CVPR 2025 Papers — Page 18
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2871 papers
Multi-Sensor Object Anomaly Detection: Unifying Appearance, Geometry, and Internal Properties
Wenqiao Li (ShanghaiTech University), Yingna Wu (ShanghaiTech University)
Object DetectionAnomaly DetectionContrastive LearningImageMultimodalityPoint Cloud
🎯 What it does: This paper proposes a multi-sensor industrial defect detection dataset called MulSen-AD and constructs a multi-modal fusion baseline method named MulSen-TripleAD based on this dataset.
Multi-subject Open-set Personalization in Video Generation
Tsai-Shien Chen (Snap Inc.), Sergey Tulyakov (Snap Inc.)
GenerationData SynthesisTransformerDiffusion modelVideoText
🎯 What it does: We propose Video Alchemist, a generative model capable of multi-subject and open-set personalization for videos.
Multi-View Pose-Agnostic Change Localization with Zero Labels
Chamuditha Jayanga Galappaththige (Queensland University of Technology), Dimity Miller (Queensland University of Technology)
SegmentationAnomaly DetectionGaussian SplattingSimultaneous Localization and MappingImageBenchmark
🎯 What it does: A label-free, pose-agnostic multi-view change detection method is proposed, utilizing 3D Gaussian Splatting to construct a change-aware 3D scene representation, thereby generating multi-view change masks.
Multi-view Reconstruction via SfM-guided Monocular Depth Estimation
Haoyu Guo (Zhejiang University), Hujun Bao (Zhejiang University)
Depth EstimationDiffusion modelPoint Cloud
🎯 What it does: This paper proposes a multi-view 3D reconstruction pipeline named Murre, which first generates a sparse point cloud using Structure from Motion (SfM), then projects it into a sparse depth map and inputs it into a monocular depth estimation network based on a diffusion model through KNN interpolation and distance maps, resulting in a multi-view consistent metric depth map. Finally, scene reconstruction is completed through TSDF/point cloud fusion.
MultiGO: Towards Multi-level Geometry Learning for Monocular 3D Textured Human Reconstruction
Gangjian Zhang (Hong Kong University of Science and Technology), Hao Wang (Hong Kong University of Science and Technology)
GenerationPose EstimationDiffusion modelGaussian SplattingImage
🎯 What it does: The MultiGO framework is proposed, achieving single-view 3D textured human reconstruction through skeletal-level enhancement, joint-level enhancement, and wrinkle-level refinement.
Multimodal Autoregressive Pre-training of Large Vision Encoders
Enrico Fini (Apple), Alaaeldin El-Nouby (Apple)
RecognitionObject DetectionGenerationRepresentation LearningTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: A multi-modal autoregressive pre-training framework AIMV2 is proposed, which uses a single visual encoder and a multi-modal decoder, capable of simultaneously reconstructing image patches and text tokens, achieving general pre-training for large visual encoders.
MultimodalStudio: A Heterogeneous Sensor Dataset and Framework for Neural Rendering across Multiple Imaging Modalities
Federico Lincetto, Pietro Zanuttigh
GenerationData SynthesisNeural Radiance FieldImageMultimodality
🎯 What it does: The MultimodalStudio system is proposed, which includes the multimodal dataset MMS-DATA and the multimodal NeRF framework MMS-FW, utilizing five different imaging modalities (RGB, monochrome, near-infrared, polarization, multispectral) for cross-modal neural rendering and information transfer.
MultiMorph: On-demand Atlas Construction
S. Mazdak Abulnaga (Massachusetts Institute of Technology), Adrian Dalca (Massachusetts Institute of Technology)
SegmentationGenerationComputational EfficiencyConvolutional Neural NetworkImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A method based on deep learning is proposed to generate multi-modal, customizable population-specific anatomical templates (atlas) with a single forward pass, significantly shortening the traditional atlas construction time.
Multiple Object Tracking as ID Prediction
Ruopeng Gao (Nanjing University), Limin Wang (Nanjing University)
Object TrackingTransformerVideo
🎯 What it does: This paper rephrases multi-object tracking (MOT) as a 'context ID prediction' task, using an end-to-end DETR framework to directly predict the ID of targets in the current frame, rather than traditional matching or ReID schemes.
Multirate Neural Image Compression with Adaptive Lattice Vector Quantization
Hao Xu (McMaster University), Xi Zhang (Nanyang Technological University)
CompressionDomain AdaptationImage
🎯 What it does: An Adaptive Lattice Vector Quantization (Adaptive LVQ) method is proposed, which can achieve variable bit rates and domain adaptation within the same network.
Multitwine: Multi-Object Compositing with Text and Layout Control
Gemma Canet Tarrés (University of Surrey), Soo Ye Kim (Adobe Research)
GenerationData SynthesisTransformerLarge Language ModelVision Language ModelDiffusion modelImageVideoTextMultimodality
🎯 What it does: The first generative model capable of simultaneously performing multi-object synthesis while considering both text and layout control is proposed.
MultiVENT 2.0: A Massive Multilingual Benchmark for Event-Centric Video Retrieval
Reno Kriz (Johns Hopkins University), Benjamin Van Durme (Johns Hopkins University)
RetrievalVision Language ModelVideoMultimodalityBenchmark
🎯 What it does: This paper constructs and benchmarks a large-scale multilingual event-centric video retrieval dataset, MULTIVENT 2.0, which includes over 218,000 videos and more than 3,900 queries for specific events, and evaluates the performance of multimodal retrieval models.
