CVPR 2026 Papers — Page 6
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 4071 papers
CICA: Coupling Confidence-Aware Pretraining with Confidence-Informed Attention for Robust Multimodal Sentiment Analysis
Haoyu Jiang (Xihua University), Yajun Du (Xihua University)
ClassificationTransformerContrastive LearningMultimodalityBenchmark
🎯 What it does: Proposed the CICA framework, combining confidence-aware pretraining with confidence-guided attention to achieve robust fusion in multimodal sentiment analysis.
CIGMA: Causal Information-Gain Mechanistic Attribution of Attention Heads in Vision Transformers
Maisha Maliha (University of Oklahoma), Dean F. Hougen (University of Oklahoma)
ClassificationRecognitionExplainability and InterpretabilityComputational EfficiencyTransformerImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: Propose the CIGMA framework, which identifies and prunes misleading attention heads in Vision Transformers through training-agnostic head-level causal attribution.
CIGPose: Causal Intervention Graph Neural Network for Whole-Body Pose Estimation
Bohao Li (Northwestern Polytechnical University), Yangming Guo (Northwestern Polytechnical University)
Pose EstimationGraph Neural NetworkImage
🎯 What it does: Propose a causal intervention-based graph neural network (CIGPose) for full-body pose estimation, addressing non-causal associations caused by visual context confusion.
CineBrain: A Large-Scale Multi-Modal Audiovisual Brain Dataset for Brain-Conditioned Video Generation
Jianxiong Gao (Fudan University), Yanwei Fu (Fudan University)
GenerationData SynthesisTransformerVision Language ModelDiffusion modelContrastive LearningMultimodalityBiomedical DataMagnetic Resonance ImagingBenchmark
🎯 What it does: Proposes a new task: synchronously acquiring fMRI and EEG signals in natural contexts, and reconstructing continuous video stimuli through multimodal brain signals.
Cinematic Audio Source Separation Using Visual Cues
Kang Zhang (Korea Advanced Institute of Science and Technology), Joon Son Chung (Korea Advanced Institute of Science and Technology)
RestorationData SynthesisConvolutional Neural NetworkFlow-based ModelVideoMultimodalityBenchmarkAudio
🎯 What it does: Proposed an audio-visual conditional movie audio source separation (AV-CASS) framework that separates mixed movie audio into three tracks: voice, sound effects, and music.
CineScene: Implicit 3D as Effective Scene Representation for Cinematic Video Generation
Kaiyi Huang, Xihui Liu
GenerationVision Language ModelDiffusion modelAuto EncoderImageVideoText
🎯 What it does: Propose the CINESCENE framework, which generates cinematic-level videos with dynamic subjects while maintaining scene consistency using implicit 3D view representations, supporting user-specified camera trajectories;
CineSRD: Leveraging Visual, Acoustic, and Linguistic Cues for Open-World Visual Media Speaker Diarization
Liangbin Huang (Hujing Digital Media and Entertainment Group), Wenji Mao (Institute of Automation Chinese Academy of Sciences)
RecognitionLarge Language ModelVideoTextMultimodalityBenchmarkAudio
🎯 What it does: Propose the CineSRD framework to address the open speaker diarization task in visual media (movies, TV series), leveraging multi-modal (visual, audio, and text) data for speaker annotation.
Circuit Mechanisms for Spatial Relation Generation in Diffusion Transformers
Binxu Wang (Harvard University), Xu Pan (Harvard University)
GenerationTransformerVision Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: This study constructs a simple text-image dataset containing two objects and their spatial relationships, trains and compares Diffusion Transformer (DiT) models under different text encoders (randomly embedded RTE and pre-trained T5), to investigate their internal mechanisms for generating correct spatial relationships.
Circular-DPO: Aligning Multi-Stage 3D Generative Models via Preference Feedback Loop
Zejian Li (Zhejiang University), Lingyun Sun (Zhejiang University)
GenerationData SynthesisFlow-based ModelMesh
🎯 What it does: Propose the Circular-DPO method, which integrates feedback from later stages into earlier stages in a multi-stage 3D generation pipeline through closed-loop feedback, achieving joint optimization based on human preferences.
CLaD: Planning with Grounded Foresight via Cross-Modal Latent Dynamics
Andrew Jeong (KAIST), Sung-Eui Yoon (KAIST)
Robotic IntelligenceTransformerVision Language ModelVision-Language-Action ModelDiffusion modelAuto EncoderImageTextMultimodality
🎯 What it does: Studied a cross-modal latent dynamics model that integrates body states (joint angles, velocities) and semantic states (visual, language) for robot grasping and manipulation, using the model to predict 'latent horizons' as conditional inputs in diffusion strategies to generate actions.
Clair Obscur: an Illumination-Aware Method for Real-World Image Vectorization
Xingyue Lin (Peking University), Liangcai Gao (Peking University)
RestorationOptimizationImage
🎯 What it does: Proposed COVec, an illumination-aware image vectorization framework based on the Clair-Obscur principle, decomposing images into three layers—albedo, shadows, and illumination—and outputting editable vector representations through SVG.
Clay-to-Stone: Phase-wise 3D Gaussian Splatting for Monocular Articulated Hand-Object Manipulation Modeling
Xingyu Liu (Beijing University of Posts and Telecommunications), Jingyu Wang (Beijing University of Posts and Telecommunications)
GenerationPose EstimationTransformerGaussian SplattingVideo
🎯 What it does: Propose a two-stage Clay-to-Stone framework, leveraging 3D Gaussian Splatting to first learn object shape and motion from monocular RGB video through deformable and semantic-associated modulation, then recover joint parameters via rigid constraints, achieving high-quality geometry reconstruction and rendering for hand-object interactions.
CLAY: Conditional Visual Similarity Modulation in Vision-Language Embedding Space
Sohwi Lim (Korea Advanced Institute Of Science And Technology), Tae-Hyun Oh (Korea Advanced Institute Of Science And Technology)
RetrievalLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: Propose a training-agnostic conditional visual similarity modulation method called CLAY, which can adaptively adjust the similarity space in the embedding space of pre-trained vision-language models (VLM) based on textual conditions, achieving conditional image retrieval.
