CVPR 2026 Papers — Page 25
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 4071 papers
One-Step Diffusion Transformer for Controllable Real-World Image Super-Resolution
Yushun Fang (Shanghai Jiao Tong University), Yanfeng Wang (Shanghai Jiao Tong University)
Super ResolutionTransformerSupervised Fine-TuningDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: Proposed a single-step diffusion transformer named ODTSR for controllable real-world image super-resolution.
One-to-All Animation: Alignment-Free Character Animation and Image Pose Transfer
Shijun Shi (Jiangnan University), Jiangning Zhang (Zhejiang University)
Image TranslationGenerationTransformerDiffusion modelAuto EncoderImageVideo
🎯 What it does: Propose the One-to-All Animation framework, which can achieve high-fidelity character animation and image pose transfer even when there is spatial and pose misalignment between the reference image and driving video.
One-to-More: High-Fidelity Training-Free Anomaly Generation with Attention Control
Haoxiang Rao (Nanjing University), Caifeng Shan (Nanjing University)
GenerationAnomaly DetectionTransformerPrompt EngineeringDiffusion modelImage
🎯 What it does: Proposes a training-free few-shot anomaly generation method called O2MAG, which generates high-fidelity local anomalies through three-branch attention grafting to achieve background preservation, text consistency, and complete mask filling.
OneCAT: Decoder-Only Auto-Regressive Model for Unified Understanding and Generation
Han Li (Shanghai Jiao Tong University), Hongkai Xiong (Shanghai Jiao Tong University)
GenerationTransformerMixture of ExpertsVision Language ModelMultimodality
🎯 What it does: This paper proposes OneCAT, a unified multimodal model that is decoder-only and solely autoregressive (AR), capable of seamlessly supporting image understanding, image generation, and image editing;
OneHOI: Unifying Human-Object Interaction Generation and Editing
Jiun Tian Hoe, Chee Seng Chan (University Of Malaya)
GenerationTransformerVision-Language-Action ModelDiffusion modelImageMultimodality
🎯 What it does: Proposed a unified single-model framework called OneHOI that can simultaneously perform the generation and editing of human-object interactions (HOI).
OneOcc: Semantic Occupancy Prediction for Legged Robots with a Single Panoramic Camera
Hao Shi (Zhejiang University), Kaiwei Wang (Zhejiang University)
SegmentationRobotic IntelligenceMixture of ExpertsImageVideoBenchmark
🎯 What it does: Propose a semantic scene completion framework called OneOcc, which uses only a single panoramic camera and is specifically designed for gait-induced body jitter and 360° continuous perception;
OneSparse: A Unified Framework for Sparse Activation Layers in Vision Models
Xingkui Zhu (Huazhong University Of Science And Technology), Xiang Bai (Xiaohongshu Inc)
ClassificationObject DetectionSegmentationConvolutional Neural NetworkTransformerMixture of ExpertsImage
🎯 What it does: Proposed the OneSparse framework, unifying MoE and memory modules into a three-stage process of dispatch–process–combine, and designed the Nexus Layer as a sparse activation layer in visual models.
OneStory: Coherent Multi-Shot Video Generation with Adaptive Memory
Zhaochong An (Meta AI), Tian Xie (Meta AI)
GenerationData SynthesisVision Language ModelDiffusion modelVideoTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes a framework named OneStory for achieving cross-shot consistency and controllable long-form storytelling in multi-shot video generation.
OneThinker: All-in-one Reasoning Model for Image and Video
Kaituo Feng, Xiangyu Yue
ClassificationRecognitionObject TrackingSegmentationGenerationTransformerReinforcement LearningVision Language ModelImageVideoChain-of-Thought
🎯 What it does: Train a unified multimodal reasoning general-purpose model called OneThinker, covering ten visual tasks involving images and videos, such as QA, captioning, spatial/temporal localization, tracking, and segmentation.
Online Data Curation for Object Detection via Marginal Contributions to Dataset-level Average Precision
Zitang Sun (Sony Group Corporation), Takeshi Ohashi (Sony Group Corporation)
Object DetectionKnowledge DistillationImage
🎯 What it does: This paper proposes an online data screening method for object detection called DetGain, which dynamically selects the most informative training samples by leveraging the marginal contribution of each image to the global average precision (mAP).
Online3R: Online Learning for Consistent Sequential Reconstruction Based on Geometry Foundation Model
Shunkai Zhou (Peking University), Hongbin Zha (Peking University)
GenerationData SynthesisPrompt EngineeringSimultaneous Localization and MappingPoint CloudSequential
🎯 What it does: This paper proposes the Online3R framework, which achieves continuous and consistent 3D reconstruction by performing online fine-tuning of a frozen geometric base model during testing using learnable visual prompts.
OnlineHMR: Video-based Online World-Grounded Human Mesh Recovery
Yiwen Zhao, László A. Jeni (Carnegie Mellon University)
Pose EstimationDepth EstimationTransformerSimultaneous Localization and MappingVideoMesh
🎯 What it does: Proposes a fully online framework, OnlineHMR, for real-time recovery of human meshes and camera trajectories in the world coordinate system from streaming video.
OnlinePG: Online Open-Vocabulary Panoptic Mapping with 3D Gaussian Splatting
Hongjia Zhai (Zhejiang University), Guofeng Zhang (Zhejiang University)
SegmentationVision Language ModelGaussian SplattingSimultaneous Localization and MappingMultimodalityPoint Cloud
🎯 What it does: Proposed the Online Open-Vocabulary Panorama Mapping System OnlinePG, which simultaneously achieves geometric reconstruction and semantic perception using 3D Gaussian Splatting
OntoAug: Rethinking Generative Data Augmentation via Ontology Guidance
Shuo Wang (Huazhong Agricultural University), Jun Luo (Huazhong Agricultural University)
ClassificationDomain AdaptationLarge Language ModelPrompt EngineeringDiffusion modelImage
🎯 What it does: Propose a generative data augmentation framework called OntoAug based on ontology, which can generate diverse backgrounds while maintaining the semantic consistency of the foreground subject, and control the generation process of diffusion models through foreground masks.
