CVPR 2025 Papers — Page 27
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2871 papers
Towards Precise Embodied Dialogue Localization via Causality Guided Diffusion
Haoyu Wang (Xi'an Jiaotong University), Wei Tang (University of Illinois at Chicago)
Diffusion modelImage
🎯 What it does: Proposes an embedded dialogue localization method based on a causal-guided diffusion model (CGD), which directly regresses coordinates and reduces reliance on high resolution;
Towards Precise Scaling Laws for Video Diffusion Transformers
Yuanyang Yin (University of Science and Technology of China), Kun Gai (Kuaishou Technology)
GenerationComputational EfficiencyHyperparameter SearchTransformerDiffusion modelVideo
🎯 What it does: This study systematically analyzes the scaling laws of video diffusion transformers and confirms their existence. A new scaling law is proposed to predict the optimal hyperparameters for any model size and computational budget.
Towards RAW Object Detection in Diverse Conditions
Zhong-Yu Li (Nankai University), Ming-Ming Cheng (Nankai University)
Object DetectionDomain AdaptationKnowledge DistillationImage
🎯 What it does: AODRaw RAW image dataset was constructed, and a method for direct pre-training in the RAW domain and using cross-domain distillation to enhance object detection performance was proposed;
Towards Realistic Example-based Modeling via 3D Gaussian Stitching
Xinyu Gao (Zhejiang University), Xiaogang Jin (Chinese University of Hong Kong)
GenerationOptimizationGaussian SplattingPoint Cloud
🎯 What it does: This paper proposes a sample-driven model reconstruction and seamless stitching method based on 3D Gaussian Splatting;
Towards Satellite Image Road Graph Extraction: A Global-Scale Dataset and A Novel Method
Pan Yin (Xi'an Jiaotong University), Deyu Meng (Xi'an Jiaotong University)
Object DetectionSegmentationImage
🎯 What it does: A global-scale satellite image road map extraction dataset and an improved model based on SAM, called SAM-Road++, are proposed.
Towards Scalable Human-aligned Benchmark for Text-guided Image Editing
Suho Ryu (Seoul National University), Joonseok Lee (Seoul National University)
Image TranslationObject DetectionGenerationLarge Language ModelDiffusion modelImageTextBenchmark
🎯 What it does: A large-scale, automated evaluation benchmark for text-driven image editing, HATIE, is proposed.
Towards Smart Point-and-Shoot Photography
Jiawan Li (Shenzhen University), Guoping Qiu (University of Nottingham)
Recommendation SystemConvolutional Neural NetworkPrompt EngineeringMixture of ExpertsImage
🎯 What it does: Developed the 'Smart Point Shooting' system, which uses real-time camera posture to guide users in composition and generate improvement suggestions.
Towards Source-Free Machine Unlearning
Sk Miraj Ahmed (Brookhaven National Laboratory), Amit K. Roy-Chowdhury (University of California)
ClassificationOptimizationImage
🎯 What it does: This paper proposes an efficient method for memory deletion of linear and mixed linear classifiers without accessing the original training data;
Towards Stable and Storage-efficient Dataset Distillation: Matching Convexified Trajectory
Wenliang Zhong (Shandong University), Weili Guan (Fuzhou University)
OptimizationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: A Matching Convex Trajectory (MCT) method is proposed to improve the dataset distillation process.
Towards Training-free Anomaly Detection with Vision and Language Foundation Models
Jinjin Zhang (Beihang University), Di Huang (Beihang University)
Anomaly DetectionTransformerLarge Language ModelVision Language ModelImageMultimodality
🎯 What it does: A training-free multimodal framework called LogSAD is proposed for simultaneously detecting structural and logical anomalies, combining GPT-4V to generate interests and combination rules, and achieving anomaly detection at different granularities (patch, interest set, combination matching).
Towards Transformer-Based Aligned Generation with Self-Coherence Guidance
Shulei Wang (Zhejiang University), Zhou Zhao (Huawei Noah's Ark Lab)
GenerationData SynthesisTransformerLarge Language ModelDiffusion modelImageText
🎯 What it does: A training-free Self-Coherence Guidance (SCG) method is proposed, which directly optimizes the cross-attention mapping of the Transformer-based Text-Guided Diffusion Model (TGDM) during the generation process to enhance the alignment capability between text and images.
Towards Unbiased and Robust Spatio-Temporal Scene Graph Generation and Anticipation
Rohith Peddi (University of Texas at Dallas), Vibhav Gogate (University of Texas at Dallas)
Object DetectionGenerationTransformerSupervised Fine-TuningVideo
🎯 What it does: The IMPARTAIL framework is proposed to eliminate long-tail bias in video scene graph generation and prediction through loss masking and curriculum learning, and introduces a robustness evaluation task.
Towards Understanding and Quantifying Uncertainty for Text-to-Image Generation
Gianni Franchi (ENSTA Paris), Andrea Pilzer (Imperial College London)
GenerationLarge Language ModelVision Language ModelDiffusion modelImageText
🎯 What it does: This paper proposes a Prompt-based UNCertainty (PUNC) method based on large visual language models to quantify the uncertainty of text-to-image (T2I) generation models when given prompts, and constructs a prompt dataset aimed at uncertainty assessment.
Towards Understanding How Knowledge Evolves in Large Vision-Language Models
Sudong Wang (Institute of Information Engineering, Chinese Academy of Sciences), Xiangyang Ji (Tsinghua University)
Knowledge DistillationRepresentation LearningTransformerVision Language ModelMultimodality
🎯 What it does: This study investigates how the internal knowledge of large-scale vision-language models (LVLM) evolves across layers, revealing critical layers and mutation layers, as well as three evolutionary stages.
