CVPR 2025 Papers — Page 5
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2871 papers
CoMatcher: Multi-View Collaborative Feature Matching
Jintao Zhang (Wuhan University), Xianwei Zheng (Chinese Academy of Sciences)
Object DetectionPose EstimationRetrievalGraph Neural NetworkImageBenchmark
🎯 What it does: Proposes a multi-view collaborative matching method CoMatcher and builds a scalable group matching pipeline.
CoMBO: Conflict Mitigation via Branched Optimization for Class Incremental Segmentation
Kai Fang (Beijing Institute of Technology), Yunchao Wei (University of Birmingham)
SegmentationKnowledge DistillationTransformerImage
🎯 What it does: This paper proposes a conflict mitigation branch optimization method named CoMBO, aimed at addressing the issues of catastrophic forgetting and new class learning conflicts in class-incremental semantic and panoptic segmentation tasks.
ComfyBench: Benchmarking LLM-based Agents in ComfyUI for Autonomously Designing Collaborative AI Systems
Xiangyuan Xue (Shanghai Artificial Intelligence Laboratory), Lei Bai (Shanghai Artificial Intelligence Laboratory)
Large Language ModelAgentic AIBenchmarkRetrieval-Augmented Generation
🎯 What it does: A ComfyBench benchmark was constructed, and ComfyAgent was proposed to automatically design and generate ComfyUI workflows using LLM.
CoMM: A Coherent Interleaved Image-Text Dataset for Multimodal Understanding and Generation
Wei Chen (Hong Kong University of Science and Technology), Long Chen (Hong Kong University of Science and Technology)
GenerationData SynthesisTransformerLarge Language ModelReinforcement LearningVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: We constructed and released CoMM—a high-quality, coherent interleaved text-image dataset—and proposed four new benchmarks for interleaved text-image generation and evaluation.
Common3D: Self-Supervised Learning of 3D Morphable Models for Common Objects in Neural Feature Space
Leonhard Sommer (University of Freiburg), Adam Kortylewski (Max Planck Institute for Informatics)
SegmentationPose EstimationNeural Radiance FieldContrastive LearningVideo
🎯 What it does: A fully self-supervised 3D deformation model called Common3D has been developed, which can learn a general 3DMM from ordinary object videos, enabling 3D shape, pose estimation, instance segmentation, and semantic correspondence from a single image.
Commonsense Video Question Answering through Video-Grounded Entailment Tree Reasoning
Huabin Liu (Shanghai Jiao Tong University), Cees G. M. Snoek (University of Amsterdam)
Large Language ModelPrompt EngineeringVision Language ModelVideoText
🎯 What it does: A video-supported entailment tree reasoning framework is proposed to explain and answer common-sense video question-and-answer problems.
Community Forensics: Using Thousands of Generators to Train Fake Image Detectors
Jeongsoo Park (University of Michigan), Andrew Owens (University of Michigan)
ClassificationData SynthesisDiffusion modelImage
🎯 What it does: A Community Forensics dataset was constructed, containing 4803 types of generative models, 2.7 million synthetic images, and corresponding real images, and this dataset was used to train an image classifier to detect AI-generated images.
Compass Control: Multi Object Orientation Control for Text-to-Image Generation
Rishubh Parihar (Indian Institute of Information Technology Hyderabad), Venkatesh Babu Radhakrishnan
GenerationData SynthesisPose EstimationDiffusion modelImage
🎯 What it does: The Compass Control method is proposed, utilizing text prompts and the directional angles of each object to achieve precise 3D pose control of multi-object scenes by creating a special 'compass token';
CompGS: Unleashing 2D Compositionality for Compositional Text-to-3D via Dynamically Optimizing 3D Gaussians
Chongjian Ge (University of California), Wei Zhan (Stability AI)
GenerationData SynthesisOptimizationDiffusion modelGaussian SplattingImageTextBenchmark
🎯 What it does: This paper proposes a text-driven combined 3D generation framework called COMPGS based on 3D Gaussian splatting, which can generate multi-object, interactive 3D scenes from a single text description.
Complementary Advantages: Exploiting Cross-Field Frequency Correlation for NIR-Assisted Image Denoising
Yuchen Wang (Beijing Institute of Technology), Hua Huang (Beijing Normal University)
RestorationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes an image denoising network called FCENet, which utilizes near-infrared (NIR) images to assist in achieving high-quality denoising of RGB images through frequency domain information fusion.
Completion as Enhancement: A Degradation-Aware Selective Image Guided Network for Depth Completion
Zhiqiang Yan (Nanjing University of Science and Technology), Jian Yang (Nanjing University of Science and Technology)
RestorationDepth EstimationConvolutional Neural NetworkImage
🎯 What it does: This work redefines the sparse depth completion task as depth enhancement, first generating coarse depth using non-CNN densification tools, and then obtaining fine depth through self-supervised degradation bridging.
Complexity Experts are Task-Discriminative Learners for Any Image Restoration
Eduard Zamfir (University of Wurzburg), Radu Timofte
RestorationTransformerMixture of ExpertsImage
🎯 What it does: A multi-task image restoration model MoCE-IR is proposed, which dynamically allocates computational resources and implements integrated restoration using variable complexity expert blocks.
Composing Parts for Expressive Object Generation
Harsh Rangwani (Adobe Research), Srikrishna Karanam (Adobe Research)
SegmentationGenerationDiffusion modelImage
🎯 What it does: This paper presents PartComposer, a training-free component-level image generation method that can generate object images based on fine-grained component attributes described in text (such as color and style).
