CVPR 2026 Papers — Page 7
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 4071 papers
Contrastive Cross-Bag Augmentation for Multiple Instance Learning-based Whole Slide Image Classification
Bo Zhang (Beijing University of Posts and Telecommunications), Wendong Wang (Beijing University of Posts and Telecommunications)
ClassificationTransformerContrastive LearningBiomedical Data
🎯 What it does: Propose a Contrastive Cross-Bag Augmentation (C Aug) to generate diverse pseudo bags and enhance the robustness of whole slide image (WSI) classification through bag-level and group-level contrastive learning.
Controllable Federated Prompt Learning at Test Time
Rui Zhu (National University of Defense Technology), Zhihe Lu (Hamad Bin Khalifa University)
Domain AdaptationFederated LearningPrompt EngineeringVision Language ModelImage
🎯 What it does: This paper proposes a triple-prompt (global, local, CLIP) dynamic balance framework called COTE for test-time adaptation in federated prompt learning on unlabeled target domain data.
Conversational Image Segmentation: Grounding Abstract Concepts with Scalable Supervision
Aadarsh Sahoo (California Institute of Technology), Georgia Gkioxari (California Institute of Technology)
SegmentationTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageTextBenchmark
🎯 What it does: Propose the Conversational Image Segmentation (CIS) task, the CONVERSEG benchmark, and the CONVERSEG-NET model based on VLM+SAM
Convexity-Aware Noise Calibration: A Self-Supervised Framework for Noise-Level-Unknown Image Denoising
Zhan Wang (China University of Petroleum (East China)), Yu Meng (China University of Petroleum (East China))
RestorationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a two-stage self-supervised image denoising framework that first accurately estimates the noise level from noisy images alone, then generates noise-clean pairs using this estimate for supervised training, achieving high-quality denoising of images with unknown noise levels.
Convolutional Neural Networks Driven by Content Similarity
Ligeng Zou (Hunan Normal University), Guihu Zhao (Central South University)
ClassificationObject DetectionSegmentationConvolutional Neural NetworkImage
🎯 What it does: Propose a novel CNN architecture named Ego, which maps content similarity to nearby positions by sorting features in the channel dimension, achieving content-driven aggregation similar to self-attention.
CoopDiff: A Diffusion-Guided Approach for Cooperation under Corruptions
Gong Chen (Tianjin University), Pengcheng Lv (Tianjin University)
Autonomous DrivingDiffusion modelPoint Cloud
🎯 What it does: Proposes CoopDiff, a collaboration-aware framework based on diffusion models, which enhances robustness in corrupted environments by denoising multi-vehicle point cloud features.
Coordinate Denoising for Non-Equilibrium Molecular Representation Learning
Qianwei Tang (Nanjing University), Furao Shen (Nanjing University)
Representation LearningGraph Neural NetworkDiffusion modelGraphBenchmark
🎯 What it does: Studied coordinate denoising methods under non-equilibrium molecular structures, proposing a denoising auxiliary task directly applicable to any molecular conformation;
CoordSpeaker: Exploiting Gesture Captioning for Coordinated Caption-Empowered Co-Speech Gesture Generation
Fengyi Fang (Tsinghua University), Wenming Yang
GenerationData SynthesisTransformerVision-Language-Action ModelDiffusion modelAuto EncoderMultimodality
🎯 What it does: Generate full-body dynamic gestures through bimodal collaborative generation using voice and text (gestural captions), supporting synchronized control of speech, emotion, style, and achieving a closed-loop from unlabeled data to complete gesture captions.
COPE: Consistent Occlusion and Prompt Enhancement Network for Occluded Person Re-identification
Siyi Sun (Xiamen University), Zhiming Luo (Xiamen University)
RetrievalTransformerPrompt EngineeringVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: Proposes a network called COPE to address feature interference and information loss in occluded person re-identification, covering three modules: Cross-Identity Consistent Occlusion (CICO), Prompt Background Filling (PBF), and Prompt Similarity Scoring (PSS).
COPO: Causal-Oriented Policy Optimization for Hallucinations of MLLMs
Peizheng Guo (Institute of Software Chinese Academy of Sciences), Gang Hua (Amazon.com Inc)
OptimizationExplainability and InterpretabilityTransformerLarge Language ModelReinforcement LearningMultimodality
🎯 What it does: This paper proposes the COPO framework, which in the post-training of multi-modal large language models, reduces the model's over-reliance on irrelevant contexts and minimizes hallucinations by incorporating token-level rewards based on causal completeness.
Copy-Transform-Paste: Zero-Shot Object-Object Alignment Guided by Vision-Language and Geometric Constraints
Rotem Gatenyo, Ohad Fried
Pose EstimationOptimizationLarge Language ModelPrompt EngineeringVision Language ModelScore-based ModelTextMesh
🎯 What it does: Proposes a text-prompt-based zero-shot 3D object-object alignment method, achieving adaptive optimization of pose and scale for two meshes through differentiable rendering and CLIP vision-language supervision combined with geometric constraints.
COPYLENS: Towards Copyrighted Characters Infringement Detection via Copyright-Aware Prompt Learning
Yaoyu Jin (Northeastern University), Bin Wang (Northeastern University)
ClassificationTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageMultimodality
🎯 What it does: Proposed a closed-loop optimization framework called COPYLENS for automatically optimizing text prompts used in copyright role detection, and constructed a large-scale annotated dataset named COPYCHARS;
CORE: Compact Object-centric REpresentations as a New Paradigm for Token Merging in LVLMs
Jingyu Lei (Zhejiang University), Der-Horng Lee (Zhejiang University)
SegmentationComputational EfficiencyRepresentation LearningConvolutional Neural NetworkTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Proposed a CORE method for visual token compression based on object-centric principles, which uses a segmentation head to generate object masks, aggregates each object in the image into a single token, and inputs them into a language model after sorting by centroid.