MUST: The First Dataset and Unified Framework for Multispectral UAV Single Object Tracking
Haolin Qin (Beijing Institute of Technology), Jianan Li (Beijing Institute of Technology)
Object TrackingTransformerVideo
🎯 What it does: The first multispectral drone single-target tracking dataset, MUST, is proposed, and based on this dataset, a unified spectral-spatial-temporal tracking framework, UNTrack, is introduced.
MUSt3R: Multi-view Network for Stereo 3D Reconstruction
Yohann Cabon (Naver Labs Europe), Vincent Leroy (Naver Labs Europe)
Pose EstimationDepth EstimationTransformerImage
🎯 What it does: This paper presents MUSt3R, which can perform dense 3D reconstruction, camera pose, and focal length estimation for multi-view image sets without calibration information.
MuTri: Multi-view Tri-alignment for OCT to OCTA 3D Image Translation
Zhuangzhuang Chen (Hong Kong University of Science and Technology), Xiaomeng Li (Hong Kong University of Science and Technology)
Image TranslationData SynthesisConvolutional Neural NetworkAuto EncoderContrastive LearningImageBiomedical Data
🎯 What it does: Proposes the MuTri multi-view triangular alignment framework, which transforms the 3D image translation from OCT to OCTA into a VQ-VAE model in a discrete finite space, utilizing three perspectives of 3D OCT, 3D OCTA, and 2D OCTA projections for multi-view guidance;
MV-DUSt3R+: Single-Stage Scene Reconstruction from Sparse Views In 2 Seconds
Zhenggang Tang (Meta Reality Labs), Zhicheng Yan (Meta Reality Labs)
Pose EstimationDepth EstimationComputational EfficiencyTransformerGaussian SplattingImage
🎯 What it does: A single-stage, pose-free, multi-view dense 3D reconstruction network MV-DUSt3R and its improved version MV-DUSt3R+ are proposed, capable of recovering large-scale scenes from multiple uncalibrated RGB images and achieving novel view synthesis in one go.
MV-MATH: Evaluating Multimodal Math Reasoning in Multi-Visual Contexts
Peijie Wang (Institute of Automation of Chinese Academy of Sciences), Cheng-Lin Liu (Institute of Automation of Chinese Academy of Sciences)
TransformerLarge Language ModelPrompt EngineeringImageTextMultimodalityBenchmarkChain-of-Thought
🎯 What it does: The MV-MATH benchmark dataset is proposed, containing 2009 real K-12 math problems, each with multiple interrelated or independent images, and systematically evaluates the mathematical reasoning capabilities of multimodal large language models (MLLMs) in a multi-image environment.
MV-SSM: Multi-View State Space Modeling for 3D Human Pose Estimation
Aviral Chharia (Carnegie Mellon University), Haoye Dong (National University of Singapore)
Pose EstimationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a Multi-View State Space Model (MV-SSM) for 3D human pose estimation.
MVBoost: Boost 3D Reconstruction with Multi-View Refinement
Xiangyu Liu (Institute of Automation, Chinese Academy of Sciences), Zhen Lei
RestorationGenerationDiffusion modelImage
🎯 What it does: This paper proposes a framework called MVBoost that utilizes a multi-view diffusion model to generate pseudo ground truth and enhances single image to 3D reconstruction, ultimately training a feedforward 3D reconstruction network without the need for real 3D data.
MVDoppler-Pose: Multi-Modal Multi-View mmWave Sensing for Long-Distance Self-Occluded Human Walking Pose Estimation
Jaeho Choi (DGIST), Amin Arbabian (Stanford University)
Pose EstimationTransformerMultimodality
🎯 What it does: This paper proposes a multimodal multi-view pose estimation framework based on millimeter-wave radar, called MVDoppler-Pose, which can achieve 3D human pose estimation in long-distance and self-occlusion scenarios.
MVGenMaster: Scaling Multi-View Generation from Any Image via 3D Priors Enhanced Diffusion Model
Chenjie Cao (Alibaba Group), Yanwei Fu (Fudan University)
GenerationData SynthesisDepth EstimationDiffusion modelImage
🎯 What it does: MVGenMaster is proposed, a framework that integrates 3D priors into multi-view diffusion models, enabling single forward inference for arbitrary reference and target views.
MVPaint: Synchronized Multi-View Diffusion for Painting Anything 3D
Wei Cheng (StepFun), Liang Pan (Shanghai AI Laboratory)
GenerationData SynthesisDiffusion modelPoint CloudMeshBenchmark
🎯 What it does: This paper presents MVPaint, a 3D texture generation framework based on synchronous multi-view diffusion, which can generate high-quality, seamless, and multi-view consistent textures for 3D meshes based on textual instructions.
MVPortrait: Text-Guided Motion and Emotion Control for Multi-view Vivid Portrait Animation
Yukang Lin (Tsinghua University), Xiu Li (Tsinghua University)
GenerationData SynthesisDiffusion modelVideoText
🎯 What it does: A two-stage text-driven multi-view portrait animation framework MVPortrait has been developed, utilizing the FLAME model to achieve control over facial movements, expressions, and viewpoints.
MVSAnywhere: Zero-Shot Multi-View Stereo
Sergio Izquierdo (Niantic), Jamie Watson (University of Edinburgh)
Depth EstimationTransformerImage
🎯 What it does: A model capable of zero-shot multi-view stereo reconstruction from arbitrary viewpoints and depth ranges is proposed.
NADER: Neural Architecture Design via Multi-Agent Collaboration
Zekang Yang (SenseTime Research and Tetras.AI), Wentao Liu (SenseTime Research and Tetras.AI)
Neural Architecture SearchGraph Neural NetworkLarge Language ModelAgentic AIImage
🎯 What it does: A neural network architecture design framework called NADER is proposed, which is based on multi-agent collaboration of large language models (LLM) and can automatically generate high-performance models in open spaces.
