CVPR 2026 Papers — Page 5
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 4071 papers
Bridging Domain Expertise and Generalization for Performance Estimation
Shuxuan Li (Sun Yat-sen University), Wei-Shi Zheng (Sun Yat-sen University)
Domain AdaptationImage
🎯 What it does: This paper proposes a FRAP framework that estimates model accuracy by fusing predictions from a base model and a task model under label distribution shift without labeled data.
Bridging Domains through Subspace-Aware Model Merging
Levy Chaves (Universidade Estadual de Campinas), Sandra Avila (Universidade Estadual de Campinas)
ClassificationDomain AdaptationTransformerImageBenchmark
🎯 What it does: The study merges models fine-tuned on different domains to construct a single model and evaluates its generalization performance on unseen domains.
Bridging Facial Understanding and Animation via Language Models
Luchuan Song (University of Rochester), Chenliang Xu (University of Rochester)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringVision-Language-Action ModelDiffusion modelAuto EncoderVideoTextMultimodality
🎯 What it does: This paper generates a diverse text-aligned facial video dataset (Open3DFaceVid) and quantizes 3D facial parameters into discrete geometric codes, constructing a unified language-facial motion bidirectional framework capable of both interpreting facial motion into natural language (Motion2Language) and generating detailed 3D facial animations from text (Language2Motion).
Bridging Fidelity-Reality with Controllable One-Step Diffusion for Image Super-Resolution
Hao Chen (Nanjing University of Science and Technology), Jiangxin Dong (Nanjing University of Science and Technology)
RestorationSuper ResolutionConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: Propose CODSR, a controllable one-step diffusion network for recovering high-quality images from low-quality images, balancing structural fidelity and perceptual quality.
Bridging Human Evaluation to Infrared and Visible Image Fusion
Jinyuan Liu (Dalian University of Technology), Xin Fan (Dalian University of Technology)
Image HarmonizationObject DetectionSegmentationReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningDiffusion modelImageMultimodality
🎯 What it does: Propose an infrared and visible light image fusion framework based on human feedback reinforcement learning (RLHF), aligning the fused images more closely with human visual preferences through human preference-guided reward model and policy optimization.
Bridging Pixels and Words: Mask-Aware Local Semantic Fusion for Multimodal Media Verification
Zizhao Chen (Xi'an Jiaotong University), Xiangru Yin (Xi'an Jiaotong University)
Anomaly DetectionTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Propose a multimodal verification framework called MaLSF based on mask-label pairs, utilizing Bidirectional Cross-Modal Verification (BCV) and Hierarchical Semantic Aggregation (HSA) to achieve active, fine-grained pixel-text alignment and conflict detection.
Bridging Privacy and Provenance: Traceable Virtual Identity Generation
Xianhan Zeng (Fudan University), Xinpeng Zhang (Fudan University)
GenerationData SynthesisSafty and PrivacyDiffusion modelImage
🎯 What it does: Proposed a framework based on diffusion models for generating virtual identity faces that are both anonymous and traceable.
Bridging RGB and Hematoxylin Components: An Interleaved Guidance and Fusion Framework for Point Supervised Nuclei Segmentation
Zihan Huan (Guilin University of Electronic Technology), Zhenbing Liu (Guilin University of Electronic Technology)
SegmentationConvolutional Neural NetworkBiomedical Data
🎯 What it does: This paper proposes a point-supervised nucleus segmentation framework called DFGNet based on dual representations (RGB and hemocyanin components), which can achieve high-precision instance segmentation using sparse point annotations.
Bridging the 2D-3D Gap: A Hierarchical Semantic-Geometric Map for Vision Language Navigation
Kailing Li (East China Normal University), Liang He (East China Normal University)
Autonomous DrivingRobotic IntelligenceTransformerVision Language ModelSimultaneous Localization and MappingImageTextMultimodalityPoint CloudBenchmarkChain-of-Thought
🎯 What it does: Proposes a Hierarchical Semantic-Geometric Map (HSGM), decoupling high-level semantic planning from low-level classical A* path planning by converting 3D environments into 2D BEV maps understandable by VLM;
Bridging the Modality Gap in Compositional Zero-Shot Learning via Sparse Alignment and Unimodal Memory Bank
Yang Zhang (Beijing Jiaotong University), Songhe Feng (Concordia University)
Knowledge DistillationRepresentation LearningTransformerSupervised Fine-TuningVision Language ModelMultimodalityBenchmark
🎯 What it does: Propose the SAM framework to address the modality gap problem in compositional zero-shot learning (CZSL).
Bridging the Perception Gap in Image Super-Resolution Evaluation
Shaolin Su (Computer Vision Center), Javier Vazquez-Corral (Computer Vision Center)
Super ResolutionImage
🎯 What it does: This paper proposes a novel evaluation framework based on relative quality difference (RQI) to measure the perceptual quality of image super-resolution models.
BriMA: Bridged Modality Adaptation for Multi-Modal Continual Action Quality Assessment
Kanglei Zhou (Tsinghua University), Liyuan Wang (Tsinghua University)
Domain AdaptationVision-Language-Action ModelMultimodalityRetrieval-Augmented Generation
🎯 What it does: Studied the problem of non-stationary modality imbalance in multi-modal continuous action quality assessment, proposing the BriMA method, which includes a memory-guided bridging completion module and a modality-aware replay module.
Bringing Your Portrait to 3D Presence
Jiawei Zhang (Nanjing University), Yan Lu (Microsoft Research Asia)
GenerationData SynthesisTransformerLarge Language ModelPrompt EngineeringGaussian SplattingImageMesh
🎯 What it does: Implemented a unified framework capable of directly reconstructing animatable 3D avatars from single portraits (including head, upper body, and full-body views).