CoRiM: Conflict-driven Risk Minimization for Dynamic Multimodal Fusion
Shihao Zou (Huazhong University of Science and Technology), Wei Wei (Huazhong University of Science and Technology)
ClassificationOptimizationRepresentation LearningHyperparameter SearchMultimodality
🎯 What it does: Propose the CoRiM framework, redefining dynamic multimodal fusion as a per-sample risk minimization task, and construct a differentiable modal conflict risk function R(w) to directly quantify conflicts during the fusion process.
CoRoGS: Contextual Gaussian Splatting for Robust Large-Deviation View Synthesis
Xin Ma (Beijing University of Posts and Telecommunications), Sheng Li (Peking University)
GenerationAutonomous DrivingGraph Neural NetworkGaussian SplattingImagePoint Cloud
🎯 What it does: This paper proposes Context-Aware Gaussian Splatting (CoRoGS), which enhances the quality of large disparity view synthesis in urban scenes by constructing a 3D Gaussian graph and using graph neural networks to achieve context-dependent Gaussian updates.
Correspondence-Attention Alignment for Multi-View Diffusion Models
Minkyung Kwon (KAIST AI), Seungryong Kim (KAIST AI)
GenerationData SynthesisComputational EfficiencyDiffusion modelImage
🎯 What it does: Propose the CAMEO method, which enhances view consistency and training efficiency by directly supervising geometric correspondences into the deep attention maps of multi-view diffusion models.
CoSMo3D: Open-World Promptable 3D Semantic Segmentation through LLM-Guided Canonical Spatial Modeling
Li Jin (Shandong University), Xueying Qin (Lightspeed)
SegmentationTransformerLarge Language ModelVision Language ModelContrastive LearningPoint Cloud
🎯 What it does: Propose a CoSMo3D open-world promptable 3D semantic segmentation method that achieves canonical space awareness by learning a latent canonical reference framework, enabling stable part segmentation under any pose.
CoT-Edit: Let CoT Guide Instruction Video Editing
Sen Liang (University of Science and Technology of China), Zhibo Chen (University of Science and Technology of China)
GenerationLarge Language ModelVision Language ModelDiffusion modelVideoTextMultimodalityChain-of-Thought
🎯 What it does: This paper proposes the Plan-Guide-Edit framework, which integrates a Chain-of-Thought (CoT) enhanced multilingual large language model (MLLM) planner, box-guided mask branch, and diffusion editor, achieving text instruction-driven video editing.
COT-FM: Cluster-wise Optimal Transport Flow Matching
Chiensheng Chiang (National Taiwan University), Tsung-Wei Ke (National Taiwan University)
GenerationData SynthesisComputational EfficiencyFlow-based ModelImageTextOrdinary Differential Equation
🎯 What it does: Propose the Cluster-wise Optimal Transport Flow Matching (COT-FM) framework, which accelerates sampling and improves generation quality by clustering the target distribution and finding corresponding source distributions for each cluster to optimize the probability paths of the Flow Matching (FM) model.
Counterfactual VLA: Self-Reflective Vision-Language-Action Model with Adaptive Reasoning
Zhenghao Peng, Marco Pavone
Autonomous DrivingOptimizationExplainability and InterpretabilityComputational EfficiencyLarge Language ModelSupervised Fine-TuningPrompt EngineeringVision-Language-Action ModelVideoTextMultimodalityTime SeriesChain-of-Thought
🎯 What it does: This paper proposes Counterfactual VLA (CF-VLA), a self-reflective framework within visual-language-action models: first generating time-segmented meta-action (high-level intent), then performing counterfactual reasoning on these predictions to identify potential unsafe or suboptimal behaviors and correct the meta-action, followed by generating the final trajectory.
CountGD++: Generalized Prompting for Open-World Counting
Niki Amini-Naieni (University of Oxford), Andrew Zisserman (University of Oxford)
Object DetectionTransformerLarge Language ModelPrompt EngineeringVision Language ModelContrastive LearningImageTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: This study proposes the COUNTGD++ model, which extends the prompting methods for open-world counting, supporting positive and negative text and visual examples, automatic pseudo-example detection, the use of external visual examples, and serving as a visual expert agent for LLMs;
Coupled Diffusion Sampling for Training-Free Multi-View Image Editing
Hadi Alzayer (Stanford University), Jiajun Wu (Stanford University)
Image HarmonizationGenerationDiffusion modelScore-based ModelImageMesh
🎯 What it does: Leverages pre-trained 2D diffusion editing models and multi-view generation models to achieve view-consistent multi-view image editing without training through coupled diffusion sampling;
Coupling Liquid Time-Constant Encoders with Modern Hopfield Memory
Bishal Ranjan Swain (Kumoh National Institute of Technology), Jaepil Ko (Kumoh National Institute of Technology)
ClassificationRecognitionRecurrent Neural NetworkTime SeriesSequential
🎯 What it does: This paper proposes coupling the Liquid Time-Constant Network (LTC) with the Modern Hopfield Network (MHN) to form a model that combines continuous-time encoding with content-addressable memory in parallel.
CoV-Align: Efficient Fine-grained Cross-Modal Alignment with Cohesive Visual Semantics Priority
Hengqi Liu (Beijing University Of Posts And Telecommunications), Xiaojie Wang (Beijing Forestry University)
RetrievalTransformerVision Language ModelContrastive LearningMultimodality
🎯 What it does: Propose the CoV-Align framework, leveraging text-free aggregation-based coarse visual semantic extraction and fine-grained cross-modal alignment to achieve efficient and accurate image-text matching.
Cov2Pose: Leveraging Spatial Covariance for Direct Manifold-aware 6-DoF Object Pose Estimation
Nassim Ali Ousalah (University of Luxembourg), Djamila Aouada (University of Luxembourg)
Pose EstimationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes an end-to-end RGB image single-object 6-DoF pose estimation method called Cov2Pose, which utilizes the spatial covariance of convolutional features as high-order statistical information, and achieves continuous pose regression through SPD matrix learning and differentiable Cholesky decomposition.