Compositional Caching for Training-free Open-vocabulary Attribute Detection
Marco Garosi (University of Trento), Massimiliano Mancini (University of Trento)
Object DetectionRetrievalTransformerLarge Language ModelVision Language ModelImageText
🎯 What it does: A training-free, open-vocabulary attribute detection method called COMCA is proposed, which constructs a scalable cache using attribute-object compatibility and enhances the predictions of VLM through soft label correction.
Compositional Targeted Multi-Label Universal Perturbations
Hassan Mahmood (Northeastern University), Ehsan Elhamifar (Northeastern University)
OptimizationAdversarial AttackImage
🎯 What it does: A combinatorial-based multi-label target universal perturbation generation framework is proposed, capable of synthesizing attack perturbations for any label combination in linear time;
Comprehensive Information Bottleneck for Unveiling Universal Attribution to Interpret Vision Transformers
Jung-Ho Hong (Korea University), Seong-Whan Lee (Korea University)
Explainability and InterpretabilityTransformerImage
🎯 What it does: The CoIBA method is proposed, which achieves a more comprehensive and global feature importance attribution by sharing information bottlenecks across multiple layers of Vision Transformer and using variational upper bounds.
Comprehensive Relighting: Generalizable and Consistent Monocular Human Relighting and Harmonization
Junying Wang (University of Southern California), Jae Shin Yoon
Image TranslationImage HarmonizationRestorationDiffusion modelImageVideo
🎯 What it does: A universal and consistent human monocular relighting and background harmonization framework (Comprehensive Relighting) is proposed, capable of achieving high-quality relighting under arbitrary body parts, lighting control, and background scenes.
ComRoPE: Scalable and Robust Rotary Position Embedding Parameterized by Trainable Commuting Angle Matrices
Hao Yu (Tsinghua University), Chun Yuan (Shenzhen University)
ClassificationObject DetectionTransformerImage
🎯 What it does: A trainable and exchangeable angle matrix rotation position encoding (ComRoPE) is proposed, achieving scalability and robustness for Transformer position encoding.
Concept Lancet: Image Editing with Compositional Representation Transplant
Jinqi Luo (University of Pennsylvania), Rene Vidal
Image TranslationGenerationVision Language ModelDiffusion modelImageBenchmark
🎯 What it does: A zero-shot, pluggable concept translation framework called CoLan is proposed, which utilizes sparse decomposition techniques to implant concepts into the latent space of diffusion models for image editing.
Concept Replacer: Replacing Sensitive Concepts in Diffusion Models via Precision Localization
Lingyun Zhang (Fudan University), Ping Chen (Fudan University)
SegmentationGenerationData SynthesisDiffusion modelImage
🎯 What it does: This paper proposes a framework for accurately locating and replacing sensitive concepts during the generation process of diffusion models, utilizing few-shot learning for concept localization, and then regenerating images with replacement concepts in localized areas.
ConceptGuard: Continual Personalized Text-to-Image Generation with Forgetting and Confusion Mitigation
Zirun Guo (Zhejiang University), Tao Jin (Zhejiang University)
GenerationData SynthesisDiffusion modelImageText
🎯 What it does: This paper presents ConceptGuard, aimed at addressing the issues of concept forgetting and concept confusion in the process of generating images from continuous text.
Condensing Action Segmentation Datasets via Generative Network Inversion
Guodong Ding (National University of Singapore), Angela Yao (National University of Singapore)
SegmentationCompressionKnowledge DistillationTransformerAuto EncoderGenerative Adversarial NetworkVideo
🎯 What it does: A distillation framework for temporal action segmentation datasets based on generative network inversion is proposed, which can compress long video segments into low-dimensional latent codes and further reduce storage through multi-instance segmentation and diversity sequence sampling.
Conditional Balance: Improving Multi-Conditioning Trade-Offs in Image Generation
Nadav Z. Cohen (Reichman University), Ariel Shamir (Reichman University)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: A conditional balancing method based on sensitivity analysis of attention layers is proposed, allowing for better balance between content and style in image generation under multiple conditions.
Conformal Prediction and MLLM aided Uncertainty Quantification in Scene Graph Generation
Sayak Nag (University of California), Amit K. Roy-Chowdhury (University of California)
GenerationExplainability and InterpretabilityGraph Neural NetworkLarge Language ModelImageMultimodality
🎯 What it does: A model-agnostic uncertainty quantification framework for scene graph generation based on conformal prediction, PC-SGG, is proposed. It utilizes a large language model for interpretability filtering of the prediction set, generating a diverse and reliable set of scene graph predictions.
Conformal Prediction for Zero-Shot Models
Julio Silva-Rodríguez (École de Technologie Supérieure), Jose Dolz (École de Technologie Supérieure)
ClassificationDomain AdaptationTransformerVision Language ModelImageMultimodality
🎯 What it does: This study investigates how to apply split conformal prediction to large-scale pre-trained vision-language models (such as CLIP) and proposes an unsupervised transfer learning framework based on optimal transport, named Conf-OT, to address the distribution shift between the source and target domains.
Conical Visual Concentration for Efficient Large Vision-Language Models
Long Xing (University of Science and Technology of China), Dahua Lin (The Chinese University of Hong Kong)
CompressionComputational EfficiencyTransformerVision Language ModelMultimodality
🎯 What it does: PyramidDrop is proposed, a method for progressively reducing visual tokens in large visual language models by layers;
ConMo: Controllable Motion Disentanglement and Recomposition for Zero-Shot Motion Transfer
Jiayi Gao (Peking University), Yang Liu (Peking University)
GenerationData SynthesisDiffusion modelVideoText
🎯 What it does: The ConMo framework is proposed to achieve zero-shot controllable motion transfer by separating the subject and camera motion and reassembling them to control motion in multi-subject videos.