BuildAnyPoint: 3D Building Structured Abstraction from Diverse Point Clouds
Tongyan Hua (Hong Kong University of Science and Technology), Wufan Zhao (Hong Kong University of Science and Technology)
GenerationTransformerDiffusion modelAuto EncoderPoint CloudMeshBenchmark
🎯 What it does: This paper proposes a general 3D building structure abstraction framework called BuildAnyPoint, which can generate high-quality, low-polygon building meshes from various point cloud distributions (such as airborne LiDAR, SfM, etc.).
Building a Precise Video Language with Human-AI Oversight
Zhiqiu Lin (Carnegie Mellon University), Deva Ramanan (Carnegie Mellon University)
GenerationReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningAgentic AIVision Language ModelContrastive LearningVideoTextMultimodality
🎯 What it does: Built a full-process system for precise video language, including structured description specifications formulated by professionals, a critical supervision framework based on human-AI collaboration, and model post-training methods utilizing pre-generated/post-generated text.
Building Robust Vision Encoders for Cross-Dataset Evaluation in Immunofluorescent Microscopy
Umar Marikkar (Surrey Institute for People-Centred AI), Sara Atito (Surrey Institute for People-Centred AI)
ClassificationRepresentation LearningTransformerContrastive LearningImageBiomedical Data
🎯 What it does: Proposed the C3R framework, which can train a robust visual encoder on immunofluorescence images across different fluorescence staining channel configurations, and achieve cross-dataset evaluation without retraining.
BuildingGPT: Auto-Regressive Building Wireframe Reconstruction Model with Reinforcement Learning
Yuzhou Liu (Institute of Automation, Chinese Academy of Sciences), Shuhan Shen (Institute of Automation, Chinese Academy of Sciences)
GenerationReinforcement Learning from Human FeedbackTransformerReinforcement LearningPoint Cloud
🎯 What it does: Utilizing an autoregressive Transformer combined with a point cloud encoder to perform complete wireframe reconstruction on building point clouds, ultimately generating complete building outlines and top structures.
Bulk RNA-seq Guided Multi-modal Detection of Anomalous Regions in Human Cancer via Spatial Transcriptomics
Hang Shi (Nanjing University of Aeronautics and Astronautics), Wei Shao (Nanjing University of Aeronautics and Astronautics)
Anomaly DetectionGraph Neural NetworkVision Language ModelImageMultimodalityGraph
🎯 What it does: Propose a multimodal method called BRGMAR that integrates spatial transcriptomics (ST), bulk RNA-seq, and pathological images to detect abnormal regions (AR) in human cancer tissues.
BulletTime: Decoupled Control of Time and Camera Pose for Video Generation
Yiming Wang (ETH Zurich), Gordon Wetzstein (Stanford University)
GenerationData SynthesisTransformerDiffusion modelAuto EncoderVideo
🎯 What it does: Proposed a 4D controllable video diffusion framework that can independently adjust world time and camera pose during video generation.
BUSSARD: Normalizing Flows for Bijective Universal Scene-Specific Anomalous Relationship Detection
Melissa Schween (Leibniz University Hannover), Bodo Rosenhahn (Leibniz University Hannover)
Anomaly DetectionFlow-based ModelAuto EncoderImage
🎯 What it does: Propose a scene graph relationship anomaly detection method called BUSSARD based on normalizing flows
Bypassing the Transport Plan: Dynamic Reweighting for Out-of-Distribution Detection with Optimal Transport
Yang Xiao (Zhejiang University), Lianyong Qi (China University of Petroleum (East China))
ClassificationAnomaly DetectionOptimizationImageBenchmark
🎯 What it does: Proposed a semi-supervised open-set OOD detection framework called DREW based on dynamic reweighting, which can directly provide reliable pseudo OOD scores during training.
C-GenReg: Training-Free 3D Point Cloud Registration by Multi-View-Consistent Geometry-to-Image Generation with Probabilistic Modalities Fusion
Yuval Haitman (Ben Gurion University), Joseph M. Francos (Ben Gurion University)
GenerationPose EstimationWorld ModelImagePoint CloudBenchmark
🎯 What it does: Designed a zero-shot, no-training 3D point cloud registration framework called C-GenReg, which generates multi-view consistent RGB images using a world generative model, extracts correspondences with a vision foundation model tailored for matching, and fuses them with geometric branch features.
C-LaV: Conditional Latent Velocity Field Denoising for Weather-Robust LiDAR Place Recognition
Xuewei Cao (University of Science and Technology of China), Yan Xia (University of Science and Technology of China)
RetrievalAutonomous DrivingTransformerFlow-based ModelPoint CloudOrdinary Differential Equation
🎯 What it does: This paper proposes a C-LaV framework for LiDAR-based pose estimation under adverse weather conditions.
C^2FG: Control Classifier-Free Guidance via Score Discrepancy Analysis
Jiayang Gao (Shanghai Jiao Tong University), Jia Wang (Shanghai Jiao Tong University)
GenerationData SynthesisTransformerDiffusion modelScore-based ModelImageStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: Proposed a time-varying control-based classifier-free guidance method called C-FG2, replacing traditional fixed weights to achieve exponentially decaying guidance strength over time, thereby improving the generation quality of conditional diffusion models.
CAD-Refiner: A Unified Framework for CAD Generation and Iterative Editing
Meng Yuan (Jilin University), Rui Ma (Jilin University)
GenerationGraph Neural NetworkTransformerLarge Language ModelVision Language ModelImageTextMultimodalityGraphSequential
🎯 What it does: Proposed the CAD-Refiner framework, achieving a unified workflow for CAD generation and editing, and supporting free-form multimodal inputs.
CADC: Content Adaptive Diffusion-Based Generative Image Compression
Xihua Sheng (City University of Hong Kong), Jing Wang (City University of Hong Kong)
CompressionVision Language ModelDiffusion modelImage
🎯 What it does: Propose a content-adaptive diffusion-based image compression algorithm CADC for generating image reconstructions with higher visual quality at extremely low bitrates.