CLCR: Cross-Level Semantic Collaborative Representation for Multimodal Learning
Chunlei Meng (Fudan University), Chun Ouyang (Fudan University)
Representation LearningConvolutional Neural NetworkTransformerVideoTextMultimodalityAudio
🎯 What it does: Propose a Cross-Layer Semantic Collaborative Representation (CLCR) framework, which decomposes the features of each modality into three semantic layers (shallow, medium, and deep). Shared and private subspaces are decomposed and controlled interactions are performed at each layer, followed by semantic synchronization and aggregation across layers, achieving semantic-level alignment and information-efficient fusion in multi-modal learning.
Cleaning the Pool: Progressive Filtering of Unlabeled Pools in Deep Active Learning
Denis Huseljic (University of Kassel), Bernhard Sick (University of Kassel)
ClassificationTransformerSupervised Fine-TuningContrastive LearningImageAudio
🎯 What it does: Propose a multi-strategy integrated active learning method called REFINE, which adopts a two-phase process: first, progressively clean the unlabeled pool through multi-round progressive filtering, retaining only samples considered valuable by at least one strategy; subsequently, select the final labeling batch using a coverage strategy from the refined candidate pool.
CLEP: Contrastive Language-Pose Pretraining
Sen Jia, Lei Li (Beijing Institute Of Technology)
GenerationPose EstimationRetrievalTransformerVision Language ModelContrastive LearningMultimodality
🎯 What it does: Propose CLEP — a contrastive learning-based pose-language pre-training framework that combines the HierFormer hierarchical pose encoder with a text encoder to achieve alignment between pose and natural language.
CLEX: Complementary Label Exchange Learning for Noisy Facial Expression Recognition
Lin Wang (South China University of Technology), Xiangmin Xu (Foshan University)
RecognitionData-Centric LearningConvolutional Neural NetworkImage
🎯 What it does: Proposes the Complementary Label Exchange Learning (CLEX) framework, which enhances robustness in noisy label sentiment recognition tasks by exchanging non-target class logits between different data augmentation views, combined with Lp norm normalization and complementary suppression loss.
ClimaOoD: Improving Anomaly Segmentation via Physically Realistic Synthetic Data
Yuxing Liu (Beijing University of Chemical Technology), Yong Liu (Beijing University of Chemical Technology)
Data SynthesisAnomaly DetectionAutonomous DrivingPrompt EngineeringDiffusion modelImage
🎯 What it does: Proposed the ClimaDrive framework and the ClimaOoD dataset, using a semantics-guided image-to-image approach to generate physically reasonable abnormal driving images under diverse weather conditions and scenarios, and trained an anomaly segmentation model on this dataset.
Clinically-Grounded Counterfactual Reasoning for Medical Video Diagnosis
Jianzhe Gao (Zhejiang University), Yizhou Wang (Peking University)
Convolutional Neural NetworkTransformerDiffusion modelGenerative Adversarial NetworkContrastive LearningVideoBiomedical Data
🎯 What it does: This paper proposes MEDVCR, a medical video diagnosis framework based on clinical rules and adversarial reasoning.
CLIP Is Shortsighted: Paying Attention Beyond the First Sentence
Marc-Antoine Lavoie (University Of Toronto Robotics Institute), Steven L. Waslander (University Of Toronto Robotics Institute)
RetrievalTransformerContrastive LearningText
🎯 What it does: Propose DeBias-CLIP, a parameter-free enhancement method that mitigates CLIP's early token bias for long texts by removing summary sentences, sub-sampling sentences, and text filling.
CLIP-like Model as a Foundational Density Ratio Estimator
Fumiya Uchiyama (University of Tokyo), Yutaka Matsuo (University of Tokyo)
ClassificationDomain AdaptationRepresentation LearningTransformerContrastive LearningImage
🎯 What it does: This paper reinterprets CLIP-like models as density ratio estimators and proposes two new applications based on this interpretation: importance-weighted learning and KL divergence estimation.
ClipGStream: Clip-Stream Gaussian Splatting for Any Length and Any Motion Multi-View Dynamic Scene Reconstruction
Jie Liang (Peking University), Ronggang Wang (Peking University)
GenerationGaussian SplattingVideo
🎯 What it does: Proposes ClipGStream, a framework based on Clip-Stream Gaussian Splatting for long-term multi-view dynamic scene reconstruction, balancing large-scale motion and scalability for long sequences.
CLIPoint3D: Language-Grounded Few-Shot Unsupervised 3D Point Cloud Domain Adaptation
Mainak Singha, Biplab Banerjee
Domain AdaptationTransformerPrompt EngineeringVision Language ModelContrastive LearningTextPoint Cloud
🎯 What it does: This paper proposes CLIPoint3D, a few-shot unsupervised 3D point cloud domain adaptation framework based on CLIP, which achieves cross-modal alignment between point clouds and language through multi-view depth projection, knowledge-driven prompt tuning, and parameter-efficient fine-tuning.
CLiViS: Unleashing Cognitive Map through Linguistic-Visual Synergy for Embodied Visual Reasoning
Kailing Li (East China Normal University), Xiaoling Wang (East China Normal University)
Graph Neural NetworkTransformerLarge Language ModelPrompt EngineeringVision Language ModelVideoTextMultimodality
🎯 What it does: Proposed a training-free, collaborative framework named CLiViS based on LLM and VLM for semantic understanding and spatiotemporal reasoning on first-person videos with complex instructions.
Cloning Deterministic Worlds: The Critical Role of Latent Geometry in Long-Horizon World Models
Zaishuo Xia (University Of California Davis), Yubei Chen (University Of California Davis)
TransformerAuto EncoderContrastive LearningWorld ModelVideoSequential
🎯 What it does: This paper proposes a geometric regularized world model (GRWM) by improving the geometric structure of the latent space, thereby significantly enhancing the accuracy of long-term prediction in deterministic environments.