Open the Motion Door: Atomic Motion Decomposition and Recomposition for Open-Vocabulary Motion Generation
Ke Fan (Shanghai Jiao Tong University), Lizhuang Ma (Shanghai Jiao Tong University)
GenerationData SynthesisLarge Language ModelVision-Language-Action ModelAuto EncoderContrastive LearningTextMultimodalitySequential
🎯 What it does: This paper proposes the Atomic Motion Decomposition and Recomposition framework, which decomposes natural language descriptions into atomic motion texts and then generates complete 3D human motion through text-motion alignment and combination feature fusion.
Open-Ended Instruction Realization with LLM-Enabled Multi-Planner Scheduling in Autonomous Vehicles
Jiawei Liu (Jilin University), Qing Guo (NKIARI)
Autonomous DrivingOptimizationTransformerLarge Language ModelTextPoint CloudBenchmark
🎯 What it does: Studied how to leverage large language models to generate scheduling scripts that coordinate multiple MPC motion planners, enabling the execution of open-ended language instructions in autonomous driving.
Open-Vocabulary Domain Generalization in Urban-Scene Segmentation
Dong Zhao, Zhun Zhong (Hefei University of Technology)
SegmentationDomain AdaptationAutonomous DrivingComputational EfficiencyTransformerVision Language ModelContrastive LearningImageTextMultimodalityBenchmark
🎯 What it does: Proposes a new task of open-vocabulary domain generalization semantic segmentation (OVDG-SS), constructs a comprehensive benchmark for urban driving scenarios, and introduces a text-image correlation refinement module based on state space, named S2-Corr, to achieve robust segmentation under unseen domains and unseen categories.
Open-world Hand-Object Interaction Video Generation Based on Structure and Contact-aware Representation
Haodong Yan (Hong Kong University of Science and Technology (Guangzhou)), Haoang Li (Hong Kong University of Science and Technology (Guangzhou))
GenerationData SynthesisTransformerDiffusion modelAuto EncoderVideoChain-of-Thought
🎯 What it does: Proposed a structure and contact-aware hand-object interaction (HOI) video generation framework that jointly generates videos and interaction representations.
OpenDance: Multimodal Controllable 3D Dance Generation with Large-scale Internet Data
Jinlu Zhang (Peking University), Yizhou Wang (Peking University)
GenerationData SynthesisTransformerVision Language ModelAuto EncoderTextMultimodalityAudio
🎯 What it does: This work constructs a large-scale multimodal dance dataset called OpenDanceSet and proposes the OpenDanceNet framework to achieve controllable 3D dance generation conditioned on music plus any combination of text, 2D keypoints, and trajectories;
OpenDPR: Open-Vocabulary Change Detection via Vision-Centric Diffusion-Guided Prototype Retrieval for Remote Sensing Imagery
Qi Guo (Wuhan University), Yanfei Zhong (Wuhan University)
SegmentationTransformerLarge Language ModelPrompt EngineeringDiffusion modelImageTextMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: Proposed OpenDPR, a training-free vision-centric diffusion-guided prototype retrieval framework for open-vocabulary change detection, further enhancing change localization under weakly supervised conditions through the S2C module.
OpenFS: Multi-Hand-Capable Fingerspelling Recognition with Implicit Signing-Hand Detection and Frame-Wise Letter-Conditioned Synthesis
Junuk Cha (Korea Advanced Institute of Science and Technology), Han-Mu Park (Korea Electronics Technology Institute)
RecognitionGenerationPose EstimationTransformerDiffusion model
🎯 What it does: Propose a multi-hand recognizable finger spelling recognition and synthesis framework OpenFS, which includes an implicit hand-finger detector and a generator based on letter frame-level conditions;
Opening the Sim-to-Real Door for Humanoid Pixel-to-Action Policy Transfer
Haoru Xue (NVIDIA), Yuke Zhu (NVIDIA)
Domain AdaptationRobotic IntelligenceRecurrent Neural NetworkReinforcement LearningImage
🎯 What it does: Developed a full RGB perception, end-to-end humanoid robot door-opening strategy (DoorMan), achieving zero-offset transfer from simulation to real-world environments;
OpenMarcie: Dataset for Multimodal Action Recognition in Industrial Environments
Hymalai Bello (DFKI Kaiserslautern), Paul Lukowicz (DFKI Kaiserslautern)
RecognitionConvolutional Neural NetworkTransformerLarge Language ModelVision-Language-Action ModelImageTextMultimodalityPoint CloudBenchmarkAudio
🎯 What it does: Built and released the OpenMarcie multimodal industrial human behavior recognition dataset, and conducted benchmark evaluations on it;
OpenMMReasoner: Pushing the Frontiers in Multimodal Reasoning with an Open and General Recipe
Kaichen Zhang (MiroMind AI), Lidong Bing (MiroMind AI)
OptimizationKnowledge DistillationData-Centric LearningTransformerSupervised Fine-TuningReinforcement LearningVision Language ModelMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Propose a unified two-stage training paradigm (SFT+RL) to construct the multimodal reasoning model OpenMMReasoner, and open-source the complete data, code, and model.
OpenT2M: No-frill Motion Generation with Open-source, Large-scale, High-quality Data
Bin Cao (Institute of Automation Chinese Academy of Sciences), Zongqing Lu (Peking University)
GenerationData SynthesisTransformerLarge Language ModelReinforcement LearningVision-Language-Action ModelTextMultimodalitySequential
🎯 What it does: Propose the OpenT2M large-scale high-quality motion dataset and train the MonoFrill text-driven motion generation model based on this dataset.
