Tongji University), Xiaodong Yue (Shanghai University)
ClassificationImage
🎯 What it does: A general evidence filtering mechanism for testing is proposed to enhance the robustness of Trustworthy Multi-View Classification (TMVC) in real-world scenarios.
Enhancing Video-LLM Reasoning via Agent-of-Thoughts Distillation
Yudi Shi (Shanghai Jiao Tong University), Weidi Xie (Shanghai Jiao Tong University)
Explainability and InterpretabilityKnowledge DistillationTransformerLarge Language ModelAgentic AIPrompt EngineeringVision Language ModelVideoChain-of-Thought
🎯 What it does: Construct a system based on Agent to automatically generate multi-step Chain-of-Thoughts and integrate them into the instruction-tuning of Video-LLM, enhancing the accuracy and interpretability of video question answering.
Enhancing Virtual Try-On with Synthetic Pairs and Error-Aware Noise Scheduling
Nannan Li (Boston University), Bryan A. Plummer (Boston University)
Image TranslationGenerationData SynthesisDiffusion modelGenerative Adversarial NetworkOptical FlowImage
🎯 What it does: A human-to-garment transformation model is proposed to enhance virtual try-on training by generating (human, synthetic garment) pairs, and a locally corrective error-aware Schrödinger bridge model (EARSB) is introduced to adjust the outputs of existing try-on models.
Enhancing Vision-Language Compositional Understanding with Multimodal Synthetic Data
Haoxin Li (Nanyang Technological University), Boyang Li (Nanyang Technological University)
GenerationData SynthesisDomain AdaptationLarge Language ModelVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: This paper proposes a method to automatically create multimodal synthetic data with subtle differences using generative models, and uses it to enhance the combinatorial reasoning ability of visual-language models.
EnliveningGS: Active Locomotion of 3DGS
Siyuan Shen (Zhejiang University), Yin Yang (University of Utah)
GenerationData SynthesisOptimizationGaussian SplattingPoint CloudMesh
🎯 What it does: This paper proposes the EnliveningGS pipeline, enabling the 3D Gaussian Splatting (3DGS) model to achieve active motion (walking, jumping, twisting, etc.) through muscle activation, and provides a complete process for muscle generation, collision, soft body dynamics, and 3DGS deformation and completion.
EntityErasure: Erasing Entity Cleanly via Amodal Entity Segmentation and Completion
Yixing Zhu (Sun Yat-sen University), Wei-Shi Zheng (Sun Yat-sen University)
RestorationSegmentationConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: A method for object erasure based on diffusion models, called EntityErasure, is proposed. It first obtains a complete entity mask through amodal entity segmentation, and then utilizes entity attention in UNet to complete the entity, thereby avoiding the generation of irrelevant entities.
EntitySAM: Segment Everything in Video
Mingqiao Ye (Adobe Research), Joon-Young Lee
Object TrackingSegmentationTransformerVideo
🎯 What it does: EntitySAM is proposed, which can automatically segment and track all entities in a video without prompts, achieving zero-shot video entity segmentation.
EntropyMark: Towards More Harmless Backdoor Watermark via Entropy-based Constraint for Open-source Dataset Copyright Protection
Ming Sun (Institute of Information Engineering, Chinese Academy of Sciences), Yuanfang Guo (Beihang University)
ClassificationObject DetectionSafty and PrivacyConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A harmless reverse gate watermark implemented through entropy constraints—EntropyMark, is proposed for copyright protection of open-source datasets.
EnvGS: Modeling View-Dependent Appearance with Environment Gaussian
Tao Xie (Zhejiang University), Xiaowei Zhou (Ant Group)
GenerationOptimizationComputational EfficiencyNeural Radiance FieldGaussian SplattingImage
🎯 What it does: An explicit 3D representation method based on environmental Gaussian primitives (EnvGS) is used to achieve high-quality, real-time view synthesis of complex reflections, jointly optimizing the base Gaussian and environmental Gaussian to reconstruct geometry, base color, and reflection color.
EnvPoser: Environment-aware Realistic Human Motion Estimation from Sparse Observations with Uncertainty Modeling
Songpengcheng Xia (Shanghai Jiao Tong University), Ling Pei (Shanghai Jiao Tong University)
Pose EstimationTransformerPoint Cloud
🎯 What it does: Proposes the EnvPoser two-stage framework, which estimates full-body motion using sparse tracking signals from HMD and controllers, along with pre-scanned environmental point clouds.
EquiPose: Exploiting Permutation Equivariance for Relative Camera Pose Estimation
Yuzhen Liu (Chinese Academy of Sciences), Qiulei Dong (Chinese Academy of Sciences)
Pose EstimationPoint Cloud
🎯 What it does: A general framework called EquiPose is proposed, which can enforce pose permutation equivariance (PPE) for any end-to-end relative camera pose estimation model and enhance its performance.