CADFS: A Big CAD Program Dataset and Framework for Computer-Aided Design with Large Language Models
Vladislav Pyatov (Applied AI Institute), Evgeny Burnaev (Applied AI Institute)
GenerationData SynthesisAI Code AssistantLarge Language ModelSupervised Fine-TuningPrompt EngineeringVision Language ModelTextMultimodalityMesh
🎯 What it does: Propose the CADFS framework, which achieves more complex CAD design history generation and reconstruction by using the FeatureScript code generator.
Calibri: Enhancing Diffusion Transformers via Parameter-Efficient Calibration
Danil Tokhchukov (Moscow State University), Konstantin Sobolev (FusionBrain Lab, AXXX)
GenerationComputational EfficiencyTransformerReinforcement LearningDiffusion modelImageTextMultimodality
🎯 What it does: This paper proposes a parameter-efficient calibration method called Calibri for Diffusion Transformers (DiT), which significantly improves generation quality and reduces inference steps by merely adjusting the output of DiT blocks through approximately 10² learnable scalar coefficients.
CaliTex: Geometry-Calibrated Attention for View-Coherent 3D Texture Generation
Chenyu Liu (Peking University), Xin Wang (Lightspeed)
SegmentationGenerationData SynthesisTransformerDiffusion modelAuto EncoderImageMesh
🎯 What it does: Propose the CaliTex framework, leveraging geometric calibration attention to achieve cross-view consistent 3D texture generation.
CamDirector: Towards Long-Term Coherent Video Trajectory Editing
Kejia Yin (University of Toronto), Juwei Lu (University of Toronto)
GenerationData SynthesisDiffusion modelVideoBenchmark
🎯 What it does: Propose a hybrid warp and history-guided autoregressive framework for long video trajectory editing, capable of generating new videos that align with user-specified camera paths while preserving the source video content;
Camera Control for Text-to-Image Generation via Learning Viewpoint Tokens
Xinxuan Lu (University of California, Irvine), Alexander C. Berg (University of California, Irvine)
GenerationTransformerSupervised Fine-TuningPrompt EngineeringVision Language ModelDiffusion modelImageText
🎯 What it does: Propose a method to achieve precise camera control in text-to-image generation by learning perspective tokens.
Camouflage-aware Image-Text Retrieval via Expert Collaboration
Yao Jiang (Sichuan University), Qijun Zhao (Sichuan University)
RetrievalTransformerMixture of ExpertsContrastive LearningImageTextMultimodality
🎯 What it does: Studied the image-text retrieval task in camouflaged scenarios, proposing a new CamoIT dataset and the CECNet model.
CamPI: Physical Adversarial Examples through Camera Power Signal Injection
Yanze Ren (Zhejiang University), Wenyuan Xu (Zhejiang University)
Adversarial AttackImage
🎯 What it does: Propose generating invisible physical adversarial samples by injecting modulated signals into the camera power supply, and optimizing adversarial parameters under both white-box and black-box scenarios.
Can a Second-View Image Be a Language? Geometric and Semantic Cross-Modal Reasoning for X-ray Prohibited Item Detection
Chuang Peng (Beijing Jiaotong University), Yunchao Wei (Beihang University)
Object DetectionAnomaly DetectionTransformerLarge Language ModelMixture of ExpertsVision Language ModelMultimodalityBiomedical DataBenchmarkChain-of-Thought
🎯 What it does: Propose DualXrayBench and the GSR model, exploring the treatment of dual-view X-ray images as a modality similar to language to enhance prohibited item detection and cross-perspective reasoning.
Can Natural Image Autoencoders Compactly Tokenize fMRI Volumes for Long-Range Dynamics Modeling?
Peter Yongho Kim (Seoul National University), Taesup Moon (Seoul National University)
ClassificationCompressionTransformerAuto EncoderTime SeriesBiomedical DataMagnetic Resonance Imaging
🎯 What it does: Compress fMRI volumes into continuous markers using a pre-trained 2D natural image autoencoder, and employ Transformer for long-term temporal modeling.
Can We Build Scene Graphs, Not Classify Them? FlowSG: Progressive Image-Conditioned Scene Graph Generation with Flow Matching
Xin Hu (UESTC), Tao He (Monash University)
GenerationGraph Neural NetworkTransformerVision Language ModelFlow-based ModelAuto EncoderImageGraphOrdinary Differential Equation
🎯 What it does: Redefine scene graph generation as a progressive, graph neural network-driven hybrid discrete-continuous flow matching process, progressively denoising and generating objects and relationships.
Can You Learn to See Without Images? Procedural Warm-Up for Vision Transformers
Zachary Shinnick (University of Adelaide), Anton van den Hengel (University of Adelaide)
ClassificationData SynthesisRepresentation LearningTransformerImage
🎯 What it does: Propose a novel preheating phase on Vision Transformer (ViT), using symbol sequences generated by formal grammar (without visual or semantic content) for masked token prediction training, and then initializing the model with this training for standard image training;
CanonCGT: Reference-Based Color Grading via Canonical Pivot Representation
Jinwon Ko (Korea University), Chang-Su Kim (Korea University)
Image HarmonizationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: Proposed a two-stage reference-based color grading framework called CanonCGT, which first canonicalizes the input image and then matches it to the reference image's color tone.
CapNav: Benchmarking Vision Language Models on Capability-conditioned Indoor Navigation
Xia Su (University of Washington), Jon Froehlich (University of Washington)
Robotic IntelligenceLarge Language ModelVision Language ModelVideoTextGraphBenchmarkChain-of-Thought
🎯 What it does: This paper proposes the CapNav benchmark to evaluate the indoor navigation performance of Vision-Language Models (VLMs) under varying robot/human mobility capabilities, and systematically tests 13 modern VLMs on this benchmark.