Consistency Posterior Sampling for Diverse Image Synthesis
Vishal Purohit (Purdue University), Xiuyuan Cheng (Duke University)
RestorationGenerationData SynthesisSuper ResolutionDiffusion modelImageStochastic Differential Equation
🎯 What it does: A posterior sampling method based on Langevin dynamics in the noise space of pre-trained generative models is proposed, which can efficiently generate diverse posterior samples.
Consistency-aware Self-Training for Iterative-based Stereo Matching
Jingyi Zhou (Fudan University), Tao Chen (Fudan University)
Depth EstimationDomain AdaptationOptical FlowImage
🎯 What it does: A framework based on consistency self-training is proposed to enhance the performance of iterative stereo matching models using unlabeled data.
Consistent and Controllable Image Animation with Motion Diffusion Models
Xin Ma (Monash University), Cunjian Chen (Monash University)
GenerationData SynthesisDiffusion modelVideo
🎯 What it does: This work proposes Cinemo, a diffusion model-based image animation method aimed at generating animated videos that maintain the appearance consistency of the input image while ensuring smooth motion.
Consistent Normal Orientation for 3D Point Clouds via Least Squares on Delaunay Graph
Rao Fu (Nanyang Technological University), Liang Yu (Alibaba Group)
OptimizationPoint Cloud
🎯 What it does: This paper proposes a globally consistent surface normal vector orientation method, which achieves the transformation from unoriented normal vectors to consistently oriented normal vectors by solving a least-squares signature field on a Delaunay graph constructed based on the shape diameter function.
Context-Aware Multimodal Pretraining
Karsten Roth (Tuebingen AI Center), Olivier J. Henaff (Google DeepMind)
Representation LearningTransformerContrastive LearningImageMultimodality
🎯 What it does: This paper proposes Context-Aware Multimodal Pretraining (LIxP), which incorporates cross-attention contextual buffering into contrastive pretraining, enabling visual-text models to perform metric-based few-shot adaptation without additional training after training.
ConText-CIR: Learning from Concepts in Text for Composed Image Retrieval
Eric Xing (Washington University in St. Louis), Nathan Jacobs (Washington University in St. Louis)
RetrievalTransformerContrastive LearningImageText
🎯 What it does: The ConText-CIR framework is proposed, which trains with Text Concept-Consistency loss and achieves efficient image-text combined retrieval through cross-modal attention.
Context-Enhanced Memory-Refined Transformer for Online Action Detection
Zhanzhong Pang (National University of Singapore), Angela Yao (National University of Singapore)
RecognitionTransformerVideo
🎯 What it does: To address the frame representation learning bias caused by the inconsistency between training and inference in Online Action Detection (OAD), a new Transformer framework called Context-Enhanced Memory-Refined Transformer (CMeRT) is proposed.
Contextual AD Narration with Interleaved Multimodal Sequence
Hanlin Wang (Nanjing University), Limin Wang (Nanjing University)
RecognitionGenerationTransformerLarge Language ModelContrastive LearningVideoTextMultimodalityAudio
🎯 What it does: A unified framework called Uni-AD is proposed, which automatically generates audio descriptions (AD) using interleaved multimodal sequences and pre-trained LLMs.
Continual SFT Matches Multimodal RLHF with Negative Supervision
Ke Zhu (Nanjing University), Jingdong Wang (Baidu)
Computational EfficiencyReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningVision Language ModelImageTextMultimodality
🎯 What it does: Proposes a negative supervision fine-tuning (nSFT) method that transforms negative feedback from RLHF into SFT loss using a single model, thereby efficiently aligning visual language models.
Continuous 3D Perception Model with Persistent State
Qianqian Wang (University of California), Angjoo Kanazawa (University of California)
Pose EstimationDepth EstimationAutonomous DrivingTransformerSimultaneous Localization and MappingImageVideoPoint Cloud
🎯 What it does: This paper proposes an online continuously updatable 3D perception model that utilizes persistent states to simultaneously predict camera pose and dense 3D point clouds while receiving RGB image streams, achieving continuous dense reconstruction.
Continuous Adverse Weather Removal via Degradation-Aware Distillation
Xin Lu (University of Science and Technology of China), Xueyang Fu (University of Science and Technology of China)
RestorationKnowledge DistillationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A gradually learning-based framework for adverse weather removal, ILAWR, is proposed, which can continuously improve performance while avoiding catastrophic forgetting as new weather types are added.
Continuous Locomotive Crowd Behavior Generation
Inhwan Bae (Gwangju Institute of Science and Technology), Hae-Gon Jeon (Gwangju Institute of Science and Technology)
GenerationData SynthesisTransformerDiffusion modelImageVideo
🎯 What it does: A continuous crowd behavior generation framework called CrowdES is designed based on diffusion models and state-switching dynamics, capable of automatically generating diverse and realistic crowd trajectories continuously from a single scene image while supporting user interactive control.
Continuous Space-Time Video Resampling with Invertible Motion Steganography
Yuantong Zhang (Wuhan University), Zhenzhong Chen (Wuhan University)
RestorationSuper ResolutionTransformerFlow-based ModelVideo
🎯 What it does: A framework for reversible motion steganography and 3D implicit feature modulation is proposed, enabling spatial and temporal continuous adjustment of video resampling (from low frame rate to high frame rate, resolution transformation).