Closed-Form Concept Erasure via Double Projections
Chi Zhang, Ping Liu
GenerationDiffusion modelImage
🎯 What it does: Proposed a closed-form double projection method to achieve training-free erasure of unwanted concepts in diffusion models, capable of removing target objects or styles while preserving other concepts in the model.
Clothe and Pose
Nakul Sharma, Minh Vo
Image TranslationImage HarmonizationGenerationPose EstimationTransformerDiffusion modelImage
🎯 What it does: Propose the Clothe and Pose task, which generates high-quality images of a user wearing specified clothing and assuming a target pose, given the user image, clothing image, and target pose, and builds a unified evaluation framework.
CLP: A Real-World Dataset of Contaminated Lens Protectors for Robust Semantic Segmentation
Sungyong Park (Soongsil University), Heewon Kim (Soongsil University)
RestorationSegmentationDomain AdaptationTransformerSupervised Fine-TuningImageBenchmark
🎯 What it does: This paper introduces the CLP dataset, providing paired clear/damaged images under real lens protector contamination (mud, raindrops, fog) and pixel-level semantic annotations for 125 object categories to evaluate robust semantic segmentation and image restoration models; it systematically evaluates multiple learning strategies (supervised, domain generalization, domain adaptation, baseline models, robustness methods, task-aware restoration) and various restoration networks on this dataset, exploring the impact of data scale and joint restoration-segmentation pipelines.
Cluster-aware Anchor Learning for Multi-View Clustering
Zhe Chen (Anhui University of Technology), Xiao-Jun Wu (Jiangnan University)
MultimodalityBenchmark
🎯 What it does: This paper proposes the Cluster-aware Anchor Learning (CAL) method, which can adaptively assign different numbers of anchors to each cluster in multi-view clustering, thereby improving clustering performance.
Cluster-Aware Neural Collapse Prompt Tuning for Long-Tailed Generalization of Vision-Language Models
Boyang Guo (Hangzhou Dianzi University), Chenggang Yan (Hangzhou Dianzi University)
ClassificationRecognitionDomain AdaptationRepresentation LearningTransformerSupervised Fine-TuningPrompt EngineeringVision Language ModelImageMultimodality
🎯 What it does: Propose a Cluster-Aware Prompt Tuning (CPT) method that locally constrains prompt learning in pre-trained vision-text models to enhance long-tail class identification while maintaining overall generalization.
Cluster-Wise Spatio-Temporal Masking for Efficient Video-Language Pretraining
Weijun Zhuang (Harbin Institute of Technology), Wangmeng Zuo (Harbin Institute of Technology)
GenerationRetrievalVision Language ModelVideoTextMultimodality
🎯 What it does: Propose ClusterSTM—a clustering-level masking strategy based on intra-frame clustering and spatiotemporal density, for efficient video-text pretraining, and introduce a video-text relevance reconstruction objective;
ClusterMark: Towards Robust Watermarking for Autoregressive Image Generators with Visual Token Clustering
Denis Lukovnikov (Ruhr University Bochum), Asja Fischer (Ruhr University Bochum)
GenerationSafty and PrivacyAuto EncoderImage
🎯 What it does: Investigated the technique of embedding detectable watermarks in autoregressive image generation models.
CME-CAD: Heterogeneous Collaborative Multi-Expert Reinforcement Learning for CAD Code Generation
Ke Niu (Fudan University), Xiangyang Xue (Fudan University)
GenerationAI Code AssistantReinforcement LearningMixture of ExpertsVision Language ModelImageTextMeshChain-of-Thought
🎯 What it does: Proposed the CME-CAD framework, which generates executable and editable CAD code using multi-expert collaborative reinforcement learning;
CMR-RD: Long-Tailed Adaptive VLM for Explainable CMR Diagnosis
Yansong Li (Beijing Normal University), Shuo Li (Shanghai Jiao Tong University)
Explainability and InterpretabilityTransformerLarge Language ModelReinforcement LearningVision Language ModelMultimodalityBiomedical DataMagnetic Resonance ImagingChain-of-Thought
🎯 What it does: Proposed CMR-RD, an interpretable cardiac magnetic resonance imaging (CMR) diagnostic visual-language model that can generate diagnostic reasoning chains aligned with imaging evidence.
Co-Me: Confidence Guided Token Merging for Visual Geometric Transformers
Yutian Chen (Carnegie Mellon University), Sebastian Scherer (Carnegie Mellon University)
Pose EstimationDepth EstimationComputational EfficiencyKnowledge DistillationTransformerImagePoint CloudBenchmark
🎯 What it does: Propose a Confidence-Guided Token Merging (Co-Me) method without retraining, significantly accelerating inference by fusing low-confidence tokens in visual geometry Transformers.
CoCoVideo: The High-Quality Commercial-Model-Based Contrastive Benchmark for AI-Generated Video Detection
Huidong Feng (Xiamen University), Ming Zeng (Xiamen University)
Anomaly DetectionConvolutional Neural NetworkTransformerLarge Language ModelContrastive LearningVideoMultimodalityBenchmark
🎯 What it does: Constructed a high-quality, contrastive commercial model-generated video dataset named CoCoVideo-26K, and proposed a detection framework named CoCoDetect that integrates contrastive learning with confidence-gated multimodal large language models for AIGC video detection.
CoD: A Diffusion Foundation Model for Image Compression
Zhaoyang Jia (University Of Science And Technology Of China), Yan Lu (Microsoft Research Asia)
CompressionTransformerDiffusion modelRectified FlowAuto EncoderImage
🎯 What it does: This paper proposes CoD (Compression-oriented Diffusion foundation model), a foundation model trained from scratch that simultaneously optimizes compression and diffusion generation, and applies it as a general-purpose foundation model to various diffusion-based image compression frameworks, significantly improving compression efficiency and image quality.