OpenVision 2: A Family of Generative Pretrained Visual Encoders for Multimodal Learning
Yanqing Liu (University of California Santa Cruz), Cihang Xie (University of California Santa Cruz)
Computational EfficiencyRepresentation LearningTransformerSupervised Fine-TuningVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: Simplify the original OpenVision model into a generative pre-training framework that uses only the image encoder and text decoder, removing the text encoder and contrastive loss, and training solely with the title generation loss.
OpenVO: Open-World Visual Odometry with Temporal Dynamics Awareness
Phuc Nguyen (University of Maryland), Ming C. Lin (University of Maryland)
Pose EstimationDepth EstimationAutonomous DrivingTransformerOptical FlowVideo
🎯 What it does: Proposed the OpenVO framework, which can achieve open-source visual odometry in uncalibrated, varying frame rate dashcam videos, outputting vehicle poses with real-scale accuracy.
OpenVoxel: Training-Free Grouping and Captioning Voxels for Open-Vocabulary 3D Scene Understanding
Sheng-Yu Huang (NVIDIA), Cheng Sun (NVIDIA)
RetrievalTransformerLarge Language ModelVision Language ModelMultimodalityRetrieval-Augmented Generation
🎯 What it does: Based on a pre-trained sparse voxel reconstruction model, an untrained grouping method is utilized to cluster voxels into object-level collections, and normalized text descriptions are generated through VLM and MLLM, ultimately achieving a 3D scene description that can be directly queried.
OPRO: Orthogonal Panel-Relative Operators for Panel-Aware In-Context Image Generation
Sanghyeon Lee (Korea Advanced Institute of Science and Technology), Jaegul Choo (Korea Advanced Institute of Science and Technology)
GenerationTransformerDiffusion modelImageBenchmark
🎯 What it does: Propose a panel relative orthogonal operator OPRO to regulate the attention mechanism of pre-trained diffusion Transformers, enabling them to maintain intra-panel consistency while transferring information across panels in multi-panel context generation.
Opti-NeuS: Neural Reconstruction for Dual-Layered Transparent and Opaque Objects
Yi Yang (Zhejiang University), Xinguo Liu (Zhejiang University)
Neural Radiance FieldImage
🎯 What it does: A three-dimensional reconstruction method is proposed for double-layer transparent and opaque objects, which can simultaneously recover the outer transparent surface and inner opaque structure without requiring special equipment or controlled environments.
Optical Diffraction-based Convolution for Semiconductor Lithography
Young-Han Son (Korea University), Tae-Eui Kam (Korea University)
OptimizationConvolutional Neural NetworkImagePhysics Related
🎯 What it does: Propose the OptiCo framework, which integrates the principles of optical diffraction (optical phase kernel) into convolutional neural networks for semiconductor lithography simulation and mask optimization.
Optical Flow Matching: Reframing Optical Flow as Continuous Transport Dynamics
Ao Luo (Southwest Jiaotong University), Xiao Wu (Southwest Jiaotong University)
Convolutional Neural NetworkOptical FlowVideoOrdinary Differential Equation
🎯 What it does: Redefine optical flow estimation as continuous transport dynamics, learning time-related velocity fields and solving for pixel trajectories via Euler ODE to obtain optical flow.
OptiMVMap: Offline Vectorized Map Construction via Optimal Multi-vehicle Perspectives
Zedong Dan (Sun Yat Sen University), Guanbin Li (Sun Yat Sen University)
Autonomous DrivingOptimizationConvolutional Neural NetworkTransformerImagePoint Cloud
🎯 What it does: This paper proposes an offline vectorized map construction framework called OptiMVMap based on the principle of 'select first, then fuse,' which utilizes multi-vehicle perspectives to compensate for the limitations of single-vehicle viewpoints, thereby generating more complete and topologically consistent high-definition maps;
OralGPT-Omni: A Versatile Dental Multimodal Large Language Model
Jing Hao (University of Hong Kong), Kuo Feng Hung (University of Hong Kong)
TransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningVision Language ModelImageVideoTextMultimodalityBiomedical DataBenchmarkChain-of-Thought
🎯 What it does: Developed a specialized language model for dental multimodal tasks, OralGPT‑Omni, and introduced a unified evaluation benchmark for the dental field, MMOral‑Uni.
OralGPT-Plus: Learning to Use Visual Tools via Reinforcement Learning for Panoramic X-ray Analysis
Yuxuan Fan (Hong Kong University of Science and Technology), Hao Tang (Peking University)
Anomaly DetectionTransformerLarge Language ModelReinforcement LearningAgentic AIVision Language ModelImageTextBiomedical DataBenchmark
🎯 What it does: Develop OralGPT-Plus, an agentic visual-language model with iterative reasoning and symmetric comparison capabilities, for diagnostic analysis of panoramic dental X-rays.
OraPO: Oracle-educated Reinforcement Learning for Data-efficient and Factual Radiology Report Generation
Zhuoxiao Chen (University of Queensland), Yuan-Fang Li (Oracle Health & AI)
GenerationReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningVision Language ModelMultimodalityBiomedical Data
🎯 What it does: Proposes OraPO – an Oracle-informed Group Relative Policy Optimization method for generating radiology reports for chest X-rays under data and computational constraints.
ORBIT: Benchmarking SfM in the Wild with 360deg Video
Sara Sabour (Google Research, DeepMind), Noah Snavely (Google Research, DeepMind)
Pose EstimationDepth EstimationSimultaneous Localization and MappingVideoBenchmark
🎯 What it does: Built a real-world scene SfM benchmark called ORBIT based on 360° videos for evaluating camera pose estimation.