Coverage Optimization for Camera View Selection
Timothy Chen (Stanford University), Mac Schwager (Stanford University)
OptimizationNeural Radiance FieldGaussian SplattingImage
🎯 What it does: Propose CONVERGE, a coverage metric based on Fisher information approximation, for active view selection;
CoVFT: Context-aware Visual Fine-tuning for Multimodal Large Language Models
Nan Zhou (Beihang University), Di Huang (Beihang University)
TransformerLarge Language ModelSupervised Fine-TuningMixture of ExpertsVision Language ModelMultimodality
🎯 What it does: Proposed CoVFT, a context-aware visual fine-tuning framework for multi-modal large language models, enabling adaptive fine-tuning of the visual encoder.
CoWTracker: Tracking by Warping instead of Correlation
Zihang Lai (University of Oxford), Andrea Vedaldi (Meta AI)
Object TrackingConvolutional Neural NetworkTransformerOptical FlowVideoBenchmark
🎯 What it does: Propose CoWTracker, a dense point tracker based on warping, which can perform point tracking tasks and migrate to optical flow estimation in an unsupervised manner.
CrackSSM: Reviving SSMs for Crack Segmentation via Dynamic Scanning
Yubin Gu (Xiamen University), Xiaoshuai Sun (Xiamen University)
SegmentationImage
🎯 What it does: Proposed the CrackSSM model, achieving efficient and accurate crack segmentation using Mamba with dynamic path scanning and a wavelet-guided decoder.
CRAFT-LoRA: Content-Style Personalization via Rank-Constrained Adaptation and Training-Free Fusion
Yu Li (Westlake University), Chi Zhang (Westlake University)
GenerationPrompt EngineeringDiffusion modelContrastive LearningImage
🎯 What it does: Proposed the CRAFT-LoRA framework, achieving decoupling and efficient fusion of content and style during LoRA training, and enabling high-quality personalized image generation on SDXL.
CRAFT: Aligning Diffusion Models with Fine-Tuning Is Easier Than You Think
Zening Sun (Hong Kong University Of Science And Technology), Zeke Xie (Hong Kong University Of Science And Technology)
GenerationLarge Language ModelSupervised Fine-TuningReinforcement LearningDiffusion modelImageTextMultimodality
🎯 What it does: Propose a lightweight diffusion model alignment method called CRAFT, which utilizes composite reward filtering to construct a high-quality sample set and performs weighted supervised fine-tuning based on this set.
CraftMesh: High-Fidelity Generative Mesh Manipulation via Poisson Seamless Fusion
James Jincheng Hu, Ligang Liu (University of Science and Technology of China)
GenerationData SynthesisDiffusion modelImageMesh
🎯 What it does: Proposed the CraftMesh framework, combining 2D image editing, 3D mesh generation, and Poisson seamless fusion to achieve controllable high-fidelity mesh editing.
CREval: An Automated Interpretable Evaluation for Creative Image Manipulation under Complex Instructions
Chonghuinan Wang (Harbin Institute of Technology), Hongxun Yao (Harbin Institute of Technology)
Image TranslationExplainability and InterpretabilityLarge Language ModelVision Language ModelImageTextMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Proposed a fully automatic, explainable QA evaluation framework called CREval for assessing the effectiveness of creative image editing under complex instructions, and constructed the CREval-Bench benchmark, which includes over 800 images, 13K evaluation questions, and covers 9 sub-dimensions across 3 categories of creative dimensions.
CREward: A Type-Specific Creativity Reward Model
Jiyeon Han (Simon Fraser University), Haedong Jeong (Sogang University)
GenerationData SynthesisExplainability and InterpretabilityReinforcement Learning from Human FeedbackTransformerReinforcement LearningVision Language ModelDiffusion modelContrastive LearningImageMultimodalityBenchmark
🎯 What it does: This paper proposes a typed reward model for visual creativity (CREward), achieving fine-grained evaluation and guidance of image creativity by decomposing creativity into three interpretable dimensions: geometry, material, and texture.
CRFT: Consistent-Recurrent Feature Flow Transformer for Cross-Modal Image Registration
Xuecong Liu (Northeastern University), Xichao Teng (National University of Defense Technology)
Convolutional Neural NetworkTransformerOptical FlowImageMultimodality
🎯 What it does: Propose a unified coarse-to-fine cascading framework CRFT, which achieves cross-modal image registration through consistent recursive feature flow learning;
CRIT: Graph-Based Automatic Data Synthesis to Enhance Cross-Modal Multi-Hop Reasoning
Junyoung Sung (Korea University), Paul Hongsuck Seo (Korea University)
Data SynthesisGraph Neural NetworkLarge Language ModelImageVideoTextBenchmarkChain-of-Thought
🎯 What it does: Propose the CRIT dataset and evaluation framework, generating cross-modal multi-hop reasoning tasks using a graph-based automated pipeline;
Critical Patch-Aware Sparse Prompting with Decoupled Training for Continual Learning on the Edge
Wonseon Lim (Chung-Ang University), Dae-Won Kim (Chung-Ang University)
Computational EfficiencyTransformerPrompt EngineeringImage
🎯 What it does: A framework named CPS-Prompt is proposed for continual learning on edge devices, which significantly reduces memory and computational overhead during training while maintaining accuracy through critical patch sampling and decoupled prompt/classifier training.
Cross from Left to Right Brain: Adaptive Text Dreamer for Vision-and-Language Navigation
Pingrui Zhang (Fudan University), Bin Zhao (Shanghai AI Laboratory)
Autonomous DrivingGraph Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningVision-Language-Action ModelMultimodality
🎯 What it does: Proposed Adaptive Text Dreamer (ATD), a dual-branch self-guided language model strategy for adaptive imagination through language in vision-language navigation (VLN), integrated with graph-structured navigation policies.
