ICCV 2025 Papers — Page 4
IEEE/CVF International Conference on Computer Vision · 2701 papers
C4D: 4D Made from 3D through Dual Correspondences
Shizun Wang (National University of Singapore), Xinchao Wang (National University of Singapore)
Object TrackingPose EstimationDepth EstimationOptimizationTransformerOptical FlowVideoPoint Cloud
🎯 What it does: Utilizing monocular video to jointly predict dense 3D point clouds, camera poses, intrinsic parameters, and motion masks, achieving complete 4D dynamic scene reconstruction;
CA-I2P: Channel-Adaptive Registration Network with Global Optimal Selection
Zhixin Cheng (University of Science and Technology of China), Tianzhu Zhang (University of Science and Technology of China)
RecognitionPose EstimationImagePoint Cloud
🎯 What it does: CA-I2P is proposed, a detection-independent image and point cloud registration network that addresses cross-modal channel differences and many-to-one matching problems through channel-level adaptive enhancement and global optimal selection.
CA2C: A Prior-Knowledge-Free Approach for Robust Label Noise Learning via Asymmetric Co-learning and Co-training
Mengmeng Sheng (Nanjing University of Science and Technology), Yazhou Yao (Beijing Institute of Technology)
ClassificationData-Centric LearningImage
🎯 What it does: This paper proposes a robust label noise learning framework CA2C without prior knowledge, utilizing asynchronous co-learning and co-training strategies of two models to achieve adaptive suppression of noise.
CABLD: Contrast-Agnostic Brain Landmark Detection with Consistency-Based Regularization
Soorena Salari (Concordia University), Yiming Xiao (Concordia University)
RecognitionSegmentationConvolutional Neural NetworkContrastive LearningBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A framework for contrast-invariant self-supervised landmark detection, CABLD, is proposed, which requires only a single template annotation to perform detection in unannotated 3D brain MR scans.
CAD-Assistant: Tool-Augmented VLLMs as Generic CAD Task Solvers
Dimitrios Mallis (University of Luxembourg), Djamila Aouada (University of Luxembourg)
Large Language ModelVision Language ModelMultimodality
🎯 What it does: Proposes CAD-Assistant, a general CAD agent based on Vision-Large-Language-Model, which implements multimodal question answering and editing using the FreeCAD API;
CAD-Recode: Reverse Engineering CAD Code from Point Clouds
Danila Rukhovich (University of Luxembourg), Djamila Aouada (University of Luxembourg)
AI Code AssistantTransformerLarge Language ModelPoint Cloud
🎯 What it does: Based on a pre-trained large language model, CAD-Recode is proposed to directly convert point clouds into executable CadQuery Python code, thereby achieving CAD reverse engineering.
CAFA: a Controllable Automatic Foley Artist
Roi Benita (Technion), Yossi Adi (Hebrew University of Jerusalem)
GenerationData SynthesisDiffusion modelVideoTextMultimodalityAudio
🎯 What it does: A controllable automatic Foley artist CAFA has been developed, based on a pre-trained text-to-audio model and injecting video information through a modality adapter, achieving audio generation under dual conditions of video and text.
Calibrating MLLM-as-a-judge via Multimodal Bayesian Prompt Ensembles
Eric Slyman (Adobe Systems), Stefan Lee (Oregon State University)
Recommendation SystemTransformerLarge Language ModelPrompt EngineeringImageTextMultimodality
🎯 What it does: A multi-modal Bayesian prompt ensemble method (MMB) is proposed to enhance the calibration and accuracy of large multi-modal language models (MLLM) in text-to-image evaluation tasks.
CaliMatch: Adaptive Calibration for Improving Safe Semi-supervised Learning
Jinsoo Bae (Korea University), Hyungrok Do (NYU Grossman School of Medicine)
ClassificationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: CaliMatch is proposed, a secure semi-supervised learning framework that calibrates multi-classifiers and OOD detectors through adaptive label smoothing and temperature scaling.
CalliReader: Contextualizing Chinese Calligraphy via an Embedding-Aligned Vision-Language Model
Yuxuan Luo (Peking University), Zhouhui Lian (Peking University)
RecognitionObject DetectionTransformerLarge Language ModelVision Language ModelImageMultimodalityBenchmark
🎯 What it does: A multimodal model called CalliReader has been developed for contextual Chinese calligraphy, aimed at complete page/region recognition, multilingual interpretation, and intent recognition.