Narrating the Video: Boosting Text-Video Retrieval via Comprehensive Utilization of Frame-Level Captions
Chan Hur (Kyungpook National University), Hyeyoung Park (Kyungpook National University)
RetrievalTransformerContrastive LearningVideoTextMultimodality
🎯 What it does: This paper proposes the NarVid framework, which significantly improves text-video retrieval performance through four modules: frame-by-frame generated narrative subtitles, cross-modal interaction, query adaptive filtering, narrative matching, and cross-view hard negative sample loss.
Navigating Image Restoration with VAR's Distribution Alignment Prior
Siyang Wang (University of Science and Technology of China), Feng Zhao (University of Science and Technology of China)
RestorationTransformerImage
🎯 What it does: This paper proposes a unified image restoration framework called VarFormer based on the VAR generative model, which can simultaneously perform six tasks: rain removal, dehazing, low-light enhancement, deblurring, and noise suppression within a single model.
Navigating the Unseen: Zero-shot Scene Graph Generation via Capsule-Based Equivariant Features
Wenhuan Huang (Soochow University), Chunping Liu (Soochow University)
Object DetectionGenerationImageMultimodality
🎯 What it does: Proposes the CAPSGG framework, which achieves zero-shot scene graph generation through Capsule networks.
Navigation World Models
Amir Bar (FAIR at Meta), Yann LeCun (Berkeley AI Research)
Autonomous DrivingOptimizationRobotic IntelligenceTransformerDiffusion modelAuto EncoderWorld ModelVideo
🎯 What it does: This paper proposes a Navigation World Model (NWM) that can predict future visual states based on past observations and navigation actions, and utilizes this model for trajectory planning and evaluation.
Nearly Zero-Cost Protection Against Mimicry by Personalized Diffusion Models
Namhyuk Ahn (Inha University), Seung-Hun Nam (NAVER WEBTOON AI)
Adversarial AttackDiffusion modelAuto EncoderImage
🎯 What it does: A low-latency, low-cost image protection framework called FastProtect is proposed, which utilizes pre-trained mixture perturbations to adaptively select perturbations during inference to defend against imitation attacks from personalized diffusion models.
NeighborRetr: Balancing Hub Centrality in Cross-Modal Retrieval
Zengrong Lin (Zhejiang University of Technology), Cong Bai (Zhejiang University of Technology)
RetrievalTransformerContrastive LearningVideoTextMultimodality
🎯 What it does: This paper proposes the NeighborRetr method, which identifies and balances good/bad hubs through sample centrality during cross-modal retrieval training, directly alleviating hubness and improving retrieval performance.
NeISF++: Neural Incident Stokes Field for Polarized Inverse Rendering of Conductors and Dielectrics
Chenhao Li (Sony Semiconductor Solutions Corporation), Yusuke Moriuchi (Sony Semiconductor Solutions Corporation)
Image
🎯 What it does: A polarization inverse rendering pipeline named NeISF++ has been developed, capable of simultaneously reconstructing the geometry, material, and lighting of both conductive and dielectric objects.
NeRFPrior: Learning Neural Radiance Field as a Prior for Indoor Scene Reconstruction
Wenyuan Zhang (Tsinghua University), Zhizhong Han (Wayne State University)
RestorationGenerationNeural Radiance FieldImage
🎯 What it does: This paper proposes the NeRFPrior method, which utilizes a NeRF trained on a single scene as a prior, combining voxel rendering and SDF learning to reconstruct high-quality indoor surfaces from multi-view RGB images.
Nested Diffusion Models Using Hierarchical Latent Priors
Xiao Zhang (University of Chicago), Michael Maire (University of Chicago)
GenerationData SynthesisTransformerDiffusion modelImage
🎯 What it does: Designed and implemented Nested Diffusion Models, utilizing hierarchical semantic latent vectors generated by multi-layer frozen pre-trained visual encoders to gradually reverse diffuse from low-dimensional semantics to high-dimensional image details;
Neural Hierarchical Decomposition for Single Image Plant Modeling
Zhihao Liu (University of Tokyo), Naoto Yokoya (University of Tokyo)
SegmentationGenerationTransformerAuto EncoderImageMeshAgriculture Related
🎯 What it does: A hierarchical box decomposition framework based on neural networks is proposed, capable of automatically generating high-quality 3D plant models from a single plant photo and refining geometric details through parametric modeling.
Neural Inverse Rendering from Propagating Light
Anagh Malik (University of Toronto), David B. Lindell (University of Toronto)
RestorationDepth EstimationNeural Radiance FieldPoint CloudPhysics Related
🎯 What it does: Performing physics-based neural inverse rendering from multi-view, time-resolved flash LiDAR measurements to recover geometry and materials, and to reproduce light propagation under new viewpoints and lighting conditions.
Neural LightRig: Unlocking Accurate Object Normal and Material Estimation with Multi-Light Diffusion
Zexin He (Chinese University of Hong Kong), Ziwei Liu (Nanyang Technological University)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: Proposes the Neural LightRig framework, which utilizes a multi-light diffusion model to generate multi-light images as auxiliary information, jointly predicting the surface normals and PBR materials of a single image.
Neural Motion Simulator Pushing the Limit of World Models in Reinforcement Learning
Chenjie Hao (University of California Davis), Yubei Chen (Open Path AI Foundation)
Robotic IntelligenceReinforcement LearningWorld ModelSequentialOrdinary Differential Equation
🎯 What it does: A world model called Neural Motion Simulator (MoSim) has been constructed to predict robot motion, enabling the prediction of future physical states from the current state and actions.