ORCA: Orchestrated Reasoning with Collaborative Agents for Document Visual Question Answering
Aymen Lassoued (Digital Research Center of Sfax), Yousri Kessentini (Computer Vision Center)
TransformerLarge Language ModelAgentic AIMixture of ExpertsVision Language ModelImageTextMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Propose the ORCA multi-agent framework, achieving single-page document visual question answering through five stages: reasoning decomposition, specialized agent collaboration, debate self-inspection, format validation, and other processes;
ORD: Object-Relation Decoupling for Generalized 3D Visual Grounding
Ronggang Huang (South China University of Technology), Xuemiao Xu (South China University of Technology)
Object DetectionTransformerVision Language ModelContrastive LearningPoint CloudMesh
🎯 What it does: This paper proposes an object-relation decoupled 3D vision localization framework called ORD, aiming to guide the alignment between language and 3D perception by leveraging the relative geometric relationships between prior anchors and the target object, thereby achieving precise localization of the target object.
Order Matters: 3D Shape Generation from Sequential VR Sketches
Yizi Chen (ETH Zurich), Loic Landrieu (LIGM)
GenerationData SynthesisTransformerDiffusion modelAuto EncoderImageMeshSequential
🎯 What it does: Propose a 3D shape generation method based on sequential VR sketches and release a new multi-class VR sketch dataset.
ORIC: Benchmarking Object Recognition under Contextual Incongruity in Large Vision-Language Models
Zhaoyang Li (University of California San Diego), Hao Su (University of California San Diego)
RecognitionLarge Language ModelSupervised Fine-TuningReinforcement LearningVision Language ModelContrastive LearningImageTextMultimodalityBenchmark
🎯 What it does: Constructed an object recognition benchmark (ORIC-Bench) tailored for context-inconsistent scenarios, and automatically generated positive and negative samples through a dual sampling strategy combining LLM and CLIP; evaluated 20 large-scale vision-language models and two open-source detectors on this benchmark, further enhancing model robustness by applying Visual-RFT (Visual Reinforcement Fine-Tuning) on 600 ORIC-style training samples.
OrienPose: Orientation-Guided Novel View Synthesis for Single-Image Unseen Object Pose Estimation
Yating Liu (Northwestern Polytechnical University), Yanning Zhang (Northwestern Polytechnical University)
GenerationPose EstimationConvolutional Neural NetworkGenerative Adversarial NetworkImageMesh
🎯 What it does: Proposes a CAD-free object pose estimation framework, OrienPose, under single-image scenarios, achieving pose prediction via direction-guided novel view synthesis.
ORION: ORthonormal Text Encoding for Universal VLM AdaptatION
Omprakash Chakraborty (École de technologie supérieure), Ismail Ben Ayed (École de technologie supérieure)
ClassificationComputational EfficiencyRepresentation LearningSupervised Fine-TuningVision Language ModelImage
🎯 What it does: Low-rank fine-tuning of the text encoder in VLM using only category names, combined with an orthogonal regularization term to enhance the discriminative power of text prototypes.
OrionEdit: Bridging Reference and Source Images for Generalized Cross-Image Editing
Zeyu Jiang (City University of Hong Kong), Yuyang Liu (City University of Hong Kong)
Image HarmonizationGenerationTransformerDiffusion modelImageBenchmark
🎯 What it does: Propose a unified cross-image editing framework called OrionEdit, which can achieve attribute transfer, style alignment, and multi-subject fusion under multiple reference images.
ORSATR-X: A Foundation Model based on Differential-and-Excitation Networks for Optical Remote Sensing Object Recognition
Canyu Mo (National University of Defense Technology), Li Liu (National University of Defense Technology)
ClassificationObject DetectionSegmentationTransformerImage
🎯 What it does: This paper studies how to migrate DINOv3 to remote sensing tasks, proposing the ORSATR-X framework, and incorporating the Weber Local Adapter (WLA) and Multi-scale Aggregation Module (MSAM) into the framework to enhance robustness to low-contrast targets and scale variations.
OrthoFuse: Training-free Riemannian Fusion of Orthogonal Style-Concept Adapters for Diffusion Models
Ali Aliev (HSE University), Maxim Rakhuba (HSE University)
GenerationDiffusion modelImage
🎯 What it does: Proposed OrthoFuse, a training-free orthogonal adapter fusion method that merges orthogonal adapters from different tasks to generate images that simultaneously preserve concepts and styles.
Orthogonal Spatial-Aware Multi-View Anchor Graph Clustering for Incomplete Remote Sensing Data
Yongshan Zhang (Sun Yat-sen University), Zhihua Cai (China University of Geosciences)
Graph Neural NetworkImage
🎯 What it does: Propose a clustering framework OSMAGC for incomplete remote sensing multi-view data, achieving clustering through spatially aware anchor graph learning and orthogonal spatial regularization.
ORV: 4D Occupancy-centric Robot Video Generation
Xiuyu Yang (Beijing Academy of Artificial Intelligence), Hao Zhao (Beijing Academy of Artificial Intelligence)
GenerationData SynthesisRobotic IntelligenceTransformerDiffusion modelGaussian SplattingVideo
🎯 What it does: Propose a 4D occupancy-based robot video generation framework, ORV, capable of generating high-fidelity, controllable single-view, multi-view, and simulation-to-real robot operation videos.
OS-Fed: One Snapshot Is All You Need
Xuwei Qian (Southeast University), Fang Dong (Southeast University)
Federated LearningComputational EfficiencyImageText
🎯 What it does: Proposes an OS-FED framework that completes federated learning by transmitting only a single compressed snapshot (image + learnable tags).
OS-Oracle: A Comprehensive Framework for Cross-Platform GUI Critic Models
Zhenyu Wu (Shanghai Jiaotong University), Zichen Ding (Shanghai AI Laboratory)
Data SynthesisSupervised Fine-TuningReinforcement LearningVision Language ModelMultimodalityBenchmark
🎯 What it does: Proposed a cross-platform GUI evaluation framework, OS-Oracle, for judging each operation step of computer use agents in mobile, Web, and desktop environments;
OSA: Echocardiography Video Segmentation via Orthogonalized State Update and Anatomical Prior-aware Feature Enhancement
Rui Wang (Shenzhen University), Jing Qin (Shenzhen University)
SegmentationConvolutional Neural NetworkVideoBiomedical DataUltrasound
🎯 What it does: This paper proposes a framework called OSA for left ventricle segmentation in cardiac ultrasound videos.