Cross-Architecture Adaptation: Cloud-Edge Continual Test-Time Adaptation with Dynamic Sampling and Heterogeneous Distillation
Zirui Xu (Xidian University), Cheng Deng (Xidian University)
Domain AdaptationKnowledge DistillationConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes a cross-architecture adaptation (CAA) framework to enable collaborative learning between cloud-based Transformers and edge CNNs during continuous testing in the temporal domain.
Cross-Axis Feature Fusion with Joint-Wise Motion Difference Prediction for Text-Based 3D Human Motion Editing
Gyojin Han (KAIST), Junmo Kim (KAIST)
GenerationTransformerVision Language ModelDiffusion modelTextTime Series
🎯 What it does: Propose a text-driven 3D human motion editing framework based on axis-aligned transformers and cross-axis fusion, combined with a soft dynamic time warping (Soft-DTW) auxiliary task to achieve joint-level editing control.
Cross-Domain Demo-to-Code via Neurosymbolic Counterfactual Reasoning
Jooyoung Kim (Sungkyunkwan University), Honguk Woo (Sungkyunkwan University)
Domain AdaptationAI Code AssistantTransformerWorld ModelVideoText
🎯 What it does: Based on a single demonstration video, a symbolic world model is constructed, and executable code strategies applicable to the deployment domain are generated through neuro-symbolic counterfactual reasoning.
Cross-domain Dual-stream Feature Disentanglement for Brain Disorder Prediction with Sparsely Labeled PET
Huabin Wang (Anhui University), Fei Liu (Monash University Malaysia)
ClassificationDomain AdaptationRepresentation LearningGraph Neural NetworkMultimodalityBiomedical DataMagnetic Resonance ImagingPositron Emission TomographyAlzheimer's Disease
🎯 What it does: Propose a dual-stream feature decoupling and alignment (DSDA) framework that leverages rich MRI annotations to guide scarce PET-based brain disease prediction.
Cross-Domain Few-Shot Segmentation via Multi-view Progressive Adaptation
Jiahao Nie (Nanyang Technological University), Shijian Lu (Nanyang Technological University)
SegmentationDomain AdaptationConvolutional Neural NetworkImage
🎯 What it does: Proposes a multi-view progressive adaptation framework (MPA) for cross-domain few-shot segmentation (CD-FSS) tasks, capable of adapting and improving segmentation performance even when only a minimal number of samples are available in the target domain.
Cross-Hand Latent Representation for Vision-Language-Action Models
Guangqi Jiang (UC San Diego), Xueyan Zou (UC San Diego)
Representation LearningRobotic IntelligenceVision-Language-Action ModelAuto EncoderMultimodality
🎯 What it does: Propose XL-VLA, a method that integrates a shared implicit action latent space into a vision-language-action (VLA) framework, enabling adaptive control across multiple grippers.
Cross-Instance Gaussian Splatting Registration via Geometry-Aware Feature-Guided Alignment
Roy Amoyal (Ben-Gurion University of Negev), Chaim Baskin (Ben-Gurion University of Negev)
Pose EstimationGaussian SplattingPoint Cloud
🎯 What it does: This paper proposes a method called Gaussian Splatting Alignment (GSA) for aligning two independent 3D Gaussian Splatting (3DGS) models through similarity transformations (rotation, translation, scaling), supporting registration of objects from the same category but different instances.
Cross-Modal Attention Calibration for LVLM Hallucination Mitigation
Jiaming Li (Sun Yat-sen University), Guanbin Li (Sun Yat-sen University)
Vision Language ModelContrastive LearningMultimodalityBenchmark
🎯 What it does: This paper proposes an untrained cross-modal attention calibration (CMAC) framework aimed at reducing hallucinations in large vision-language models (LVLMs) during generation tasks.
Cross-Modal Emotion Transfer for Emotion Editing in Talking Face Video
Chanhyuk Choi (Ulsan National Institute of Science and Technology), Taehwan Kim (Ulsan National Institute of Science and Technology)
GenerationTransformerContrastive LearningVideoMultimodalityAudio
🎯 What it does: Propose a cross-modal emotion transfer framework C-MET, which edits and generates more rich and delicate facial expressions in speaker videos by learning the mapping between audio and visual emotion semantic vectors;
Cross-modal Fuzzy Alignment Network for Text-Aerial Person Retrieval and A Large-scale Benchmark
Yifei Deng (Anhui University), Jin Tang (Anhui University)
RetrievalTransformerVision Language ModelImageTextMultimodalityBenchmarkChain-of-Thought
🎯 What it does: This paper proposes a cross-modal fuzzy alignment network (CFAN) for the text-aerial person retrieval task and constructs a large-scale AERI-PEDES dataset.
Cross-Modal Guided Visual Synthesis for Data-Efficient Multimodal Depression Recognition
Shanliang Yang (Shandong University of Technology), Xiaoxiao Wang (Shandong University of Technology)
ClassificationTransformerAuto EncoderTextMultimodalityBiomedical DataAudio
🎯 What it does: Proposed the Cross-Modal Guided Visual Synthesis (CMG-VS) framework, which generates visual features conditioned on speech and text to enhance the robustness of multi-modal depression recognition.
Cross-modal Identity Mapping: Minimizing Information Loss in Modality Conversion via Reinforcement Learning
Haonan Jia (Taobao & Tmall Group of Alibaba), Kaifu Zhang (Taobao & Tmall Group of Alibaba)
RetrievalRepresentation LearningTransformerLarge Language ModelReinforcement LearningVision Language ModelImageTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: Propose an unlabeled reinforcement learning framework called Cross-modal Identity Mapping (CIM), which quantifies information loss by measuring the representation consistency of retrieved images (GRC) and the relevance between queries and retrieved images (QIR), using these as rewards to guide large vision-language models to generate more fine-grained and accurate image descriptions.
