CVPR 2026 Papers — Page 38
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 4071 papers
UAV-CB: A Complex-Background RGB-T Dataset and Local Frequency Bridge Network for UAV Detection
Shenghui Huang (Pengcheng Laboratory), Ke Chen (Pengcheng Laboratory)
Object DetectionDomain AdaptationConvolutional Neural NetworkImageMultimodalityBenchmark
🎯 What it does: Proposed a new RGB-T UAV detection dataset named UAV-CB and designed the Local Frequency Bridge Network (LFBNet) model.
UAVLight: A Benchmark for Illumination-Robust 3D Reconstruction in Unmanned Aerial Vehicle (UAV) Scenes
Kang Du (Hong Kong University of Science and Technology), Zeyu Wang (Hong Kong University of Science and Technology)
Neural Radiance FieldGaussian SplattingImagePoint CloudBenchmark
🎯 What it does: Proposed and released the UAVLight benchmark, specifically designed for evaluating 3D reconstruction and relighting by drones under different natural lighting conditions.
UCAN: Unified Convolutional Attention Network for Expansive Receptive Fields in Lightweight Super-Resolution
Cao Thien Tan (Ho Chi Minh City Open University), Nguyen Duc Dung (AI Tech Lab, Ho Chi Minh City University Of Technology)
Super ResolutionKnowledge DistillationConvolutional Neural NetworkTransformerImage
🎯 What it does: Design and implement a lightweight unified convolutional attention network (UCAN), which expands the receptive field while maintaining efficiency by fusing convolution and attention mechanisms for single-image super-resolution.
UCMNet: Uncertainty-Aware Context Memory Network for Under-Display Camera Image Restoration
Daehyun Kim (Hanyang University), Tae Hyun Kim (Hanyang University)
RestorationConvolutional Neural NetworkTransformerImage
🎯 What it does: Proposes UCMNet, an uncertainty-guided context memory network for restoring high-frequency details in under-display camera (UDC) images.
UDAPose: Unsupervised Domain Adaptation for Low-Light Human Pose Estimation
Haopeng Chen (University of Mississippi), Bo Wang (University of Mississippi)
Pose EstimationDomain AdaptationTransformerDiffusion modelImage
🎯 What it does: Achieving unsupervised domain adaptation for human pose estimation under low-light conditions
UETrack: A Unified and Efficient Framework for Single Object Tracking
Ben Kang (Dalian University of Technology), Huchuan Lu (City University of Hong Kong)
Object TrackingKnowledge DistillationTransformerMixture of ExpertsVision Language ModelMultimodalityBenchmark
🎯 What it does: Propose a unified and efficient single-object tracking framework called UETrack, which supports multi-modal inputs such as RGB, Depth, Thermal, Event, and Language;
UFO: Unifying Feed-Forward and Optimization-based Methods for Large Driving Scene Modeling
Kaiyuan Tan (Xiaomi EV), Hangjun Ye (UIUC)
Autonomous DrivingTransformerGaussian SplattingPoint Cloud
🎯 What it does: Proposes UFO, a recursive framework that maintains and updates a persistent set of scene tokens to achieve 4D driving scene modeling without explicit rendering.
UFVideo: Towards Unified Fine-Grained Video Cooperative Understanding with Large Language Models
Hewen Pan (Huazhong University of Science and Technology), Shengshan Hu (Huazhong University of Science and Technology)
RecognitionSegmentationTransformerLarge Language ModelVision Language ModelVideoTextBenchmark
🎯 What it does: Propose UFVideo, a unified multi-granularity video large language model that can simultaneously perform global video understanding, pixel-level video localization/segmentation, and temporal event localization;
UI-Lens: Assessing General MLLMs' Potential to Automate UI Display Quality Assurance
Wei Xiang (Zhejiang University), Shi Chen (Zhejiang University)
Anomaly DetectionVision Language ModelMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Constructed the UI-Lens benchmark to systematically evaluate the capability of multimodal large language models (MLLMs) in mobile UI defect detection.
UIKA: Fast Universal Head Avatar from Pose-Free Images
Zijian Wu (Nanjing University), Hao Zhu (Nanjing University)
GenerationData SynthesisTransformerGaussian SplattingImage
🎯 What it does: Propose an end-to-end fast and general-purpose head animation avatar generation framework called UIKA, which can accept any number of input images without pose information. It utilizes pixel-level UV correspondence to project multi-view information into a unified UV space, then fuses learnable UV tokens with Transformer, ultimately generating a real-time renderable 3D Gaussian head model that supports linearly blended skinning animation.
ULF-Loc: Unbiased Landmark Feature for Robust Visual Localization with 3D Gaussian Splatting
Yingdong Gu (Wuhan University), Jiayuan Li (Wuhan University)
Pose EstimationComputational EfficiencyGaussian SplattingImage
🎯 What it does: Propose the ULF-Loc framework, achieving high-precision visual localization through unbiased landmark features and local geometric consistency verification.
Ultra Diffusion Poser: Diffusion-Based Human Motion Tracking from Sparse Inertial Sensors and Ranging-based Between-sensor Distances
Dominik Hollidt (ETH Zurich), Christian Holz (ETH Zurich)
Pose EstimationRecurrent Neural NetworkDiffusion modelMultimodalityTime Series
🎯 What it does: Proposed Ultra Diffusion Poser (UDP), a full-body pose estimation framework based on diffusion models, which utilizes sparse IMU + UWB distance measurements. It first reconstructs the 3D sensor layout using MDS, then incorporates UWB-diffusion guidance during diffusion sampling to correct the pose.