Continuous, Subject-Specific Attribute Control in T2I Models by Identifying Semantic Directions
Stefan Andreas Baumann (LMU Munich), Björn Ommer (LMU Munich)
GenerationData SynthesisDiffusion modelContrastive LearningImageText
🎯 What it does: This paper proposes a method for fine-grained attribute adjustment of T2I diffusion models by adding fine-grained, continuous, and subject-specific attribute control vectors in the CLIP embedding space of text input.
ControlFace: Harnessing Facial Parametric Control for Face Rigging
Wooseok Jang (Korea University), Seungryong Kim (KAIST)
Image TranslationGenerationConvolutional Neural NetworkDiffusion modelImageVideo
🎯 What it does: ControlFace is proposed, a facial redirection method that utilizes 3DMM rendering to control parameters such as facial pose, expression, and lighting, capable of generating high-quality realistic images without altering identity and details.
Controllable Human Image Generation with Personalized Multi-Garments
Yisol Choi (Korea Advanced Institute of Science and Technology), Jinwoo Shin (Korea Advanced Institute of Science and Technology)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: The BootComp framework is proposed to achieve controllable human image generation under multiple clothing conditions.
Convex Combination Star Shape Prior for Data-driven Image Semantic Segmentation
Xinyu Zhao (Beijing Normal University), Jun Liu (Beijing Normal University)
SegmentationTransformerSupervised Fine-TuningImageMagnetic Resonance Imaging
🎯 What it does: This paper proposes a Convex Combination Star (CCS) shape prior and seamlessly integrates it into data-driven image segmentation networks (such as SAM/SAM2) through a variational model and network module to enhance segmentation accuracy.
Convex Relaxation for Robust Vanishing Point Estimation in Manhattan World
Bangyan Liao (Zhejiang University), Peidong Liu (Westlake University)
OptimizationImage
🎯 What it does: This paper proposes a globally optimal algorithm for vanishing point estimation in a Manhattan world using convex relaxation methods;
CORE4D: A 4D Human-Object-Human Interaction Dataset for Collaborative Object REarrangement
Yun Liu (Tsinghua University), Li Yi (Tsinghua University)
Object DetectionData SynthesisPose EstimationVideoPoint CloudMesh
🎯 What it does: This paper presents CORE4D, a 4D human-object-human interaction dataset aimed at collaborative object rearrangement, covering both real capture and synthetic redirection.
CorrBEV: Multi-View 3D Object Detection by Correlation Learning with Multi-modal Prototypes
Ziteng Xue (Yanshan University), Zhipeng Zhang (Beijing Jiaotong University)
Object DetectionAutonomous DrivingTransformerContrastive LearningMultimodalityPoint Cloud
🎯 What it does: A pluggable CorrBEV framework is designed to enhance the representation capability of multi-view camera 3D detection in occluded scenes by introducing visual and linguistic prior prototypes.
Correcting Deviations from Normality: A Reformulated Diffusion Model for Multi-Class Unsupervised Anomaly Detection
Farzad Beizaee (École Nationale Supérieure de l'Électronique et de ses Applications), Jose Dolz (École Nationale Supérieure de l'Électronique et de ses Applications)
Anomaly DetectionDiffusion modelAuto EncoderImage
🎯 What it does: A rewritten diffusion model DeCo-Diff is proposed, which only corrects abnormal regions while keeping normal areas unchanged in multi-class unsupervised anomaly detection.
Correlative and Discriminative Label Grouping for Multi-Label Visual Prompt Tuning
Lei-Lei Ma (Anhui University), Haifeng Zhao (Anhui University)
ClassificationObject DetectionTransformerPrompt EngineeringMixture of ExpertsImage
🎯 What it does: A multi-label visual prompt tuning framework (ML-VPT) is proposed, which achieves adaptive learning of label-level representations through grouping based on label co-occurrence and distinction, as well as group-aware mixture of experts.
CoSDH: Communication-Efficient Collaborative Perception via Supply-Demand Awareness and Intermediate-Late Hybridization
Junhao Xu (Beihang University), Di Huang (Beihang University)
Object DetectionAutonomous DrivingAuto EncoderPoint Cloud
🎯 What it does: The CoSDH framework is proposed, utilizing supply-demand awareness and late-stage mixed fusion to achieve efficient communication for collaborative 3D object detection.
CoSER: Towards Consistent Dense Multiview Text-to-Image Generator for 3D Creation
Bonan Li (University of Chinese Academy of Sciences), Xinchao Wang (National University of Singapore)
GenerationData SynthesisDiffusion modelImageText
🎯 What it does: This paper proposes CoSER, a multi-view text-to-image generation framework designed to produce high-quality, perspective-consistent 3D assets.
COSMIC: Clique-Oriented Semantic Multi-space Integration for Robust CLIP Test-Time Adaptation
Fanding Huang (Tsinghua University), Zhi Wang (Tsinghua University)
Domain AdaptationGraph Neural NetworkTransformerContrastive LearningImage
🎯 What it does: A training-independent testing adaptation framework called COSMIC is proposed, which integrates multi-granularity features of CLIP and DINOv2 through a Dual Semantics Graph, and utilizes Clique Guided Hyper-class for high-quality filtering and clustering of the cache, significantly improving the zero-shot performance of CLIP in distribution shift scenarios.