Cross-modal Representation Learning for Diffusion-generated Image Detection
Tao Gong (University of Science and Technology of China), Nenghai Yu (University of Science and Technology of China)
Anomaly DetectionRepresentation LearningConvolutional Neural NetworkTransformerDiffusion modelContrastive LearningImageMultimodalityBenchmark
🎯 What it does: This paper proposes a method that utilizes two modalities, RGB and neighborhood pixel relationship (NPR), to learn a forgery-aware embedding space for diffusion-generated images through cross-modal contrastive learning (CMCL) and cross-modal mutual distillation (CMMD), constructing a powerful forgery image detector;
Cross-Scale Pansharpening via ScaleFormer and the PanScale Benchmark
Ke Cao (HFIPS, Chinese Academy of Sciences), Jie Zhang (HFIPS, Chinese Academy of Sciences)
Super ResolutionTransformerImageBenchmark
🎯 What it does: Propose a cross-scale panchromatic image fusion method called ScaleFormer and the PanScale benchmark to address challenges in high-resolution cross-scale fusion.
Cross-Slice Knowledge Transfer via Masked Multi-Modal Heterogeneous Graph Contrastive Learning for Spatial Gene Expression Inference
Zhiceng Shi (Yunnan University), Wenwen Min (Yunnan University)
Graph Neural NetworkContrastive LearningMultimodalityGraphBiomedical Data
🎯 What it does: Predict spatial gene expression of tissue slices by constructing cross-slice multimodal heterogeneous graphs combined with contrastive learning.
Cross-Subject EEG-to-Video Reconstruction and Beyond
Runduo Han (Dalian University of Technology), Hongchen Tan (Dalian University of Technology)
GenerationData SynthesisDomain AdaptationConvolutional Neural NetworkTransformerVision Language ModelDiffusion modelAuto EncoderGenerative Adversarial NetworkVideoBiomedical Data
🎯 What it does: Proposed SAM-Net for cross-subject EEG-to-Video reconstruction, addressing cross-subject EEG signal differences and new subject adaptation issues, while generating coherent high-quality videos.
Cross-View Distillation and Adaptive Masking for Incomplete Multi-View Multi-Label Classification
Yadong Liu (Harbin Institute of Technology), Jie Wen (Shanghai University)
ClassificationMultimodalityBenchmark
🎯 What it does: This paper addresses the view imbalance problem in dual missing multi-view multi-label learning by proposing the Cross-view Distillation and Adaptive Masking (CDAM) framework, which first enhances weak view representations through cross-view distillation and then achieves robust fusion by adaptively masking low-quality views.
Cross-View Splatter: Feed-Forward View Synthesis with Georeferenced Images
Matias Turkulainen (Aalto University), Daniyar Turmukhambetov (Niantic Spatial)
GenerationData SynthesisDepth EstimationTransformerGaussian SplattingImagePoint Cloud
🎯 What it does: Proposes Cross-View Splatter, a feedforward model that can utilize GPS-tagged ground camera images and satellite orthoimages to predict 3D Gaussian splats in a unified coordinate system, supporting multiple inputs and unknown 6DoF views.
CrossEarth-Gate: Fisher-Guided Adaptive Tuning Engine for Efficient Adaptation of Cross-Domain Remote Sensing Semantic Segmentation
Shilei Cao (Sun Yat-sen University), Haohuan Fu (National Supercomputing Center in Shenzhen)
SegmentationDomain AdaptationTransformerSupervised Fine-TuningImage
🎯 What it does: Proposed CrossEarth-Gate, a parameter-efficient fine-tuning (PEFT) engine based on Fisher information for dynamic module selection, enabling efficient adaptation for cross-domain remote sensing semantic segmentation.
CrossHOI-Bench: A Unified Benchmark for HOI Evaluation across Vision-Language Models and HOI-Specific Methods
Qinqian Lei (National University of Singapore), Robby T. Tan (National University of Singapore)
Object DetectionVision Language ModelImageBenchmark
🎯 What it does: This paper proposes CrossHOI-Bench, a unified multiple-choice benchmark for human-object interaction (HOI) detection, designed to fairly evaluate large vision-language models (VLMs) and specialized HOI methods.
CrossHOI: Learning Cross-View Representations for Monocular 3D Human-Object Interaction Reconstruction
Pei Geng (Nanjing University of Science and Technology), Jian Yang (Nanjing University of Science and Technology)
Pose EstimationRepresentation LearningConvolutional Neural NetworkTransformerVision-Language-Action ModelImage
🎯 What it does: Studied the problem of 3D human-robot interaction (HOI) reconstruction under monocular images, and proposed a method to improve reconstruction accuracy and contact prediction by generating new perspective features.
CrossVL: Complexity-Aware Feature Routing and Paired Curriculum for Cross-View Vision-Language Detection
Zhipeng Liu (University of Exeter), Chunbo Luo (University of Exeter)
Object DetectionTransformerVision Language ModelMultimodality
🎯 What it does: Designed a cross-view visual language detection framework named CrossVL, aiming to address the issue of performance degradation caused by geometric differences between ground and aerial perspectives.
CrowdGaussian: Reconstructing High-Fidelity 3D Gaussians for Human Crowd from a Single Image
Yizheng Song (Nanjing University), Hao Zhu (Nanjing University)
GenerationPose EstimationDepth EstimationDiffusion modelGaussian SplattingImagePoint Cloud
🎯 What it does: Proposes the CrowdGaussian two-stage framework, which reconstructs a complete 3D Gaussian field from occluded crowd images using LORM, and enhances details and performs reverse distillation via CrowdRefiner (a single-step diffusion model).
CROWn: A Unified Framework for Anti-Aliased Downsampling and Phase-Calibrated Fusion in 3D Medical Segmentation
Xingru Huang (Hangzhou Dianzi University), Xiaoshuai Zhang (Ocean University of China)
SegmentationConvolutional Neural NetworkBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: Proposed the CROWn framework for 3D medical image segmentation, specifically addressing the challenges of aliasing undersampling and cross-scale phase misalignment caused by directional spacing and varying reconstructions.