Ultra-Fast Neural Video Compression
Jiahao Li (Microsoft Research Asia), Yan Lu (Microsoft Research Asia)
CompressionConvolutional Neural NetworkTransformerVideo
🎯 What it does: Propose a block-based video coding framework DCVC-UF, which utilizes a cross-frame interaction module to compress multiple frames into a single latent representation, and achieves parallel decoding through frame-specific decoders, significantly improving encoding and decoding throughput.
Ultra-Low Bitrate Perceptual Image Compression with Shallow Encoder
Tianyu Zhang (University of Science and Technology of China), Chang Wen Chen (Hong Kong Polytechnic University)
CompressionConvolutional Neural NetworkDiffusion modelAuto EncoderImage
🎯 What it does: Propose the AEIC framework, utilizing a shallow encoder and first-order diffusion decoder to achieve ultra-low bitrate (<0.05 bpp) image compression, balancing encoding efficiency and perceptual quality;
UltraFlux: Data-Model Co-Design for High-quality Native 4K Text-to-Image Generation across Diverse Aspect Ratios
Tian Ye (Hong Kong University of Science and Technology), Lei Zhu (Hong Kong University of Science and Technology)
GenerationTransformerPrompt EngineeringDiffusion modelAuto EncoderImageText
🎯 What it does: Designed and implemented UltraFlux, a native 4K multi-aspect-ratio text-to-image generation model trained on the large-scale 1M MultiAspect-4K-1M dataset.
Ultrasound-CLIP: Semantic-Aware Contrastive Pre-training for Ultrasound Image-Text Understanding
Jiayun Jin (Hangzhou City University), Binbin Zhou (Hangzhou City University)
ClassificationRetrievalTransformerVision Language ModelContrastive LearningMultimodalityBiomedical DataUltrasound
🎯 What it does: Constructed the first large-scale ultrasound image-text dataset US-365K, and proposed the Ultrasound-CLIP semantic-aware contrastive learning framework, which uses the UDT semantic ontology to guide image-text alignment.
Unblur-SLAM: Dense Neural SLAM for Blurry Inputs
Qi Zhang (University of Amsterdam), Martin R. Oswald (University of Amsterdam)
Convolutional Neural NetworkGaussian SplattingSimultaneous Localization and MappingImage
🎯 What it does: Propose an online SLAM system, Unblur-SLAM, which simultaneously handles motion blur and defocus blur, and achieves clear 3D reconstruction from blurred inputs through a single-frame deblurring network, 3D Gaussian splatting, and adaptive optimization.
Uncertainty-Aware Exploratory Direct Preference Optimization for Multimodal Large Language Models
Huatian Zhang (University of Science and Technology of China), Yongdong Zhang (University of Science and Technology of China)
OptimizationExplainability and InterpretabilityReinforcement Learning from Human FeedbackTransformerLarge Language ModelVision Language ModelDiffusion modelMultimodalityBenchmark
🎯 What it does: Improve direct preference optimization (DPO) by introducing an exploration mechanism based on knowledge uncertainty to reduce visual hallucinations in multimodal large language models (MLLMs).
Uncertainty-Aware Knowledge Distillation for Multimodal Large Language Models
Jingchen Sun (NEC Laboratories America, Inc.), Changyou Chen (NEC Laboratories America, Inc.)
Knowledge DistillationTransformerLarge Language ModelVision Language ModelMultimodality
🎯 What it does: Propose a uncertainty-aware knowledge distillation framework named Beta-KD, which distills large multimodal large language models (e.g., MobileVLM, Qwen-VL) into smaller student models.
Uncertainty-Aware Modality Fusion for Unaligned RGB-T Salient Object Detection
Mianzhao Wang (Tianjin University of Technology), Shengyong Chen (Tianjin University of Technology)
SegmentationTransformerImageMultimodality
🎯 What it does: Proposed an RGB-T salient object detection framework UMFNet without explicit registration.
Uncertainty-driven 3D Gaussian Splatting Active Mapping via Anisotropic Visibility Field
Shangjie Xue (Georgia Institute of Technology), Danfei Xu (Georgia Institute of Technology)
OptimizationComputational EfficiencyNeural Radiance FieldGaussian SplattingSimultaneous Localization and MappingImagePoint Cloud
🎯 What it does: Propose the GAVIS framework, which constructs a view-dependent anisotropic visibility field using 3D Gaussian Splatting (3DGS) to quantify the uncertainty of synthetic views, and realizes active mapping based on information gain.
Uncertainty-guided Compositional Alignment with Part-to-Whole Semantic Representativeness in Hyperbolic Vision-Language Models
Hayeon Kim (Seoul National University), Se Young Chun (Seoul National University)
ClassificationRetrievalRepresentation LearningVision Language ModelContrastive LearningMultimodality
🎯 What it does: Propose a vision-language model based on hyperbolic space called UNCHA, which quantifies the semantic representativeness of each part relative to the whole using a hyperbolic uncertainty metric, and integrates it into contrastive loss and reasoning loss to achieve more precise part-whole alignment and hierarchical semantic learning.
Underground Plant Exploration: Non-Destructive 3D Root Assessment with GPR Based on Point Graph Neural Network
Yuwei Zhou (Rochester Institute of Technology), Guoyu Lu (SUNY Binghamton)
Object DetectionSegmentationGraph Neural NetworkPoint CloudAgriculture Related
🎯 What it does: Propose a non-destructive 3D underground root modeling framework based on ground-penetrating radar (GPR), which first preprocesses B-scan signals, uses a hyperbola detection network to extract root positions, and then generates high-resolution root point clouds using a point graph neural network combined with an upsampling module.
Understanding and Enforcing Weight Disentanglement in Task Arithmetic
Shangge Liu (Nanjing University), Dacheng Tao (Nanyang Technological University)
ClassificationRepresentation LearningTransformerImage
🎯 What it does: Investigate the theoretical foundations of task arithmetic, propose Task-Feature Specialization (TFS), and demonstrate its relationship with weight decoupling and weight vector orthogonality, followed by designing an orthogonal regularization method called OrthoReg to improve task arithmetic performance.