OSMO: Open-vocabulary Self-eMOtion Tracking
Mohamed Abdelfattah (École Polytechnique Fédérale de Lausanne), Edoardo Remelli (École Polytechnique Fédérale de Lausanne)
RecognitionExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelVideoTextMultimodalityRetrieval-Augmented GenerationChain-of-ThoughtAudio
🎯 What it does: Constructed the first large-scale self-emotion tracking dataset OSMO and proposed a multimodal large language model OSIRIS capable of real-time emotion tracking for users.
OSPO: Object-Centric Self-Improving Preference Optimization for Text-to-Image Generation
Yoonjin Oh (Korea University), Sungwoong Kim (Korea University)
GenerationOptimizationTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningPrompt EngineeringDiffusion modelMultimodality
🎯 What it does: Propose OSPO, a self-improving preference optimization framework that enhances fine-grained alignment in text-to-image generation by generating its own object-level preference data and attention-based object masks, thereby reducing object misreporting.
Otil: Accelerating Diffusion Model Inference via Communication-Efficient Multi-GPU Parallelism
Xin Li (Zhejiang University), Wenzhi Chen (Zhejiang University)
GenerationComputational EfficiencyDiffusion modelImage
🎯 What it does: Propose a multi-GPU parallel inference framework named Otil, which accelerates diffusion model inference by synchronizing only the most informative latent variable blocks.
Out of Sight, Out of Track: Adversarial Attacks on Propagation-based Multi-Object Trackers via Query State Manipulation
Halima Bouzidi (University of California Irvine), Mohammad Al Faruque (University of California Irvine)
Object TrackingAdversarial AttackVideoBenchmark
🎯 What it does: Proposes FADE, an adversarial attack framework targeting Tracking-by-Propagation (TBP) multi-object trackers, achieving trajectory failure through query budget consumption and temporal memory corruption.
Outlier-Robust Diffusion Solvers for Inverse Problems
Yang Zheng (University of Electronic Science and Technology of China), Zhaoqiang Liu (University of Electronic Science and Technology of China)
RestorationOptimizationDiffusion modelImage
🎯 What it does: Propose two robust inverse problem solving methods based on diffusion models—Robust-GD and Robust-CG, designed for reconstructing measurement data with outliers.
Ov3R: Open-Vocabulary Semantic 3D Reconstruction from RGB Videos
Ziren Gong, Matteo Poggi
SegmentationGenerationPose EstimationTransformerVision Language ModelSimultaneous Localization and MappingVideoText
🎯 What it does: Built an open-vocabulary semantic 3D reconstruction framework called Ov3R based on CLIP, which can generate globally consistent 3D reconstructions and open-vocabulary semantic segmentations in real-time from RGB videos.
OVI-MAP: Open-Vocabulary Instance-Semantic Mapping
Zilong Deng (ETH Zurich), Daniel Barath (ETH Zurich)
SegmentationDepth EstimationComputational EfficiencyTransformerVision Language ModelImageMultimodalityPoint Cloud
🎯 What it does: Proposes OVI-MAP, an online open-vocabulary 3D semantic instance mapping framework, which first generates stable 3D instance maps through category-free instance reconstruction, and then assigns zero-shot semantics to each instance using a vision-language model;
OVOD-Agent: A Markov-Bandit Framework for Proactive Visual Reasoning and Self-Evolving Detection
Chujie Wang (Wuhan University), Chu He (Wuhan University)
Object DetectionReinforcement LearningImageChain-of-Thought
🎯 What it does: Propose OVOD-Agent, a lightweight Markov-Bandit framework that transforms open-vocabulary object detection from passive matching to active visual reasoning and self-evolving detection;
OVSegDT: Segmenting Transformer for Open-Vocabulary Object Goal Navigation
Tatiana Zemskova (Cognitive AI Systems Lab), Aleksandr Panov (Cognitive AI Systems Lab)
SegmentationAutonomous DrivingTransformerReinforcement LearningVision-Language-Action ModelContrastive LearningImage
🎯 What it does: This paper proposes OVSegDT, a lightweight Transformer model that achieves open-vocabulary goal navigation using RGB input and target binary segmentation masks.
P-Flow: Prompting Visual Effects Generation
Rui Zhao (National University of Singapore), Mike Zheng Shou (National University of Singapore)
GenerationData SynthesisPrompt EngineeringFlow-based ModelVideoText
🎯 What it does: Proposes a training-free framework called P-Flow, which customizes dynamic visual effects in videos by optimizing text prompts during inference.
P2GS: Physical Prior-guided Gaussian Splatting for Photometrically Consistent Urban Reconstruction
Kota Shimomura (Chubu University), Hironobu Fujiyoshi (Chubu University)
Autonomous DrivingOptimizationGaussian SplattingImagePhysics Related
🎯 What it does: This paper proposes a physics-constrained high dynamic range Gaussian scattering method called P2GS, achieving exposure-invariant radiance field reconstruction for multi-exposure urban scenes.
PA-Attack: Guiding Gray-Box Attacks on LVLM Vision Encoders with Prototypes and Attention
Hefei Mei (City University of Hong Kong), Minjing Dong (City University of Hong Kong)
Adversarial AttackTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Targeting Large Vision-Language Models with shared visual encoders, the authors propose a gray-box attack method called PA-Attack, which can weaken the model's performance across multiple tasks by generating adversarial perturbations within the visual encoder without accessing the large language model's parameters.