CAPT: Confusion-Aware Prompt Tuning for Reducing Vision-Language Misalignment
Maoyuan Shao (Minzu University of China), Chuang Zhu (Beijing University of Posts and Telecommunications)
ClassificationDomain AdaptationPrompt EngineeringMixture of ExpertsVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Propose the CAPT (Confusion-Aware Prompt Tuning) framework, which leverages the model's inherent misclassification patterns to construct a confusion bank, and separately explores confusion relationships at the semantic and sample levels. By unifying multi-grained differences through expert fusion, the alignment accuracy of CLIP in fine-grained classification and cross-domain generalization is enhanced.
Captain Safari: A World Engine with Pose-Aligned 3D Memory
Yu-Cheng Chou (Johns Hopkins University), Junfei Xiao (Johns Hopkins University)
GenerationTransformerDiffusion modelWorld ModelVideoRetrieval-Augmented Generation
🎯 What it does: Designed and implemented a world engine named Captain Safari, which generates long-duration, 3D-consistent FPV videos along given camera trajectories through pose-aligned 3D memory retrieval.
CaptionFormer: Unified Segmentation, Tracking, and Captioning for Spatio-Temporal Objects
Gabriel Fiastre (Inria, Ecole Normale Superieure, Cnrs, Psl Research University), Cordelia Schmid (Inria, Ecole Normale Superieure, Cnrs, Psl Research University)
Object DetectionObject TrackingSegmentationGenerationTransformerPrompt EngineeringVision Language ModelVideoTextMultimodality
🎯 What it does: This paper proposes a unified model called CaptionFormer, which can simultaneously achieve object detection, instance segmentation, trajectory tracking, and object-level natural language description in videos.
CaptionQA: Is Your Caption as Useful as the Image Itself?
Shijia Yang (ADVANCED MICRO DEVICES, INC.), Chenfeng Xu (UT AUSTIN)
GenerationTransformerLarge Language ModelImageTextMultimodalityBenchmark
🎯 What it does: Introduce the CaptionQA benchmark to assess the practicality of generated captions in multi-domain downstream tasks.
CAR-SAM: Cross-Attention Reconstruction for Post-Training Quantization of the Segment Anything Model
Houji Wen (Nanjing University of Science and Technology), Jun Li (Nanjing University of Science and Technology)
Object DetectionSegmentationComputational EfficiencyTransformerImageVideo
🎯 What it does: Proposes a unified post-training quantization framework, CAR-SAM, specifically designed for quantizing the decoder cross-attention structure of the Segment Anything Model (SAM) and its successor version SAM2.
CARD: A Multi-Modal Automotive Dataset for Dense 3D Reconstruction in Challenging Road Topography
Gasser Elazab (ICARIAD SE), Olaf Hellwich (Vision & Robotics GmbH)
Autonomous DrivingSimultaneous Localization and MappingMultimodalityPoint CloudBenchmark
🎯 What it does: Create and release the CARD multimodal driving dataset, focusing on irregular road surfaces such as slopes and potholes, providing quasi-real-time synchronized panoramic video, dual LiDAR, and tire trajectory data, and generating high-density 3D ground truth and road geometry annotations.
CARD: Correlation Aware Restoration with Diffusion
Niki Nezakati (University of California Riverside), Vishwanath Saragadam (University of California Riverside)
RestorationDiffusion modelImageBenchmark
🎯 What it does: Proposed a training-agnostic CARD method that achieves high-quality image restoration in correlated noise scenarios by whitening noise and using closed-form sampling with DDRM in the whitened domain, and created the CIN-D dataset for evaluating correlated noise.
CARE What Fails: Contrastive Anchored-REflection for Verifiable Multimodal Reasoning
Yongxin Wang (Mohamed bin Zayed University of Artificial Intelligence), Xiaodan Liang (Mohamed bin Zayed University of Artificial Intelligence)
Reinforcement Learning from Human FeedbackReinforcement LearningContrastive LearningMultimodality
🎯 What it does: This paper proposes a post-training framework called CARE, which converts failed examples in multimodal reasoning tasks into supervision signals by leveraging verifiable rewards, primarily through anchored contrastive objectives and reflection-guided resampling;
CARE-Edit: Condition-Aware Routing of Experts for Contextual Image Editing
Yucheng Wang (Hong Kong University of Science and Technology), Dan Xu (Hong Kong University of Science and Technology)
Image HarmonizationGenerationTransformerSupervised Fine-TuningMixture of ExpertsDiffusion modelMultimodality
🎯 What it does: Propose CARE-Edit, a condition-aware expert routing framework for unified multimodal image editing tasks (text-driven, mask-driven, reference image-driven, etc.), reducing conflicts between different editing conditions through dynamic allocation of computational resources.
CARE: A Molecular-Guided Foundation Model with Adaptive Region Modeling for Whole Slide Image Analysis
Di Zhang (Xi'an Jiaotong University), Zeyu Gao (University of Cambridge)
ClassificationSegmentationExplainability and InterpretabilityTransformerVision Language ModelContrastive LearningMultimodalityBiomedical Data
🎯 What it does: This paper proposes CARE, a foundational model for whole-slide images, which employs adaptive region partitioning and leverages molecular data to guide cross-modal pretraining, achieving efficient and clinically interpretable pathological analysis.
CaReFlow: Cyclic Adaptive Rectified Flow for Multimodal Fusion
Sijie Mai (South China Normal University), Shiqin Han (South China Normal University)
ClassificationRectified FlowMultimodalityOrdinary Differential Equation
🎯 What it does: Propose the CaReFlow framework, which utilizes cyclic adaptive rectified flow to achieve distribution mapping and fusion from source modality to target modality.