Coded-E2LF: Coded Aperture Light Field Imaging from Events
Tomoya Tsuchida (Nagoya University), Hajime Nagahara (Osaka University)
RestorationConvolutional Neural NetworkSequential
🎯 What it does: Proposed a fully event-based method called Coded-E2LF for 4D light field reconstruction using programmable apertures and pure event cameras.
CodeDance: A Dynamic Tool-integrated MLLM for Executable Visual Reasoning
Qi Song (Beihang University), Yunqing Zhao (ByteDance)
AI Code AssistantReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningVision-Language-Action ModelMultimodalityBenchmark
🎯 What it does: Propose CodeDance, which utilizes executable code as a general mediator for dynamic tool invocation and visual reasoning.
CodeMMR: Bridging Natural Language, Code, and Image for Unified Retrieval
Jiahui Geng (MBZUAI), Fakhri Karray (MBZUAI)
RetrievalTransformerLarge Language ModelSupervised Fine-TuningContrastive LearningImageTextMultimodalityBenchmark
🎯 What it does: Propose the MMCoIR multimodal multilingual code retrieval benchmark and train a unified retrieval model CodeMMR, achieving cross-modal retrieval across natural language, code, and image modalities;
CodePercept: Code-Grounded Visual STEM Perception for MLLMs
Tongkun Guan (Shanghai Jiao Tong University), Xiaokang Yang (Shanghai Jiao Tong University)
Image TranslationGenerationData SynthesisLarge Language ModelReinforcement LearningVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: Propose using executable code as a visual perception medium, constructing the ICC-1M image-caption-code triplet dataset, and enhancing the visual perception capabilities of multimodal large language models in STEM visual reasoning through the CodePercept framework.
CodeV: Code with Images for Faithful Visual Reasoning via Tool-Aware Policy Optimization
Xinhai Hou (University of Michigan), Bryan Wang (Amazon.com)
AI Code AssistantSupervised Fine-TuningReinforcement LearningAgentic AIVision Language ModelMultimodality
🎯 What it does: Proposed CodeV, a code-based visual reasoning agent, and designed Tool-Aware Policy Optimization (TAPO) to explicitly reward the authenticity of tool usage;
CoFiDA-M: Concept-Aware Feature Modulation for Cross-Domain Adaptation with Image-Only Inference
Nurjahan Sultana (Manchester Metropolitan University), Wenqi Lu (Manchester Metropolitan University)
ClassificationDomain AdaptationKnowledge DistillationImageBiomedical Data
🎯 What it does: Propose the CoFiDA-M framework, which leverages MONET semantic concept information available during training. A teacher network is trained to edit features via FiLM, and the edited features along with classification results are distilled to a student network that uses only images, achieving cross-domain skin lesion classification.
COG: Confidence-aware Optimal Geometric Correspondence for Unsupervised Single-reference Novel Object Pose Estimation
Yuchen Che (Institute of Science Tokyo), Asako Kanezaki (Institute of Science Tokyo)
Pose EstimationImage
🎯 What it does: Proposes a confidence-aware optimal transport (OT)-based unsupervised single-view novel object pose estimation framework named COG, achieving precise pose regression by leveraging soft correspondences and semantic priors.
CogDriver: Integrating Cognitive Inertia for Temporally Coherent Planning in Autonomous Driving
Pei Liu (Hong Kong University of Science and Technology), Jun Ma (Hong Kong University of Science and Technology)
Autonomous DrivingKnowledge DistillationTransformerLarge Language ModelVision-Language-Action ModelMultimodality
🎯 What it does: This paper proposes the CogDriver framework, aiming to achieve cognitive inertia (temporal inertia) in autonomous driving systems, enabling vehicles to maintain temporally consistent and coherent internal representations, thereby reducing decision jitter and completing multi-step complex operations.
CogniEdit: Dense Gradient Flow Optimization for Fine-Grained Image Editing
Yan Li (Hongkong University of Science and Technology), Qi Tian (Huawei Inc)
GenerationOptimizationTransformerLarge Language ModelDiffusion modelImageMultimodalityBenchmark
🎯 What it does: This paper proposes the CogniEdit framework, integrating multimodal LLM, dynamic token focusing, and dense GRPO optimization to achieve instruction-driven fine-grained image editing.
CogniVerse: Revolutionizing Multi-Modal Retrieval-Augmented Generation with Cognitive Reflection and Geometric Reasoning
Xiang Fang (Huazhong University of Science and Technology), Changshuo Wang (University College London)
GenerationRetrievalGraph Neural NetworkMultimodalityGraphRetrieval-Augmented Generation
🎯 What it does: Proposes a multimodal retrieval-augmented generation framework named CogniVerse, addressing core bottlenecks in MMRAG such as noisy retrieval, cross-modal semantic imbalance, lack of adaptive reasoning, and incoherent generation.
CoIn: Coverage and Informativeness-Guided Token Reduction for Efficient Large Multimodal Models
Chenxi Du (Tsinghua University), Ju Ren (Tsinghua University)
Computational EfficiencyVision Language ModelImageTextMultimodality
🎯 What it does: Propose CoIn, a no-training, model-agnostic visual token selection framework that uses information (visual saliency + cross-modal alignment) and coverage (subset selection based on volume) to jointly determine retained visual tokens, significantly reducing LMM inference costs.
CoIn3D: Revisiting Configuration-Invariant Multi-Camera 3D Object Detection
Zhaonian Kuang (Xi'an Jiaotong University), Gang Hua (Amazon Alexa AI)
Object DetectionAutonomous DrivingTransformerGaussian SplattingImage
🎯 What it does: Propose the CoIn3D framework, leveraging spatial-aware feature modulation (SFM) and camera-aware data augmentation (CDA) to address differences in multi-camera configurations, enhancing cross-configuration 3D object detection generalization performance.