PackUV: Packed Gaussian UV Maps for 4D Volumetric Video
Aashish Rai (Brown University), Srinath Sridhar (UMass Amherst)
CompressionGaussian SplattingOptical FlowVideo
🎯 What it does: Propose PackUV to compress 3D Gaussian attributes into multi-layer UV atlases, achieving efficient storage and streaming of 4D volume videos. Additionally, present the PackUV-GS method for directly fitting multi-view videos in the UV space, and release the large-scale PackUV-2B dataset.
PaCo-RL: Advancing Reinforcement Learning for Consistent Image Generation with Pairwise Reward Modeling
Bowen Ping (Xi'an Jiaotong University), Hangwei Qian (Xi'an Jiaotong University)
GenerationReinforcement Learning from Human FeedbackTransformerReinforcement LearningVision Language ModelFlow-based ModelImageMultimodalityChain-of-Thought
🎯 What it does: Proposes the PaCo-RL framework for achieving consistent image generation through reinforcement learning, incorporating a specialized pairwise consistency reward model PaCo-Reward and an efficient RL algorithm PaCo-GRPO.
PACT: Phase-Like Transition Constraints in Adapter-Based Continual Learning of Vision-Language Models
Xuan Wang (Tsinghua University), Jungong Han (Tsinghua University)
TransformerVision Language ModelImage
🎯 What it does: Propose the PACT framework, which in visual-language model continual learning applies phase-class transfer constraints to task-specific adapters after convergence, creating two smooth states of 'freezing' and 'melting' to enhance cross-task transferability and stability.
PAD-Hand: Physics-Aware Diffusion for Hand Motion Recovery
Elkhan Ismayilzada (Michigan State University), Zijun Cui (Michigan State University)
Pose EstimationConvolutional Neural NetworkTransformerDiffusion modelImageMeshPhysics Related
🎯 What it does: Propose a physics-aware diffusion framework to correct the initial hand motion estimation from images for physical consistency and provide physical consistency variance for each joint.
PAF: Perturbation-Aware Filtering for Open-Set Semi-Supervised Learning
Yinan Han (Nanjing University of Science and Technology), Qing-Yuan Jiang (Nanjing University)
ClassificationAnomaly DetectionRepresentation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: Propose the Perturbation-Aware Filtering (PAF) method, which dynamically filters OOD samples in open-set semi-supervised learning by measuring representation instability under semantic-preserving perturbations, and design a two-stage training framework: the first stage enhances representation robustness using labeled data and self-supervised rotation tasks, while the second stage employs PAF to filter OOD samples and performs instability-weighted pseudo-label training.
PAI-Bench: A Comprehensive Benchmark For Physical AI
Fengzhe Zhou (Georgia Tech), Humphrey Shi (Georgia Tech)
Vision Language ModelVideoTextMultimodalityBenchmarkPhysics Related
🎯 What it does: Propose PAI-Bench, a unified and comprehensive benchmark for evaluating the perception and prediction capabilities of video generation, conditional video generation, and video understanding tasks in physical AI scenarios.
PALM: Progress-Aware Policy Learning via Affordance Reasoning for Long-Horizon Robotic Manipulation
Yuanzhe Liu (University of Pennsylvania), Ismini Lourentzou (University of Illinois Urbana-Champaign)
Robotic IntelligenceTransformerSupervised Fine-TuningVision-Language-Action ModelDiffusion modelAuto EncoderMultimodality
🎯 What it does: Proposes PALM, a vision-language-action (VLA) framework that integrates structured prospective affordance reasoning with progress-aware strategies for long-horizon robotic manipulation tasks.
PAM: A Pose-Appearance-Motion Engine for Sim-to-Real HOI Video Generation
Mingju Gao (Peking University), Hao Zhao (Tsinghua University)
GenerationDiffusion modelVideoMultimodality
🎯 What it does: Propose a three-stage Pose-Appearance-Motion (PAM) engine that generates high-quality, realistic hand-object interaction videos using only initial and target gestures, along with object geometry information.
PAMotion: Physics-Aware Motion Generation for Full-Body Interaction with Multiple Objects
Yan Di (Harbin Institute of Technology), Xiangyang Ji (Tsinghua University)
GenerationVision-Language-Action ModelDiffusion modelVideoTextSequentialPhysics Related
🎯 What it does: Proposed a physics-aware diffusion framework called PAMotion for generating realistic full-body motion sequences with multi-object interactions based on text instructions.
PanDA: Unsupervised Domain Adaptation for Multimodal 3D Panoptic Segmentation in Autonomous Driving
Yining Pan (Singapore University Of Technology And Design), Na Zhao (Singapore University Of Technology And Design)
Domain AdaptationAutonomous DrivingMultimodalityPoint Cloud
🎯 What it does: Proposes the PanDA framework, which specifically addresses the problem of unsupervised domain adaptation for multi-modal 3D panoptic segmentation in the context of autonomous driving.
PaNDaS: Learnable Shape Interpolation Modeling with Localized Control
Thomas Besnier (University of Lille), Mohamed Daoudi (Ecole Polytechnique)
GenerationGraph Neural NetworkDiffusion modelMesh
🎯 What it does: A deep learning framework called PaNDaS, which combines global and local features, is proposed for achieving local non-rigid deformation and interpolation on triangle meshes.
Pano360: Perspective to Panoramic Vision with Geometric Consistency
Zhengdong Zhu (South China University of Technology), Zhiheng Zhou (South China University of Technology)
GenerationTransformerContrastive LearningImage
🎯 What it does: Propose a Transformer-based Pano360 framework that performs global alignment of multi-image stitching using 3D perspective geometric consistency and automatically generates seamless panoramas.