Neural Video Compression with Context Modulation
Chuanbo Tang (University of Science and Technology of China), Dong Liu (University of Science and Technology of China)
CompressionOptical FlowVideo
🎯 What it does: A deep context modulation neural video compression framework named DCMVC is proposed, which utilizes bidirectional information from reference frames and reference features to generate higher quality temporal context, thereby improving video compression rates.
Neuro-3D: Towards 3D Visual Decoding from EEG Signals
Zhanqiang Guo (Shanghai Artificial Intelligence Laboratory), Chunfeng Song (Shanghai Artificial Intelligence Laboratory)
ClassificationRecognitionGenerationDiffusion modelAuto EncoderContrastive LearningImageVideoPoint Cloud
🎯 What it does: This study proposes decoding 3D visual information from EEG signals and constructs a corresponding EEG-3D dataset.
Neuro-Symbolic Evaluation of Text-to-Video Models using Formal Verification
S P Sharan (University of Texas at Austin), Sandeep Chinchali (University of Texas at Austin)
GenerationData SynthesisTransformerLarge Language ModelPrompt EngineeringVision Language ModelVideoText
🎯 What it does: This paper proposes NeuS-V, a method for evaluating the alignment of text-video pairs by converting text prompts into temporal logic (TL) specifications and constructing automata for generated videos, followed by formal verification.
Neuron: Learning Context-Aware Evolving Representations for Zero-Shot Skeleton Action Recognition
Yang Chen (Hong Kong Polytechnic University), Dacheng Tao (Nanyang Technological University)
RecognitionPose EstimationGraph Neural NetworkTransformerLarge Language ModelPrompt EngineeringVideo
🎯 What it does: This paper addresses the zero-shot skeleton action recognition task and proposes the Neuron framework, which achieves fine-grained alignment through dynamically evolving dual-modal skeleton-semantic micro-prototypes, significantly enhancing the recognition capability for unseen actions.
NexusGS: Sparse View Synthesis with Epipolar Depth Priors in 3D Gaussian Splatting
Yulong Zheng (Ocean University of China), Yong Du (Ocean University of China)
Data SynthesisDepth EstimationGaussian SplattingOptical FlowPoint Cloud
🎯 What it does: This paper presents NexusGS, a novel view synthesis method based on 3D Gaussian Splatting for sparse views, which achieves accurate geometric reconstruction and high-quality rendering by utilizing epipolar depth priors between views.
NightAdapter: Learning a Frequency Adapter for Generalizable Night-time Scene Segmentation
Qi Bi (Westlake University), Yefeng Zheng (Westlake University)
SegmentationDomain AdaptationAutonomous DrivingTransformerSupervised Fine-TuningImage
🎯 What it does: This paper proposes NightAdapter, which achieves efficient generalization of semantic segmentation in nighttime scenes by adapting the Vision Foundation Model in the frequency domain.
NitroFusion: High-Fidelity Single-Step Diffusion through Dynamic Adversarial Training
Dar-Yen Chen (University of Surrey), Yi-Zhe Song (University of Surrey)
GenerationData SynthesisDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: Achieved single-step diffusion model NitroFusion through dynamic adversarial training, generating high-fidelity images in one stroke.
NLPrompt: Noise-Label Prompt Learning for Vision-Language Models
Bikang Pan (ShanghaiTech University), Ye Shi (ShanghaiTech University)
ClassificationRecognitionTransformerPrompt EngineeringVision Language ModelImageMultimodality
🎯 What it does: Research on prompt learning of visual-language models in noisy label environments, proposing a more robust training method.
NN-Former: Rethinking Graph Structure in Neural Architecture Representation
Ruihan Xu (Peking University), Shiliang Zhang (Harbin Institute of Technology)
Neural Architecture SearchGraph Neural NetworkTransformerGraph
🎯 What it does: A neural network predictor called NN-Former, which integrates GNN and Transformer, has been designed and implemented for one-time predictions of accuracy and inference latency of neural architectures.
nnWNet: Rethinking the Use of Transformers in Biomedical Image Segmentation and Calling for a Unified Evaluation Benchmark
Yanfeng Zhou (DAMO Academy, Alibaba Group), Minfeng Xu (DAMO Academy, Alibaba Group)
SegmentationConvolutional Neural NetworkTransformerImageBiomedical DataBenchmark
🎯 What it does: A new UNet variant called WNet is designed, which organically combines Transformers with convolutional modules to achieve continuous transmission of global and local features and multi-scale fusion for biomedical image segmentation.
No Pains, More Gains: Recycling Sub-Salient Patches for Efficient High-Resolution Image Recognition
Rong Qin (Nankai University), Jufeng Yang (Nankai University)
RecognitionComputational EfficiencyTransformerImage
🎯 What it does: Proposes dual-buffer sub-significant patch selection and dual-attention multi-instance learning to enhance the accuracy and efficiency of high-resolution image recognition.
No Thing, Nothing: Highlighting Safety-Critical Classes for Robust LiDAR Semantic Segmentation in Adverse Weather
Junsung Park (Korea Advanced Institute of Science and Technology), Hyunjung Shim (Korea Advanced Institute of Science and Technology)
SegmentationDomain AdaptationAutonomous DrivingKnowledge DistillationConvolutional Neural NetworkPoint Cloud
🎯 What it does: This paper proposes a robust improvement method for LiDAR semantic segmentation in adverse weather conditions, focusing on enhancing the segmentation accuracy of 'things' classes (such as people, vehicles, etc.) that are critical for safety in traffic.
Noise Calibration and Spatial-Frequency Interactive Network for STEM Image Enhancement
Hesong Li (Beijing Institute of Technology), Ying Fu (Beijing Institute of Technology)
RestorationData SynthesisConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: By statistically calibrating the noise in real STEM images, more realistic synthetic data is constructed, and a Spatial-Frequency Interaction Network (SFIN) is proposed to achieve denoising and enhancement of STEM images.