Erase Diffusion: Empowering Object Removal Through Calibrating Diffusion Pathways
Yi Liu (Alibaba Group), Ran Lin (Alibaba Group)
RestorationGenerationDiffusion modelImage
🎯 What it does: A diffusion model specifically designed for object removal, called EraDiff, is proposed, along with the Chain-Rectifying Optimization (CRO) and Self-Rectifying Attention (SRA) mechanisms. The aim is to guide the model to train and sample along the diffusion path from the object to the background, thereby enhancing the effect of eliminating the target while maintaining visual coherence.
Erasing Undesirable Influence in Diffusion Models
Jing Wu (Monash University), Mehrtash Harandi (Qualcomm)
OptimizationMeta LearningDiffusion modelImage
🎯 What it does: An algorithm named EraseDiff is proposed, aimed at removing unnecessary information from diffusion models while maintaining the model's practicality.
ERUPT: Efficient Rendering with Unposed Patch Transformer
Maxim V. Shugaev (BlueHalo), Naresh P. Cuntoor (Carnegie Mellon University)
GenerationPose EstimationComputational EfficiencyTransformerDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes ERUPT, a model capable of learning implicit camera poses and performing efficient view synthesis with only a small number of uncalibrated images.
ESC: Erasing Space Concept for Knowledge Deletion
Tae-Young Lee (Korea University), Gyeong-Moon Park (Korea University)
Safty and PrivacyComputational EfficiencyConvolutional Neural NetworkTransformerReinforcement LearningImage
🎯 What it does: This paper proposes a Knowledge Deletion (KD) framework aimed at achieving the complete elimination of specified knowledge in trained models, meeting users' privacy and practical needs.
ESCAPE: Equivariant Shape Completion via Anchor Point Encoding
Burak Bekci (Technical University Munich), Mahdi Saleh (Technical University Munich)
GenerationAutonomous DrivingOptimizationTransformerPoint Cloud
🎯 What it does: A rotation-invariant shape completion framework ESCAPE based on anchor point distance encoding is proposed.
Escaping Plato's Cave: Towards the Alignment of 3D and Text Latent Spaces
Souhail Hadgi (Ecole Polytechnique), Maks Ovsjanikov
RetrievalContrastive LearningTextMultimodalityPoint Cloud
🎯 What it does: This paper studies the alignment of the latent spaces of independently trained 3D encoders and text encoders after training. It proposes to first use Canonical Correlation Analysis (CCA) to project into a low-dimensional subspace, followed by affine or local CKA alignment, thereby achieving cross-modal retrieval and matching.
Estimating Body and Hand Motion in an Ego-sensed World
Brent Yi (University of California Berkeley), Angjoo Kanazawa (University of California Berkeley)
Pose EstimationDiffusion modelSimultaneous Localization and MappingImageVideo
🎯 What it does: A system named EgoAllo was designed and implemented, which estimates human posture, height, and hand parameters using the SLAM pose and camera images from a head-mounted device, and projects them into a global coordinate system.
ETAP: Event-based Tracking of Any Point
Friedhelm Hamann (Technische Universitat Berlin), Guillermo Gallego (Technische Universitat Berlin)
Object TrackingTransformerImageVideo
🎯 What it does: The first method for arbitrary point tracking (ETAP) using a pure event camera is proposed, capable of achieving robust tracking in high dynamic range and high motion speed scenarios.
Ev-3DOD: Pushing the Temporal Boundaries of 3D Object Detection with Event Cameras
Hoonhee Cho (Korea Advanced Institute of Science and Technology), Kuk-Jin Yoon (Korea Advanced Institute of Science and Technology)
Object DetectionAutonomous DrivingMultimodalityPoint Cloud
🎯 What it does: A system called Ev-3DOD is proposed, which utilizes event cameras to achieve 3D object detection and can estimate object positions in the blind spots of LiDAR/cameras in real-time.
Eval3D: Interpretable and Fine-grained Evaluation for 3D Generation
Shivam Duggal (Massachusetts Institute of Technology), Wei-Chiu Ma (Cornell University)
GenerationExplainability and InterpretabilityVision Language ModelTextPoint CloudBenchmark
🎯 What it does: This paper proposes Eval3D, a fine-grained and interpretable 3D generation evaluation tool based on the consistency of multimodal foundational models.
Evaluating Model Perception of Color Illusions in Photorealistic Scenes
Lingjun Mao (University of California), Alane Suhr (University of California)
GenerationData SynthesisTransformerVision Language ModelImageChain-of-Thought
🎯 What it does: This study investigates the perception of color illusions by visual language models (VLM) in real-world scenarios and constructs a large-scale Realistic Color Illusion Dataset (RCID).