Towards Universal AI-Generated Image Detection by Variational Information Bottleneck Network
Haifeng Zhang (Chongqing University of Posts and Telecommunications), Bin Xiao (Chongqing University of Posts and Telecommunications)
ClassificationRecognitionAnomaly DetectionConvolutional Neural NetworkDiffusion modelGenerative Adversarial NetworkContrastive LearningImageMultimodality
🎯 What it does: This paper proposes VIB-Net, a method that utilizes variational information bottleneck to filter CLIP multimodal features for general AI-generated image detection.
Towards Universal Dataset Distillation via Task-Driven Diffusion
Ding Qi (Tongji University), Cairong Zhao (Tongji University)
ClassificationObject DetectionSegmentationKnowledge DistillationDiffusion modelImage
🎯 What it does: This paper proposes UniDD, a unified dataset distillation framework capable of performing data distillation across three visual tasks: image classification, object detection, and semantic segmentation.
Towards Universal Soccer Video Understanding
Jiayuan Rao (Shanghai Jiao Tong University), Weidi Xie (Shanghai Jiao Tong University)
ClassificationRecognitionGenerationTransformerLarge Language ModelVision Language ModelVideoMultimodality
🎯 What it does: A large-scale multimodal football video dataset, SoccerReplay-1988, and a unified visual encoder, MatchVision, have been proposed, establishing a unified framework for football video understanding tasks such as event classification, commentary generation, and foul recognition.
Towards Visual Discrimination and Reasoning of Real-World Physical Dynamics: Physics-Grounded Anomaly Detection
Wenqiao Li (ShanghaiTech University), Yingna Wu (ShanghaiTech University)
Anomaly DetectionVision Language ModelVideoPhysics Related
🎯 What it does: We propose Phys-AD, a physical-driven video dataset for industrial anomaly detection, which includes 22 object categories, over 6,400 video segments, and 47 types of anomalies.
Towards Zero-Shot Anomaly Detection and Reasoning with Multimodal Large Language Models
Jiacong Xu (Johns Hopkins University), Isht Dwivedi (Honda Research Institute USA)
Anomaly DetectionTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageMultimodalityBiomedical DataMagnetic Resonance ImagingComputed TomographyChain-of-Thought
🎯 What it does: This paper proposes a dedicated multimodal large language model, Anomaly-OV, capable of visual anomaly detection and reasoning under zero-shot conditions. It also constructs a dataset of 125k visual instruction fine-tuning data, Anomaly-Instruct-125k, and an evaluation benchmark, VisA-D&R.
Tra-MoE: Learning Trajectory Prediction Model from Multiple Domains for Adaptive Policy Conditioning
Jiange Yang (Nanjing University), Limin Wang (Nanjing University)
Robotic IntelligenceTransformerReinforcement LearningMixture of ExpertsVideoBenchmark
🎯 What it does: This paper proposes a trajectory prediction model called Tra-MoE, which is jointly trained on multi-domain video data, and based on this, achieves visual alignment and reinforcement of robot policies through an adaptive 2D trajectory mask.
Track Any Anomalous Object:A Granular Video Anomaly Detection Pipeline
Yuzhi Huang (Xiamen University), Yixuan Yuan (Chinese University of Hong Kong)
Object DetectionObject TrackingSegmentationAnomaly DetectionVideo
🎯 What it does: The TAO (Track Any Object) framework is proposed, which first uses object detection to obtain candidate boxes. After filtering through anomaly scores and robust filtering, these boxes are provided as prompts to SAM2 for pixel-level segmentation, achieving fine-grained detection and localization of anomalies in videos.
Track4Gen: Teaching Video Diffusion Models to Track Points Improves Video Generation
Hyeonho Jeong (Adobe), Duygu Ceylan (Adobe)
Object TrackingGenerationDiffusion modelOptical FlowVideo
🎯 What it does: Combining video diffusion models with point tracking tasks, trajectory supervision is added during the generation process to enhance video consistency.
Tracktention: Leveraging Point Tracking to Attend Videos Faster and Better
Zihang Lai (University of Oxford), Andrea Vedaldi (University of Oxford)
Image TranslationDepth EstimationConvolutional Neural NetworkTransformerOptical FlowVideo
🎯 What it does: This paper proposes the Tracktention layer, which explicitly aligns video features using point tracking information to enhance the temporal consistency of video predictions.
TraF-Align: Trajectory-aware Feature Alignment for Asynchronous Multi-agent Perception
Zhiying Song (Tsinghua University), Jun Li (Tsinghua University)
Object DetectionAutonomous DrivingConvolutional Neural NetworkTransformerPoint Cloud
🎯 What it does: Proposes the TraF-Align framework, which utilizes trajectory field prediction and trajectory-aware attention to align asynchronous multi-agent point clouds at the feature level, achieving delay compensation, spatial alignment, and semantic consistency, ultimately completing multi-agent collaborative perception.
Training Data Provenance Verification: Did Your Model Use Synthetic Data from My Generative Model for Training?
Yuechen Xie (Zhejiang University), Mingli Song (Zhejiang University)
ClassificationData SynthesisConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: Proposes the TrainProVe method to determine whether a suspicious model has been trained using synthetic data generated by the defender's text-to-image diffusion model in a black-box setting.