CARI4D: Category Agnostic 4D Reconstruction of Human-Object Interaction
Xianghui Xie (NVIDIA), Stan Birchfield (NVIDIA)
Object TrackingGenerationPose EstimationDepth EstimationOptimizationTransformerContrastive LearningVideoMeshBenchmark
🎯 What it does: Achieve category-agnostic 4D (spatiotemporal) human-object interaction reconstruction from monocular RGB videos, outputting size, shape, pose, and contact information while maintaining spatial and temporal consistency throughout the entire video.
CaricHarmony: Contrastive Diffusion Paths for Identity-Preserving Caricature Synthesis
Dongyu Wang (University of Surrey), Yi-Zhe Song (University of Surrey)
Image TranslationGenerationTransformerDiffusion modelContrastive LearningImage
🎯 What it does: Propose an untrained image generation framework called CaricHarmony, which can achieve exaggerated caricature effects while preserving the subject's identity based on freehand hand-drawn sketches.
CARLoS: Retrieval via Concise Assessment Representation of LoRAs at Scale
Shahar Sarfaty (Tel Aviv University), Amit H. Bermano (Tel Aviv University)
GenerationRetrievalVision Language ModelDiffusion modelContrastive LearningImageTextMultimodality
🎯 What it does: Propose a 3D representation based on the generation effects of LoRA (CARLoS), and achieve efficient LoRA retrieval using this representation.
CASPA: Graph-Structured Concept Anchors for Modality-Agnostic Adaptation in Vision-Language Models
Abhiroop Chatterjee (Jadavpur University), Emmett Ientilucci (Rochester Institute of Technology)
ClassificationRecognitionDomain AdaptationComputational EfficiencyRepresentation LearningTransformerPrompt EngineeringVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Proposes a bidirectional semantic adapter called CASPA based on concept anchors, enabling efficient adaptation across modalities and tasks while keeping the CLIP backbone frozen.
CASR: A Robust Cyclic Framework for Arbitrary Large-Scale Super-Resolution with Distribution Alignment and Self-Similarity Awareness
Wenhao Guo (Beijing University of Posts and Telecommunications), RuiDe Li (Beijing University of Posts and Telecommunications)
Super ResolutionSupervised Fine-TuningDiffusion modelImage
🎯 What it does: Propose a cyclic single-network framework CASR that achieves arbitrary-scale super-resolution using superpixel structure alignment and self-similar perception modules.
CaST-Bench: Benchmarking Causal Chain-Grounded Spatio-Temporal Reasoning for Video Question Answering
Mingfang Zhang (Woven by Toyota), Quan Kong (Woven by Toyota)
Large Language ModelVision Language ModelVideoBenchmarkRetrieval-Augmented Generation
🎯 What it does: Constructed CaST-Bench, a causal chain geospatial-temporal reasoning benchmark for video question answering, containing 2,066 multiple-choice questions requiring retrieval of causal evidence;
CAST: Context-Aware Dynamic Latent Space Transformation for Interactive Text-to-Image Retrieval
Xuanzuo Lin, Jianfeng Dong (Zhejiang Gongshang University)
RetrievalLarge Language ModelContrastive LearningMultimodality
🎯 What it does: Propose the CAST framework, which achieves multi-round interactive text-to-image retrieval through context-aware dynamic latent space transformation.
CaT-GS: Efficient 3DGS Rendering for Large-Scale Scenes with Inter-frame Caching and Tile Scheduling
Tingjia Zhang (Shanghai Jiao Tong University), Guihai Chen (Shanghai Jiao Tong University)
Computational EfficiencyGaussian SplattingImage
🎯 What it does: Proposed an efficient 3D Gaussian Splatting rendering pipeline called CaT-GS, primarily addressing computational redundancy and GPU load imbalance in large-scale scenes.
Catalyst4D: High-Fidelity 3D-to-4D Scene Editing via Dynamic Propagation
Shifeng Chen (Beihang University), Di Huang (Beihang University)
GenerationGaussian SplattingOptical FlowVideo
🎯 What it does: Migrate the static 3D Gaussian editing method to dynamic 4D Gaussian scenes, achieving high-quality editing with temporal consistency;
CATNet: Collaborative Alignment and Transformation Network for Cooperative Perception
Gong Chen (Tianjin University), Xin Xie (Tianjin University)
Autonomous DrivingMultimodalityBenchmark
🎯 What it does: Proposes the CATNet framework, which employs three modules: spatiotemporal recursive synchronization, dual-band denoising, and adaptive feature selection to address the issues of latency and noise in collaborative perception.
CaTok: Taming Mean Flows for One-Dimensional Causal Image Tokenization
Yitong Chen (Fudan University), Yu-Gang Jiang (Fudan University)
GenerationTransformerDiffusion modelFlow-based ModelRectified FlowAuto EncoderImage
🎯 What it does: Proposed a 1D visual causal tokenizer named CATOK based on MeanFlow, which can learn causal 1D tokens under the diffusion autoencoder framework and achieve single-step and multi-step sampling;
Causal Motion Diffusion Models for Autoregressive Motion Generation
Qing Yu (LY Corporation), Kent Fujiwara (LY Corporation)
GenerationTransformerVision Language ModelDiffusion modelAuto EncoderTextSequential
🎯 What it does: Proposed Causal Motion Diffusion Models (CMDM), achieving real-time, long-sequence motion generation based on text, balancing the quality of diffusion models with the causality of autoregressive models.
Causality in Video Diffusers is Separable from Denoising
Xingjian Bai (Massachusetts Institute of Technology), Zongze Wu (Adobe Research)
GenerationComputational EfficiencyTransformerDiffusion modelVideo
🎯 What it does: Propose the Separable Causal Diffusion (SCD) framework, which separates causal reasoning from multi-step denoising in video diffusion. First, a one-time causal Transformer encoder generates temporal context, followed by a lightweight frame-level diffusion decoder to complete denoising.