CameraCtrl II: Dynamic Scene Exploration via Camera-controlled Video Diffusion Models
Hao He (Chinese University of Hong Kong), Hongsheng Li (Chinese University of Hong Kong)
GenerationData SynthesisTransformerDiffusion modelVideo
🎯 What it does: The CameraCtrl II framework is proposed, which implements a diffusion model based on camera control, capable of continuous and controllable video generation and exploration in dynamic scenes.
Can Generative Geospatial Diffusion Models Excel as Discriminative Geospatial Foundation Models?
Yuru Jia (KU Leuven), Andrea Nascetti (KTH)
ClassificationSegmentationMixture of ExpertsDiffusion modelImageBenchmark
🎯 What it does: Transform the generative diffusion model into a self-supervised remote sensing foundation model, and perform pre-training and fine-tuning on various discriminative tasks.
Can Knowledge be Transferred from Unimodal to Multimodal? Investigating the Transitivity of Multimodal Knowledge Editing
Lingyong Fang (Shanghai Jiao Tong University), Gongshen Liu (Shanghai Jiao Tong University)
TransformerLarge Language ModelSupervised Fine-TuningImageTextMultimodalityBenchmark
🎯 What it does: The concept of Transmissibility in Multimodal Knowledge Editing (TMKE) is proposed, and a corresponding benchmark is constructed to evaluate existing multimodal large models and various knowledge editing methods.
Can We Achieve Efficient Diffusion Without Self-Attention? Distilling Self-Attention into Convolutions
Ziyi Dong (Sun Yat-sen University), Liang Lin (Sun Yat-sen University)
GenerationComputational EfficiencyKnowledge DistillationConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: A method based on Pyramid Convolution Blocks (∆ ConvBlock) is proposed, replacing the original self-attention module with convolution operations, and achieving an efficient diffusion model through knowledge distillation.
Can3Tok: Canonical 3D Tokenization and Latent Modeling of Scene-Level 3D Gaussians
Quankai Gao (University of Southern California), Jae Shin Yoon (Adobe Research)
GenerationData SynthesisTransformerAuto EncoderGaussian SplattingPoint Cloud
🎯 What it does: A scene-level variational autoencoder Can3Tok based on 3D Gaussian splatting is designed to achieve low-dimensional latent encoding and reconstruction of large-scale scenes.
CanFields: Consolidating Diffeomorphic Flows for Non-Rigid 4D Interpolation from Arbitrary-Length Sequences
Miaowei Wang (University of Edinburgh), Amir Vaxman (University of Edinburgh)
OptimizationFlow-based ModelPoint CloudOrdinary Differential Equation
🎯 What it does: This paper proposes a 4D point cloud interpolation method based on deformable flow fields, called CanFields, which can recover continuous and temporally coherent deformation models from independent sampled 3D point cloud sequences of arbitrary lengths, achieving joint optimization of fine geometry and motion.
CanonSwap: High-Fidelity and Consistent Video Face Swapping via Canonical Space Modulation
Xiangyang Luo (Tsinghua University), Shao-Lun Huang (Tsinghua University)
Image TranslationGenerationData SynthesisGenerative Adversarial NetworkOptical FlowImageVideo
🎯 What it does: The CanonSwap framework is proposed, which first maps the target video's face to a canonical space to remove motion information, then performs identity swapping in that space and restores the original motion, achieving high-fidelity and temporally consistent video face replacement.
CaO2: Rectifying Inconsistencies in Diffusion-Based Dataset Distillation
Haoxuan Wang (University of Illinois Chicago), Yan Yan (University of Illinois Chicago)
Knowledge DistillationTransformerDiffusion modelImage
🎯 What it does: A two-stage Diffusion-based dataset distillation framework called CaO2 is proposed to address the issues of target inconsistency and conditional inconsistency in existing methods.
CAP: Evaluation of Persuasive and Creative Image Generation
Aysan Aghazadeh (University of Pittsburgh), Adriana Kovashka (University of Pittsburgh)
GenerationTransformerLarge Language ModelDiffusion modelContrastive LearningImageTextMultimodality
🎯 What it does: Three evaluation metrics for advertising image generation are proposed: creativity, alignment, and persuasiveness. It is suggested to use large language models to expand implicit prompts to enhance the generation quality of text-to-image models.