Noise Diffusion for Enhancing Semantic Faithfulness in Text-to-Image Synthesis
Boming Miao (Beijing Normal University), Yao Zhu (Tsinghua University)
GenerationData SynthesisOptimizationLarge Language ModelVision Language ModelDiffusion modelImageText
🎯 What it does: A framework (Noise Diffusion) is proposed to optimize the initial noise of diffusion models using large visual language models (LVLM), guiding the update of initial noise through VQA scores to enhance the semantic fidelity of text-to-image synthesis.
Noise Modeling in One Hour: Minimizing Preparation Efforts for Self-supervised Low-Light RAW Image Denoising
Feiran Li (Sony Research), Daisuke Iso (Sony Research)
RestorationImage
🎯 What it does: This paper proposes a noise synthesis pipeline for low-light RAW image denoising that only requires the collection of dark frames, avoiding system gain calibration and signal-independent noise analysis.
Noise-Consistent Siamese-Diffusion for Medical Image Synthesis and Segmentation
Kunpeng Qiu (National University of Singapore), Yongxin Guo (City University of Hong Kong)
SegmentationGenerationData SynthesisDiffusion modelImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A dual-branch diffusion model called Siamese-Diffusion is proposed, which generates high morphological fidelity and diversified medical images using Mask-Diffusion and Image-Diffusion. It guides Mask-Diffusion to converge to higher quality local minima in the parameter space through Noise Consistency Loss, ultimately sampling only with Mask-Diffusion.
Noise-Resistant Video Anomaly Detection via RGB Error-Guided Multiscale Predictive Coding and Dynamic Memory
Han Hu (East China University of Science and Technology), Siyuan Fan (East China University of Science and Technology)
Anomaly DetectionConvolutional Neural NetworkRecurrent Neural NetworkContrastive LearningVideo
🎯 What it does: This paper proposes an end-to-end video anomaly detection method that combines an RGB error-guided multi-scale prediction coding framework (EG-MPC) with a dynamic memory module (DMM). It first predicts the next frame and then uses the DMM for frame reconstruction, thereby obtaining anomaly scores during testing.
NoiseCtrl: A Sampling-Algorithm-Agnostic Conditional Generation Method for Diffusion Models
Longquan Dai (Nanjing University of Science and Technology), Jinhui Tang (Nanjing University of Science and Technology)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: A training-free conditional generation method named NoiseCtrl is proposed, which achieves conditional control over image generation by replacing standard Gaussian noise with directional noise during the reverse sampling process of the diffusion model.
Non-Natural Image Understanding with Advancing Frequency-based Vision Encoders
Wang Lin (Zhejiang University), Jingyuan Chen (Zhejiang University)
ClassificationRetrievalTransformerVision Language ModelContrastive LearningImageText
🎯 What it does: A frequency modulation visual encoder FM-ViT is proposed and integrated into MLLM EDGE, specifically to enhance the understanding and reasoning of non-natural images such as geometric shapes, charts, and function graphs.
Nonisotropic Gaussian Diffusion for Realistic 3D Human Motion Prediction
Cecilia Curreli (Technical University of Munich), Daniel Cremers (Technical University of Munich)
GenerationData SynthesisPose EstimationGraph Neural NetworkDiffusion modelAuto EncoderSequential
🎯 What it does: We propose SkeletonDiffusion, a probabilistic human motion prediction model that uses nonisotropic Gaussian diffusion in latent space, capable of generating diverse and realistic future action sequences.
NoPain: No-box Point Cloud Attack via Optimal Transport Singular Boundary
Zezeng Li (Dalian University of Technology), Weimin Wang (Dalian University of Technology)
Adversarial AttackAuto EncoderPoint Cloud
🎯 What it does: A no-box point cloud adversarial attack framework called NoPain is proposed, which identifies singular boundaries on the data manifold by calculating the semi-discrete optimal transport (OT) mapping of noise to the feature space, and generates transferable adversarial samples by sampling on these boundaries.
Not All Parameters Matter: Masking Diffusion Models for Enhancing Generation Ability
Lei Wang (Nankai University), Jian Yang (Nankai University)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: A MaskUNet is designed, which dynamically selects effective parameters by applying a learnable binary mask to the pre-trained U-Net during inference, improving image generation quality without updating any original parameters.
Not Just Text: Uncovering Vision Modality Typographic Threats in Image Generation Models
Hao Cheng (Hong Kong University of Science and Technology), Renjing Xu (Hong Kong University of Science and Technology)
GenerationAdversarial AttackDiffusion modelImageBenchmark
🎯 What it does: This paper reveals the security risks of generating harmful content through font attacks in visual modality inputs of image generation models and proposes a corresponding dataset.
Not Only Text: Exploring Compositionality of Visual Representations in Vision-Language Models
Davide Berasi (Fondazione Bruno Kessler), Nicola Strisciuglio
ClassificationRepresentation LearningVision Language ModelDiffusion modelImageMultimodality
🎯 What it does: This paper studies the composability of visual embeddings in Visual-Language Models (VLM) and proposes a Geodesically Decomposable Embeddings (GDE) framework based on Riemannian geometry to decompose visual concepts and suppress noise and sparsity.
NoT: Federated Unlearning via Weight Negation
Yasser H. Khalil (Huawei Noah's Ark Lab), Xi Chen (Huawei Technologies Canada Inc)
Federated LearningConvolutional Neural NetworkTransformerSupervised Fine-TuningImage
🎯 What it does: A federated model no-learning algorithm (NoT) is proposed that requires no additional storage and does not need to access forgotten data, achieving effective forgetting of target data by inverting the weights of specific layers of the model.