Pano3DComposer: Feed-Forward Compositional 3D Scene Generation from Single Panoramic Image
Zidian Qiu (Sun Yat-sen University), Ancong Wu (Sun Yat-sen University)
GenerationData SynthesisDepth EstimationGaussian SplattingImageMesh
🎯 What it does: Designed a fast panoramic 3D scene synthesis framework called Pano3DComposer based on a single panoramic image, which utilizes a pluggable object-world transformation predictor to position offline-generated 3D objects in the panoramic coordinate system, and achieves high-precision layout through a coarse-to-fine alignment mechanism.
PanoEnv: Exploring 3D Spatial Intelligence in Panoramic Environments with Reinforcement Learning
Zekai Lin (University of Glasgow), Xu Zheng (Hong Kong University of Science and Technology)
Data SynthesisReinforcement LearningVision Language ModelImageMultimodalityBenchmark
🎯 What it does: Propose PanoEnv, a large-scale panoramic visual question answering (VQA) benchmark for evaluating and enhancing the 3D spatial reasoning capabilities of Vision-Language Models (VLMs).
PanoVGGT: Feed-Forward 3D Reconstruction from Panoramic Imagery
Yijing Guo (ShanghaiTech University), Yujiao Shi (ShanghaiTech University)
GenerationPose EstimationDepth EstimationTransformerImagePoint Cloud
🎯 What it does: Proposed a Transformer-based end-to-end framework called PanoVGGT, which can infer camera pose, dense depth, and globally consistent 3D point clouds in a single forward pass from unordered panoramic images.
Pantheon360: Taming Digital Twin Generation via 3D-Aware 360deg Video Diffusion
Ting-Hsuan Chen (University Of Southern California), Liu Ren (Bosch Research)
GenerationData SynthesisVision Language ModelDiffusion modelVideoPoint Cloud
🎯 What it does: Propose the Pantheon360 framework, which can generate controllable, temporally consistent 360° videos from sparse or single 360° input images, and supports user-defined camera trajectories.
Paparazzo: Active Mapping of Moving 3D Objects
Davide Allegro (University of Padova), Vincent Lepetit (École Nationale des Ponts et Chaussées)
Object TrackingPose EstimationGaussian SplattingSimultaneous Localization and MappingPoint CloudMesh
🎯 What it does: Address the active 3D mapping task in dynamic scenes by designing a learning-agnostic framework called Paparazzo, which leverages EKF to predict moving object trajectories and selects the most informative viewpoints, achieving active observation and reconstruction of moving rigid bodies.
Paper2Figure: A Multi-Agent Collaborative System for Figure Generation Towards Academic Research Paper
Siwei Han (UNC-Chapel Hill), Huaxiu Yao (UNC-Chapel Hill)
GenerationTransformerLarge Language ModelAgentic AIPrompt EngineeringImageText
🎯 What it does: Propose a dual multi-agent system (Paper2Figure), which automatically generates, refines, and edits academic paper figures from text descriptions through an intermediate language FigScript and an interactive Web platform.
PaQ-DETR: Learning Pattern and Quality-Aware Dynamic Queries for Object Detection
Zhengjian Kang (New York University), Ye Zhang (University of Pittsburgh)
Object DetectionTransformerImage
🎯 What it does: PaQ-DETR is proposed within the Transformer-based DETR framework, combining dynamic query generation with shared patterns and quality-aware one-to-many assignment, significantly enhancing query utilization and supervision balance.
Parallax to Align Them All: An OmniParallax Attention Mechanism for Distributed Multi-View Image Compression
Haotian Zhang (Peking University), Jiaqi Zhang (Peking University)
CompressionImage
🎯 What it does: Propose an end-to-end distributed multi-view image compression framework called ParaHydra, which utilizes the OmniParallax Attention Mechanism (OPAM) to adaptively fuse multi-view information during decoding, achieving higher compression efficiency;
Parallel Jacobi Decoding for Fast Autoregressive Image Generation
Boya Liao (Westlake University), Huan Wang (Westlake University)
GenerationComputational EfficiencyTransformerImage
🎯 What it does: Proposed a Parallel Jacobi Decoding (PJD) method, significantly improving the inference speed of image generation based on autoregressive (AR) Transformers without requiring additional training.
Parallel Rigidity Matters for Bundle Adjustment
Lalit Manam (Mitsubishi Electric Research Labs), Venu Madhav Govindu (Indian Institute of Science)
Pose EstimationSimultaneous Localization and MappingImage
🎯 What it does: This paper investigates the unique solvability issue in Bundle Adjustment, utilizing parallel rigidity theory to perform topological analysis on bipartite graphs (camera-3D points), and proposes an efficient algorithm (GPRBA) based on view graphs (camera-camera relationships) to extract parallel rigid subgraphs, thereby eliminating misaligned cameras and 3D points caused by independent scaling between subgraphs.
Parallelised Differentiable Straightest Geodesics for 3D Meshes
Hippolyte Verninas (Imperial College London), Simone Foti (Imperial College London)
OptimizationComputational EfficiencyMesh
🎯 What it does: This paper proposes a differentiable, GPU-parallelized implementation of three-dimensional mesh straightest geodesics, achieving exponential mapping, geodesic tracking, and parallel transport;
ParallelVLM: Lossless Video-LLM Acceleration with Visual Alignment Aware Parallel Speculative Decoding
Quan Kong (Zhejiang University), Cong Wang (Zhejiang University)
Computational EfficiencyVision Language ModelVideo
🎯 What it does: This paper proposes a parallel prediction-validation inference framework called ParallelVLM to accelerate the autoregressive decoding of video large language models.
Parameter-Efficient Adaptation for MLLMs via Implicit Modality Decomposition
Mingfang Zhang (Beihang University), Jiaxin Chen (Beihang University)
Computational EfficiencyLarge Language ModelSupervised Fine-TuningMultimodality
🎯 What it does: Propose an implicit modal decomposition (IMoD) method for parameter-efficient fine-tuning of multi-modal large language models, enabling the model to balance learning between text and non-text modalities while maintaining low parameter count and mergeability.