CapeLLM: Support-Free Category-Agnostic Pose Estimation with Multimodal Large Language Models
Junho Kim (EverEx), Byung-Hoon Kim (Yonsei University)
Pose EstimationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: A framework called CapeLLM is proposed, which can achieve category-independent pose estimation using only query images and text descriptions without any supporting images.
CaptionSmiths: Flexibly Controlling Language Pattern in Image Captioning
Kuniaki Saito (OMRON SINIC X), Yoshitaka Ushiku
GenerationTransformerLarge Language ModelVision Language ModelImageText
🎯 What it does: A single image captioning model called CaptionSmiths is proposed, which can continuously control the length, descriptiveness (information density), and vocabulary uniqueness of captions by inserting interpolable conditional vectors into the input of the language model.
CAPTURE: Evaluating Spatial Reasoning in Vision Language Models via Occluded Object Counting
Atin Pothiraj (University of North Carolina Chapel Hill), Mohit Bansal (University of North Carolina Chapel Hill)
Object DetectionTransformerVision Language ModelDiffusion modelImageMultimodalityBenchmark
🎯 What it does: This paper proposes the CAPTURE benchmark to evaluate the ability of visual language models to count modal patterns in occluded scenes.
Capturing head avatar with hand contacts from a monocular video
Haonan He (Hong Kong University of Science and Technology), Jie Song (Hong Kong University of Science and Technology)
GenerationPose EstimationDepth EstimationVideo
🎯 What it does: This paper proposes a high-fidelity 3D avatar framework that can simultaneously reconstruct the full head appearance and the interaction between the hands and face from monocular video.
CarGait: Cross-Attention based Re-ranking for Gait recognition
Gavriel Habib (OriginAI), Nir Darshan (OriginAI)
RecognitionRetrievalConvolutional Neural NetworkTransformerContrastive LearningImage
🎯 What it does: Building on existing single-stage gait recognition models, CarGait is proposed to achieve bidirectional interaction between the probe and candidate fine-grained gait strips through cross-attention, enabling the re-ranking of the top K results.
CARIM: Caption-Based Autonomous Driving Scene Retrieval via Inclusive Text Matching
Minjoo Ki (Yonsei University), Jinhan Lee (Naver Labs)
RetrievalAutonomous DrivingLarge Language ModelVision Language ModelContrastive LearningVideoText
🎯 What it does: The CARIM method is proposed to achieve self-driving scene retrieval based on text matching, capable of retrieving videos that meet all query conditions.
CARL: Causality-guided Architecture Representation Learning for an Interpretable Performance Predictor
Han Ji (Sichuan University), Yanan Sun (Sichuan University)
Explainability and InterpretabilityRepresentation LearningNeural Architecture SearchGraph Neural NetworkImage
🎯 What it does: This paper proposes a causal intervention-based architecture representation learning method called CARL, which enhances the generalization and interpretability of performance predictors in neural architecture search by splitting key and redundant features in the latent space and generating cross-intervention samples.
CARP: Visuomotor Policy Learning via Coarse-to-Fine Autoregressive Prediction
Zhefei Gong (Westlake University), Donglin Wang (Westlake University)
Robotic IntelligenceTransformerReinforcement LearningDiffusion modelImage
🎯 What it does: This paper proposes a Coarse-to-Fine Autoregressive Policy (CARP) for robot visual motion planning, achieving efficient and precise action sequence generation through multi-scale action discretization and hierarchical autoregressive prediction.
CasP: Improving Semi-Dense Feature Matching Pipeline Leveraging Cascaded Correspondence Priors for Guidance
Peiqi Chen (Wuhan University), Yongjun Zhang (Wuhan University)
Pose EstimationComputational EfficiencyTransformerImage
🎯 What it does: This paper proposes CasP—a semi-dense feature matching pipeline based on cascade correspondence priors, aimed at improving matching speed and robustness.
Cassic: Towards Content-Adaptive State-Space Models for Learned Image Compression
Shiyu Qin (Tsinghua University), Yaowei Wang (Harbin Institute of Technology)
CompressionConvolutional Neural NetworkImage
🎯 What it does: A learning-based image compression framework called Cassic has been developed, which is based on a content-adaptive visual state space model, improving the scanning order and entropy model.
CAT: A Unified Click-and-Track Framework for Realistic Tracking
Yongsheng Yuan (Dalian University of Technology), Huchuan Lu (Dalian University of Technology)
Object TrackingTransformerMixture of ExpertsVideo
🎯 What it does: A unified click-tracking framework CAT is proposed, which utilizes a single point click to complete target initialization and achieve continuous tracking.