Notes-guided MLLM Reasoning: Enhancing MLLM with Knowledge and Visual Notes for Visual Question Answering
Wenlong Fang (South China Normal University), Yun Xue (South China Normal University)
RecognitionRetrievalTransformerLarge Language ModelVision Language ModelImageTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: A knowledge-driven visual question answering (KB-VQA) framework called NoteMR is proposed, based on knowledge notes and visual notes for MLLM reasoning.
Novel View Synthesis with Pixel-Space Diffusion Models
Noam Elata (Technion), Ron Sokolovsky (Apple)
GenerationData SynthesisDepth EstimationDiffusion modelImage
🎯 What it does: This paper proposes and trains an end-to-end pixel-space diffusion model called VIVID, which generates target views from a single input image at arbitrary angles, completely replacing traditional depth estimation, photometric mapping, and completion pipelines.
NSD-Imagery: A Benchmark Dataset for Extending fMRI Vision Decoding Methods to Mental Imagery
Reese Kneeland (University of Minnesota), Thomas Naselaris (University of Minnesota)
Diffusion modelContrastive LearningImageMultimodalityMagnetic Resonance ImagingBenchmark
🎯 What it does: This paper presents the NSD-Imagery dataset, which expands the existing NSD visual fMRI data and uses this dataset to evaluate and compare the generalization performance of various fMRI-to-Image decoding models on mental imagery.
NTClick: Achieving Precise Interactive Segmentation With Noise-tolerant Clicks
Chenyi Zhang (Beijing Jiaotong University), Yunchao Wei (Beijing Jiaotong University)
SegmentationTransformerImage
🎯 What it does: This paper proposes NTClick, which achieves high-precision interactive segmentation using noise-tolerant clicks and implements a coarse-to-fine segmentation through a two-stage network.
NTR-Gaussian: Nighttime Dynamic Thermal Reconstruction with 4D Gaussian Splatting Based on Thermodynamics
Kun Yang (Northwestern Polytechnical University), Qing Wang (National University of Defense Technology)
RestorationData SynthesisGaussian SplattingOptical FlowImage
🎯 What it does: A nighttime dynamic thermal map reconstruction method based on 4D Gaussian Splatting, NTR-Gaussian, is proposed, which can predict the temperature distribution of outdoor scenes at different time points.
Nullu: Mitigating Object Hallucinations in Large Vision-Language Models via HalluSpace Projection
Le Yang (Xi'an Jiaotong University), Chao Shen (Xi'an Jiaotong University)
Object DetectionGenerationOptimizationTransformerLarge Language ModelVision Language ModelTextMultimodality
🎯 What it does: To address the issue of object hallucination in large visual-language models, the Nullu method is proposed. It obtains a low-rank subspace HalluSpace through PCA/SVD decomposition of the feature differences between ground truth and hallucination samples, and then projects the model's MLP weights onto its orthogonal complement space to suppress hallucination information during inference.
Number it: Temporal Grounding Videos like Flipping Manga
Yongliang Wu (Southeast University), Xu Yang (Southeast University)
RetrievalTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringVision Language ModelVideoText
🎯 What it does: A lightweight method for video large language models (Vid-LLMs) is proposed to achieve precise video temporal grounding (VTG) by overlaying numerical identifiers (Number-Prompt, NumPro) on each frame.
NVComposer: Boosting Generative Novel View Synthesis with Multiple Sparse and Unposed Images
Lingen Li (Chinese University of Hong Kong), Ying Shan (Tencent PCG)
GenerationData SynthesisPose EstimationDiffusion modelImageVideo
🎯 What it does: A generative new view synthesis model NVComposer is proposed, which can generate high-quality new view images using any number of uncalibrated views without external multi-view alignment.
NVILA: Efficient Frontier Visual Language Models
Zhijian Liu (NVIDIA), Yao Lu (MIT)
CompressionComputational EfficiencyTransformerSupervised Fine-TuningVision Language ModelImageVideoTextMultimodality
🎯 What it does: A visual language model named NVILA is proposed, which achieves efficient and accurate multimodal understanding by first enlarging the image/video resolution and then compressing the visual tokens.
O-TPT: Orthogonality Constraints for Calibrating Test-time Prompt Tuning in Vision-Language Models
Ashshak Sharifdeen (Mohamed bin Zayed University of AI), Muhammad Haris Khan (Mohamed bin Zayed University of AI)
ClassificationDomain AdaptationOptimizationPrompt EngineeringVision Language ModelImageText
🎯 What it does: This paper proposes a calibration method for Visual Language Models (VLM) in Test-time Prompt Tuning (TPT), called O-TPT, which enhances the angular separation of text features by introducing orthogonal constraints during the prompt learning process, thereby reducing the model's uncertainty error.
Object Detection using Event Camera: A MoE Heat Conduction based Detector and A New Benchmark Dataset
Xiao Wang (Anhui University), Yonghong Tian (Peking University)
Object DetectionConvolutional Neural NetworkTransformerMixture of ExpertsImageBenchmark
🎯 What it does: This paper proposes an event camera object detection framework based on Mixture-of-Experts thermal conduction (MvHeat-DET) and releases the EvDET200K dataset for the first time, which contains 10,054 entries and 200,000 bounding boxes.
Object-aware Sound Source Localization via Audio-Visual Scene Understanding
Sung Jin Um (Kyung Hee University), Jung Uk Kim (Kyung Hee University)
RecognitionObject DetectionConvolutional Neural NetworkLarge Language ModelContrastive LearningMultimodalityAudio
🎯 What it does: This paper proposes a sound and visual source localization framework based on a multimodal large language model (MLLM), utilizing foreground/background text generated by MLLM for fine localization.