COSMOS: Cross-Modality Self-Distillation for Vision Language Pre-training
Sanghwan Kim (Technical University of Munich), Zeynep Akata (Technical University of Munich)
ClassificationSegmentationRetrievalKnowledge DistillationTransformerVision Language ModelContrastive LearningTextMultimodality
🎯 What it does: This paper proposes COSMOS, which utilizes cross-modal self-distillation for pre-training visual language models;
CoSpace: Benchmarking Continuous Space Perception Ability for Vision-Language Models
Yiqi Zhu, Yang Liu
RecognitionObject DetectionData-Centric LearningTransformerVision Language ModelImageMultimodalityBenchmark
🎯 What it does: Designed and constructed the CoSpace benchmark specifically to evaluate the capabilities of visual language models in continuous spatial perception (multi-view continuous images).
CoT-VLA: Visual Chain-of-Thought Reasoning for Vision-Language-Action Models
Qingqing Zhao (NVIDIA), Donglai Xiang (NVIDIA)
Robotic IntelligenceTransformerVision-Language-Action ModelVideoMultimodalityChain-of-Thought
🎯 What it does: CoT-VLA is a model that incorporates visual chain reasoning within the visual-language-action (VLA) framework, first predicting sub-goal images as intermediate reasoning steps, and then generating a sequence of actions to complete the task.
CountLLM: Towards Generalizable Repetitive Action Counting via Large Language Model
Ziyu Yao (Peking University), Lei Li (University of Washington)
RecognitionData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringVideoMultimodality
🎯 What it does: A framework for counting repeated actions based on large language models (LLM) called CountLLM is proposed, treating the repeated action counting task as video question answering (VQA). It extracts features through a video encoder, generates periodic representations using a periodic transformer, and inputs them along with explicit periodic text prompts into a pre-trained LLM to generate counting results.
COUNTS: Benchmarking Object Detectors and Multimodal Large Language Models under Distribution Shifts
Jiansheng Li (Tsinghua University), Peng Cui (Tsinghua University)
Object DetectionDomain AdaptationTransformerLarge Language ModelSupervised Fine-TuningImageMultimodalityBenchmark
🎯 What it does: A large-scale, fine-grained annotated OOD dataset called COUNTS has been constructed, and based on it, two evaluation benchmarks, O(OD)2 and OODG, have been proposed to assess the generalization ability of object detectors and multimodal large language models under natural distribution shifts.
CPath-Omni: A Unified Multimodal Foundation Model for Patch and Whole Slide Image Analysis in Computational Pathology
Yuxuan Sun (Zhejiang University), Lin Yang (Westlake University)
ClassificationSegmentationRetrievalTransformerLarge Language ModelContrastive LearningImageTextMultimodalityBiomedical Data
🎯 What it does: CPath‑Omni is proposed, a unified multimodal foundation model capable of simultaneously handling classification, VQA, annotation, and visual reference tasks for both pathological patches and whole slide images (WSI).
Crab: A Unified Audio-Visual Scene Understanding Model with Explicit Cooperation
Henghui Du (Renmin University of China), Di Hu (Renmin University of China)
RecognitionObject DetectionSegmentationData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelVideoMultimodalityAudio
🎯 What it does: A unified audio-video scene understanding model, Crab, is proposed, capable of simultaneously working on various audio-video tasks such as temporal localization, spatial localization, spatiotemporal reasoning, and pixel-level understanding.
CraftsMan3D: High-fidelity Mesh Generation with 3D Native Diffusion and Interactive Geometry Refiner
Weiyu Li (Hong Kong University of Science and Technology), Xiaoxiao Long (Hong Kong University of Science and Technology)
GenerationData SynthesisTransformerDiffusion modelPoint CloudMesh
🎯 What it does: CraftsMan3D proposes a two-stage 3D generation framework, first using a 3D native DiT model to generate a rough mesh from a single image or text, and then adding high detail and supporting hand-drawn editing with a normal-based interactive geometric refinement module.
Creating Your Editable 3D Photorealistic Avatar with Tetrahedron-constrained Gaussian Splatting
Hanxi Liu (Peking University), Zhouhui Lian (Peking University)
GenerationData SynthesisGaussian SplattingVideo
🎯 What it does: Proposes the TetGS model for generating editable 3D full-body avatars from monocular videos.
CRISP: Object Pose and Shape Estimation with Test-Time Adaptation
Jingnan Shi (Massachusetts Institute of Technology), Luca Carlone (Massachusetts Institute of Technology)
Pose EstimationDomain AdaptationTransformerImage
🎯 What it does: A unified framework named CRISP has been designed and implemented, capable of simultaneously estimating the 6D pose and implicit 3D shape of objects from a single RGB-D image without category information, and an adaptive self-training method CRISP ST has been proposed to correct domain shifts during testing.
Critic-V: VLM Critics Help Catch VLM Errors in Multimodal Reasoning
Di Zhang (Fudan University), Dongzhan Zhou (Shanghai Artificial Intelligence Laboratory)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningVision Language ModelTextMultimodality
🎯 What it does: The Critic-V framework is proposed, which splits the reasoning process of visual language models into a Reasoner and an external Critic, iteratively improving the reasoning path through natural language criticism.
CroCoDL: Cross-device Collaborative Dataset for Localization
Hermann Blum (Lamarr Institute for AI and Security), Zuria Bauer (ETH Zurich)
Pose EstimationRetrievalRobotic IntelligenceSimultaneous Localization and MappingMultimodalityBenchmark
🎯 What it does: The CroCoDL dataset has been constructed and released, providing multimodal visual localization data recorded across 10 different scenes and 4 types of devices (robots, handheld smartphones, MR headsets, NavVis scanners), and generating high-precision pseudo Ground-Truth based on LaMAR.