CryoHype: Reconstructing a thousand cryo-EM structures with transformer-based hypernetworks
Jeffrey Gu (Princeton University), Ellen D. Zhong (Princeton University)
Representation LearningProtein Structure PredictionTransformerBiomedical Data
🎯 What it does: Proposed CryoHype, a Transformer-based hypernetwork that dynamically generates weights for implicit neural representations, enabling high-resolution 3D reconstruction of cryo-EM data with extreme compositional heterogeneity.
CryoKRAQEN: Kernel-Regularized Annealing for Quantized Embedding Networks in Cryo-EM Heterogeneous Reconstruction
Wenyuan Gao (ShanghaiTech University), Xuming He (ShanghaiTech University)
Biomedical Data
🎯 What it does: Proposes a decoder-only framework called CryoKRAQEN based on tri-plane implicit representation, using kernel-guided annealed quantization to address image-structure uncertainty in Cryo-EM heterogeneous reconstruction.
cryoSENSE: Compressive Sensing Enables High-throughput Microscopy with Sparse and Generative Priors on the Protein Cryo-EM Image Manifold
Zain Shabeeb (Georgia Institute of Technology), Amirali Aghazadeh (Georgia Institute of Technology)
Protein Structure PredictionDiffusion modelBiomedical Data
🎯 What it does: Proposed and validated cryoSENSE, a high-throughput cryo-electron microscopy imaging framework based on compressed sensing and generative priors.
CSF: Black-box Fingerprinting via Compositional Semantics for Text-to-Image Models
Junhoo Lee (Seoul National University), Nojun Kwak (Seoul National University)
GenerationPrompt EngineeringVision Language ModelDiffusion modelImageText
🎯 What it does: This paper proposes a fully black-box method—Compositional Semantic Fingerprinting (CSF)—that identifies the source of text generation models by utilizing compositional ambiguous prompts.
CTCal: Rethinking Text-to-Image Diffusion Models via Cross-Timestep Self-Calibration
Xiefan Guo (Beihang University), Di Huang (Beihang University)
GenerationSupervised Fine-TuningDiffusion modelAuto EncoderImageTextBenchmark
🎯 What it does: Propose a Cross-Temporal Self-Calibration (CTCAL) framework that utilizes high-quality cross-attention maps obtained from smaller time steps (with less noise) to guide text–image alignment learning at larger time steps (with more noise), thereby enhancing the precise consistency between text and images.
CubeComposer: Spatio-Temporal Autoregressive 4K 360deg Video Generation from Perspective Video
Lingen Li (CUHK), Ying Shan (Tencent)
GenerationTransformerDiffusion modelFlow-based ModelAuto EncoderVideo
🎯 What it does: Proposes CubeComposer, a self-recursive diffusion model capable of directly generating 4K 360° panoramic videos from normal perspective videos.
Cubic Discrete Diffusion: Discrete Visual Generation on High-Dimensional Representation Tokens
Yuqing Wang (University of Hong Kong), Xihui Liu (University of Hong Kong)
GenerationRepresentation LearningTransformerDiffusion modelImage
🎯 What it does: Propose the Cubic Discrete Diffusion (CubiD) model, which discretizes high-dimensional visual representation features (768–1024 dimensions) and generates images.
CUBic: Coordinated Unified Bimanual Perception and Control Framework
Xingyu Wang (Beihang University), Zhaoxin Fan (Beihang University)
Robotic IntelligenceConvolutional Neural NetworkTransformerDiffusion modelBenchmark
🎯 What it does: Propose the CUBic framework to achieve unified coordination of dual-arm perception and control.
CUE: Concept-Aware Multi-Label Expansion to Mitigate Concept Confusion in Long-Tailed Learning
Ruichi Zhang (Xiamen University), Yang Lu (Xiamen University)
ClassificationRepresentation LearningData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringVision Language ModelImageMultimodality
🎯 What it does: Propose a Concept-Aware Multi-Label Expansion (CUE) method, leveraging vision-language models (VLM) and large language models (LLM) to generate instance-level and class-level relevant labels, incorporating multi-label supervision to address concept confusion issues arising during long-tailed fine-tuning.
CUPID: Generative 3D Reconstruction via Joint Object and Pose Modeling
Binbin Huang (University of Hong Kong), Shenghua Gao (University of Hong Kong)
GenerationData SynthesisPose EstimationFlow-based ModelGaussian SplattingImageMesh
🎯 What it does: Proposes CUPID, a streaming generative reconstruction framework that jointly generates 3D object structures and camera poses;
CURE: Curriculum-guided Multi-task Training for Reliable Anatomy Grounded Report Generation
Pablo Messina (Pontificia Universidad Catolica De Chile), Bernard Ghanem (Kaust)
GenerationExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelBiomedical Data
🎯 What it does: Proposes the CURE framework, which performs fine-grained fine-tuning of multi-task medical vision-language models through error-aware curriculum learning to enhance the reliability and interpretability of anatomical localization and report generation.
Curriculum Group Policy Optimization: Adaptive Sampling for Unleashing the Potential of Text-to-Image Generation
Baoteng Li (Beijing University of Posts and Telecommunications), Zhanyu Ma (Beijing University of Posts and Telecommunications)
GenerationReinforcement LearningDiffusion modelFlow-based ModelImageTextBenchmark
🎯 What it does: This paper proposes a Curriculum Guided by Reward Variance (CGPO), which enhances the efficiency of reinforcement learning in text-to-image generation by dynamically adjusting the sample sampling probability and class weighting.
Curvature-Aware Captioning: Leveraging Geodesic Attention for 3D Scene Understanding
Ziyao He (East China Normal University), Xian Wei (East China Normal University)
GenerationTransformerReinforcement LearningVision Language ModelPoint Cloud
🎯 What it does: For the 3D dense description task, the Curvature-Aware Captioning (CAC) framework is proposed, adopting Oblique manifold geometric attention in the encoding stage and Lorentz hyperbolic geometry bidirectional attention in the decoding stage to achieve geometric consistency of point cloud features and hierarchical semantic fusion.