CausalLens: Sensitivity-Guided Multi-Head Causal Intervention for Hallucination Mitigation in Large Vision-Language Models
Junyang Ji (Tsinghua University), Zhihai He (Southern University of Science and Technology)
Explainability and InterpretabilityComputational EfficiencyRepresentation LearningTransformerVision Language ModelMultimodalityBenchmark
🎯 What it does: This paper proposes CausalLens, a training-agnostic and single-forward inference method, which enhances the visual grounding capability of vision-language models and significantly reduces hallucinations by amplifying visual channels based on the visual sensitivity of attention heads and performing residual correction after projection.
CausalVAD: De-confounding End-to-End Autonomous Driving via Causal Intervention
Jiacheng Tang (Fudan University), Jian Pu (Fudan University)
Autonomous DrivingMultimodalityPoint Cloud
🎯 What it does: Designed and implemented the CausalVAD framework, which employs sparse causal intervention (SCIS) to perform causal disentanglement at multiple stages in the perception, prediction, and planning phases of an end-to-end autonomous driving model, thereby eliminating the effects of causal confounding and data bias.
CC-VQA: Conflict- and Correlation-Aware Method for Mitigating Knowledge Conflict in Knowledge-Based Visual Question Answering
Yuyang Hong (School of Artificial Intelligence, UCAS), Jieping Ye (Alibaba Cloud Computing)
RetrievalTransformerVision Language ModelContrastive LearningImageTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: Proposes a training-free framework that combines visual-centric conflict reasoning with relevance-guided encoding and decoding to alleviate conflicts between parameter knowledge and retrieved context in knowledge-based visual question answering.
CCCaption: Dual-Reward Reinforcement Learning for Complete and Correct Image Captioning
Zhijiang Tang, Jianqiang Huang (Chinese Academy Of Sciences)
GenerationLarge Language ModelReinforcement LearningVision Language ModelMultimodality
🎯 What it does: Propose the CCCaption framework, which utilizes dual-reward reinforcement learning to achieve completeness and correctness of image captions, and constructs a multimodal query dataset named CCaption-44k with 44k samples;
CCF: Complementary Collaborative Fusion for Domain Generalized Multi-Modal 3D Object Detection
Yuchen Wu (Singapore University of Technology and Design), Na Zhao (Singapore University of Technology and Design)
Object DetectionAutonomous DrivingTransformerImagePoint Cloud
🎯 What it does: Research and improve dual-branch (image + LiDAR) 3D object detection models to enhance robustness in cross-domain scenarios (rainy days, nighttime, Boston).
CD-Buffer: Complementary Dual-Buffer Framework for Test-Time Adaptation in Adverse Weather Object Detection
Youngjun Song (Yonsei University), Dosik Hwang (Yonsei University)
Object DetectionDomain AdaptationImage
🎯 What it does: Proposed the CD-Buffer dual-buffer framework for adaptive adjustment during real-time testing of object detection models under adverse weather conditions.
CDICS: Delving Into Fine-Grained Attribute for In-Context Segmentation via Compositional Prompts and Phased Decoupling
Zhiyu Li (University of Science and Technology of China), Nenghai Yu (University of Science and Technology of China)
SegmentationTransformerPrompt EngineeringImage
🎯 What it does: Proposes the CDICS framework, utilizing three types of combination prompts (semantic, parts, and color) to achieve fine-grained controllable context segmentation;
Cell-Type Prototype-Informed Neural Network for Gene Expression Estimation from Pathology Images
Kazuya Nishimura (University of Osaka), Yasuhiro Kojima (National Cancer Center)
Convolutional Neural NetworkTransformerImageBiomedical Data
🎯 What it does: This paper proposes a Cell-Type Prototype-Informed Neural Network (CPNN), which predicts gene expression from pathology images (whole-slide images or spatial transcriptome images) by leveraging cell-type prototypes from public single-cell RNA sequencing data and learning cell composition weights in images.
CF-IPT: Cross-Modal Fusion Interactive Prompt Tuning of Vision-Language Pre-Trained Model for Multisource Remote Sensing Data Classification
Jinheng Ji (Xidian University), Yunsong Li (Xidian University)
ClassificationTransformerPrompt EngineeringVision Language ModelContrastive LearningMultimodality
🎯 What it does: This paper proposes an interactive prompt tuning framework based on CLIP, named CF-IPT, for multi-source remote sensing image classification.
CFG-Ctrl: Control-Based Classifier-Free Diffusion Guidance
Hanyang Wang (Tsinghua University), Yueqi Duan (Tsinghua University)
GenerationTransformerDiffusion modelFlow-based ModelImageMultimodality
🎯 What it does: This paper studies the Classifier-Free Guidance (CFG-Ctrl) framework from the perspective of control theory and proposes the Sliding Mode Control (SMC-CFG) method to enhance semantic alignment and generation quality in flow matching generative models.
CG-Floor: Centroid-Guided Diffusion for Large-Scale Floorplan Generation
Hongjin Lian (Tianjin University), Kun Li (Cardiff University)
GenerationTransformerDiffusion modelAuto EncoderImageTextGraph
🎯 What it does: Designed and implemented a system capable of generating large-scale, semantically consistent, and editable floor plans from text or scene graphs, with support for 3D conversion.
CG-Reasoner: Centroid-Guided Positional Reasoning Segmentation for Medical Imaging with a Robust Visual-Text Consistency Metric
Lakshmikar Reddy Polamreddy, Ming Ma (Yeshiva University)
SegmentationExplainability and InterpretabilityConvolutional Neural NetworkLarge Language ModelSupervised Fine-TuningVision Language ModelMultimodalityBiomedical DataMagnetic Resonance ImagingComputed TomographyUltrasound
🎯 What it does: Proposed a unified cross-modal framework CG-Reasoner that can simultaneously perform medical image segmentation and localization reasoning, achieving interpretable diagnostic reports through the fusion of visual and language information.