Understanding and Mitigating Hallucinations in Multimodal Chain-of-Thought Models
Ji Ma (Northwestern Polytechnical University), Yanning Zhang (Northwestern Polytechnical University)
Explainability and InterpretabilityMultimodalityChain-of-Thought
🎯 What it does: Analyze and mitigate hallucination phenomena in multimodal chain-of-thought (MCoT) models
Understanding Counting Mechanisms in Large Language and Vision-Language Models
Hosein Hasani (Sharif University of Technology), Mahdieh Soleymani Baghshah (Sharif University of Technology)
Explainability and InterpretabilityTransformerImageMultimodality
🎯 What it does: This paper investigates the internal counting mechanisms of large language models and multimodal models in counting tasks using mechanism interpretability methods.
Understanding Task Transfer in Vision-Language Models
Bhuvan Sachdeva (Microsoft Research India), Vineeth N. Balasubramanian (Microsoft Research India)
Representation LearningTransformerSupervised Fine-TuningVision Language ModelImageVideoTextMultimodality
🎯 What it does: Investigate the zero-shot performance of Vision-Language models on other visual perception tasks after single-task fine-tuning, and construct a cross-task transfer graph;
Understanding Temporal Logic Consistency in Video-Language Models through Cross-Modal Attention Discriminability
Chengzhi Li (Beijing Institute of Technology), Zhongbin Guo (Beijing Institute of Technology)
Explainability and InterpretabilityTransformerSupervised Fine-TuningVision Language ModelContrastive LearningVideoTextMultimodality
🎯 What it does: Conduct an explanatory analysis of temporal logic consistency in video-large language models, revealing that insufficient temporal discriminability of cross-modal attention heads leads to inconsistencies.
Understanding the Role of Hallucination in Reinforcement Post-Training of Multimodal Reasoning Models
Gengwei Zhang (University of North Carolina at Chapel Hill), Tianlong Chen (University of North Carolina at Chapel Hill)
TransformerLarge Language ModelReinforcement LearningVision Language ModelImageTextMultimodality
🎯 What it does: In the paper, the authors propose the Hallucination-as-Cue framework, which systematically investigates the model's reasoning process and hallucination behavior under damaged inputs by introducing modality-specific corruptions (empty images, random images, text deletion) during the reinforcement learning post-training of multi-modal large language models, and uses this diagnostic perspective to analyze the true mechanisms of RL training.
Understanding, Accelerating, and Improving MeanFlow Training
Jin-Young Kim, Dominik Narnhofer (ETH Zurich)
GenerationComputational EfficiencyTransformerDiffusion modelFlow-based ModelImage
🎯 What it does: This paper systematically analyzes the interaction between instantaneous velocity (v) and average velocity (u) in MeanFlow training, proposing to improve one-shot generation performance by first accelerating the learning of v and gradually weighting the time intervals of u.
Uni-DAD: Unified Distillation and Adaptation of Diffusion Models for Few-step Few-shot Image Generation
Yara Bahram (ETS Montreal), Eric Granger (ETS Montreal)
GenerationDomain AdaptationKnowledge DistillationDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a single-stage unified distillation and adaptation framework, Uni-DAD, for rapidly generating high-quality images in few-step and few-sample scenarios.
Uni-Encoder Meets Multi-Encoders: Representation Before Fusion for Brain Tumor Segmentation with Missing Modalities
Peibo Song (Shandong University), Si Yong Yeo (Nanyang Technological University)
SegmentationRepresentation LearningTransformerMultimodalityBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper proposes a two-stage heterogeneous method called UniME for precise segmentation of brain tumors under missing modality conditions.
Uni-Hema: Unified Model for Digital Hematopathology
Abdul Rehman (Information Technology University), Waqas Sultani
ClassificationObject DetectionSegmentationConvolutional Neural NetworkTransformerVision Language ModelImageTextBiomedical Data
🎯 What it does: A unified multi-task, multi-modal (image + text) model named Uni-Hema was constructed for tasks in digital hemopathology, including cell detection, classification, segmentation, morphology prediction, and visual question answering.
UNI-OOD: Unified Object- and Image-level Out-of-Distribution Detection via Cross-Context Attentive Vision-Language Modeling
Yuchuan Li (Queen's University), Il-Min Kim (Queen's University)
Anomaly DetectionTransformerVision Language ModelContrastive LearningImageText
🎯 What it does: Proposed a unified object-level and image-level OOD detection framework, UNI-OOD, which utilizes dual CLIP encoders and cross-context attention mechanisms to achieve both types of detection in a single model.
Uni3R: Unified 3D Reconstruction and Semantic Understanding via Generalizable Gaussian Splatting from Unposed Multi-View Images
Xiangyu Sun (Sungkyunkwan University), Eunbyung Park (Yonsei University)
SegmentationGenerationDepth EstimationTransformerAuto EncoderGaussian SplattingImage
🎯 What it does: Built a unified 3D Gaussian field that can directly generate from multi-view images without pose information, achieving single forward pass inference for geometry reconstruction, semantic segmentation, and depth prediction.