Evaluating Vision-Language Models as Evaluators in Path Planning
Mohamed Aghzal (George Mason University), Ziyu Yao (George Mason University)
Autonomous DrivingOptimizationTransformerLarge Language ModelVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: A benchmark called PATHEVAL is proposed to evaluate the performance of Vision-Language Models in path planning scenarios.
EvEnhancer: Empowering Effectiveness, Efficiency and Generalizability for Continuous Space-Time Video Super-Resolution with Events
Shuoyan Wei (Beijing Jiaotong University), Huihui Bai (Beijing Jiaotong University)
RestorationSuper ResolutionTransformerVideo
🎯 What it does: This paper proposes EvEnhancer, an end-to-end method for continuous space-time video super-resolution (C-STVSR) that utilizes high temporal resolution and high dynamic range data from event cameras, combined with frame images.
Event Ellipsometer: Event-based Mueller-Matrix Video Imaging
Ryota Maeda (POSTECH), Seung-Hwan Baek (POSTECH)
Video
🎯 What it does: This paper presents an event ellipsometer capable of capturing Mueller matrix videos of dynamic scenes at a speed of 30 frames per second.
Event Fields: Capturing Light Fields at High Speed, Resolution, and Dynamic Range
Ziyuan Qu (Dartmouth), Adithya Pediredla (Dartmouth)
Data SynthesisDepth EstimationImageVideo
🎯 What it does: The Event Fields framework is proposed, utilizing event cameras and two optical designs (spatial multiplexing with mirror prisms and temporal multiplexing with galvanometer mirrors) to capture light fields with high frame rates, high resolutions, and wide dynamic ranges. Based on this, new applications such as SloMoRF, instant real-time depth estimation, and HDR light field reconstruction are realized.
Event-based Video Super-Resolution via State Space Models
Zeyu Xiao (National University of Singapore), Xinchao Wang (National University of Singapore)
RestorationSuper ResolutionRecurrent Neural NetworkVideo
🎯 What it does: Proposes MamEVSR, an end-to-end framework that combines event cameras with video super-resolution;
Event-Equalized Dense Video Captioning
Kangyi Wu (Xi'an Jiaotong University), Sanping Zhou (Xi'an Jiaotong University)
Object DetectionGenerationTransformerVideoText
🎯 What it does: A framework called Event-Equalized Dense Video Captioning (E DVC) is proposed to evenly locate and generate descriptions of all events in untrimmed videos, addressing the issue of temporal bias.
EventFly: Event Camera Perception from Ground to the Sky
Lingdong Kong (National University of Singapore), Benoit R. Cottereau (Centre de Recherche en Cognition et Neuroergonomie)
Object DetectionSegmentationDomain AdaptationAutonomous DrivingImageVideoBenchmark
🎯 What it does: The EventFly framework is proposed to achieve adaptive learning for cross-platform event camera dense perception.
EventGPT: Event Stream Understanding with Multimodal Large Language Models
Shaoyu Liu (Xidian University), Ming Li (Guangdong Laboratory of Artificial Intelligence and Digital Economy)
Object DetectionSegmentationGenerationTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: The first event stream multimodal large language model, EventGPT, has been constructed to understand and generate event camera data.
EventPSR: Surface Normal and Reflectance Estimation from Photometric Stereo Using an Event Camera
Bohan Yu (Peking University), Imari Sato (The University of Tokyo)
Depth EstimationOptimizationImage
🎯 What it does: Using event cameras to simultaneously estimate the surface normals and reflection parameters (metalness, roughness) in photometric stereo.
EventSplat: 3D Gaussian Splatting from Moving Event Cameras for Real-time Rendering
Toshiya Yura (Sony Semiconductor Solutions Corporation), Igor Gilitschenski (University of Toronto)
GenerationOptimizationComputational EfficiencyGaussian SplattingSimultaneous Localization and MappingImageVideoPoint Cloud
🎯 What it does: This paper proposes a method for 3D Gaussian Splatting using data from event cameras, achieving a new perspective synthesis from event data to real-time visualization.
Every SAM Drop Counts: Embracing Semantic Priors for Multi-Modality Image Fusion and Beyond
Guanyao Wu (Dalian University of Technology), Risheng Liu (Dalian University of Technology)
Image TranslationSegmentationKnowledge DistillationImageMultimodality
🎯 What it does: This paper proposes a complete framework for multimodal (infrared + visible) image fusion using the semantic prior of the Segment Anything Model (SAM), while also considering the adaptability to downstream tasks.
Everything to the Synthetic: Diffusion-driven Test-time Adaptation via Synthetic-Domain Alignment
Jiayi Guo (Tsinghua University), Gao Huang (Tsinghua University)
Domain AdaptationDiffusion modelImageMultimodality
🎯 What it does: A new test-time domain adaptation framework (SDA) is proposed, which utilizes diffusion models to generate synthetic domains and fine-tune the source model, while mapping target data to the same synthetic domain to achieve domain alignment for cross-domain TTA.