Training-free Dense-Aligned Diffusion Guidance for Modular Conditional Image Synthesis
Zixuan Wang (Sichuan University), Yinjie Lei (Sichuan University)
GenerationData SynthesisDiffusion modelOptical FlowImageText
🎯 What it does: A modular conditional image synthesis framework is proposed, achieving fine control over three basic conditions: text, layout, and drag-and-drop, through three dense alignment modules: Dense Concept Alignment, Dense Geometry Alignment, and Dense Motion Alignment.
Training-free Neural Architecture Search through Variance of Knowledge of Deep Network Weights
Ondrej Tybl (Czech Technical University in Prague), Lukas Neumann (Czech Technical University in Prague)
Neural Architecture SearchImage
🎯 What it does: This paper proposes a training-free network architecture search agent VKDNW, which uses the variance of the Fisher information matrix to evaluate the classification accuracy of the network.
Trajectory Mamba: Efficient Attention-Mamba Forecasting Model Based on Selective SSM
Yizhou Huang (Brunel University), Kezhi Wang (Brunel University)
Autonomous DrivingComputational EfficiencyRecurrent Neural NetworkReinforcement LearningMultimodalityTime Series
🎯 What it does: A lightweight multimodal trajectory prediction framework called Tamba based on a Selective State Space Model (SSM) has been designed and implemented.
Transfer Your Perspective: Controllable 3D Generation from Any Viewpoint in a Driving Scene
Tai-Yu Pan (Ohio State University), Wei-Lun Chao (Cornell University)
GenerationData SynthesisAutonomous DrivingDiffusion modelGenerative Adversarial NetworkPoint Cloud
🎯 What it does: A method called Transfer Your Perspective (TYP) is proposed, which utilizes single-vehicle perception data and semantic information to generate realistic point clouds from other perspectives, thereby achieving collaborative perception without the need for actual multi-vehicle data collection.
Transformers without Normalization
Jiachen Zhu (Meta), Zhuang Liu (Meta)
ClassificationOptimizationTransformerLarge Language ModelDiffusion modelImageAudio
🎯 What it does: A Transformer alternative without normalization layers is proposed and validated—Dynamic Tanh (DyT), which replaces traditional layer normalization or RMSNorm with a learnable scale α and the tanh mapping.
TransPixeler: Advancing Text-to-Video Generation with Transparency
Luozhou Wang (Hong Kong University of Science and Technology), Ying-Cong Chen (Hong Kong University of Science and Technology)
GenerationData SynthesisTransformerDiffusion modelOptical FlowVideoText
🎯 What it does: The TransPixeler method is proposed to generate RGBA videos containing both RGB and alpha channels from text, extending the capabilities of pre-trained RGB video generation models.
Traversing Distortion-Perception Tradeoff using a Single Score-Based Generative Model
Yuhan Wang (Chinese University of Hong Kong), Xiaojun Yuan (University of Electronic Science and Technology of China)
RestorationGenerationDiffusion modelScore-based ModelImage
🎯 What it does: Using a single pre-trained score network, flexible traversal of the distortion-perception trade-off is achieved in inverse problems such as image denoising by adjusting the variance scale of the reverse diffusion process.
TreeMeshGPT: Artistic Mesh Generation with Autoregressive Tree Sequencing
Stefan Lionar (Sea AI Lab), Gim Hee Lee (National University of Singapore)
GenerationData SynthesisTransformerPoint CloudMesh
🎯 What it does: This paper presents TreeMeshGPT, a Transformer generator that utilizes Autoregressive Tree Sequencing technology to generate high-quality, artistic triangular meshes based on point cloud conditions.
Tripartite Weight-Space Ensemble for Few-Shot Class-Incremental Learning
Juntae Lee (Qualcomm AI Research), Sungrack Yun (Qualcomm AI Research)
ClassificationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a Tri-weighted Ensemble (Tri-WE) method for Few-Shot Class Incremental Learning (FSCIL), combined with Amplified Data Knowledge Distillation (ADKD) to achieve adaptive updates and mitigate catastrophic forgetting during the incremental phase.
TriTex: Learning Texture from a Single Mesh via Triplane Semantic Features
Dana Cohen-Bar, Yoni Kasten
Image TranslationData SynthesisConvolutional Neural NetworkMesh
🎯 What it does: A feed-forward texture transfer model called TriTex is proposed, which uses a single textured mesh for training and can map the texture of this mesh onto any new shape of the same category.
TSAM: Temporal SAM Augmented with Multimodal Prompts for Referring Audio-Visual Segmentation
Abduljalil Radman (Aalto University), Jorma Laaksonen (Aalto University)
SegmentationTransformerPrompt EngineeringVideoTextMultimodalityAudio
🎯 What it does: This paper proposes TSAM, a model that extends SAM for the Referring Audio-Visual Segmentation (Ref-AVS) task, utilizing multi-modal (text, audio, visual) prompts to achieve dynamic video segmentation.
TSD-SR: One-Step Diffusion with Target Score Distillation for Real-World Image Super-Resolution
Linwei Dong (Zhejiang University), Changqing Zou (Zhejiang Lab)
RestorationSuper ResolutionDiffusion modelScore-based ModelImage
🎯 What it does: A single-step real-time image super-resolution framework TSD-SR based on a pre-trained text-image diffusion model is constructed, mapping directly from low-quality images to high-quality images;
TSP-Mamba: The Travelling Salesman Problem Meets Mamba for Image Super-resolution and Beyond
Kun Zhou (SmartMore Corporation), Jiangbo Lu (SmartMore Corporation)
RestorationSuper ResolutionConvolutional Neural NetworkRecurrent Neural NetworkTransformerImage
🎯 What it does: This paper proposes a TSP-based local scanning path Mamba model, TSP-Mamba, for lightweight single-image super-resolution.