UniAVGen: Unified Audio and Video Generation with Asymmetric Cross-Modal Interactions
Guozhen Zhang (Nanjing University), Limin Wang (Nanjing University)
GenerationTransformerDiffusion modelFlow-based ModelVideoMultimodalityAudio
🎯 What it does: Developed a unified dual-branch diffusion Transformer framework, UniAVGen, for human-centric audio-visual synchronization generation with multi-task support;
UNICBench: UNIfied Counting Benchmark for MLLM
Chenggang Rong (Northwestern Polytechnical University), Junyu Gao
TransformerLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodalityBenchmarkAudio
🎯 What it does: Proposed a unified cross-modal counting benchmark (UNICBench), covering image, text, and audio modalities, and constructed three-layer counting capability and difficulty labels;
UnicEdit-10M: A Dataset and Benchmark Breaking the Scale-Quality Barrier via Unified Verification for Reasoning-Enriched Edits
Keming Ye (Zhejiang University), Shengyu Zhang (Zhejiang University)
Image TranslationData SynthesisTransformerLarge Language ModelVision Language ModelDiffusion modelImageTextBenchmark
🎯 What it does: Constructed a unified image editing dataset UnicEdit-10M with 10M scale, and proposed a unified post-verification process Qwen-Verify and a comprehensive evaluation benchmark UnicBench.
UniChange: Unifying Change Detection with Multimodal Large Language Model
Xu Zhang (Nankai University), Xiang Li (Nankai University)
SegmentationTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: Designed a unified change detection framework called UniChange based on a multimodal large language model, which can simultaneously perform binary change detection and semantic change detection, and supports joint training with multi-source data.
UniComp: Rethinking Video Compression Through Informational Uniqueness
Chao Yuan (Meituan Inc.), Lin Ma (Meituan Inc.)
CompressionTransformerVideo
🎯 What it does: Proposes a video compression framework called UniComp based on information uniqueness, aiming to maximize the fidelity of video representations under limited computational budgets.
UniCompress: Token Compression for Unified Vision-Language Understanding and Generation
Ziyao Wang (Sony AI), Lingjuan Lyu (Sony AI)
CompressionTransformerLarge Language ModelVision Language ModelMultimodality
🎯 What it does: Proposes UNICOMPRESS, a pluggable visual token compression scheme that enables unified vision-language models to perform reasoning with shorter visual sequences in understanding and generation tasks.
UniCorrn: Unified Correspondence Transformer Across 2D and 3D
Prajnan Goswami, Huaizu Jiang
Pose EstimationRetrievalTransformerContrastive LearningImageMultimodalityPoint Cloud
🎯 What it does: Propose UniCorrn, a unified correspondence transformer model that can share weights across 2D-2D, 2D-3D, and 3D-3D tasks, achieving cross-modal keypoint correspondence.
UniDAC: Universal Metric Depth Estimation for Any Camera
Girish Chandar Ganesan (Michigan State University), Xiaoming Liu (University of North Carolina at Chapel Hill)
Depth EstimationTransformerContrastive LearningImage
🎯 What it does: UniDAC proposes a general monocular metric depth estimation framework, achieving robust reasoning for arbitrary cameras by decomposing depth into relative depth and spatially varying scale estimation.
UniDef: Universal Defense Against Unauthorized Image Manipulation
Mingwen Shao (Shenzhen University of Advanced Technology), Chao Dong (Shenzhen University of Advanced Technology)
Adversarial AttackDiffusion modelImage
🎯 What it does: To address unauthorized image editing in diffusion models, this paper proposes a generic defense framework called UniDef, which can prevent images from being tampered with across various models and tasks.
UniDex: A Robot Foundation Suite for Universal Dexterous Hand Control from Egocentric Human Videos
Gu Zhang (Tsinghua University), Huazhe Xu (Tsinghua University)
Robotic IntelligenceVision-Language-Action ModelDiffusion modelImageTextMultimodalityPoint Cloud
🎯 What it does: Built the UniDex suite, which includes: ① UniDex-Dataset based on egocentric human video preprocessing; ② Unified Function-Actuator-Aligned Space (FAAS) and the pre-trained 3D Vision-Language-Action (VLA) policy UniDex-VLA on this space; ③ Portable data collection device UniDex-Cap for collecting human-robot co-training data.
UniEdit-I: Training-free Image Editing for Unified VLM via Iterative Understanding, Editing and Verifying
Chengyu Bai (Peking University), Shanghang Zhang (Peking University)
GenerationTransformerLarge Language ModelVision Language ModelDiffusion modelFlow-based ModelMultimodalityOrdinary Differential Equation
🎯 What it does: Developed a novel training-agnostic, closed-loop image editing framework called UniEdit-I, which enables dynamic and controllable editing of images within the semantic latent space of a unified vision-language model.
Unified Camera Positional Encoding for Controlled Video Generation
Cheng Zhang (Monash University), Jianfei Cai (Monash University)
GenerationTransformerDiffusion modelVideoText
🎯 What it does: This paper proposes a Unified Camera Position Encoding (UCPE), which enables video diffusion Transformers to precisely control camera parameters such as pose, focal length, and distortion by combining relative ray encoding and absolute orientation encoding, thereby achieving high-quality, controllable text-to-video generation.
Unified Customized Generation by Disentangled Reward Modeling
Shaojin Wu (ByteDance), Qian He (ByteDance)
GenerationData SynthesisTransformerSupervised Fine-TuningReinforcement LearningDiffusion modelFlow-based ModelAuto EncoderImageMultimodalityBenchmark
🎯 What it does: Proposed the USO unified model, achieving unified generation for subject customization, style customization, and joint customization of both, while enhancing generation quality through cross-task co-decomposition and auxiliary style rewards.
Unified Generation and Self-Verification for Vision-Language Models via Advantage Decoupled Preference Optimization
Xinyu Qiu (Zhejiang University), Linchao Zhu (Zhejiang University)
OptimizationReinforcement LearningVision Language ModelMultimodalityBenchmark
🎯 What it does: Propose a unified advantage disentangled preference optimization (ADPO) framework that enables vision-language models to simultaneously generate answers and self-verify within the same strategy.