ColaVLA: Leveraging Cognitive Latent Reasoning for Hierarchical Parallel Trajectory Planning in Autonomous Driving
Qihang Peng (Tsinghua University), Hongsheng Li (Chinese University Of Hong Kong)
Autonomous DrivingExplainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningVision-Language-Action ModelImageMultimodalityPoint CloudChain-of-Thought
🎯 What it does: Proposes ColaVLA, a unified vision-language-action framework that compresses multimodal inputs through cognitive latent reasoning and generates safe, interpretable continuous trajectories in real-time via a hierarchical parallel planner.
CoLC: Communication-Efficient Collaborative Perception with LiDAR Completion
Yushan Han (Key Laboratory of Big Data and Artificial Intelligence in Transportation Ministry of Education), Yidong Li (Key Laboratory of Big Data and Artificial Intelligence in Transportation Ministry of Education)
Autonomous DrivingTransformerPoint Cloud
🎯 What it does: Propose CoLC, a collaborative perception framework that achieves communication efficiency by maintaining the advantages of early fusion through foreground-aware sampling (FAPS) and LiDAR completion module (CEEF), and using Dense-Guided Dual Alignment (DGDA) for feature consistency training.
Collaborative Multi-Mode Pruning for Vision-Language Models
Zimeng Wu (Beihang University), Jiaxin Chen (Beihang University)
Computational EfficiencyTransformerVision Language ModelMultimodality
🎯 What it does: Proposes the Collaborative Multi-Mode Pruning (CoMP) framework, which jointly prunes parameters and tokens in vision-language models.
CoLoGen: Progressive Learning of Concept-Localization Duality for Unified Image Generation
Yuxin Song (Baidu Inc), Jingdong Wang (Baidu Inc)
GenerationMixture of ExpertsVision Language ModelDiffusion modelImageMultimodality
🎯 What it does: CoLoGen proposes a unified multimodal image generation framework that learns complementary representations of concepts (semantics) and localization (spatial) through progressive training, and fuses multi-task requirements via dynamic routing experts.
Color When It Counts: Grayscale-Guided Online Triggering for Always-On Streaming Video Sensing
Weitong Cai (Queen Mary University of London), Zhensong Zhang (Huawei Noah's Ark Lab)
OptimizationComputational EfficiencyTransformerVision Language ModelVideoBenchmark
🎯 What it does: Propose the ColorTrigger framework, which uses an always-on grayscale camera to monitor video and triggers color acquisition only when significant changes in grayscale features are detected, achieving low-power real-time video perception.
Color-Encoded Illumination for High-Speed Volumetric Scene Reconstruction
David Novikov (Weizmann Institute of Science), Mark Sheinin (Weizmann Institute of Science)
GenerationFlow-based ModelGaussian SplattingImageVideo
🎯 What it does: Encoding high-frequency temporal information into a single frame of a traditional low-speed camera using color flickering illumination to achieve volumetric reconstruction of fast-moving scenes;
CoLoR: The Devil is in Scene Coordinate Regression for Large-Scale Visual Localization
Xindong Mao (Beihang University), Jin Zheng (Beihang University)
Pose EstimationDepth EstimationRetrievalContrastive LearningImage
🎯 What it does: Proposed the CoLoR framework, which leverages multi-view and single-view segmentation of scene points, employing two-stage training to provide strong supervision for all points, thereby achieving large-scale visual localization.
ColorFLUX: A Structure-Color Decoupling Framework for Old Photo Colorization
Bingchen Li (University of Science and Technology of China), Zhibo Chen (University of Science and Technology of China)
RestorationKnowledge DistillationSupervised Fine-TuningReinforcement LearningPrompt EngineeringDiffusion modelImage
🎯 What it does: Propose a framework for colorizing old grayscale photographs based on the FLUX diffusion model, named ColorFLUX. It employs structure-color decoupling learning, visual semantic prompts, and progressive direct preference optimization (DPO), enabling the recovery of color, brightness, and saturation after denoising and scratch removal.
Common Inpainted Objects In-N-Out of Context
Tianze Yang (University of Georgia), Jin Sun (University of Georgia)
ClassificationImage TranslationImage HarmonizationRestorationObject DetectionGenerationData SynthesisKnowledge DistillationVision Language ModelDiffusion modelImage
🎯 What it does: Propose the COinCO dataset, which uses a diffusion-based repair model to replace individual objects in COCO images, generating 97,722 images with both contextually consistent and inconsistent scenarios, and provides fine-grained context labels (position, size, co-occurrence) for each image.
CoMo: Learning Continuous Latent Motion from Internet Videos for Scalable Robot Learning
Jiange Yang (Nanjing University), Limin Wang (Nanjing University)
Robotic IntelligenceTransformerDiffusion modelAuto EncoderContrastive LearningVideo
🎯 What it does: Train an unsupervised continuous latent motion model to learn continuous motion representations from internet videos for robot learning;
CompBench: Benchmarking Complex Instruction-guided Image Editing
Bohan Jia (East China Normal University), Shaohui Lin (East China Normal University)
Large Language ModelDiffusion modelMultimodalityBenchmark
🎯 What it does: Constructed and publicly released CompBench, a large-scale benchmark dataset for complex instruction-driven image editing, and systematically evaluated the capabilities of existing editing models through this benchmark.
CompetitorFormer: Mitigating Query Conflicts for 3D Instance Segmentation via Competitive Strategy
Duanchu Wang (Xi'an Jiaotong University), Di Wang (Xi'an Jiaotong University)
SegmentationTransformerPoint CloudBenchmark
🎯 What it does: This paper addresses the query conflict problem in Transformer-based 3D instance segmentation by proposing the competitive framework CompetitorFormer, which introduces a query competition layer, relative relation encoding, and ranking cross-attention before each decoding layer to explicitly model competitive relationships among queries;
Complementary Prototype Mapping for Efficient Multimodal Anomaly Detection
Yuan Zhao (Dalian University of Technology), Lihe Zhang (Dalian University of Technology)
Anomaly DetectionTransformerAuto EncoderMultimodalityBiomedical DataAlzheimer's Disease
🎯 What it does: Proposed the Complementary Prototype Mapping (CPMAD) framework, which uses dynamically extracted consensus and complementary prototypes to guide multi-modal unsupervised defect detection, significantly reducing cross-modal mapping ambiguity and improving localization accuracy.