CGHair: Compact Gaussian Hair Reconstruction with Card Clustering
Haimin Luo (Carnegie Mellon University), Fernando De la Torre (Carnegie Mellon University)
GenerationCompressionGaussian SplattingImage
🎯 What it does: This paper proposes CGHair, a compact pipeline for rapidly reconstructing high-fidelity hair from multi-view images, which utilizes hierarchical clustering to map hair strands to hair cards and shares Gaussian texture codebooks, achieving high-compression 3D Gaussian rendering.
CGL: Advancing Continual GUI Learning via Reinforcement Fine-Tuning
Zhenquan Yao (Harbin Institute Of Technology), Wangmeng Zuo (Harbin Institute Of Technology)
Supervised Fine-TuningReinforcement Learning
🎯 What it does: Proposed a Continual Graphical User Interface Learning (CGL) framework that addresses the problem of adapting to new tasks without forgetting old tasks in GUI continual learning by enhancing the synergy between supervised fine-tuning (SFT) and reinforcement learning (RL).
CGU-Bayes: Causal Graph Uncertainty-Guided Bayesian Inference for Domain Generalization
Naiyu Yin (Lehigh University), Qiang Ji (Rensselaer Polytechnic Institute)
Domain AdaptationImage
🎯 What it does: Propose a Bayesian inference-based causal graph uncertainty quantification framework, CGU-BAYES, for domain generalization prediction.
Chain of Event-Centric Causal Thought for Physically Plausible Video Generation
Zixuan Wang (Sichuan University), Yinjie Lei (Sichuan University)
GenerationTransformerLarge Language ModelPrompt EngineeringDiffusion modelVideoTextBenchmarkPhysics RelatedRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Designed an event-centric physically feasible video generation framework that decomposes complex physical phenomena into causal chain events and generates visual videos through cross-modal prompts.
Chain of World: World Model Thinking in Latent Motion
Fuxiang Yang (Harbin Institute of Technology), Baorui Ma (Li Auto)
Robotic IntelligenceTransformerSupervised Fine-TuningVision-Language-Action ModelAuto EncoderWorld ModelVideoMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Propose the CoWVLA architecture, which combines world models with potential actions, utilizing chain-based world reasoning and structural-motion decoupled potential action representations to achieve continuous dynamic reasoning and terminal keyframe prediction.
Chain-of-Frames: Advancing Video Understanding in Multimodal LLMs via Frame-Aware Reasoning
Sara Ghazanfari (New York University), Siddharth Garg (New York University)
TransformerLarge Language ModelSupervised Fine-TuningVision Language ModelVideoTextMultimodalityChain-of-Thought
🎯 What it does: Proposed the Chain-of-Frames (CoF) reasoning chain, achieving explicit temporal reasoning in single-stage inference by leveraging video frame numbers;
Chain-of-Models Pre-Training: Rethinking Training Acceleration of Vision Foundation Models
Jiawei Fan (Intel Labs China), Anbang Yao (Intel Labs China)
Computational EfficiencyKnowledge DistillationTransformerContrastive LearningImage
🎯 What it does: Propose Chain-of-Models Pre-Training (CoM-PT), achieving lossless accelerated pre-training of visual foundation models through constructing model chains and reverse knowledge transfer.
Chain-of-Thought Guided Multi-Modal Object Re-Identification
Ya Gao (Anhui University), Jin Tang (Anhui University)
RetrievalTransformerLarge Language ModelVision Language ModelContrastive LearningMultimodalityChain-of-Thought
🎯 What it does: Designed and implemented the CoT-ReID framework, leveraging multi-modal chain-of-thought (CoT) text to guide feature extraction, cross-modal consistency, and decision fusion in visual-textual collaborative learning, thereby enhancing multi-modal ReID performance.
CHAL: Causal-guided Hierarchical Anomaly-aware Learning for Moving Infrared Small Target Detection
Weiwei Duan (University of Electronic Science and Technology of China), Mao Ye (University of Electronic Science and Technology of China)
Object DetectionAnomaly DetectionConvolutional Neural NetworkImageVideo
🎯 What it does: Propose a causality-guided hierarchical anomaly-aware framework named CHAL for detecting small moving targets in infrared imagery.
ChangeBridge: Spatiotemporal Image Generation with Multimodal Controls for Remote Senisng
Zhenghui Zhao (Wuhan University), Zhuo Zheng (University of Tokyo)
GenerationData SynthesisTransformerDiffusion modelImageMultimodalityBenchmark
🎯 What it does: Proposed a spatiotemporal image generation model called ChangeBridge, which can generate later scenes based on early remote sensing observations and multi-modal controls (text coordinates, semantic masks, instance layouts);
Changes in Real Time: Online Scene Change Detection with Multi-View Fusion
Chamuditha Jayanga Galappaththige (Queensland University Of Technology Centre For Robotics), Dimity Miller (Queensland University Of Technology Centre For Robotics)
SegmentationPose EstimationGaussian SplattingSimultaneous Localization and MappingVideoPoint Cloud
🎯 What it does: Propose an online scene change detection method that can detect changes in real-time camera streams without requiring pose annotations, unsupervised labels, and achieves change detection through multi-view consistency;
Charge: A Comprehensive Novel View Synthesis Benchmark and Dataset to Bind Them All
Michal Nazarczuk (Huawei Noah's Ark Lab), Eduardo Pérez-Pellitero (Huawei Noah's Ark Lab)
GenerationData SynthesisDepth EstimationNeural Radiance FieldGaussian SplattingImageVideoMultimodalityBenchmark
🎯 What it does: This paper proposes the Charge dataset and corresponding benchmark for evaluating high-quality novel view synthesis algorithms in both static and dynamic scenes, covering four camera configurations: dense, multi-view, sparse, and monocular trajectory;
Chart-FR1: Visual Focus-Driven Fine-Grained Reasoning on Dense Charts
Hongkun Pan (Zhejiang University), Wei Chen (State Key Lab Of Cad Cg Zhejiang University)
Reinforcement LearningVision Language ModelMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Proposed Chart-FR1, a high information density chart fine-grained reasoning model achieved through visual focused chain-of-thought reasoning and reinforcement learning.