Unified Latent Space for Understanding and Generation via Semantic Auto-encoder
Xiaojie Li (Kesheng BioTech), Daquan Zhou (Peking University)
RestorationGenerationTransformerAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: Propose and train a semantic autoencoder (S-AE) based on pre-trained DINOv3, achieving detail reconstruction while maintaining semantic discriminability through stage-wise training and semantic regularization, constructing a unified latent space.
Unified Multimodal Models as Auto-Encoders
Zhiyuan Yan (Shenzhen Graduate School, Peking University), Li Yuan (Baidu)
RecognitionObject DetectionGenerationReinforcement LearningAuto EncoderMultimodality
🎯 What it does: Propose a unified autoencoder framework that connects image-text understanding (I2T) with text-image generation (T2I) through text mediation, and achieves bidirectional collaborative improvement via post-training reconstruction reinforcement learning (Unified-GRPO).
Unified Number-Free Text-to-Motion Generation Via Flow Matching
Guanhe Huang (King's College London), Oya Celiktutan (King's College London)
GenerationTransformerFlow-based ModelAuto EncoderTextTime Series
🎯 What it does: Propose Unified Motion Flow (UMF) to achieve human action generation for any number of people, utilizing text prompts to control actions while supporting both single-person and multi-person interactions.
Unified Personalized Understanding, Generating and Editing
Yu Zhong (Zhejiang University), Yueting Zhuang (Zhejiang University)
GenerationTransformerPrompt EngineeringMixture of ExpertsVision Language ModelMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: Propose OmniPersona, an end-to-end unified framework for personalized understanding, generation, and editing;
Unified Primitive Proxies for Structured Shape Completion
Zhaiyu Chen (Technical University of Munich), Xiao Xiang Zhu (Technical University of Munich)
RestorationGenerationTransformerPoint Cloud
🎯 What it does: Propose UniCo, a unified structured shape completion model that can simultaneously predict complete quadratic surface primitives (geometry, semantics, inlier membership) and full point clouds in a single forward pass, directly generating primitive sets usable for assembly.
Unified Spatiotemporal Token Compression for Video-LLMs at Ultra-Low Retention
Junhao Du (Shanghai Jiao Tong University), Guo Lu (Shanghai Jiao Tong University)
CompressionTransformerLarge Language ModelVision Language ModelVideoBenchmark
🎯 What it does: This paper proposes a unified spatiotemporal token compression framework that can compress visual tokens in video large language models to as low as 2% without significantly affecting performance.
Unified Spherical Frontend: Learning Rotation-Equivariant Representations of Spherical Images from Any Camera
Mukai Yu (Carnegie Mellon University), László A. Jeni (Carnegie Mellon University)
ClassificationObject DetectionSegmentationRepresentation LearningConvolutional Neural NetworkImage
🎯 What it does: Propose a unified spherical front-end (USF) framework that maps images from any camera to a sphere via geometric projection, enabling resampling, convolution, and pooling on the sphere to accomplish visual tasks under transformations such as distortion and rotation.
Unified Vector Floorplan Generation via Markup Representation
Kaede Shiohara (University of Tokyo), Toshihiko Yamasaki (University of Tokyo)
GenerationTransformerImageTextGraph
🎯 What it does: Designed a unified floorplan markup language (Floorplan Markup Language, FML) and proposed an FMLM model based on autoregressive Transformer, achieving multiple vector floorplan generation tasks in a single model, including unconditional, boundary, graph, boundary+graph, and room count conditions.
UniFusion: A Unified Image Fusion Framework with Robust Representation and Source-Aware Preservation
Xingyuan Li (Zhejiang University), Jinyuan Liu (Dalian Maritime University)
TransformerAuto EncoderContrastive LearningImageMultimodalityBiomedical Data
🎯 What it does: Propose a unified image fusion framework named UniFusion, leveraging DINOv3 as a general-purpose semantic feature extractor, combined with adaptive Adapter, reconstruction alignment mechanism, and dual-layer optimization to achieve high-quality cross-task fusion.
Unifying Language-Action Understanding and Generation for Autonomous Driving
Xinyang Wang (Zhejiang University), Wei Chen (Li Auto)
Autonomous DrivingTransformerLarge Language ModelVision-Language-Action ModelImageVideoTextMultimodality
🎯 What it does: Designed and implemented the LinkVLA model, unifying language instructions and trajectory actions into a single discrete codebook, and achieving efficient, aligned closed-loop driving through bidirectional learning and coarse-to-fine generation.
Unifying Perception and Action: A Hybrid-Modality Pipeline with Implicit Visual Chain-of-Thought for Robotic Action Generation
Xiangkai Ma (Nanjing University), Sanglu Lu (Nanjing University)
Robotic IntelligenceTransformerVision Language ModelVision-Language-Action ModelAuto EncoderVideoMultimodalityChain-of-Thought
🎯 What it does: Propose a Vision-Integrated Trajectory Alignment (VITA) framework that unifies visual perception and action generation within a single model by leveraging a shared discrete latent space, enabling end-to-end robot control.
Unifying Precise Keyframes and Semantic Control via Multi-level Diffusion
Linjun Wu (Zhejiang University), Xiaogang Jin (Zhejiang University)
GenerationConvolutional Neural NetworkTransformerVision Language ModelVision-Language-Action ModelDiffusion modelTextPoint CloudSequential
🎯 What it does: Propose a multi-level diffusion model that synchronizes control using text and sparse keyframes for human motion generation and editing, achieving precise spatiotemporal constraints and semantic consistency.
UniGame: Turning a Unified Multimodal Model Into Its Own Adversary
Zhaolong Su (William & Mary), Jindong Wang (William & Mary)
TransformerSupervised Fine-TuningGenerative Adversarial NetworkContrastive LearningMultimodality
🎯 What it does: Propose a self-adversarial post-training framework called UniGame, which leverages a generation branch to actively identify and challenge vulnerabilities in the understanding branch, thereby enhancing the consistency between understanding and generation in a unified multimodal model (UMM).