Complet4R: Geometric Complete 4D Reconstruction
Weibang Wang (Tsinghua University), Hang Zhao (Tsinghua University)
Object TrackingGenerationDepth EstimationTransformerVideoPoint CloudBenchmark
🎯 What it does: Propose the Complet4R framework to achieve spatiotemporally consistent and geometrically complete 4D reconstruction and point tracking from monocular video.
ComPose: A Unified Completion-Pose Framework for Robust Category-Level Object Pose Estimation
Huan Ren (University of Science and Technology of China), Tianzhu Zhang (University of Science and Technology of China)
Pose EstimationConvolutional Neural NetworkTransformerPoint CloudBenchmark
🎯 What it does: Proposes a unified ComPose framework that tightly integrates point cloud completion with pose estimation, using a keypoint progressive completion module to generate complete geometry, thereby enhancing pose prediction.
Composing Concepts from Images and Videos via Concept-prompt Binding
Xianghao Kong (Hong Kong University of Science and Technology), Anyi Rao (Hong Kong University of Science and Technology)
GenerationTransformerPrompt EngineeringVision Language ModelDiffusion modelImageVideoText
🎯 What it does: This paper proposes a one-shot visual concept combination method called BiCo, which achieves flexible fusion of image and video concepts by binding visual concepts to text prompts and combining them as needed.
Composite-Attribute Person Re-Identification via Pose-Guided Disentanglement
Kartik Patwari (University of California Davis), Kah Kuen Fu (Amazon)
Pose EstimationRetrievalConvolutional Neural NetworkTransformerContrastive LearningImageTextBenchmark
🎯 What it does: Propose the Composite-Attributes Person Re‑ID (CA‑ReID) task, which requires simultaneously achieving identity matching and attribute constraints given a query image and a phrase or composite attribute description;
Compositional Text-to-Image Generation Via Region-aware Bimodal Direct Preference Optimization
Zhuohan Liu (Fudan University), Zuxuan Wu (Fudan University)
GenerationSupervised Fine-TuningReinforcement LearningVision Language ModelDiffusion modelMultimodalityBenchmark
🎯 What it does: In the text-to-image generation field, the BIDPO framework is proposed, which uses dual-modal preference optimization and region-aware guidance to perform post-training fine-tuning on Stable Diffusion XL, and constructs the BICOMP preference dataset through an automated pipeline.
Compositional Transformation Reasoning for Composed Video Retrieval
Sihong Huang (South China University of Technology), Xiaoyong Wei (Sichuan University)
RetrievalTransformerLarge Language ModelVision Language ModelContrastive LearningVideoTextMultimodality
🎯 What it does: Proposes a zero-shot composite video retrieval framework called MoRe based on a multimodal large language model, capable of performing fine-grained transformation reasoning across three dimensions: entities, actions, and scenes.
Compressed-Domain-Aware Online Video Super-Resolution
Yuhang Wang (Beijing Institute of Technology), Xiaoyao Yang (Beijing Institute of Technology)
Super ResolutionCompressionOptical FlowVideo
🎯 What it does: Propose a framework named CDA-VSR for online video super-resolution that leverages compressed domain information.
Computation and Communication Efficient Federated Unlearning via On-server Gradient Conflict Mitigation and Expression
Minh-Duong Nguyen (VinUniversity), Dung D. Le (University of Sydney)
Federated LearningSafty and PrivacyConvolutional Neural NetworkImage
🎯 What it does: Proposes a two-phase federated forgetting framework named FOUL, first marking the knowledge of clients to be forgotten through causal/non-causal feature decoupling during the learning phase, then performing data-free joint forgetting on the server side using gradient matching, ultimately achieving efficient and accurate client forgetting.
Computational Speckle Pattern Interferometry
Shengxi Wu (Carnegie Mellon University), Matthew O'Toole (Carnegie Mellon University)
Optical FlowVideoPhysics Related
🎯 What it does: Proposed Computational Speckle Image Interferometry (CSPI), which can estimate pixel-level displacement and motion from a single frame.
Computer Vision with a Superpixelation Camera
Sasidharan Mahalingam (Portland State University), Atul Ingle (Portland State University)
Object DetectionSegmentationDepth EstimationConvolutional Neural NetworkTransformerImage
🎯 What it does: Designed and implemented a camera system (SuperCam) capable of real-time adaptive sampling and directly outputting superpixel images without generating complete images
Conan: Progressive Learning to Reason Like a Detective over Multi-Scale Visual Evidence
Kun Ouyang (Peking University), Xu Sun (Peking University)
Explainability and InterpretabilityRepresentation LearningTransformerLarge Language ModelReinforcement LearningVision-Language-Action ModelVideoTextMultimodalityChain-of-Thought
🎯 What it does: Proposed the Conan framework based on multi-step video reasoning with multi-scale evidence, constructed the large-scale Conan-91K dataset, and trained phased cold start and AIR RLVR models.
Concept Regions Matter: Benchmarking CLIP with a New Cluster-Importance Approach
Aishwarya Agarwal (IIIT Hyderabad), Vineet Gandhi (IIIT Hyderabad)
Explainability and InterpretabilityTransformerVision Language ModelContrastive LearningImageMultimodalityBenchmark
🎯 What it does: Propose the Cluster-based Concept Importance (CCI) method based on CLIP patch embedding clustering to explain the regions the model focuses on and assess their contribution to predictions, while constructing a new COVAR evaluation benchmark to systematically study CLIP's robustness under factors such as background, perspective, and scale.
Concept-Aware Batch Sampling Improves Language-Image Pretraining
Adhiraj Ghosh (Tubingen AI Center, University of Tbingen), Matthias Bethge (Tubingen AI Center, University of Tbingen)
RetrievalRepresentation LearningData-Centric LearningLarge Language ModelVision Language ModelContrastive LearningMultimodality
🎯 What it does: Constructed the DATACONCEPT dataset containing 128M image-text pairs, fine-grained concept labels, and conceptually rewritten captions, and proposed an online controllable concept-aware batch sampling framework CABS, which is divided into two strategies: diversity maximization (CABS-DM) and frequency maximization (CABS-FM).