ChArtist: Generating Pictorial Charts with Unified Spatial and Subject Control
Shishi Xiao, Gromit Yeuk-Yin Chan
GenerationData SynthesisTransformerDiffusion modelImage
🎯 What it does: Propose ChArtist, which leverages diffusion models combined with spatial (chart skeleton) and subject (reference image) control to automatically generate visualized charts with data authenticity and visual consistency.
ChartNet: A Million-Scale, High-Quality Multimodal Dataset for Robust Chart Understanding
Jovana Kondic (MIT), Rogerio Feris (MIT-IBM Watson AI Lab)
Data SynthesisLarge Language ModelSupervised Fine-TuningVision Language ModelMultimodalityBenchmarkChain-of-Thought
🎯 What it does: This paper introduces the ChartNet dataset, which automatically generates 1.5 million multimodal chart samples containing executable plotting code, rendered images, data tables, natural language summaries, and chain-of-thought (CoT) question-answering using code-driven synthesis and quality filtering techniques. Experiments are conducted on multiple chart understanding tasks based on this dataset.
ChartR: Evaluating Reasoning Accuracy and Robustness in Chart Question Answering
Xiaojun Chen (Shenzhen University), Liang-Jie Zhang (Shenzhen University)
TransformerLarge Language ModelImageTextMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Designed the ChartR benchmark to evaluate the accuracy and robustness of multimodal large language models in chart reasoning chains.
CHEEM: Continual Learning by Reuse, New, Adapt and Skip - A Hierarchical Exploration-Exploitation Approach
Chinmay Savadikar (North Carolina State University), Tianfu Wu (North Carolina State University)
Computational EfficiencyRepresentation LearningMeta LearningNeural Architecture SearchTransformerMixture of ExpertsImage
🎯 What it does: Propose a sample-free class-incremental continual learning framework named CHEEM, leveraging the internal parameter memory of Vision Transformers and external task centroid memory to achieve task-adaptive network structures.
ChimeraLoRA: Multi-Head LoRA-Guided Synthetic Datasets
Hoyoung Kim (POSTECH), Jungseul Ok (POSTECH)
ClassificationData SynthesisTransformerDiffusion modelImage
🎯 What it does: Propose the multi-head LoRA framework ChimeraLoRA, combining class-shared LoRA A and image-specific LoRA B to generate diverse and detail-rich synthetic data under few-sample guidance.
CHIPS: Efficient CLIP Adaptation via Curvature-aware Hybrid Influence-based Data Selection
Xinlin Zhuang (MBZUAI), Imran Razzak (MBZUAI)
ClassificationRetrievalDomain AdaptationTransformerVision Language ModelContrastive LearningMultimodalityBiomedical DataBenchmark
🎯 What it does: This paper proposes an efficient data selection method called CHIPS for the CLIP model, aiming to achieve the effect of continuous pre-training with a small amount of data in vertical domains;
CHIRP dataset: towards long-term, individual-level, behavioral monitoring of bird populations in the wild
Alex Hoi Hang Chan (University of Konstanz), Hemal Naik (Max Planck Institute of Animal Behavior)
RecognitionObject DetectionObject TrackingSegmentationPose EstimationConvolutional Neural NetworkTransformerVideoBenchmark
🎯 What it does: Created the CHIRP dataset and the CORVID identification pipeline for long-term monitoring of individual bird behaviors in the wild.
ChordEdit: One-Step Low-Energy Transport for Image Editing
Liangsi Lu (Guangdong University of Technology), Yang Shi (Guangdong University of Technology)
Image TranslationGenerationComputational EfficiencyImageBenchmark
🎯 What it does: Proposes ChordEdit, a training-free and inversion-free one-step text-to-image editing method that achieves precise editing of fast one-step text-to-image (T2I) models through low-energy control fields.
Choreographing a World of Dynamic Objects
Yanzhe Lyu (Stanford University), Jiajun Wu (Stanford University)
GenerationData SynthesisOptimizationKnowledge DistillationRobotic IntelligenceDiffusion modelScore-based ModelFlow-based ModelGaussian SplattingVideoTextMesh
🎯 What it does: Propose CHORD, a generic 4D generative pipeline capable of dynamically animating scenes with multi-object interactions and generating physically consistent motion trajectories for robot manipulation.
Chorus: Multi-Teacher Pretraining for Holistic 3D Gaussian Scene Encoding
Yue Li (University of Amsterdam), Martin R. Oswald (University of Amsterdam)
SegmentationKnowledge DistillationRepresentation LearningContrastive LearningGaussian SplattingImagePoint Cloud
🎯 What it does: Through multi-teacher pre-training, align the 3D Gaussian Splatting (3DGS) scene encoder with various 2D foundational models, generating transferable panoramic 3D features.
ChronoGS: Disentangling Invariants and Changes in Multi-Period Scenes
Zhongtao Wang (Peking University), Guoping Wang (Peking University)
GenerationGaussian SplattingImageTime SeriesBenchmark
🎯 What it does: Propose a temporally modulated Gaussian representation method called ChronoGS, which can jointly reconstruct multi-period scenes and simultaneously capture invariant structures as well as time-varying geometry and appearance within the same model.
CI-VID: A Coherent Interleaved Text-Video Dataset
Yiming Ju (Beijing Academy Of Artificial Intelligence), Tengfei Pan (Beijing Academy Of Artificial Intelligence)
TransformerSupervised Fine-TuningDiffusion modelVideoTextMultimodality
🎯 What it does: This paper proposes and releases the CI-VID dataset, consisting of over 340k samples. Each sample includes 3.1 video clips along with interleaved textual descriptions, achieving semantic coherence across edits and entity consistency in text-video sequences.