EvOcc: Accurate Semantic Occupancy for Automated Driving Using Evidence Theory
Jonas Kälble (Bosch), Eddy Ilg (University of Technology Nuremberg)
Object DetectionSegmentationAutonomous DrivingPoint Cloud
🎯 What it does: This paper proposes the EvOcc framework, which utilizes evidence theory to construct a 3D semantic occupancy map from partially labeled LiDAR data, and uses this high-quality occupancy map to supervise the occupancy prediction model of multi-view cameras.
EVolSplat: Efficient Volume-based Gaussian Splatting for Urban View Synthesis
Sheng Miao, Yiyi Liao
GenerationData SynthesisAutonomous DrivingConvolutional Neural NetworkGaussian SplattingPoint Cloud
🎯 What it does: This paper proposes an end-to-end, volume-based efficient 3D Gaussian splatting (EVolSplat) model for rapidly generating high-quality new view renderings of urban scenes under sparse vehicle camera inputs.
Evolving High-Quality Rendering and Reconstruction in a Unified Framework with Contribution-Adaptive Regularization
You Shen (Xiamen University), Liujuan Cao (Xiamen University)
RestorationGenerationGaussian SplattingPoint Cloud
🎯 What it does: This paper proposes a unified 3D Gaussian scattering model, CarGS, which can achieve high-quality view synthesis and surface reconstruction simultaneously within the same network.
EVOS: Efficient Implicit Neural Training via EVOlutionary Selector
Weixiang Zhang (Tsinghua University), Zhi Wang (Tsinghua University)
Super ResolutionComputational EfficiencyImage
🎯 What it does: This paper proposes an evolutionary selection-based sampling acceleration method called EVOS, which significantly reduces the number of forward passes required for training implicit neural representations (INR) and improves training efficiency.
EVPGS: Enhanced View Prior Guidance for Splatting-based Extrapolated View Synthesis
Jiahe Li (Meitu Inc), Ting Liu (Meitu Inc)
GenerationData SynthesisDiffusion modelGaussian SplattingPoint CloudMesh
🎯 What it does: Under limited view coverage, an efficient Gaussian scattering model is utilized to achieve extrapolated view synthesis, proposing a two-stage coarse-fine adjustment perspective prior-guided framework EVPGS;
Exact: Exploring Space-Time Perceptive Clues for Weakly Supervised Satellite Image Time Series Semantic Segmentation
Hao Zhu (Institute of Computing Technology, Chinese Academy of Sciences), Feng Dai (Institute of Computing Technology, Chinese Academy of Sciences)
RecognitionObject DetectionSegmentationTransformerContrastive LearningImageTime SeriesAgriculture Related
🎯 What it does: This paper proposes a weakly supervised semantic segmentation framework called Exact, which utilizes only image-level labels for crop semantic segmentation in satellite image time series (SITS), addressing two major challenges: spatial noise disturbance and temporal semantic misalignment.
ExpertAF: Expert Actionable Feedback from Video
Kumar Ashutosh (University of Texas at Austin), Kristen Grauman (University of Texas at Austin)
Pose EstimationRetrievalTransformerLarge Language ModelSupervised Fine-TuningVideoTextMultimodality
🎯 What it does: A system called ExpertAF is proposed, which can automatically generate executable expert feedback from learner videos, providing both textual comments and corrective demonstration videos.
Explainable Saliency: Articulating Reasoning with Contextual Prioritization
Nuo Chen (University of Minnesota), Qi Zhao (University of Minnesota)
SegmentationExplainability and InterpretabilityConvolutional Neural NetworkVision Language ModelImage
🎯 What it does: Designed and implemented the XSal model, which performs explicit reasoning through a Vision-Language model, combining semantic prototypes and context prioritization to generate interpretable visual attention maps and natural language explanations.
Explaining Domain Shifts in Language: Concept Erasing for Interpretable Image Classification
Zequn Zeng (Xidian University), Jiawei Ma (City University of Hong Kong)
Domain AdaptationExplainability and InterpretabilityTransformerLarge Language ModelVision Language ModelImage
🎯 What it does: This paper proposes a language-guided concept elimination framework (LanCE) to enhance the generalization ability of concept bottleneck models in cross-domain tasks by removing the influence of domain-specific concepts.
Explaining in Diffusion: Explaining a Classifier with Diffusion Semantics
Tahira Kazimi (Virginia Tech), Pinar Yanardag (Virginia Tech)
ClassificationExplainability and InterpretabilityVision Language ModelDiffusion modelImageText
🎯 What it does: Proposes the DiffEx method, which utilizes visual-language models to generate domain-specific hierarchical semantic vocabularies and employs text-to-image diffusion models for untrained semantic editing of classifiers, thereby explaining the decision-making process of classifiers.