Object-Centric Prompt-Driven Vision-Language-Action Model for Robotic Manipulation
Xiaoqi Li (Peking University), Hao Dong (Peking University)
Robotic IntelligenceTransformerLarge Language ModelPrompt EngineeringVision-Language-Action ModelMultimodality
🎯 What it does: This paper proposes a method that utilizes multimodal visual cues from CrayonRobo (color indicators for contact points, posture, and subsequent motion direction) to guide robots in precise control from keyframes to long-term tasks.
Object-Shot Enhanced Grounding Network for Egocentric Video
Yisen Feng (Harbin Institute of Technology), Liqiang Nie (Harbin Institute of Technology)
Object DetectionRetrievalTransformerContrastive LearningVideoTextMultimodality
🎯 What it does: This paper proposes a network called OSGNet, which is used to accurately locate video segments in first-person (egocentric) videos based on natural language queries.
ObjectMover: Generative Object Movement with Video Prior
Xin Yu (University of Hong Kong), Xiaojuan Qi (University of Hong Kong)
GenerationData SynthesisTransformerDiffusion modelImageVideo
🎯 What it does: This paper proposes ObjectMover, a single-stage generative method based on video diffusion models, designed to smoothly move target objects within a single image while maintaining consistency in object identity, lighting, shadows, perspective, and other attributes, as well as completing occlusion filling and material recognition tasks.
Occlusion-aware Text-Image-Point Cloud Pretraining for Open-World 3D Object Recognition
Khanh Nguyen (University of Western Australia), Ajmal Mian (University of Western Australia)
RecognitionObject DetectionDomain AdaptationComputational EfficiencyConvolutional Neural NetworkTransformerContrastive LearningMultimodalityPoint Cloud
🎯 What it does: This study proposes an occlusion-aware text-image-point cloud pre-training framework (OccTIP) and designs an efficient DuoMamba network to address the recognition problem of point clouds in the open world.
OccMamba: Semantic Occupancy Prediction with State Space Models
Heng Li (University of Science and Technology of China), Yanyong Zhang (University of Science and Technology of China)
SegmentationAutonomous DrivingConvolutional Neural NetworkMultimodalityPoint Cloud
🎯 What it does: This paper proposes the first outdoor semantic occupancy prediction network, OccMamba, based on the Mamba state space model, which can directly and efficiently handle millions of dense voxels.
OCRT: Boosting Foundation Models in the Open World with Object-Concept-Relation Triad
Luyao Tang (Xiamen University), Kun Zhang (Carnegie Mellon University)
Object DetectionDomain AdaptationAdversarial AttackGraph Neural NetworkSupervised Fine-TuningContrastive LearningImage
🎯 What it does: This paper proposes a framework based on the Object-Concept-Relation Trinity (OCR T), utilizing unsupervised Slot-Attention to iteratively decompose images into object-level representations, which are then projected into a sparse concept space and filtered for key information through importance weighting. Subsequently, a variable-degree concept graph is constructed for high-order relational reasoning, and finally, relational tokens are injected into the fine-tuning process of SAM and CLIP to enhance their generalization and robustness in scenarios of distribution drift, weak supervision, and attacks.
Octopus: Alleviating Hallucination via Dynamic Contrastive Decoding
Wei Suo (Northwestern Polytechnical University), Yanning Zhang (Northwestern Polytechnical University)
GenerationOptimizationTransformerReinforcement LearningVision Language ModelMultimodality
🎯 What it does: To address the hallucination problem in large vision-language models, this paper proposes a dynamic contrastive decoding framework called Octopus, which can identify the type of hallucination at each generation step and select different contrastive decoding strategies as needed.
ODA-GAN: Orthogonal Decoupling Alignment GAN Assisted by Weakly-supervised Learning for Virtual Immunohistochemistry Staining
Tong Wang (Dalian University of Technology), Hongming Xu (Dalian University of Technology)
Image TranslationGenerationData SynthesisGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: A model named ODA-GAN is proposed for unpaired virtual IHC staining, which converts H&E images to IHC images.
Odd-One-Out: Anomaly Detection by Comparing with Neighbors
Ankan Bhunia (University of Edinburgh), Hakan Bilen (University of Edinburgh)
Anomaly DetectionKnowledge DistillationImageBenchmark
🎯 What it does: A scene-specific 'Odd-One-Out' anomaly detection task is proposed, which constructs a 3D voxel representation through multi-view images, compares across instances to identify occasionally appearing anomalous objects, and releases two new benchmark datasets.
ODE: Open-Set Evaluation of Hallucinations in Multimodal Large Language Models
Yahan Tu (Beijing Jiaotong University), Jitao Sang (Beijing Jiaotong University)
GenerationData SynthesisGraph Neural NetworkTransformerLarge Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: Proposed and implemented the ODE (Open-Set Dynamic Evaluation) protocol for dynamically generating open-set samples to evaluate the hallucinations of multimodal large language models (in terms of existence and attributes).
ODHSR: Online Dense 3D Reconstruction of Humans and Scenes from Monocular Videos
Zetong Zhang (ETH Zurich), Martin R. Oswald (University of Amsterdam)
Pose EstimationDepth EstimationOptimizationGaussian SplattingSimultaneous Localization and MappingOptical FlowVideo
🎯 What it does: An online end-to-end framework ODHSR has been developed, capable of synchronously performing camera trajectory estimation, human pose estimation, and dense photorealistic 3D reconstruction of scenes and humans in monocular videos.
OFER: Occluded Face Expression Reconstruction
Pratheba Selvaraju (University of Massachusetts), Ilya Zharkov (Max Planck Institute for Intelligent Systems)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: The paper proposes a multi-hypothesis 3D face reconstruction method based on conditional diffusion models, capable of generating diverse and expressive face models under occlusion conditions.