Concept-Aware LoRA for Domain-Aligned Segmentation Dataset Generation
Minho Park (KAIST), Sungha Choi (Kyung Hee University)
SegmentationData SynthesisDomain AdaptationDiffusion modelImage
🎯 What it does: Propose Concept‑Aware LoRA (CA‑LoRA), a fine-tuning method for text-to-image models, to generate semantic segmentation datasets that are aligned with the target domain and rich in information.
Concept-Guided Fine-Tuning: Steering ViTs away from Spurious Correlations to Improve Robustness
Yehonatan Elisha (Tel Aviv University), Noam Koenigstein (Tel Aviv University)
Domain AdaptationExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImage
🎯 What it does: This study proposes Concept-Guided Fine-Tuning (CFT), which significantly enhances the model's robustness under out-of-distribution conditions by aligning the internal feature maps of ViT with automatically generated concept-level semantic masks in an unsupervised manner.
ConceptPose: Training-Free Zero-Shot Object Pose Estimation using Concept Vectors
Liming Kuang (Technical University of Munich), Benjamin Busam (Technical University of Munich)
Pose EstimationTransformerLarge Language ModelVision Language ModelImage
🎯 What it does: Propose ConceptPose, a zero-shot, model-free, and no-training object pose estimation framework
ConceptPrism: Concept Disentanglement in Personalized Diffusion Models via Residual Token Optimization
Minseo Kim (KAIST), Junmo Kim (KAIST)
GenerationSupervised Fine-TuningVision Language ModelDiffusion modelImage
🎯 What it does: Propose the ConceptPrism framework, which jointly optimizes target tokens and residual tokens through cross-image comparison to achieve concept disentanglement in personalized text-to-image generation.
Condensed Test-Time Adaptation of VLMs for Action Recognition
Wenxuan Ge (Nanjing University of Science and Technology), Jinhui Tang (Nanjing Forestry University)
RecognitionDomain AdaptationVision Language ModelVideo
🎯 What it does: A training-free test-time adaptation framework named CONDA is proposed for zero-shot action recognition, addressing the non-transferability issue of traditional two-step mapping chains.
Conditional Factuality Controlled LLMs with Generalization Certificates via Conformal Sampling
Kai Ye (Case Western Reserve University), Shuo Li (Case Western Reserve University)
Explainability and InterpretabilityTransformerLarge Language ModelTextMultimodality
🎯 What it does: This paper proposes Conditional Factuality Control (CFC), a post-hoc conformal method that achieves conditional coverage in large language model generation using variable thresholds.
ConeSep: Cone-based Robust Noise-Unlearning Compositional Network for Composed Image Retrieval
Zixu Li (Shandong University), Liqiang Nie (Harbin Institute Of Technology)
RetrievalVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: This paper addresses the problem of noisy triplet correspondence in compositional image retrieval, proposing a three-stage network called ConeSep based on cone space. It achieves precise identification and processing of noisy samples through geometric fidelity quantization, negative boundary learning, and boundary-based directional unlearning.
Confidence-Guided Multi-Scale Aggregation for Sparse-View High-Resolution 3D Gaussian Splatting
Qinzheng Zhou (Huazhong University of Science and Technology), Zhihang Li (Alibaba Group)
RestorationGaussian SplattingImagePoint Cloud
🎯 What it does: This paper systematically investigates the impact of different input resolutions on the performance of 3D Gaussian Splatting (3DGS) for high-resolution 3D Gaussian point cloud reconstruction under sparse views, and proposes a confidence-based multi-scale aggregation framework (CAGS) to achieve coarse-to-fine hierarchical optimization for noise suppression and detail quality enhancement.
Conflict-Aware Adaptive Cross-Reconstruction for Multimodal Sentiment Analysis
Yan Wang (Shanxi University), Xingwang Zhao (Shanxi University)
ClassificationTransformerMultimodality
🎯 What it does: Propose a Conflict-Aware Adaptive Cross-Modal Reconstruction (CACR) framework for multimodal sentiment analysis, addressing semantic ambiguity caused by emotional conflicts between different modalities in the same sample.
Confusion-Aware Spectral Regularizer for Long-Tailed Recognition
Ziquan Zhu (University of Leicester), Tianjin Huang (University of Exeter)
ClassificationImageBenchmark
🎯 What it does: This paper proposes a long-tailed image classification method called CAR based on frequency-weighted confusion matrix spectral regularization.
Consensus Entropy: Harnessing Multi-VLM Agreement for Self-Verifying and Self-Improving OCR
Yulong Zhang (Fudan University), Gongshen Liu (East China Normal University)
RecognitionVision Language ModelImageTextBenchmark
🎯 What it does: Propose an unsupervised Consensus Entropy (CE) metric and the CE-OCR framework, which is an OCR solution that uses outputs from multiple Vision-Language Models (VLMs) for self-inspection and self-correction.
Consensus vs. Controversy: Mapping the Decision Space Where Architectures Diverge
Minhyeok Lee (Chung-Ang University)
ClassificationExplainability and InterpretabilityComputational EfficiencyRepresentation LearningConvolutional Neural NetworkTransformerImageBenchmark
🎯 What it does: Propose a framework to map decision space differences among various visual model architectures (CNN, ViT, MLP) on the ImageNet validation set, and identify approximately 10% of 'controversial images' that cause significant model prediction discrepancies;
ConsID-Gen: View-Consistent and Identity-Preserving Image-to-Video Generation
Mingyang Wu (Texas A&M University), Zhengzhong Tu (eBay Inc)
GenerationTransformerVision Language ModelDiffusion modelImageVideoTextBenchmark
🎯 What it does: Proposes ConsID-Gen, a perspective-assisted Image-to-Video (I2V) generation framework that combines a dual-stream visual encoder, text-visual alignment module, and Diffusion Transformer, enabling the generation of high-quality videos while maintaining object identity and geometric consistency; simultaneously constructs the ConsIDVid dataset and ConsIDVid-Bench evaluation benchmark to measure the identity preservation performance of I2V generation from multiple perspectives.