Explicit Depth-Aware Blurry Video Frame Interpolation Guided by Differential Curves
Zaoming Yan (East China Normal University), Haichuan Song (East China Normal University)
RestorationDepth EstimationTransformerOptical FlowVideo
🎯 What it does: A deep perception fuzzy video frame interpolation framework DC-BVFI based on differential curve theory is proposed, which utilizes UBNet to convert two-dimensional fuzzy frames into three-dimensional camera space point clouds, and estimates scene flow through MPNet, thereby explicitly modeling depth changes for more accurate intermediate frame synthesis.
Exploiting Deblurring Networks for Radiance Fields
Haeyun Choi (KT), Sunghyun Cho (POSTECH)
RestorationData SynthesisPose EstimationNeural Radiance FieldGaussian SplattingSimultaneous Localization and MappingImage
🎯 What it does: The DeepDeblurRF framework is proposed to quickly construct high-quality light radiation fields from blurred training images.
Exploiting Temporal State Space Sharing for Video Semantic Segmentation
Syed Ariff Syed Hesham (Nanyang Technological University), Xudong Jiang (Institute for Infocomm Research A Star)
SegmentationRecurrent Neural NetworkTransformerVideo
🎯 What it does: A TV3S architecture based on state space models is proposed, which achieves efficient video semantic segmentation by independently and parallelly processing spatial blocks and incorporating optional gating and window shifting.
Exploration-Driven Generative Interactive Environments
Nedko Savov (INSAIT), Luc Van Gool (INSAIT)
GenerationTransformerReinforcement LearningWorld ModelVideo
🎯 What it does: This paper proposes a method that utilizes an automated exploration agent to generate interactive data in a large number of virtual game environments, and based on this, trains a transferable multi-environment world model (GenieRedux-G).
Exploring CLIP's Dense Knowledge for Weakly Supervised Semantic Segmentation
Zhiwei Yang (Fudan University), Zhijian Song (Fudan University)
SegmentationTransformerLarge Language ModelVision Language ModelContrastive LearningImage
🎯 What it does: Proposes ExCEL, which generates CAM through patch-text alignment of CLIP, enhancing weakly supervised semantic segmentation performance.
Exploring Contextual Attribute Density in Referring Expression Counting
Zhicheng Wang (Huazhong University of Science and Technology), Zhiguo Cao (Huazhong University of Science and Technology)
Object DetectionTransformerMultimodality
🎯 What it does: A Referring Expression Counting (REC) framework based on Contextual Attribute Density (CAD), named CAD-GD, is proposed and implemented on the multimodal detection model GroundingDINO.
Exploring Historical Information for RGBE Visual Tracking with Mamba
Chuanyu Sun (Dalian University of Technology), Xin Yang (Dalian University of Technology)
Object TrackingTransformerVideoMultimodality
🎯 What it does: A RGBE visual tracking framework called MamTrack based on Mamba is proposed, which includes a Fusion Mamba module for cross-modal fusion and a historical decoder that utilizes long sequence modeling.
Exploring Intrinsic Normal Prototypes within a Single Image for Universal Anomaly Detection
Wei Luo (Tsinghua University), Wenyong Yu (Huazhong University of Science and Technology)
Anomaly DetectionTransformerImage
🎯 What it does: A novel anomaly detection framework called INP-Former based on Vision Transformer is proposed, which utilizes the 'Intrinsic Normal Prototypes (INPs)' within a single test image for adaptive reconstruction, enabling anomaly detection and localization without relying on external normal samples.
Exploring Scene Affinity for Semi-Supervised LiDAR Semantic Segmentation
Chuandong Liu (Wuhan University), Gui-Song Xia (Wuhan University)
SegmentationAutonomous DrivingPoint CloudBenchmark
🎯 What it does: A semi-supervised LiDAR semantic segmentation method based on scene affinity (AIScene) is proposed, which combines a teacher-student framework with pseudo-labels for learning.
Exploring Semantic Feature Discrimination for Perceptual Image Super-Resolution and Opinion-Unaware No-Reference Image Quality Assessment
Guanglu Dong (Sichuan University), Chao Ren (Sichuan University)
RestorationSuper ResolutionGenerative Adversarial NetworkImage
🎯 What it does: A Semantic Feature Discrimination (SFD) framework is proposed, utilizing pixel-level semantic features from CLIP and text-guided global features for adversarial discrimination, thereby enhancing the perceptual quality of super-resolution (SR) and directly reusing the discriminator in no-reference image quality assessment (OU NR-IQA) to achieve high-accuracy unlabelled quality evaluation.