UniGen-1.5: Enhancing Image Generation and Editing through Reward Unification in RL
Rui Tian (Fudan University), Afshin Dehghan (Apple)
GenerationTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningVision Language ModelImageTextMultimodality
🎯 What it does: Propose UniGen-1.5, a unified multimodal large language model capable of image understanding, text-to-image generation, and image editing.
UniGenDet: A Unified Generative-Discriminative Framework for Co-Evolutionary Image Generation and Generated Image Detection
Yanran Zhang (Tsinghua University), Jie Zhou (Tsinghua University)
GenerationAnomaly DetectionTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelFlow-based ModelImageTextMultimodality
🎯 What it does: Proposed a unified generative-discriminative framework called UniGenDet, achieving synergistic evolution between image generation and generated image detection.
UniGeoRS: A Unified Benchmark for Tri-view Geo-Localization
Xiao Liang (Beijing Institute of Technology), Kang Ma (Beijing Institute of Technology)
RetrievalConvolutional Neural NetworkContrastive LearningImageBenchmark
🎯 What it does: Proposed a unified tri-perspective geolocation dataset named UniGeoRS, and introduced a cross-perspective attention matching enhancement module named CAME to improve cross-perspective feature alignment and retrieval accuracy.
UniGeoSeg: Towards Unified Open-World Segmentation for Geospatial Scenes
Shuo Ni (Beijing Institute of Technology), Jing Zhang (Wuhan University)
SegmentationTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageMultimodalityBenchmark
🎯 What it does: Proposed the GeoSeg-1M million-scale remote sensing instruction-driven segmentation dataset and the GeoSeg-Bench benchmark, along with the unified framework UniGeoSeg;
UniLDiff: Unlocking the Power of Diffusion Priors for All-in-One Image Restoration
Zihan Cheng (Xiamen University), Yanyun Qu (Xiamen University)
RestorationMixture of ExpertsDiffusion modelAuto EncoderImage
🎯 What it does: Propose UniLDiff, a unified latent diffusion model integrating Degradation-Aware Feature Fusion (DAFF) and Detail-Aware Expert Module (DAEM) for one-time image restoration.
UniLight: A Unified Representation for Lighting
Zitian Zhang (Universite Laval), Valentin Deschaintre (Adobe Research)
GenerationRetrievalRepresentation LearningTransformerVision Language ModelDiffusion modelContrastive LearningImageTextMultimodality
🎯 What it does: The paper proposes a unified light representation called UNILIGHT, which can map various light modalities such as text, images, environment maps, and irradiance maps into a shared latent space.
UniLS: End-to-End Audio-Driven Avatars for Unified Listening and Speaking
Xuangeng Chu (Shanda AI Research Tokyo), Bo Zheng (Shanda AI Research Tokyo)
GenerationTransformerSupervised Fine-TuningVideoAudio
🎯 What it does: Developed UniLS, an end-to-end audio-driven conversational facial animation framework capable of simultaneously generating speaking and listening expressions.
UniM: A Unified Any-to-Any Interleaved Multimodal Benchmark
Yanlin Li (NUS), Wynne Hsu (NUS)
Large Language ModelAgentic AIMultimodalityBenchmark
🎯 What it does: This paper proposes the UNIM benchmark, constructing a unified any-to-any interwoven multimodal dataset (7 modalities, 31k examples, 30 domains), and designs a three-dimensional evaluation system along with the baseline model UNIMA.
UniMERNet: A Universal Network for Real-World Mathematical Expression Recognition
Zhuangcheng Gu (Shanghai AI Laboratory), Conghui He (Shanghai AI Laboratory)
RecognitionTransformerVision Language ModelImageText
🎯 What it does: Designed UniMERNet, specifically for recognizing mathematical expressions in diverse real-world scenarios.
UniMMAD: Unified Multi-Modal and Multi-Class Anomaly Detection via MoE-Driven Feature Decompression
Yuan Zhao (Dalian University of Technology), Xiaoqi Zhao (Nanyang Technological University)
Anomaly DetectionMixture of ExpertsMultimodalityBenchmark
🎯 What it does: Propose UniMMAD, a unified multi-modal multi-class unsupervised anomaly detection framework that can handle different sensor combinations and classes within a single model, supporting continual learning.
UniPart: Part-Level 3D Generation with Unified 3D Geom-Seg Latents
Xufan He (Nanjing University of Science and Technology), Dong Du (Nanjing University of Science and Technology)
GenerationTransformerDiffusion modelFlow-based ModelAuto EncoderImageMesh
🎯 What it does: Propose a unified Geom-Seg VecSet latent representation and a two-stage latent diffusion framework UniPart to achieve controllable part-level 3D generation based on images.
UniPercept: A Unified Diffusion Model for Generalizable Visual Perception
Zuyan Zhao, Xilin Chen
SegmentationPose EstimationDepth EstimationTransformerDiffusion modelFlow-based ModelRectified FlowImagePoint Cloud
🎯 What it does: Proposed a unified diffusion foundation-adapter framework called UniPercept, which can efficiently generalize and rapidly adapt to various visual perception tasks.
UniPixie: Unified and Probabilistic 3D Physics Learning via Flow Matching
Qilin Huang (University of Pennsylvania), Lingjie Liu (MIT)
GenerationTransformerFlow-based ModelImagePhysics Related
🎯 What it does: Propose UNIPIXIE, a unified controllable continuous physical attribute prediction framework that can generate adjustable material properties from a single visual input and supports multiple physics solvers;
UniPR: Unified Object-level Real-to-Sim Perception and Reconstruction from a Single Stereo Pair
Chuanrui Zhang (NTU), Ziwei Wang (NTU)
Object DetectionGenerationPose EstimationTransformerVision Language ModelAuto EncoderImage
🎯 What it does: UniPR proposes an end-to-end single stereo image pair perception and reconstruction framework that can simultaneously detect, locate, and reconstruct the real-scale 3D shapes of all objects in the scene.