Cropper: Vision-Language Model for Image Cropping through In-Context Learning
Seung Hyun Lee (Google DeepMind), Feng Yang (Google)
SegmentationRetrievalTransformerVision Language ModelImageAgriculture Related
🎯 What it does: This paper proposes the Cropper framework, which utilizes context learning from large-scale visual language models to achieve image cropping.
Cross-Modal 3D Representation with Multi-View Images and Point Clouds
Ziyang Zhou (Xi'an Jiaotong University), Ruofei Zhang (Hong Kong Polytechnic University)
ClassificationGenerationRetrievalRepresentation LearningTransformerContrastive LearningImageMultimodalityPoint Cloud
🎯 What it does: Proposes the OpenView framework, which integrates multi-view images and point clouds to learn a unified 3D representation, applicable for cross-modal retrieval, zero-shot classification, 3D generation, clustering, and other multi-tasks.
Cross-Modal and Uncertainty-Aware Agglomeration for Open-Vocabulary 3D Scene Understanding
Jinlong Li (University of Trento), Nicu Sebe (University of Trento)
SegmentationKnowledge DistillationDiffusion modelContrastive LearningPoint Cloud
🎯 What it does: Fuses features from various 2D foundational models into a 3D point cloud model to achieve cross-modal knowledge distillation and complete open vocabulary 3D scene understanding.
Cross-modal Causal Relation Alignment for Video Question Grounding
Weixing Chen (Sun Yat-sen University), Liang Lin (Sun Yat-sen University)
RecognitionRetrievalOptimizationTransformerContrastive LearningGaussian SplattingVideoMultimodality
🎯 What it does: A cross-modal causal relationship alignment framework (CRA) is proposed, utilizing Gaussian Smoothing Grounding, Cross-Modal Alignment, and Explicit Causal Intervention to achieve weakly supervised video question answering and temporal localization.
Cross-Modal Distillation for 2D/3D Multi-Object Discovery from 2D Motion
Saad Lahlali (Universite Paris-Saclay), Quoc-Cuong Pham (Universite Paris-Saclay)
Object DetectionAutonomous DrivingKnowledge DistillationOptical FlowMultimodalityPoint Cloud
🎯 What it does: This paper proposes a 3D multi-object discovery method DIOD-3D based on 2D motion information, and achieves mutual supervision between 2D and 3D through the cross-modal distillation framework xMOD, enabling unsupervised multi-modal object discovery.
Cross-modal Information Flow in Multimodal Large Language Models
Zhi Zhang (University of Amsterdam), Ekaterina Shutova (University of Amsterdam)
Explainability and InterpretabilityTransformerLarge Language ModelVision Language ModelTextMultimodality
🎯 What it does: By using the attention knockout method in multimodal large language models (MLLM), this study systematically tracks and analyzes the flow of visual and linguistic information across different layers, revealing a two-stage fusion process from vision to language and the hierarchical mechanism of final answer generation.
Cross-Modal Interactive Perception Network with Mamba for Lung Tumor Segmentation in PET-CT Images
Jie Mei (Hunan University), Dong Dai (Tianjin Medical University Cancer Institute and Hospital)
SegmentationImageBiomedical DataComputed TomographyPositron Emission Tomography
🎯 What it does: This study first collected and annotated 21,930 pairs of PET-CT lung images, constructing the publicly available dataset PCLT20K; subsequently, a cross-modal interactive perception network (CIPA) based on the Mamba state space model was proposed for lung tumor segmentation in PET-CT images.
Cross-Rejective Open-Set SAR Image Registration
Shasha Mao (Xidian University), Yimeng Zhang (Xidian University)
ClassificationRecognitionTransformerContrastive LearningImage
🎯 What it does: To address the problem of SAR image registration, a Cross Reject Open Set Registration method (CroR-OSIR) is proposed, which achieves accurate selection and localization of matching points through cross open set classification and supervised contrastive learning in two image domains.
Cross-View Completion Models are Zero-shot Correspondence Estimators
Honggyu An (KAIST), Seungryong Kim (Korea University)
Depth EstimationTransformerImage
🎯 What it does: This paper studies and utilizes the cross-view completion model's cross-attention mechanism, proving that it can directly serve as a zero-shot correspondence estimator.
CrossOver: 3D Scene Cross-Modal Alignment
Sayan Deb Sarkar (Stanford University), Iro Armeni (Stanford University)
RetrievalTransformerContrastive LearningImageTextMultimodalityPoint Cloud
🎯 What it does: A CrossOver framework is proposed for cross-modal 3D scene alignment and retrieval, supporting various modalities such as RGB, point clouds, CAD models, floor plans, and text.
CrossSDF: 3D Reconstruction of Thin Structures From Cross-Sections
Thomas Walker (University of Edinburgh), Oisin Mac Aodha (University of Edinburgh)
GenerationOptimizationNeural Radiance FieldPoint CloudMeshBiomedical DataComputed Tomography
🎯 What it does: Proposes the CrossSDF method, which utilizes 2D cross-sections to generate a 3D Signed Distance Field (SDF) to address the reconstruction problem of sparse planes and thin structures.
CryptoFace: End-to-End Encrypted Face Recognition
Wei Ao (Michigan State University), Vishnu Naresh Boddeti (Michigan State University)
RecognitionSafty and PrivacyComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: This paper presents CryptoFace, a system that implements end-to-end encrypted facial recognition using fully homomorphic encryption.