Curvature-Aware Zeroth-Order Optimization for Memory-Efficient Test-Time Adaptation
Junming Zhang (Shanghai Jiao Tong University), Fei Wen (Shanghai Jiao Tong University)
Domain AdaptationOptimizationImageBenchmark
🎯 What it does: This paper proposes a curvature-aware zeroth-order optimization (CAZO) method, achieving memory-efficient adaptation of pre-trained models during testing, avoiding high memory consumption caused by backpropagation.
CURVE: A Benchmark for Cultural and Multilingual Long Video Reasoning
Darshan Singh (Google DeepMind), Shachi Dave (Google DeepMind)
Large Language ModelVideoTextMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Proposes the CURVE benchmark for multicultural, multilingual long video reasoning and evaluates existing video LLMs.
Customized Fusion: A Closed-Loop Dynamic Network for Adaptive Multi-Task-Aware Infrared-Visible Image Fusion
Zengyi Yang (Hefei University of Technology), Huafeng Li (Kunming University of Science and Technology)
Image HarmonizationConvolutional Neural NetworkReinforcement LearningImage
🎯 What it does: This paper proposes a closed-loop dynamic network (CLDyN), which couples a visualization fusion network (VFN) with a task-specific semantic compensation module (RSC) to achieve adaptive semantic compensation after infrared-visible image fusion. This allows the generation of multi-task customizable fusion results according to different downstream task requirements.
CustomTex: High-fidelity Indoor Scene Texturing via Multi-Reference Customization
Weilin Chen (Xiamen University), Liujuan Cao (Xiamen University)
GenerationTransformerDiffusion modelScore-based ModelNeural Radiance FieldImage
🎯 What it does: Propose a self-supervised method based on dual distillation, utilizing multiple reference images to achieve instance-level high-fidelity texturization for indoor scenes.
Cut to the Chase: Training-free Multimodal Summarization via Chain-of-Events
Xiaoxing You (Hangzhou Dianzi University), Jun Yu (Harbin Institute of Technology)
GenerationGraph Neural NetworkTransformerLarge Language ModelVision Language ModelVideoTextMultimodalityChain-of-Thought
🎯 What it does: Propose an unsupervised multi-modal summarization framework called CoE, which generates text summaries by constructing a Hierarchical Event Graph (HEG) to achieve hierarchical event modeling, cross-modal alignment, and temporal reasoning.
CVA: Context-aware Video-text Alignment for Video Temporal Grounding
Sungho Moon (Daegu Gyeongbuk Institute of Science and Technology), Sunghoon Im (Daegu Gyeongbuk Institute of Science and Technology)
TransformerVision Language ModelContrastive LearningVideoTextMultimodality
🎯 What it does: Designed and implemented three modules (QCD, CBD, CTE) to enhance video-text alignment and temporal localization performance.
Cycle-Consistent Tuning for Layered Image Decomposition
Zheng Gu (Shenzhen University), Hui Huang (Shenzhen University)
Image TranslationData SynthesisTransformerSupervised Fine-TuningVision Language ModelDiffusion modelImage
🎯 What it does: This paper proposes a diffusion model-based image layer decomposition framework, focusing on the task of logo-object separation, achieving multi-layer output from a single image through cyclic consistency training and self-improving data generation.
CycleBEV: Regularizing View Transformation Networks via View Cycle Consistency for Bird's-Eye-View Semantic Segmentation
Jeongbin Hong (Electronics and Telecommunications Research Institute), Kyoung-Wook Min (Electronics and Telecommunications Research Institute)
SegmentationAutonomous DrivingTransformerPoint Cloud
🎯 What it does: Designed a regularized framework called CycleBEV based on view cycle consistency to enhance the performance of view transformation (VT) models in bird's-eye-view (BEV) semantic segmentation.
CycleManip: Enabling Cycle-based Manipulation via Effective History Perception and Understanding
Yi-Lin Wei (Sun Yat Sen University), Wei-Shi Zheng
Robotic IntelligenceTransformerVision Language ModelDiffusion modelImagePoint CloudTime SeriesBenchmark
🎯 What it does: Propose the CycleManip framework, enabling robots to perform cyclic operations with a predetermined number of cycles through end-to-end imitation learning, and provide a complete set of cyclic task benchmarks and automatic evaluation tools.
D-Convexity: A Unified Differentiable Convex Shape Prior via Quasi-Concavity for Data-driven Image Segmentation
Shengzhe Chen (Arizona State University), Hao Yan (Arizona State University)
SegmentationConvolutional Neural NetworkBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper proposes a threshold-free convex prior based on a function called quasi-convexity, which is transformed into differentiable low-order inequalities for end-to-end trained image segmentation networks.
D-Prism: Differentiable Primitives for Structured Dynamic Modeling
Xingyuan Yu (Zhejiang University), Guofeng Zhang (Zhejiang University)
GenerationNeural Radiance FieldGaussian SplattingImageMesh
🎯 What it does: Propose the D-Prism framework to dynamically reconstruct the geometry and rigid motion of structured objects from monocular images, while supporting post-editing of actions.
D^3FER: Dual Channel and Dual Branch Network for Robust Facial Expression Recognition under Dual Challenges
Hui Tang (Nanjing University of Science and Technology), Zhong Jin (Geely Automobile Research Institute)
RecognitionData-Centric LearningContrastive LearningImage
🎯 What it does: Propose a dual-channel dual-branch network D FER, which jointly models expression recognition under dual challenges of visual perturbations and label noise through a unified framework combining weak/strong augmentation, dynamic queue, momentum-updated Query-Key architecture, and adversarial contrastive learning.
D$^2$-FOSA: Dual-Diffusion Guided EEG-to-Image Reconstruction with Frequency-Oriented Semantic Alignment
Chenglong Yu (Beihang University), Yang Li (Beihang University)
GenerationRetrievalConvolutional Neural NetworkGraph Neural NetworkTransformerVision Language ModelDiffusion modelImageBiomedical Data
🎯 What it does: Propose an end-to-end EEG-image reconstruction framework named D2-FOSA, achieving reconstruction and retrieval of images from electroencephalogram (EEG) signals.