Exploring Simple Open-Vocabulary Semantic Segmentation
Zihang Lai (Visual Geometry Group University of Oxford)
SegmentationTransformerContrastive LearningImageText
🎯 What it does: S-Seg proposes a vocabulary-free semantic segmentation framework that utilizes only pseudo-masks and language alignment, achieving pixel-level segmentation without manual annotations.
Exploring Sparse MoE in GANs for Text-conditioned Image Synthesis
Jiapeng Zhu (Ant Group), Yujun Shen (CUHK)
GenerationData SynthesisMixture of ExpertsGenerative Adversarial NetworkImageText
🎯 What it does: This paper presents Aurora, a GAN text-to-image generator based on sparse Mixture-of-Experts, which achieves high-resolution image generation through a two-stage training process (64×64 base + 512×512 upsampling).
Exploring Temporally-Aware Features for Point Tracking
Inès Hyeonsu Kim (KAIST), Seungryong Kim (KAIST)
Object TrackingTransformerContrastive LearningVideo
🎯 What it does: A temporal-aware feature backbone called Chrono for point tracking is proposed, which directly adds a temporal adapter on the pre-trained DINOv2;
Exploring the Deep Fusion of Large Language Models and Diffusion Transformers for Text-to-Image Synthesis
Bingda Tang (New York University), Saining Xie (New York University)
GenerationData SynthesisTransformerLarge Language ModelDiffusion modelRectified FlowImageTextMultimodality
🎯 What it does: A systematic empirical study of the text-to-image generation method that deeply integrates large language models (LLM) with diffusion transformers (DiT) is conducted, comparing it with a shallow fusion baseline, and providing a reproducible training and evaluation process.
Exploring Timeline Control for Facial Motion Generation
Yifeng Ma (Tsinghua University), Liefeng Bo (Alibaba Group)
GenerationTransformerDiffusion modelVideoText
🎯 What it does: A new control signal for facial action generation, called timeline control, is proposed, enabling the synthesis of facial expressions with fine timing.
Exploring Visual Vulnerabilities via Multi-Loss Adversarial Search for Jailbreaking Vision-Language Models
Shuyang Hao, Yujun Cai
Computational EfficiencyAdversarial AttackConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a new computer vision method aimed at improving the accuracy and efficiency of image processing.
Exposure-slot: Exposure-centric Representations Learning with Slot-in-Slot Attention for Region-aware Exposure Correction
Donggoo Jung (Hanyang University), Tae Hyun Kim (Toronto Metropolitan University)
RestorationRepresentation LearningConvolutional Neural NetworkRecurrent Neural NetworkPrompt EngineeringImage
🎯 What it does: Proposes the Exposure-slot framework, which achieves exposure correction of multi-exposure images through a U-shaped encoder-decoder.
Extrapolating and Decoupling Image-to-Video Generation Models: Motion Modeling is Easier Than You Think
Jie Tian (Huazhong University of Science and Technology), Yu Cheng (Chinese University of Hong Kong)
GenerationData SynthesisTransformerDiffusion modelImageVideo
🎯 What it does: This paper proposes a three-stage Extrapolating and Decoupling framework to improve motion controllability and motion amplitude in image-to-video (I2V) generation, achieved through a lightweight adapter, training-free extrapolation, and parameter decoupling.
Extreme Rotation Estimation in the Wild
Hana Bezalel (Tel Aviv University), Hadar Averbach-Elor (Cornell University)
Pose EstimationTransformerDiffusion modelImageBenchmark
🎯 What it does: A method based on Transformer and a new ExtremeLandmarkPairs dataset are proposed for estimating the relative 3D rotation of Internet image pairs, suitable for extreme perspectives, non-overlapping, different intrinsic parameters, and lighting conditions.
EZSR: Event-based Zero-Shot Recognition
Yan Yang (Australian National University), Liu Liu (Huawei)
RecognitionTransformerContrastive LearningImage
🎯 What it does: A zero-shot recognition framework for event cameras based on CLIP has been designed, utilizing scalar modulation and training an event encoder with large-scale synthetic event data to achieve alignment between event data and text embeddings.
F-LMM: Grounding Frozen Large Multimodal Models
Size Wu (Nanyang Technological University), Chen Change Loy (Nanyang Technological University)
Object DetectionSegmentationConvolutional Neural NetworkTransformerVision Language ModelImageMultimodality
🎯 What it does: Utilizing the frozen large multimodal model (LMM) word-image attention maps, a lightweight CNN mask decoder and SAM mask refiners are constructed to achieve visual alignment without fine-tuning the LMM parameters.
F^3OCUS - Federated Finetuning of Vision-Language Foundation Models with Optimal Client Layer Updating Strategy via Multi-objective Meta-Heuristics
Pramit Saha (University of Oxford), J. Alison Noble (University of Oxford)
OptimizationFederated LearningTransformerVision Language ModelMultimodalityBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper proposes a federated fine-tuning framework for visual language models called FOCUS 3, which combines client layer importance scoring with server-side multi-objective meta-heuristic optimization to achieve efficient layer selection and global layer diversity balance.