CSC-PA: Cross-image Semantic Correlation via Prototype Attentions for Single-network Semi-supervised Breast Tumor Segmentation
Zhenhui Ding (Shenzhen University), Jing Qin (Hong Kong Polytechnic University)
SegmentationConvolutional Neural NetworkImageUltrasound
🎯 What it does: A single-network semi-supervised breast ultrasound image segmentation framework CSC-PA is proposed, which enhances segmentation performance by utilizing cross-image semantic correlation.
CTRL-D: Controllable Dynamic 3D Scene Editing with Personalized 2D Diffusion
Kai He (University of Toronto), Igor Gilitschenski (University of Toronto)
GenerationData SynthesisOptimizationDiffusion modelImageVideo
🎯 What it does: The CTRL-D framework simplifies dynamic 3D scene editing into 2D editing on a single image, followed by one-click fine-tuning of InstructPix2Pix and two-stage optimization of deformable 3D Gaussians to achieve controllable local editing.
CTRL-O: Language-Controllable Object-Centric Visual Representation Learning
Aniket Didolkar (Mila - Quebec Artificial Intelligence Institute), Aishwarya Agrawal (Mila - Quebec Artificial Intelligence Institute)
Object DetectionGenerationRepresentation LearningTransformerVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: CTRL-O is proposed, a visual representation learning model that achieves controllable object-level representation through natural language queries;
Cubify Anything: Scaling Indoor 3D Object Detection
Justin Lazarow (Apple), Afshin Dehghan (Apple)
Object DetectionTransformerPoint Cloud
🎯 What it does: A large-scale indoor 3D object detection dataset CA-1M is proposed, along with the CuTR model based on Transformer.
Curriculum Coarse-to-Fine Selection for High-IPC Dataset Distillation
Yanda Chen (Harbin Institute of Technology), Liqiang Nie (Harbin Institute of Technology)
Knowledge DistillationImage
🎯 What it does: This paper proposes a course-based coarse-to-fine selection method named CCFS for ensemble data distillation under high per-class sample (high-IPC) conditions, enhancing the distillation effect by gradually selecting suitable real images combined with synthetic images.
Curriculum Direct Preference Optimization for Diffusion and Consistency Models
Florinel-Alin Croitoru (University of Bucharest), Mubarak Shah (University of Central Florida)
GenerationOptimizationReinforcement Learning from Human FeedbackDiffusion modelImageText
🎯 What it does: For text-to-image generation, a Curriculum Learning-based Direct Preference Optimization method (Curriculum DPO) is proposed, and DPO is transferred to consistency models (Consistency-DPO).
CustAny: Customizing Anything from A Single Example
Lingjie Kong (Fudan University), Yanwei Fu (Fudan University)
Image TranslationGenerationDiffusion modelContrastive LearningImageVideo
🎯 What it does: Proposes the CustAny framework, which can customize any object in a zero-shot manner while maintaining identity consistency, given only a reference image and text prompt;
Customized Condition Controllable Generation for Video Soundtrack
Fan Qi (Tianjin University of Technology), Changsheng Xu (Chinese Academy of Sciences)
GenerationData SynthesisDiffusion modelContrastive LearningVideoAudio
🎯 What it does: This paper proposes a customized conditional video dubbing generation framework based on a latent diffusion model, capable of simultaneously generating music and sound effects that match the video content.
CustomKD: Customizing Large Vision Foundation for Edge Model Improvement via Knowledge Distillation
Jungsoo Lee (Qualcomm AI Research), Fatih Porikli (Qualcomm AI Research)
Domain AdaptationKnowledge DistillationConvolutional Neural NetworkTransformerImage
🎯 What it does: A new knowledge distillation framework called CustomKD is proposed, utilizing large-scale visual foundation models (such as DINOv2 and OpenCLIP) as teachers. Through alternating training of feature customization and distillation, it significantly enhances the performance of edge models (such as MobileNetV3 and WideResNet) in scenarios with unlabeled data (Unsupervised Domain Adaptation UDA and Semi-Supervised Learning SSL).
CXPMRG-Bench: Pre-training and Benchmarking for X-ray Medical Report Generation on CheXpert Plus Dataset
Xiao Wang (Anhui University), Jin Tang (Anhui University)
GenerationConvolutional Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningContrastive LearningImageTextMultimodalityBenchmark
🎯 What it does: A complete benchmark was established on the CheXpert Plus dataset, evaluating 19 mainstream X-ray image report generation models, 14 large language models, and 2 vision-language models, and a multi-stage pre-trained model MambaXray-VL was proposed for X-ray image report generation.
D^2iT: Dynamic Diffusion Transformer for Accurate Image Generation
Weinan Jia (University of Science and Technology of China), Zhendong Mao (Institute of Artificial Intelligence)
GenerationData SynthesisTransformerDiffusion modelAuto EncoderImage
🎯 What it does: A two-stage dynamic compression diffusion generation framework D2iT is designed, which first uses dynamic VAE for adaptive encoding of regions, and then employs a dynamic diffusion Transformer for multi-granularity noise prediction, achieving a unification of local authenticity and global consistency.
D^3-Human: Dynamic Disentangled Digital Human from Monocular Video
Honghu Chen (University of Science and Technology of China), Juyong Zhang (University of Science and Technology of China)
SegmentationGenerationOptimizationVideoMesh
🎯 What it does: The D3-Human method is proposed, which utilizes monocular video for high-quality decoupled reconstruction of dynamic human figures, generating 3D meshes for clothing and the human body separately, and can be directly used for clothing transfer and animation production.