ConsistCompose: Unified Multimodal Layout Control for Image Composition
Xuanke Shi (SenseTime Research), Dahua Lin (SenseTime Research)
GenerationData SynthesisTransformerPrompt EngineeringVision Language ModelDiffusion modelFlow-based ModelImageMultimodalityBenchmark
🎯 What it does: Propose the ConsistCompose framework, which directly embeds layout coordinates into language prompts to achieve unified multi-modal layout-controllable image generation and multi-instance composition.
Consistency Beyond Contrast: Enhancing Open-Vocabulary Object Detection Robustness via Contextual Consistency Learning
Bozhao Li (Harbin Institute of Technology, Shenzhen), Jingyong Su (Harbin Institute of Technology, Shenzhen)
Object DetectionData SynthesisTransformerLarge Language ModelDiffusion modelContrastive LearningImageTextBenchmark
🎯 What it does: This paper proposes the Contextual Consistency Learning (CCL) framework, which enhances the robustness of open-vocabulary object detection by strengthening internal consistency across different backgrounds.
Consistent Instance Field for Dynamic Scene Understanding
Junyi Wu (University of Illinois Chicago), Ziyan Wu (United Imaging Intelligence)
SegmentationGaussian SplattingImage
🎯 What it does: Propose a Consistent Instance Field (CIF) framework that models dynamic scenes as continuous probabilistic spatiotemporal fields, integrating occupancy probability and conditional instance distributions, and achieving visualization and learning through differentiable rasterized Gaussian Splatting.
ConsisVLA-4D: Advancing Spatiotemporal Consistency in Efficient 3D-Perception and 4D-Reasoning for Robotic Manipulation
Wei Li (Harbin Institute of Technology), Liqiang Nie (Harbin Institute of Technology)
Robotic IntelligenceTransformerVision-Language-Action ModelImage
🎯 What it does: Proposed the ConsisVLA-4D framework, achieving efficient consistency in robot manipulation through 3D perception and 4D spatiotemporal reasoning
Contact-Aware Neural Dynamics
Changwei Jing (University of California San Diego), Sha Yi (University of California San Diego)
Robotic IntelligenceSupervised Fine-TuningDiffusion modelPoint CloudTime Series
🎯 What it does: By training a neural forward dynamics model in simulation and fine-tuning it on a small amount of real grasp data, the method leverages tactile signals to achieve implicit alignment of contact dynamics, thereby enhancing the prediction and control performance of robots in contact-rich grasp and manipulation tasks.
Content-Adaptive Hierarchical Hyperprior for Neural Video Coding
Junqi Liao (University of Science and Technology of China), Xiaoyan Sun (University of Science and Technology of China)
CompressionVideo
🎯 What it does: Proposed a neural video coding framework based on hierarchical hyperprior to achieve content-adaptive hierarchical structure optimization.
Content-Aware Dynamic Patchification for Efficient Video Diffusion
Sheng Li (University of Pittsburgh), Yifan Gong (Adobe Research)
GenerationTransformerDiffusion modelAuto EncoderVideoText
🎯 What it does: Propose DynaPatch: a content-based dynamic patching framework that adaptively selects patch sizes in Diffusion Transformer video generation based on spatiotemporal complexity, significantly reducing the number of tokens.
Content-Aware Frequency Encoding for Implicit Neural Representations with Fourier-Chebyshev Features
Junbo Ke (Hunan Normal University), You-Wei Wen (Hunan Normal University)
Representation LearningNeural Radiance FieldImageMesh
🎯 What it does: Propose content-aware frequency encoding (CAFE) and its extended version CAFE+, which dynamically synthesize frequencies through parallel linear layers and Hadamard product, thereby mitigating the spectral bias and fixed Fourier basis limitations of INR.
Context-Nav: Context-Driven Exploration and Viewpoint-Aware 3D Spatial Reasoning for Instance Navigation
Won Shik Jang (Gwangju Institute of Science and Technology), Ue-Hwan Kim (Gwangju Institute of Science and Technology)
Robotic IntelligenceTransformerVision Language ModelVision-Language-Action ModelTextMultimodalityBenchmark
🎯 What it does: In sparse unmarked 3D scenes, leveraging long natural language descriptions to guide agents to first perform semantic-driven frontier exploration on a global value map, then confirm target instances through perspective-aware 3D relationship verification.
Continual Distillation of Teachers from Different Domains
Nicolas Michel (University of Tokyo), Toshihiko Yamasaki (University of Tokyo)
Domain AdaptationKnowledge DistillationImage
🎯 What it does: Propose a Continual Distillation framework that enables a single student model to learn sequentially from multiple teacher models without retaining previous teacher models.
Continual Learning for fMRI-Based Brain Disorder Diagnosis via Functional Connectivity Matrices Generative Replay
Qianyu Chen (Nanyang Technological University), Shujian Yu (VU Amsterdam)
ClassificationKnowledge DistillationGraph Neural NetworkTransformerAuto EncoderGenerative Adversarial NetworkBiomedical DataMagnetic Resonance Imaging
🎯 What it does: Proposes a continual learning framework FORGE for cross-site fMRI brain disease diagnosis, utilizing generative replay to achieve knowledge preservation.
Continuous Exposure-Time Modeling for Realistic Atmospheric Turbulence Synthesis
Junwei Zeng (Nanjing University of Aeronautics and Astronautics), Songcan Chen (Nanjing University of Aeronautics and Astronautics)
Data SynthesisVideoPhysics Related
🎯 What it does: Propose a synthetic model for atmospheric turbulence under continuous time exposure and construct the ET-Turb large-scale synthetic dataset