Tuning the Frequencies: Robust Training for Sinusoidal Neural Networks
Tiago Novello (IMPA), Luiz Velho (IMPA)
RestorationImage
🎯 What it does: This paper proposes a robust training framework named TUNER, which achieves more efficient initialization and spectral control for multilayer perceptrons (INR) with sine activation functions.
Turbo3D: Ultra-fast Text-to-3D Generation
Hanzhe Hu (Carnegie Mellon University), Kai Zhang (Adobe Research)
GenerationData SynthesisComputational EfficiencyKnowledge DistillationTransformerDiffusion modelGaussian SplattingPoint CloudMesh
🎯 What it does: Turbo3D is a high-speed text-to-3D system that can generate high-quality 3D Gaussian splatting (3DGS) assets from text prompts in less than 1 second (on a single A100 GPU), using a two-stage pipeline: the first stage is a very fast four-step multi-view generator, and the second stage is a single-step multi-view reconstructor.
TurboFill: Adapting Few-step Text-to-image Model for Fast Image Inpainting
Liangbin Xie (University of Macau), Chao Dong (Shenzhen University)
RestorationGenerationDiffusion modelGenerative Adversarial NetworkImageBenchmark
🎯 What it does: A fast image inpainting model named TurboFill has been developed, achieving high-quality image inpainting with a small number of diffusion steps.
Twinner: Shining Light on Digital Twins in a Few Snaps
Jesus Zarzar (King Abdullah University of Science and Technology), David Novotny (Meta AI)
GenerationOptimizationTransformerDiffusion modelImageVideo
🎯 What it does: Twinner is proposed, a forward large-scale reconstruction model capable of predicting object geometry, PBR materials, and environmental lighting from a small number of pose images.
Two by Two: Learning Multi-Task Pairwise Objects Assembly for Generalizable Robot Manipulation
Yu Qi (Shanghai Qi Zhi Institute), Huazhe Xu (Shanghai AI Laboratory)
Pose EstimationRobotic IntelligenceGraph Neural NetworkPoint Cloud
🎯 What it does: This paper presents the first daily object pairing assembly dataset 2BY2 and designs a two-step SE(3) pose estimation network to achieve precise pairing assembly.
Two is Better than One: Efficient Ensemble Defense for Robust and Compact Models
Yoojin Jung (Inha University), Byung Cheol Song (Inha University)
CompressionOptimizationComputational EfficiencyAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: The Efficient Ensemble Defense (EED) method is proposed, which generates diverse sub-models by pruning a single base model using different importance scores, and then integrates them to enhance robustness against adversarial attacks while maintaining the compressed model size.
Type-R: Automatically Retouching Typos for Text-to-Image Generation
Wataru Shimoda (CyberAgent), Kota Yamaguchi (CyberAgent)
RestorationGenerationLarge Language ModelDiffusion modelImageTextBenchmark
🎯 What it does: A post-processing pipeline named Type-R is proposed, which automatically detects and corrects typographical errors in text-generated images.
U-Know-DiffPAN: An Uncertainty-aware Knowledge Distillation Diffusion Framework with Details Enhancement for PAN-Sharpening
Sungpyo Kim (Korea Advanced Institute of Science and Technology), Munchurl Kim (Korea Advanced Institute of Science and Technology)
RestorationKnowledge DistillationDiffusion modelImage
🎯 What it does: Developed a diffusion model U-Know-DiffPAN based on uncertainty knowledge distillation for PAN sharpening of high-resolution multispectral images.
UA-Pose: Uncertainty-Aware 6D Object Pose Estimation and Online Object Completion with Partial References
Ming-Feng Li (Carnegie Mellon University), Cheng-Hao Kuo (Amazon)
Pose EstimationImagePoint CloudMesh
🎯 What it does: This paper proposes UA-Pose, a framework for 6D pose estimation using partial reference information (limited RGBD images or a single RGB image) and completing the 3D model of the object online during the testing phase.
UCM-VeID V2: A Richer Dataset and A Pre-training Method for UAV Cross-Modality Vehicle Re-Identification
Xingyue Liu (National University of Defense Technology), Ping Zhong (National University of Defense Technology)
RecognitionRetrievalDomain AdaptationConvolutional Neural NetworkGenerative Adversarial NetworkContrastive LearningImageMultimodality
🎯 What it does: Proposed the UCM-VeID V2 dataset and designed a self-supervised pre-training method to address the modality bias problem in cross-modal vehicle ReID.
UCOD-DPL: Unsupervised Camouflaged Object Detection via Dynamic Pseudo-label Learning
Weiqi Yan (Xiamen University), Liujuan Cao (Xiamen University)
Object DetectionContrastive LearningImage
🎯 What it does: This paper proposes an unsupervised pseudo-label learning method based on a teacher-student framework for latent object detection.
UHD-processer: Unified UHD Image Restoration with Progressive Frequency Learning and Degradation-aware Prompts
Yidi Liu (University of Science and Technology of China), Zheng-Jun Zha (University of Science and Technology of China)
RestorationSuper ResolutionPrompt EngineeringAuto EncoderImage
🎯 What it does: A unified model named UHD-Processor is proposed, capable of performing various denoising tasks such as denoising, dehazing, deburring, low-light enhancement, and rain/snow removal for ultra-high-resolution images all at once.