Unique Lives, Shared World: Learning from Single-Life Videos
Tengda Han (Google DeepMind), Dima Damen (Google DeepMind)
Depth EstimationRepresentation LearningContrastive LearningVideo
🎯 What it does: This paper proposes the single-person life learning paradigm, training a visual model using only first-person videos from a single individual and investigating its ability to learn the geometry of the world.
UniRain: Unified Image Deraining with RAG-based Dataset Distillation and Multi-objective Reweighted Optimization
Qianfeng Yang (Dalian Polytechnic University), Jiangxin Dong (Nanjing University of Science and Technology)
RestorationMixture of ExpertsVision Language ModelImageRetrieval-Augmented Generation
🎯 What it does: The paper proposes UniRain, a unified image de-raining framework capable of handling both raindrops and rain streaks during the day and night.
UniRefiner: Teaching Pre-trained ViTs to Self-Dispose Dross via Contrastive Register
Congpei Qiu (Xi'an Jiaotong University), Tong Zhang (University Of Chinese Academy Of Sciences)
SegmentationDepth EstimationTransformerSupervised Fine-TuningContrastive LearningImage
🎯 What it does: Post-training fine-tuning of pre-trained Vision Transformers by identifying and separating three categories of pseudo tokens (Fixed Pattern FP, Global Proxy GP, Attention Hijacking AH), utilizing a contrastive registration mechanism to guide these pseudo tokens into the registration region, thereby significantly enhancing spatial representation quality.
UniSER: A Foundation Model for Unified Soft Effects Removal
Jingdong Zhang (Texas A&M University), Xiaohang Zhan (Adobe Research)
RestorationData SynthesisTransformerDiffusion modelAuto EncoderImage
🎯 What it does: Propose UniSER, a unified foundation model for removing various soft effects (such as lens flare, haze, shadows, reflections) in real-world scenarios, supporting fine-grained control over removal intensity in arbitrary regions; it also enables the reverse addition or enhancement of soft effects.
UniSH: Unifying Scene and Human Reconstruction in a Feed-Forward Pass
Mengfei Li (Hong Kong University of Science and Technology), Yike Guo (Hong Kong University of Science and Technology)
RestorationPose EstimationDepth EstimationKnowledge DistillationTransformerNeural Radiance FieldVideoPoint Cloud
🎯 What it does: UniSH simultaneously recovers scene geometry, camera parameters, and human SMPL models in a single forward inference, achieving panoramic 4D human-machine joint reconstruction.
UniSpector: Towards Universal Open-set Defect Recognition via Spectral-Contrastive Visual Prompting
Geonuk Kim (LG Energy Solution), Junho Yim (LG Energy Solution)
Object DetectionAnomaly DetectionTransformerPrompt EngineeringContrastive LearningImageBenchmark
🎯 What it does: Proposes a vision-based open defect detection framework called UniSpector, addressing the prompt embedding collapse issue in industrial inspection, enabling the identification of unknown defects without retraining.
UniT: Unified Multimodal Chain-of-Thought Test-time Scaling
Leon Liangyu Chen, Felix Juefei-Xu
GenerationData SynthesisOptimizationComputational EfficiencyTransformerAgentic AIVision Language ModelMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Propose the UniT framework, which utilizes a unified multimodal model to achieve self-iteration and optimization in multi-round generation and editing through chain-of-thought (Chain-of-Thought) during inference, enabling self-inspection, subgoal decomposition, and content memory.
UniTEX: Universal High Fidelity Generative Texturing for 3D Shapes
Yixun Liang (Hong Kong University of Science and Technology), Ping Tan (Hong Kong University of Science and Technology)
GenerationTransformerSupervised Fine-TuningDiffusion modelImageMesh
🎯 What it does: Propose a two-stage 3D texture generation framework called UniTEX, which first generates multi-view images using a large-scale Diffusion Transformer, and then directly regresses complete textures in 3D space through a Large Texturing Model.
UnityVideo: Unified Multi-Modal Multi-Task Learning for Enhancing World-Aware Video Generation
Jiehui Huang (Kuaishou Technology), Jiaya Jia
SegmentationGenerationDepth EstimationTransformerDiffusion modelVideoMultimodality
🎯 What it does: Propose UnityVideo, a unified multimodal multitask video generation and understanding framework that can simultaneously handle text-to-video, controllable generation, and multimodal estimation tasks.
UniVBench: Towards Unified Evaluation for Video Foundation Models
Jianhui Wei (Zhejiang University), Zuozhu Liu (Zhejiang University)
TransformerLarge Language ModelAgentic AIVision Language ModelVideoTextMultimodalityBenchmark
🎯 What it does: Propose UniVBench, a unified evaluation framework for video foundation models, covering four core capabilities: video understanding, generation, editing, and reconstruction.
Universal 3D Shape Matching via Coarse-to-Fine Language Guidance
Qinfeng Xiao (Hong Kong Polytechnic University), Kit-lun Yick (Hong Kong Polytechnic University)
RetrievalTransformerLarge Language ModelContrastive LearningMesh
🎯 What it does: Designed a category-agnostic, coarse-to-fine language-guided 3D shape matching framework called UniMatch.
Universal Guideline-Driven Image Clustering via a Hybrid LLM Agent
Wenliang Zhong (University of Texas at Arlington), Junzhou Huang (University of Texas at Arlington)
TransformerLarge Language ModelAgentic AIVision Language ModelMultimodality
🎯 What it does: Proposes a generic image clustering framework that handles diverse image clustering scenarios through text guidance, covering tasks from general to fine-grained, criteria from global to local, and distributions from balanced to long-tailed.