Parameter-efficient Continual Learning for Enhancing Plasticity without Forgetting under Limited Model Capacity
Yitian Chen (Central South University), Xinning Chen (Central South University)
Computational EfficiencyConvolutional Neural NetworkReinforcement LearningImage
🎯 What it does: Proposed the GRAPA framework, achieving parameter-efficient continual learning under resource constraints, balancing model stability and plasticity.
Parameter-Efficient Semantic Augmentation for Enhancing Open-Vocabulary Object Detection
Weihao Cao (Beijing Jiaotong University), Liping Jing (Beijing Jiaotong University)
Object DetectionTransformerVision Language ModelAuto EncoderContrastive LearningMultimodality
🎯 What it does: Proposed a parameter-efficient semantic enhancement framework called HSA-DINO to improve the performance of open-vocabulary object detection in vertical domains while maintaining the original open-vocabulary capability.
Parameterized Prompt for Incremental Object Detection
Zijia An (State Key Laboratory of AI Safety, Institute of Computing Technology, Chinese Academy of Sciences), Yongjun Xu (State Key Laboratory of AI Safety, Institute of Computing Technology, Chinese Academy of Sciences)
Object DetectionTransformerPrompt EngineeringImage
🎯 What it does: Proposes a parameterized prompt and prompt fusion framework to address the prompt pool confusion problem in incremental object detection.
ParaUni: Enhance Generation in Unified Multimodal Model with Reinforcement-driven Hierarchical Parallel Information Interaction
Jiangtong Tan (University of Science and Technology of China), Feng Zhao (University of Science and Technology of China)
GenerationReinforcement LearningVision Language ModelDiffusion modelMultimodality
🎯 What it does: Propose the ParaUni framework, which parallelly extracts visual information from all layers of a Vision-Language Model (VLM), aggregates them through a Layer Integration Module (LIM) as conditional inputs to a diffusion model, and introduces a Layer-wise Dynamic Adjustment Mechanism (LDAM) during the reinforcement learning phase to apply noise perturbations to specific layers based on different rewards, thereby improving image generation quality and multi-reward optimization.
ParkGaussian: Surround-view 3D Gaussian Splatting for Autonomous Parking
Xiaobao Wei (Xiaomi EV), Hangjun Ye (Xiaomi EV)
Autonomous DrivingGaussian SplattingSimultaneous Localization and MappingImageBenchmark
🎯 What it does: Proposed the ParkGaussian framework for 3D Gaussian smoothing reconstruction in underground parking lots, and constructed the first dedicated benchmark dataset for parking lots called ParkRecon3D.
Parse, Search, and Confirmation: Training-Free Aerial Vision-and-Dialog Navigation with Chain-of-Thought Reasoning and Structured Spatial Memory
Yu Qi (Hefei University of Technology), Meng Wang (Hefei University of Technology)
Autonomous DrivingTransformerLarge Language ModelVision-Language-Action ModelImageChain-of-Thought
🎯 What it does: Proposes a training-free aerial vision-dialogue navigation framework PSC-AVDN.
PARSE: Part-Aware Relational Spatial Modeling
Yinuo Bai (ShanghaiTech University), Jingyi Yu (ShanghaiTech University)
GenerationData SynthesisGraph Neural NetworkTransformerSupervised Fine-TuningVision Language ModelDiffusion modelMeshGraphRetrieval-Augmented Generation
🎯 What it does: Proposed the PARSE framework, which utilizes part-based assembly graphs (PAG) and part-aware spatial configuration solvers to convert object part-level geometric relationships into feasible scenarios, and constructed a 10,000 indoor 3D scene dataset named PARSE-10K with part-level contact graphs.
Part$^{2}$GS: Part-aware Modeling of Articulated Objects using 3D Gaussian Splatting
Tianjiao Yu (University of Illinois Urbana-Champaign), Ismini Lourentzou (University of Illinois Urbana-Champaign)
GenerationOptimizationGaussian SplattingImagePoint Cloud
🎯 What it does: Propose Part GS 2, a self-supervised method based on 3D Gaussian splatting, for reconstructing multi-part objects with physically consistent motion from multi-view images captured under two different poses.
PartDiffuser: Part-wise 3D Mesh Generation via Discrete Diffusion
Yichen Yang (Beihang University), Baochang Zhang (Beihang University)
GenerationTransformerDiffusion modelPoint CloudMesh
🎯 What it does: Propose PartDiffuser, a semi-autoregressive diffusion framework for generating 3D meshes from point clouds.
Partial Weakly-Supervised Oriented Object Detection
Mingxin Liu (Shanghai Jiao Tong University), Xue Yang (Shanghai Jiao Tong University)
Object DetectionConvolutional Neural NetworkImage
🎯 What it does: Proposes a partially weakly supervised object detection framework called PWOOD, which achieves efficient detection by utilizing only horizontal boxes or single-point weak annotations along with massive unlabeled data.
ParticleGS: Learning Neural Gaussian Particle Dynamics from Videos for Prior-free Physical Motion Extrapolation
Jinsheng Quan (Zhejiang University), Yawei Luo (Zhejiang University)
GenerationGaussian SplattingVideoPhysics RelatedOrdinary Differential Equation
🎯 What it does: Propose ParticleGS, a physics-based particle dynamics framework based on 3D Gaussian Splatting, which utilizes neural ODEs to learn the continuous physical evolution of scenes, achieving physically consistent rendering of future frames.
Particulate: Feed-Forward 3D Object Articulation
Ruining Li (University of Oxford), Andrea Vedaldi (University of Oxford)
GenerationTransformerPoint CloudMeshGraph
🎯 What it does: This paper proposes a one-time forward network called PARTICULATE, which directly predicts the complete kinematic structure of objects using a single static 3D grid, including part segmentation, motion trees, and motion constraints, thereby enabling the rapid generation of fully movable 3D models that can be physically simulated.