Face Forgery Video Detection via Temporal Forgery Cue Unraveling
Zonghui Guo (Ocean University of China), Shiguang Shan (Chinese Academy of Sciences)
Anomaly DetectionTransformerVideo
🎯 What it does: A facial forgery video detection framework TFCU is proposed, which progressively excavates temporal forgery clues (instantaneous anomalies, gradual inconsistencies, cumulative distortions) to enhance detection robustness.
FaceBench: A Multi-View Multi-Level Facial Attribute VQA Dataset for Benchmarking Face Perception MLLMs
Xiaoqin Wang (Shenzhen University), Linlin Shen (Shenzhen University)
RecognitionTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageMultimodalityBenchmark
🎯 What it does: This paper presents the FaceBench dataset and the Face-LLaVA baseline model for evaluating the capabilities of multimodal large language models in facial attribute recognition and question answering.
FactCheXcker: Mitigating Measurement Hallucinations in Chest X-ray Report Generation Models
Alice Heiman (Stanford University), Pranav Rajpurkar (Harvard University)
Object DetectionGenerationTransformerLarge Language ModelPrompt EngineeringImageTextBiomedical DataComputed Tomography
🎯 What it does: The FactCheXcker framework is proposed to correct measurement hallucinations in chest X-ray reports through a query-code-update process.
Factored-NeuS: Reconstructing Surfaces, Illumination, and Materials of Possibly Glossy Objects
Yue Fan (Chongqing University), Yiqun Wang (Chongqing University)
RestorationNeural Radiance FieldImage
🎯 What it does: This study proposes an inverse rendering framework called Factored-NeuS, which can simultaneously recover the geometry, material, and lighting information of objects from multi-view images, particularly achieving fine reconstruction for glossy objects.
FADA: Fast Diffusion Avatar Synthesis with Mixed-Supervised Multi-CFG Distillation
Tianyun Zhong (Zhejiang University), Zhou Zhao (Zhejiang University)
GenerationData SynthesisComputational EfficiencyKnowledge DistillationDiffusion modelOrdinary Differential EquationAudio
🎯 What it does: A fast speech-driven avatar synthesis framework FADA based on diffusion models has been designed, which significantly improves inference speed while maintaining high quality.
FADE: Frequency-Aware Diffusion Model Factorization for Video Editing
Yixuan Zhu (Tsinghua University), Jie Zhou (Tsinghua University)
GenerationData SynthesisTransformerDiffusion modelVideo
🎯 What it does: This paper proposes an untrained frequency-aware video editing framework FADE, which utilizes frequency chunking and spectral-guided modulation in a pre-trained text-to-video diffusion model to achieve high-fidelity editing of appearance and motion.
FaithDiff: Unleashing Diffusion Priors for Faithful Image Super-resolution
Junyang Chen (Nanjing University of Science and Technology), Jiangxin Dong (Nanjing University of Science and Technology)
RestorationSuper ResolutionTransformerDiffusion modelAuto EncoderImage
🎯 What it does: This paper proposes a method called FaithDiff for trustworthy image super-resolution using latent diffusion models (LDM), which can restore high-quality images while maintaining structural consistency.
FALCON: Fairness Learning via Contrastive Attention Approach to Continual Semantic Scene Understanding
Thanh-Dat Truong (University of Arkansas), Khoa Luu (University of Arkansas)
Object DetectionSegmentationKnowledge DistillationTransformerContrastive LearningImage
🎯 What it does: This paper proposes FALCON, a fairness learning and contrastive attention method for continuous semantic scene understanding, aimed at addressing catastrophic forgetting, background shift, and fairness issues in continuous semantic segmentation.
FAM Diffusion: Frequency and Attention Modulation for High-Resolution Image Generation with Stable Diffusion
Haosen Yang (Samsung AI Center), Brais Martinez (Samsung AI Center)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: A training-free, single-channel high-resolution image generation method called FAM Diffusion is proposed, which enhances the generation quality at higher resolutions through frequency modulation and attention modulation modules without modifying the original diffusion model.
Fancy123: One Image to High-Quality 3D Mesh Generation via Plug-and-Play Deformation
Qiao Yu (Huazhong University of Science and Technology), Min Chen (South China University of Technology)
GenerationDiffusion modelImageMesh
🎯 What it does: This paper presents a framework named Fancy123, which achieves the generation of high-quality 3D meshes from a single image, focusing on enhancing multi-view consistency and fidelity of the input image through 2D and 3D deformation modules, and significantly improving texture clarity through a 'projection' operation.