Universal-to-Specific: Dynamic Knowledge-Guided Multiple Instance Learning for Few-Shot Whole Slide Image Classification
Junjian Li (Central South University), Jianxin Wang (Central South University)
ClassificationTransformerLarge Language ModelVision Language ModelBiomedical Data
🎯 What it does: This paper proposes a dynamic knowledge-guided multi-instance learning framework called DyKo for whole slide image (WSI) classification under few-shot scenarios.
UniVerse: A Unified Modulation Framework for Segmentation-Free, Disentangled Multi-Concept Personalization
Quynh Phung (University of Maryland, College Park), Jia-Bin Huang (University of Maryland, College Park)
GenerationTransformerVision Language ModelDiffusion modelImageTextMultimodalityBenchmark
🎯 What it does: Proposes UniVerse, a unified modulation framework that can extract visual and textual conditions from reference images and text without requiring segmentation masks, enabling the decomposition and combination of multiple concepts;
UniVerse: Empower Unified Generation with Reasoning and Knowledge
Kaiyue Sun (HKU), Xihui Liu (HKU)
GenerationData SynthesisTransformerLarge Language ModelPrompt EngineeringVision Language ModelMultimodalityChain-of-Thought
🎯 What it does: Proposed a large-scale 120k semantically annotated image-text dataset named UniVerse, containing implicit prompts, reasoning chains, and explicit prompt triplets, and trained a unified multimodal model using this dataset to achieve text-to-image generation with enhanced reasoning and knowledge capabilities.
Unlearning without Forgetting: Securely Removing Targeted Concepts from Large-Scale Vision-Language Open-Vocabulary Detectors
Zhongze Wu (Central South University), Jun Long (Central South University)
Object DetectionSafty and PrivacyComputational EfficiencyRepresentation LearningTransformerVision Language ModelFlow-based ModelMultimodality
🎯 What it does: To address the machine unlearning problem in open-vocabulary detection models, the SafeDetect method is proposed, which utilizes zero-space projection to update only the preserved concept's orthogonal subspace, thereby safely forgetting specified concepts.
Unleashing Stealthy Backdoor Pandemic by Infecting a Single Diffusion Model
Mohaiminul Al Nahian (Binghamton University), Adnan Siraj Rakin (Florida International University)
Adversarial AttackConvolutional Neural NetworkTransformerDiffusion modelAuto EncoderImage
🎯 What it does: Implant a hidden backdoor into a single diffusion model, enabling the synthetic data it generates to automatically propagate the backdoor during subsequent classifier training;
Unleashing the Intrinsic Visual Representation Capability of Multimodal Large Language Models
Hengzhuang Li (HUST), Hai Jin (HUST)
Representation LearningTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelMultimodality
🎯 What it does: This paper proposes a training framework called LaVer, enabling multi-modal large language models to directly learn more discriminative visual representations through masked image modeling in the visual encoding space, addressing the modality imbalance issue where visual information is neglected in deeper layers.
Unleashing the Power of Chain-of-Prediction for Monocular 3D Object Detection
Zhihao Zhang (Michigan State University), Xiaoming Liu (Michigan State University)
Object DetectionAutonomous DrivingTransformerImageChain-of-Thought
🎯 What it does: Propose MonoCoP, a monocular 3D object detection framework that adaptively leverages geometric correlations among 3D attributes (size, orientation, depth).
Unleashing Vision-Language Semantics for Deepfake Video Detection
Jiawen Zhu (Singapore Management University), Guansong Pang (Singapore Management University)
Anomaly DetectionTransformerPrompt EngineeringVision Language ModelContrastive LearningVideoText
🎯 What it does: Proposed a CLIP-based deepfake video detection framework named VLAForge.
Unleashing VLA Potentials in Autonomous Driving via Explicit Learning from Failures
Yuechen Luo (Tsinghua University), Fuxi Wen (Tsinghua University)
Autonomous DrivingSupervised Fine-TuningReinforcement LearningVision Language ModelVision-Language-Action Model
🎯 What it does: Propose the ELF-VLA framework, which enhances the RL fine-tuning process of VLA by providing structured failure diagnosis feedback through a teacher model
Unlocking 3D Affordance Segmentation with 2D Semantic Knowledge
Yu Huang (Shanghai Jiaotong University), Wei Shen (Shanghai Jiaotong University)
SegmentationComputational EfficiencyTransformerPrompt EngineeringAuto EncoderMultimodalityPoint Cloud
🎯 What it does: This paper proposes a pre-training strategy called Cross-Modal Affinity Transfer (CMAT), which transfers the semantic structure of 2D visual foundation models to 3D point cloud encoders, achieving more functionally semantically meaningful 3D affordance segmentation;
Unlocking Motion from Large Vision Models with a Semantic and Kinematic Duality for Gait Recognition
Zhanbo Huang (Michigan State University), Yu Kong (Michigan State University)
RecognitionTransformerSupervised Fine-TuningVideoText
🎯 What it does: Propose the GaitMax framework, which unifies semantic-level global representations with kinematic-level fine-grained temporal representations. It extracts features using a large-scale vision model and achieves global and fine-grained temporal modeling of gait through multi-branch attention.
Unlocking Positive Transfer in Incrementally Learning Surgical Instruments: A Self-reflection Hierarchical Prompt Framework
Yu Zhu (Chinese University of Hong Kong), Pheng-Ann Heng (Chinese University of Hong Kong)
SegmentationConvolutional Neural NetworkGraph Neural NetworkPrompt EngineeringVideoBiomedical Data
🎯 What it does: Designed a self-reflective hierarchical prompting framework for forgetting-free learning of incrementally emerging instrument categories in surgical videos, achieving forward and backward knowledge transfer.