OffsetOPT: Explicit Surface Reconstruction without Normals
Huan Lei (AIML, University of Adelaide)
OptimizationTransformerPoint CloudMesh
🎯 What it does: This paper proposes an explicit surface reconstruction method called OffsetOPT, which can directly reconstruct meshes from 3D point clouds without the need for point normals.
Olympus: A Universal Task Router for Computer Vision Tasks
Yuanze Lin (University of Oxford), Philip Torr (Microsoft)
RecognitionObject DetectionSegmentationGenerationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageVideoTextMultimodality
🎯 What it does: Developed the Olympus framework, using a multimodal large language model (MLLM) as a task router, combined with specialized image/video/3D task models to complete 20 types of visual tasks.
Omni-ID: Holistic Identity Representation Designed for Generative Tasks
Guocheng Qian (Snap Research), Kfir Aberman (Snap Research)
GenerationData SynthesisTransformerImage
🎯 What it does: A global identity representation called Omni-ID is proposed for generative tasks, capable of encoding a fixed-size structured identity code from multiple facial images with different poses and expressions.
Omni-RGPT: Unifying Image and Video Region-level Understanding via Token Marks
Miran Heo (NVIDIA), Ryo Hachiuma (NVIDIA)
RecognitionObject DetectionTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageVideoTextMultimodality
🎯 What it does: Designed and implemented Omni-RGPT, a unified multimodal large language model capable of region-based understanding and interaction in images and videos.
Omni-Scene: Omni-Gaussian Representation for Ego-Centric Sparse-View Scene Reconstruction
Dongxu Wei (Westlake University), Peidong Liu (Westlake University)
Autonomous DrivingTransformerGaussian SplattingPoint Cloud
🎯 What it does: This paper proposes Omni-Scene, which utilizes a combination of pixel-level and voxel-level Omni-Gaussian representations to achieve high-quality 3D reconstruction and view synthesis in ego-centric scenes with limited view overlap for autonomous driving.
Omnia de EgoTempo: Benchmarking Temporal Understanding of Multi-Modal LLMs in Egocentric Videos
Chiara Plizzari, Federico Tombari
TransformerLarge Language ModelVideoTextMultimodalityBenchmark
🎯 What it does: This paper proposes and evaluates a temporal reasoning dataset specifically for first-person video question answering, called EgoTempo, and conducts systematic experiments on the performance of multimodal large language models (MLLM) on this task.
Omnidirectional Multi-Object Tracking
Kai Luo (Hunan University), Kailun Yang (Hunan University)
Object TrackingImageVideo
🎯 What it does: A multi-object tracking framework called OmniTrack is proposed, specifically designed for panoramic images, addressing challenges such as geometric distortion, resolution loss, and uneven lighting in panoramic views.
OmniDocBench: Benchmarking Diverse PDF Document Parsing with Comprehensive Annotations
Linke Ouyang (Shanghai AI Laboratory), Conghui He
TransformerLarge Language ModelVision Language ModelTextBenchmark
🎯 What it does: Proposes the OmniDocBench document parsing benchmark, which includes high-quality multi-label annotations for 9 types of PDFs and provides an end-to-end and multi-dimensional evaluation framework.
OmniDrive: A Holistic Vision-Language Dataset for Autonomous Driving with Counterfactual Reasoning
Shihao Wang (NVIDIA), Jose M. Alvarez (NVIDIA)
Autonomous DrivingTransformerLarge Language ModelVision Language ModelMultimodality
🎯 What it does: Proposes the OmniDrive dataset and two LLM driving agent frameworks, Omni-L and Omni-Q, which achieve high-quality 3D visual-language question answering and planning data through adversarial reasoning.
OmniFlow: Any-to-Any Generation with Multi-Modal Rectified Flows
Shufan Li (University of California Los Angeles), Aditya Grover (University of California Los Angeles)
GenerationData SynthesisTransformerFlow-based ModelRectified FlowImageTextMultimodalityAudio
🎯 What it does: OmniFlow is constructed as a unified multimodal generative model capable of performing generative tasks from any input to any output, such as text to image, text to audio, audio to image, etc.
OmniGen: Unified Image Generation
Shitao Xiao (Beijing Academy of Artificial Intelligence), Zheng Liu (Beijing Academy of Artificial Intelligence)
GenerationTransformerFlow-based ModelRectified FlowAuto EncoderImageTextMultimodality
🎯 What it does: We propose OmniGen, a unified image generation model based on VAE and Transformer, capable of performing various tasks such as text generation, image editing, and visual condition-based generation within a single framework.
OmniGuard: Hybrid Manipulation Localization via Augmented Versatile Deep Image Watermarking
Xuanyu Zhang (Peking University), Jian Zhang (Peking University)
Image TranslationImage HarmonizationRestorationData SynthesisSafty and PrivacyTransformerAuto EncoderImage
🎯 What it does: This paper proposes a hybrid visual watermarking framework named OmniGuard for image tampering localization and copyright protection.
OmniManip: Towards General Robotic Manipulation via Object-Centric Interaction Primitives as Spatial Constraints
Mingjie Pan (Peking University), Hao Dong (Peking University)
Pose EstimationOptimizationRobotic IntelligenceTransformerLarge Language ModelVision Language ModelPoint CloudMesh
🎯 What it does: This paper presents OmniManip, an open-source vocabulary manipulation method that utilizes object-centered interaction primitives, combining high-level reasoning from VLM and 3D spatial constraints to achieve a dual closed-loop robotic operating system without the need for VLM fine-tuning.