UIBDiffusion: Universal Imperceptible Backdoor Attack for Diffusion Models
Yuning Han (Columbia University), Yingjie Lao (Clemson University)
GenerationAdversarial AttackDiffusion modelImage
🎯 What it does: A universal and undetectable backdoor attack method called UIBDiffusion is proposed, which utilizes Universal Adversarial Perturbations (UAP) to generate covert triggers to hijack diffusion models.
UltraFusion: Ultra High Dynamic Imaging using Exposure Fusion
Zixuan Chen (Shanghai AI Laboratory), Tianfan Xue (The Chinese University of Hong Kong)
RestorationGenerationDiffusion modelOptical FlowImageVideoBenchmark
🎯 What it does: This paper proposes an UltraFusion method that models exposure fusion as guided filling, generating ghost-free, high-quality HDR images using low-light/high-light images under approximately 9 stops of exposure difference.
UMFN: Unified Multi-Domain Face Normalization for Joint Cross-domain Prototype Learning and Heterogeneous Face Recognition
Meng Pang (Nanchang University), Hong Rao (Nanchang University)
RecognitionDomain AdaptationConvolutional Neural NetworkAuto EncoderGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: A Unified Multi-Domain Face Normalization Network (UMFN) is proposed, which can simultaneously eliminate various interferences such as pose, expression, occlusion, and lighting under different domains, generating a standard face prototype with a frontal neutral expression while retaining identity information.
UMotion: Uncertainty-driven Human Motion Estimation from Inertial and Ultra-wideband Units
Huakun Liu (Nara Institute of Science and Technology), Kiyoshi Kiyokawa (Nara Institute of Science and Technology)
Pose EstimationRecurrent Neural NetworkSimultaneous Localization and MappingTime Series
🎯 What it does: This paper proposes an uncertainty-driven online state estimation framework that combines IMU and UWB sensors for real-time estimation of 3D human shape and posture.
Unbiased Video Scene Graph Generation via Visual and Semantic Dual Debiasing
Yanjun Li (University of Science and Technology of China), Lizhi Xu (University of Science and Technology of China)
Object DetectionGenerationTransformerVideo
🎯 What it does: This paper addresses the problem of unbiased video scene graph generation (VidSGG) and proposes a dual debiasing framework named VISA, which can simultaneously eliminate visual bias and semantic bias.
Unbiasing through Textual Descriptions: Mitigating Representation Bias in Video Benchmarks
Nina Shvetsova (Goethe University Frankfurt), Christian Rupprecht (University of Oxford)
ClassificationRecognitionRetrievalRepresentation LearningTransformerLarge Language ModelVision Language ModelVideoTextBenchmark
🎯 What it does: Utilize a visual-language model to generate frame-level textual descriptions and extract concepts such as objects and actions through a large language model, systematically detecting and eliminating representation biases (e.g., object bias, single-frame bias) in video datasets.
Unboxed: Geometrically and Temporally Consistent Video Outpainting
Zhongrui Yu (ETH Zurich), Abdelaziz Djelouah (Disney Research Studios)
RestorationSegmentationGenerationDiffusion modelGaussian SplattingOptical FlowVideo
🎯 What it does: A multi-stage video outpainting method that combines 3D Gaussian Splatting and a pre-trained video diffusion model is proposed, achieving high-quality and temporally consistent frame extensions.
Uncertain Multimodal Intention and Emotion Understanding in the Wild
Qu Yang (Wuhan University), Mang Ye (Wuhan University)
ClassificationRecognitionTransformerPrompt EngineeringContrastive LearningImageVideoTextMultimodalityAudio
🎯 What it does: A large-scale multimodal dataset MINE was constructed, and the BEAR framework was proposed to achieve joint understanding of sentiment and intent.
Uncertainty Meets Diversity: A Comprehensive Active Learning Framework for Indoor 3D Object Detection
Jiangyi Wang (Singapore University of Technology and Design), Na Zhao (Singapore University of Technology and Design)
Object DetectionPoint Cloud
🎯 What it does: A proactive learning framework for indoor 3D object detection is proposed, integrating uncertainty and diversity as two major sampling criteria.
Uncertainty Weighted Gradients for Model Calibration
Jinxu Lin (University of Sydney), Chang Xu (University of Sydney)
ClassificationOptimizationConvolutional Neural NetworkImage
🎯 What it does: A gradient-weighted training method based on sample uncertainty (BSCE-GRA) is proposed to achieve better model calibration.
Uncertainty-guided Perturbation for Image Super-Resolution Diffusion Model
Leheng Zhang (University of Electronic Science and Technology of China), Shuhang Gu (University of Electronic Science and Technology of China)
RestorationSuper ResolutionDiffusion modelImage
🎯 What it does: A noise-weighted and improved diffusion model called UPSR, based on uncertainty guidance, is proposed for single image super-resolution tasks.
Uncertainty-Instructed Structure Injection for Generalizable HD Map Construction
Xiaolu Liu (Zhejiang University), Jianke Zhu (Zhejiang University)
Object DetectionAutonomous DrivingKnowledge DistillationTransformerPoint Cloud
🎯 What it does: Proposes the UIGenMap model, which utilizes uncertainty-guided structural injection to achieve more generalized high-precision HD map vectorization construction.
UnCommon Objects in 3D
Xingchen Liu (Meta AI), David Novotny
GenerationData SynthesisPose EstimationDiffusion modelGaussian SplattingSimultaneous Localization and MappingVideoPoint Cloud
🎯 What it does: This paper proposes and constructs a large-scale real object 3D dataset called uCO3D, and utilizes this dataset to train various 3D learning models, enhancing the performance of multi-view reconstruction, image synthesis, and text-to-3D generation.