D2Cache: Second-Order Delta Caching for Higher Video Diffusion Acceleration
Enhuai Liu (University of Sydney), Chang Xu (University of Sydney)
GenerationComputational EfficiencyDiffusion modelVideo
🎯 What it does: Proposed a training-agnostic second-order delta caching method called D2Cache to accelerate video diffusion model inference, significantly reducing cumulative errors.
D2Dewarp: Dual Dimensions Geometric Representation Learning Based Document Image Dewarping
Heng Li (Harbin Institute of Technology), Qingcai Chen (Harbin Institute of Technology)
RestorationRepresentation LearningConvolutional Neural NetworkImage
🎯 What it does: Proposed the D2Dewarp method, utilizing dual-dimensional geometric representation learning to achieve document image dewarping.
D2FANet: Enhancing Video Object Detection with Dual-Domain Feature Aggregation Network
Qiang Qi (Qingdao University of Science and Technology), Xiao Wang (Qingdao University of Science and Technology)
Object DetectionConvolutional Neural NetworkTransformerVideo
🎯 What it does: Propose a dual-domain feature aggregation network called D2FANet for video object detection, which integrates spatiotemporal and frequency domain features to enhance motion perception and temporal consistency.
D2T2 - Multimodal Automated Planning for Brachytherapy
Lance C. Moore (University of California, San Diego), Nuno Vasconcelos (University of California, San Diego)
OptimizationTransformerAuto EncoderImageBiomedical DataComputed Tomography
🎯 What it does: Propose an end-to-end Transformer architecture D2T2 that directly predicts the dwell time required for ovarian dose distribution and generates the dose distribution via a physical model.
D3D-VLP: Dynamic 3D Vision-Language-Planning Model for Embodied Grounding and Navigation
Zihan Wang (National University of Singapore), Gim Hee Lee (National University of Singapore)
Robotic IntelligenceLarge Language ModelReinforcement LearningVision-Language-Action ModelImageTextMultimodalityChain-of-Thought
🎯 What it does: Proposed a dynamic 3D vision-language-planning model (D3D-VLP), unifying planning, localization, and navigation within a single autoregressive 3D VLM;
DA-Mamba: Learning Domain-Aware State Space Model for Global-Local Alignment in Domain Adaptive Object Detection
Haochen Li (Intelligent Software Research Center, Institute of Software, CAS), Ling Li (Intelligent Software Research Center, Institute of Software, CAS)
Object DetectionDomain AdaptationConvolutional Neural NetworkPrompt EngineeringGenerative Adversarial NetworkImage
🎯 What it does: Propose DA-Mamba, a CNN-SSM hybrid architecture, inserting image-aware SSM and object-aware SSM into the detection backbone and detection head respectively, achieving global-local domain-invariant feature alignment.
DA-VAE: Plug-in Latent Compression for Diffusion via Detail Alignment
Xin Cai (Chinese University of Hong Kong), Tianfan Xue (Chinese University of Hong Kong)
GenerationCompressionDiffusion modelAuto EncoderImageText
🎯 What it does: Propose DA-VAE, a structured latent encoder that extends channels on a pre-trained VAE, enhancing compression rate through a detail alignment mechanism, and achieving high-resolution generation via lightweight fine-tuning.
DABO: Difficulty-Aware Bayesian Optimization with Diffusion-Learned Priors
Mengyang Li (Tianjin Normal University), Pinlong Zhao (Hangzhou Dianzi University)
Hyperparameter SearchTransformerDiffusion modelTime Series
🎯 What it does: Constructed the DABO framework, achieving difficulty-aware scheduling for HPO through hierarchical difficulty modeling, curve generation via conditional diffusion models, and difficulty-aware PFN with adaptive acquisition functions.
DAGE: Dual-Stream Architecture for Efficient and Fine-Grained Geometry Estimation
Tuan Duc Ngo (Adobe Research), Joon-Young Lee (UMass Amherst)
Pose EstimationDepth EstimationKnowledge DistillationTransformerSimultaneous Localization and MappingImageVideoPoint Cloud
🎯 What it does: Proposes DAGE—a dual-stream Transformer architecture that leverages low-resolution global attention to achieve cross-view consistency and camera pose estimation, while preserving details through high-resolution single-frame paths, thus enabling end-to-end prediction of high-resolution, fine-grained, and scale-consistent 3D geometry and camera poses.
Dance Across Shifts: Forward-Facilitation Continual Test-Time Adaptation through Dynamic Style Bridging
Zhilin Zhu (Harbin Institute of Technology), Xiaopeng Hong (Harbin Institute of Technology)
ClassificationDomain AdaptationTransformerPrompt EngineeringDiffusion modelContrastive LearningImage
🎯 What it does: Propose a forward-promoting continuous test-time adaptation (CTTA) paradigm, which dynamically aligns generated class representative images with real-time target domain samples at input, statistical, and representation levels through dynamic style bridging, providing reliable supervision signals in unsupervised online environments and significantly enhancing model adaptability to evolving distributions.
DARC: Dual Adjustment Reasoning with Counterfactuals for Trustworthy Chest X-ray Classification
Zhifang Liao, Yucheng Song (Central South University)
ClassificationExplainability and InterpretabilityImageBiomedical DataBenchmark
🎯 What it does: Proposes the DARC framework, which jointly employs backdoor adjustment and counterfactual reasoning to systematically eliminate spurious correlations caused by non-pathological confounders and pathological co-occurrence in chest X-rays.
Dark3R: Learning Structure from Motion in the Dark
Andrew Y. Guo (University of Toronto), David B. Lindell (University of Toronto)
Pose EstimationDepth EstimationKnowledge DistillationNeural Radiance FieldImage
🎯 What it does: Achieve structure from motion (SfM) and novel view synthesis under extremely low-light conditions, enabling the estimation of camera poses and 3D geometry from raw images with low signal-to-noise ratio (SNR).