Fast and Accurate Gigapixel Pathological Image Classification with Hierarchical Distillation Multi-Instance Learning
Jiuyang Dong (Harbin Institute of Technology), Yongbing Zhang (Harbin Institute of Technology)
ClassificationComputational EfficiencyKnowledge DistillationConvolutional Neural NetworkImageBiomedical Data
🎯 What it does: A hierarchical distillation multi-instance learning framework (HDMIL) is proposed, which first trains a self-distillation dynamic multi-instance network (DMIN) using high-resolution WSI to generate patch importance masks, and then trains a lightweight pre-screening network (LIPN) using low-resolution WSI, significantly reducing inference time while maintaining or even improving classification performance.
Fast3R: Towards 3D Reconstruction of 1000+ Images in One Forward Pass
Jianing Yang (Meta FAIR), Matt Feiszli (Meta FAIR)
Pose EstimationComputational EfficiencyTransformerVision Language ModelImage
🎯 What it does: We propose Fast3R, a multi-view 3D reconstruction framework based on Transformer, capable of processing over 1000 unordered, unposed images in a single forward pass;
Faster Parameter-Efficient Tuning with Token Redundancy Reduction
Kwonyoung Kim (Yonsei University), Kwanghoon Sohn (Korea Institute of Science and Technology)
Computational EfficiencyTransformerSupervised Fine-TuningImage
🎯 What it does: The FPET framework is proposed, which inserts a differentiable token rearrangement module in the middle layers of ViT, utilizing a lightweight adapter to learn token similarity, reducing token redundancy, thereby achieving faster inference and lower training costs while maintaining parameter efficiency.
FASTer: Focal token Acquiring-and-Scaling Transformer for Long-term 3D Objection Detection
Chenxu Dang (Huazhong University of Science and Technology), Jie Ma (Huazhong University of Science and Technology)
Object DetectionAutonomous DrivingTransformerPoint Cloud
🎯 What it does: For multi-frame LiDAR point clouds, we propose FASTer, a lightweight long-term 3D object detection framework that dynamically acquires focus tokens and scales them within the transformer.
FastVLM: Efficient Vision Encoding for Vision Language Models
Pavan Kumar Anasosalu Vasu (Apple), Hadi Pouransari (Apple)
Computational EfficiencyTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: Proposes FastVLM, an efficient vision-language model;
FATE: Full-head Gaussian Avatar with Textural Editing from Monocular Video
Jiawei Zhang (Nanjing University), Hao Zhu (Nanjing University)
RestorationGenerationData SynthesisNeural Radiance FieldGenerative Adversarial NetworkVideo
🎯 What it does: This paper reconstructs editable full-head 3D avatars from monocular video, achieving 360° panoramic rendering and texture editing.
FDS: Frequency-Aware Denoising Score for Text-Guided Latent Diffusion Image Editing
Yufan Ren (Ecole Polytechnique Federale de Lausanne), Sabine Süsstrunk (Ecole Polytechnique Federale de Lausanne)
GenerationOptimizationDiffusion modelScore-based ModelImage
🎯 What it does: A frequency-aware denoising scoring method based on discrete wavelet transform is proposed, which can optimize specific frequency sub-bands in text-guided latent diffusion image editing, thereby preserving detail and color consistency.
Feat2GS: Probing Visual Foundation Models with Gaussian Splatting
Yue Chen (Zhejiang University), Yuliang Xiu (Max Planck Institute for Intelligent Systems)
Data SynthesisRepresentation LearningGaussian SplattingImagePoint Cloud
🎯 What it does: The Feat2GS framework is proposed, which utilizes 2D features from visual foundation models to read 3D Gaussian parameters through a lightweight MLP, assessing the geometric and texture 3D cognitive abilities of VFM from unlabeled multi-view images.
Feature Information Driven Position Gaussian Distribution Estimation for Tiny Object Detection
Jinghao Bian (Xidian University), Guangming Shi (Xidian University)
Object DetectionGaussian SplattingImage
🎯 What it does: A feature enhancement module based on pixel information entropy and positional Gaussian distribution is proposed to improve small object detection performance.
Feature Selection for Latent Factor Models
Rittwika Kansabanik (Florida State University), Adrian Barbu (Florida State University)
ClassificationData-Centric LearningGenerative Adversarial NetworkImage
🎯 What it does: A class-specific feature selection method based on low-rank latent factor models was developed, utilizing signal-to-noise ratio (SNR) to identify important features for multi-class tasks.
Feature Spectrum Learning for Remote Sensing Change Detection
Qi Zang (Xidian University), Zhun Zhong (Hefei University of Technology)
SegmentationConvolutional Neural NetworkTransformerSupervised Fine-TuningImage
🎯 What it does: This paper studies a technique for eliminating pseudo-changes in remote sensing time series images—Feature Spectrum Learning (FeaSpect). It achieves style alignment by applying spectral transformation (FFT) in the feature space, thereby improving the accuracy of change detection (CD).