Understanding Fine-tuning CLIP for Open-vocabulary Semantic Segmentation in Hyperbolic Space
Zelin Peng (Shanghai Jiao Tong University), Wei Shen (Shanghai Jiao Tong University)
SegmentationContrastive LearningImage
🎯 What it does: This paper proposes HyperCLIP, a fine-tuning strategy for scaling the CLIP text encoder in hyperbolic space to achieve open-domain semantic segmentation.
Understanding Multi-layered Transmission Matrices
Anat Levin (Technion), Marina Alterman (Technion)
ImagePhysics Related
🎯 What it does: This paper studies the application of multilayer optical devices (SLM) in wavefront shaping through transfer matrix theory and multilayer planar models, quantifying the relationship between the required number of layers and the focusable field of view.
Understanding Multi-Task Activities from Single-Task Videos
Yuhan Shen (Northeastern University), Ehsan Elhamifar (Northeastern University)
SegmentationDomain AdaptationConvolutional Neural NetworkLarge Language ModelPrompt EngineeringVideo
🎯 What it does: This paper proposes a Multi-Task Temporal Action Segmentation (MT-TAS) framework aimed at addressing the challenges posed by action interleaving, task switching, and background interference in single-person multi-task environments.
UNEM: UNrolled Generalized EM for Transductive Few-Shot Learning
Long Zhou (Politecnico di Milano), Ismail Ben Ayed (Universitè Paris-Saclay)
ClassificationOptimizationVision Language ModelImage
🎯 What it does: This paper proposes a framework called UNEM that gradually unfolds the Expectation-Maximization (EM) algorithm into a neural network for transductive few-shot classification, enhancing performance by learning to adjust class balance and temperature hyperparameters.
Uni-Renderer: Unifying Rendering and Inverse Rendering Via Dual Stream Diffusion
Zhifei Chen (Hong Kong University of Science and Technology), Ying-Cong Chen
GenerationData SynthesisDiffusion modelImage
🎯 What it does: Proposes Uni-Renderer, a framework that utilizes a dual-stream diffusion model to unify rendering and inverse rendering;
Uni4D: Unifying Visual Foundation Models for 4D Modeling from a Single Video
David Yifan Yao (University of Illinois at Urbana-Champaign), Shenlong Wang (University of Illinois at Urbana-Champaign)
Pose EstimationDepth EstimationOptimizationLarge Language ModelSimultaneous Localization and MappingVideo
🎯 What it does: This paper proposes the Uni4D framework, which utilizes pre-trained visual foundation models and a three-stage energy optimization to reconstruct camera trajectories, static and dynamic 3D geometry, and dense 3D dynamic motion from a single arbitrary video without training or fine-tuning.
UNIALIGN: Scaling Multimodal Alignment within One Unified Model
Bo Zhou (Nanjing University of Science and Technology), Wenguan Wang (Zhejiang University)
ClassificationComputational EfficiencyKnowledge DistillationTransformerMixture of ExpertsContrastive LearningImageVideoMultimodalityPoint CloudAudio
🎯 What it does: A unified multimodal alignment model UNIALIGN is proposed, capable of aligning various modalities (images, text, audio, video, 3D point clouds, depth maps) within a single encoder and completing all tasks in one training phase.
UniAP: Unifying Inter- and Intra-Layer Automatic Parallelism by Mixed Integer Quadratic Programming
Hao Lin (Nanjing University), Wu-Jun Li (Nanjing University)
OptimizationTransformerText
🎯 What it does: Proposes the UniAP method, which unifies and simultaneously optimizes cross-layer (pipeline) and intra-layer (tensor, data, FSDP) parallel strategies through mixed-integer quadratic programming (MIQP) to achieve optimal throughput for distributed training.
UNIC-Adapter: Unified Image-instruction Adapter with Multi-modal Transformer for Image Generation
Lunhao Duan (Wuhan University), Gui-Song Xia (Wuhan University)
GenerationTransformerDiffusion modelImageTextMultimodality
🎯 What it does: Designed and implemented a unified image instruction adapter (UNIC-Adapter) for controllable image generation under multimodal conditions within a single model.
UNICL-SAM: Uncertainty-Driven In-Context Segmentation with Part Prototype Discovery
Dianmo Sheng (University of Science and Technology of China), Nenghai Yu (University of Science and Technology of China)
SegmentationGraph Neural NetworkTransformerImage
🎯 What it does: A context segmentation framework driven by uncertainty using a graph model and partial prototype learning is proposed.
Unified Dense Prediction of Video Diffusion
Lehan Yang (University of Virginia), Ming-Hsuan Yang (University of California, Merced)
SegmentationGenerationDepth EstimationTransformerDiffusion modelAuto EncoderVideo
🎯 What it does: A unified network called UDP Diff is proposed, which can simultaneously generate videos based on text prompts along with their corresponding entity segmentation and depth maps, embedding these dense predictions into the video generation in the form of colormaps.
Unified Medical Lesion Segmentation via Self-referring Indicator
Shijie Chang (Dalian University of Technology), Tiancheng Wang (Dalian University of Technology)
SegmentationConvolutional Neural NetworkTransformerBiomedical DataMagnetic Resonance ImagingComputed TomographyUltrasound
🎯 What it does: A unified medical lesion segmentation framework SR-ICL is proposed, capable of achieving multi-task segmentation using a small number of reference images.