D^3: Scaling Up Deepfake Detection by Learning from Discrepancy
Yongqi Yang (Wuhan University), Yu Wu (Princeton University)
ClassificationAnomaly DetectionConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A deep fake detection framework called D3 is proposed, which utilizes a dual-branch parallel network and image distortion difference signals to learn general fake features.
D^3CTTA: Domain-Dependent Decorrelation for Continual Test-Time Adaption of 3D LiDAR Segmentation
Jichun Zhao (University of Chinese Academy of Sciences), Dong Gong (University of New South Wales)
SegmentationDomain AdaptationAutonomous DrivingPoint Cloud
🎯 What it does: This paper proposes a Continuous Testing Time Adaptation (CTTA) framework DCTTA, designed for LiDAR semantic segmentation, which can online update the model in continuously changing environments.
D2SP: Dynamic Dual-Stage Purification Framework for Dual Noise Mitigation in Vision-based Affective Recognition.
Haoran Wang (Fudan University), Wenqiang Zhang (Fudan University)
RecognitionConvolutional Neural NetworkRecurrent Neural NetworkTransformerImageVideo
🎯 What it does: A dynamic two-stage denoising framework D2SP is proposed, which first removes low-quality samples in a coarse-grained manner and then corrects mislabeling in a fine-grained manner, thereby improving dynamic facial expression recognition performance.
DA-VPT: Semantic-Guided Visual Prompt Tuning for Vision Transformers
Li Ren (University of Central Florida), Kien Hua (University of Central Florida)
ClassificationSegmentationTransformerPrompt EngineeringContrastive LearningImage
🎯 What it does: To achieve parameter-efficient fine-tuning of Vision Transformers, we propose Distribution-Aware Visual Prompt Tuning (DA-VPT).
DaCapo: Score Distillation as Stacked Bridge for Fast and High-quality 3D Editing
Yufei Huang (Zhejiang University), Zhen Lei (Chinese Academy of Sciences)
Image TranslationData SynthesisComputational EfficiencyKnowledge DistillationDiffusion modelScore-based ModelPoint CloudMesh
🎯 What it does: This paper proposes a method for fast and high-quality 3D scene editing using diffusion bridge technology, which can directly transform the source scene into the target scene.
DAGSM: Disentangled Avatar Generation with GS-enhanced Mesh
Jingyu Zhuang (Sun Yat-sen University), Guanbin Li (Sun Yat-sen University)
GenerationDiffusion modelGaussian SplattingMesh
🎯 What it does: A text-based separated digital human generation framework DAGSM is proposed, which first generates a naked human body and then gradually generates various types of clothing, mapping the human body and clothing to GS-enhanced mesh (GS-Mesh) separately, thus achieving high-quality, editable, and animatable digital human models.
DAMM-Diffusion: Learning Divergence-Aware Multi-Modal Diffusion Model for Nanoparticles Distribution Prediction
Junjie Zhou (Nanjing University of Aeronautics and Astronautics), Wei Shao (Nanjing University of Aeronautics and Astronautics)
GenerationData SynthesisDiffusion modelMultimodalityBiomedical Data
🎯 What it does: A diffusion model called DAMM-Diffusion, which integrates single-modal and multi-modal branches, is proposed for predicting the distribution of nanoparticles under tumor microenvironment conditions (vascular and cell nuclei).
DarkIR: Robust Low-Light Image Restoration
Daniel Feijoo (Cidaut AI), Marcos V. Conde (Computer Vision Lab)
RestorationConvolutional Neural NetworkImage
🎯 What it does: A lightweight multi-task network called DarkIR is proposed for simultaneous denoising, deblurring, and low-light enhancement in nighttime/low-light environments.
DART: Disease-aware Image-Text Alignment and Self-correcting Re-alignment for Trustworthy Radiology Report Generation
Sang-Jun Park (Korea University), Tae-Eui Kam (Korea University)
GenerationRetrievalTransformerContrastive LearningImageTextMultimodalityElectronic Health Records
🎯 What it does: The DART framework is proposed to generate reliable radiology reports through disease-aware image-text alignment and self-correcting realignment.
DashGaussian: Optimizing 3D Gaussian Splatting in 200 Seconds
Youyu Chen (Harbin Institute of Technology), Yinyu Nie (Huawei Noah's Ark Lab)
OptimizationGaussian SplattingPoint Cloud
🎯 What it does: This paper presents DashGaussian, a pluggable scheduling scheme that significantly accelerates the optimization of 3D Gaussian Splatting (3DGS) by dynamically adjusting the rendering resolution and the number of Gaussian atoms.
Data Distributional Properties As Inductive Bias for Systematic Generalization
Felipe del Rio (Pontificia Universidad Catolica de Chile), Alvaro Soto (Pontificia Universidad Catolica de Chile)
TransformerMultimodality
🎯 What it does: Exploring how the characteristics of training data distribution (diversity, burstiness, latent intervention) enhance the system generalization of multimodal models.
Data Synthesis with Diverse Styles for Face Recognition via 3DMM-Guided Diffusion
Yuxi Mi (Fudan University), Shuigeng Zhou (Tencent)
RecognitionGenerationData SynthesisDiffusion modelImage
🎯 What it does: This paper proposes a diffusion model named MorphFace, which can generate identity-preserving and stylistically diverse synthetic faces to enhance the diversity of training data for facial recognition models.