Unified Reconstruction of Static and Dynamic Scenes from Events
Qiyao Gao, Boxin Shi
RestorationData SynthesisConvolutional Neural NetworkRecurrent Neural NetworkSupervised Fine-TuningImageVideo
🎯 What it does: A unified event camera scene reconstruction framework URSEE is proposed, which is divided into two stages for static scenes: convolutional integration + denoising, and an end-to-end video reconstruction for dynamic scenes using voxel-grid + parallel channels.
Unified Uncertainty-Aware Diffusion for Multi-Agent Trajectory Modeling
Guillem Capellera (Institut de Robótica i Informática Industrial), Antonio Agudo (Institut de Robótica i Informática Industrial)
TransformerDiffusion modelTime Series
🎯 What it does: A unified uncertainty-aware diffusion model U2Diff has been designed and implemented, capable of predicting, interpolating, and recovering missing trajectories of multiple agents, while providing state-level uncertainty estimates at each step.
UniGoal: Towards Universal Zero-shot Goal-oriented Navigation
Hang Yin (Tsinghua University), Jiwen Lu (Tsinghua University)
Robotic IntelligenceGraph Neural NetworkTransformerLarge Language ModelVision Language ModelGraph
🎯 What it does: A unified graph structure framework called UniGoal is proposed for zero-shot general goal-oriented navigation, supporting three types of target: object categories, instance images, and text descriptions.
UniGraspTransformer: Simplified Policy Distillation for Scalable Dexterous Robotic Grasping
Wenbo Wang (Microsoft Research), Baining Guo (Microsoft Research)
Knowledge DistillationRobotic IntelligenceTransformerReinforcement LearningPoint Cloud
🎯 What it does: This paper proposes UniGraspTransformer, a general-purpose Transformer network that simplifies the training process of multi-object manipulation through offline policy distillation.
UniHOPE: A Unified Approach for Hand-Only and Hand-Object Pose Estimation
Yinqiao Wang (Chinese University of Hong Kong), Chi-Wing Fu (Chinese University of Hong Kong)
Pose EstimationTransformerDiffusion modelImage
🎯 What it does: This paper proposes UniHOPE, a unified method for 3D hand and handheld object pose estimation that can simultaneously handle scenes with only hands and hand-object interactions.
UniK3D: Universal Camera Monocular 3D Estimation
Luigi Piccinelli (ETH Zurich), Luc Van Gool (ETH Zurich)
Depth EstimationAutonomous DrivingTransformerImage
🎯 What it does: We propose UniK3D, a general method that can directly infer metric 3D geometry from a single image under any camera model (including fisheye, panoramic, etc.) without the need for prior camera parameters.
UniMamba: Unified Spatial-Channel Representation Learning with Group-Efficient Mamba for LiDAR-based 3D Object Detection
Xin Jin (Chang'an University), Junchi Yan (Shanghai Jiao Tong University)
Object DetectionAutonomous DrivingRepresentation LearningConvolutional Neural NetworkPoint Cloud
🎯 What it does: A unified LiDAR 3D detection backbone network called UniMamba is proposed, which integrates 3D convolution with Mamba (state space model) to achieve efficient local and global context modeling.
UniNet: A Contrastive Learning-guided Unified Framework with Feature Selection for Anomaly Detection
Shun Wei (Nanjing University of Information Science and Technology), Xiaolong Xu (Nanjing University)
Anomaly DetectionContrastive LearningImageVideoBiomedical Data
🎯 What it does: This paper proposes a unified contrastive learning-driven multi-domain anomaly detection framework called UniNet, which combines a student-teacher model, a lightweight multi-scale embedding module, domain-relevant feature selection, and feature similarity comparison with margin loss, further achieving robust anomaly score calculation through a weighted decision mechanism.
UniPhy: Learning a Unified Constitutive Model for Inverse Physics Simulation
Himangi Mittal (Carnegie Mellon University), Shubham Tulsiani (Carnegie Mellon University)
OptimizationTime SeriesPhysics Related
🎯 What it does: A unified latent conditional neural constitutive model, UniPhy, has been trained to infer material properties from the motion trajectories of objects and to re-simulate the behavior of the material under different initial conditions.
UniPose: A Unified Multimodal Framework for Human Pose Comprehension, Generation and Editing
Yiheng Li (Chinese Academy of Sciences), Xilin Chen (Chinese Academy of Sciences)
GenerationPose EstimationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: The UniPose framework is proposed, achieving simultaneous understanding, generation, and editing of human poses within a unified multimodal model.
UniPre3D: Unified Pre-training of 3D Point Cloud Models with Cross-Modal Gaussian Splatting
Ziyi Wang (Tsinghua University), Jiwen Lu (Tsinghua University)
ClassificationSegmentationTransformerContrastive LearningGaussian SplattingPoint Cloud
🎯 What it does: A unified 3D point cloud pre-training framework called UniPre3D is proposed, which is applicable to both object-level and scene-level point clouds.
UniReal: Universal Image Generation and Editing via Learning Real-world Dynamics
Xi Chen (University of Hong Kong), Hengshuang Zhao (University of Hong Kong)
GenerationData SynthesisTransformerDiffusion modelImageVideo
🎯 What it does: Proposes UniReal, a unified framework that treats various image generation and editing tasks as discrete video frames, achieving multi-image input/output through diffusion transformers;
UniRestore: Unified Perceptual and Task-Oriented Image Restoration Model Using Diffusion Prior
I-Hsiang Chen (National Taiwan University), Ming-Hsuan Yang (University of California Merced)
RestorationSegmentationPrompt EngineeringDiffusion modelImage
🎯 What it does: We propose UniRestore, a unified image restoration model that simultaneously meets perceptual quality and downstream task requirements.
UniScene: Unified Occupancy-centric Driving Scene Generation
Bohan Li (Shanghai Jiao Tong University), Xin Jin (Tsinghua University)
GenerationData SynthesisAutonomous DrivingTransformerDiffusion modelGaussian SplattingVideoMultimodalityPoint Cloud
🎯 What it does: This paper proposes UniScene, a unified framework that first generates semantic placeholder information and then generates multi-view videos and LiDAR point clouds based on this, meeting the requirements for controllable and fully annotated data synthesis.
UniSTD: Towards Unified Spatio-Temporal Learning across Diverse Disciplines
Chen Tang (MMLab, Chinese University of Hong Kong), Xiangyu Yue (Chinese University of Hong Kong)
Autonomous DrivingRobotic IntelligenceTransformerMixture of ExpertsTime SeriesSequential
🎯 What it does: A unified Transformer-based spatiotemporal learning framework called UniSTD is proposed, which supports the parallel training of up to 10 cross-domain tasks within the same model.
Unity in Diversity: Video Editing via Gradient-Latent Purification
Junyu Gao (University of Science and Technology Beijing), Yufan Hu (Institute of Automation Chinese Academy of Sciences)
GenerationOptimizationDiffusion modelVideoText
🎯 What it does: A unified Gradient-Latent Variable Purification (GLP) framework is proposed for text-driven video editing, which collects gradient and latent variable information during the optimization process and aggregates consistent directions in a local coordinate system to reduce gradient noise and determine the optimal stopping point.
UniVAD: A Training-free Unified Model for Few-shot Visual Anomaly Detection
Zhaopeng Gu (Chinese Academy of Sciences), Jinqiao Wang (Chinese Academy of Sciences)
Anomaly DetectionGraph Neural NetworkImageBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: This paper proposes UniVAD, a training-agnostic, unified few-shot visual anomaly detection method that can detect anomalies in industrial, logical, and medical fields without the need to train models for each domain, requiring only a small number of normal samples during testing.
Universal Actions for Enhanced Embodied Foundation Models
Jinliang Zheng (Tsinghua University), Xianyuan Zhan (Tsinghua University)
Robotic IntelligenceVision-Language-Action ModelMultimodality
🎯 What it does: A robot foundation model called UniAct is proposed, which is based on a Universal Action Space and can learn shared atomic behaviors from multi-source heterogeneous action data, quickly adapting to new robots through a lightweight decoder.
Universal Domain Adaptation for Semantic Segmentation
Seun-An Choe (Kyung Hee University), Gyeong-Moon Park (Korea University)
SegmentationDomain AdaptationImage
🎯 What it does: This paper proposes a universal domain adaptation framework for unsupervised semantic segmentation, called UniMAP, which can achieve adaptation between the source domain and the target domain in an unknown category setting.
Universal Scene Graph Generation
Shengqiong Wu (National University of Singapore), Tat-seng Chua (National University of Singapore)
Object DetectionGenerationTransformerContrastive LearningImageVideoTextMultimodalityPoint Cloud
🎯 What it does: This paper proposes the concept of Unified Scene Graph (USG) and designs an end-to-end USG-Par model, achieving unified scene graph parsing for any multimodal input (images, text, video, 3D).
Unlearning through Knowledge Overwriting: Reversible Federated Unlearning via Selective Sparse Adapter
Zhengyi Zhong (National University of Defense Technology), Wei Yang Bryan Lim (Nanyang Technological University)
Federated LearningSafty and PrivacyComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: A reversible federated no-learning framework FUSED is proposed, which constructs sparse adapters at key layers to cover forgotten knowledge without altering the original model;
Unleashing In-context Learning of Autoregressive Models for Few-shot Image Manipulation
Bolin Lai (Georgia Institute of Technology), Tong Xiao (Meta)
Image TranslationGenerationTransformerLarge Language ModelPrompt EngineeringImageTextMultimodality
🎯 What it does: A multi-modal autoregressive model called InstaManip is proposed, which learns quickly and applies new image editing operations by leveraging the context of text and visual examples.
Unleashing the Potential of Consistency Learning for Detecting and Grounding Multi-Modal Media Manipulation
Yiheng Li (Chinese Academy of Sciences), Zhen Lei (Chinese Academy of Sciences)
ClassificationRecognitionAnomaly DetectionTransformerContrastive LearningImageTextMultimodality
🎯 What it does: A multi-modal media manipulation detection and localization framework based on consistency learning, named CSCL, is proposed, specifically targeting fine-grained forgery detection and localization tasks for images and text.
Unleashing the Potential of Multi-modal Foundation Models and Video Diffusion for 4D Dynamic Physical Scene Simulation
Zhuoman Liu (Hong Kong Polytechnic University), Di Zhang (Kuaishou Technology)
GenerationOptimizationTransformerLarge Language ModelDiffusion modelGaussian SplattingOptical FlowVideoMultimodalityPhysics Related
🎯 What it does: Utilize multimodal foundational models to query material types and initial parameters, and achieve automated physical simulation of 4D dynamic scenes through video diffusion and optical flow-guided optimization of the differentiable physics point method (MPM).
Unlocking Generalization Power in LiDAR Point Cloud Registration
Zhenxuan Zeng (Northwestern Polytechnical University), Peng Wang (Northwestern Polytechnical University)
Object DetectionAutonomous DrivingTransformerPoint Cloud
🎯 What it does: This paper proposes the UGP framework, which removes cross-attention and incorporates progressive self-attention and BEV features to enhance the generalization ability of LiDAR point cloud registration in cross-distance and cross-dataset scenarios.