Category-Specific Selective Feature Enhancement for Long-Tailed Multi-Label Image Classification
Ruiqi Du (Xidian University), Jingjing Ma (Xidian University)
ClassificationRecognitionTransformerImage
🎯 What it does: This paper proposes a framework based on Category-Specific Selective Feature Enhancement (CSSFE) for the long-tail multi-label image classification task, aiming to improve the representation and recognition capabilities of rare categories.
CATP-LLM: Empowering Large Language Models for Cost-Aware Tool Planning
Duo Wu (Tsinghua University), Zhi Wang (Tsinghua University)
OptimizationTransformerLarge Language ModelReinforcement LearningText
🎯 What it does: Developed the CATP-LLM framework, which supports LLM for cost-aware tool planning and can generate tool plans for non-sequential parallel execution.
CATSplat: Context-Aware Transformer with Spatial Guidance for Generalizable 3D Gaussian Splatting from A Single-View Image
Wonseok Roh (Korea University), Sangpil Kim (Korea University)
RestorationGenerationDepth EstimationTransformerVision Language ModelGaussian SplattingImageVideoPoint Cloud
🎯 What it does: This paper proposes a Transformer framework called CATSplat that utilizes text and spatial priors to achieve generalizable 3D scene reconstruction and novel view synthesis from a single-view image through 3D Gaussian splatting.
Causal Disentanglement and Cross-Modal Alignment for Enhanced Few-Shot Learning
Tianjiao Jiang (Australian Institute for Machine Learning, University of Adelaide), Javen Qinfeng Shi (Australian Institute for Machine Learning, University of Adelaide)
ClassificationDomain AdaptationMeta LearningTransformerContrastive LearningImageMultimodality
🎯 What it does: Proposes the Causal CLIP Adapter (CCA), which utilizes ICA for causal de-mixing of CLIP features and enhances few-shot learning performance through unidirectional and bidirectional cross-modal alignment.
Causal-Entity Reflected Egocentric Traffic Accident Video Synthesis
Lei-Lei Li (Xi'an Jiaotong University), Tat-Seng Chua (National University of Singapore)
GenerationData SynthesisAutonomous DrivingDiffusion modelVideo
🎯 What it does: This paper proposes a diffusion model called Causal-VidSyn for synthesizing first-person traffic accident videos.
Causality-guided Prompt Learning for Vision-language Models via Visual Granulation
Mengyu Gao (Chinese Academy of Sciences), Qiulei Dong (Chinese Academy of Sciences)
ClassificationRecognitionDomain AdaptationTransformerPrompt EngineeringVision Language ModelDiffusion modelContrastive LearningImageMultimodality
🎯 What it does: This paper proposes CaPL, a causal-guided text prompt learning method that enhances CLIP's performance on fine-grained tasks by utilizing visual fine-grained decomposition and visual granulation techniques.
CAVIS: Context-Aware Video Instance Segmentation
Seunghun Lee (Daegu Gyeongbuk Institute of Science and Technology), Sunghoon Im (Daegu Gyeongbuk Institute of Science and Technology)
Object TrackingSegmentationTransformerContrastive LearningVideo
🎯 What it does: This paper proposes the Context-Aware Video Instance Segmentation (CAVIS) framework, which significantly improves the accuracy of video instance segmentation and tracking by integrating contextual information around the target with core features.
CC-OCR: A Comprehensive and Challenging OCR Benchmark for Evaluating Large Multimodal Models in Literacy
Zhibo Yang (Huazhong University of Science and Technology), Junyang Lin (Alibaba Group)
RecognitionTransformerVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: Designed and released the CC-OCR benchmark to evaluate the capabilities of large multimodal models in OCR-related tasks (multi-scene text reading, multilingual text reading, document parsing, key information extraction);
CCL-LGS: Contrastive Codebook Learning for 3D Language Gaussian Splatting
Lei Tian (Dalian University of Technology), Xu Jia (Dalian University of Technology)
SegmentationGenerationContrastive LearningGaussian SplattingPoint Cloud
🎯 What it does: Construct a perspective-consistent 3D language Gaussian splatting semantic field to address the issue of semantic inconsistency across perspectives.
CCMNet: Leveraging Calibrated Color Correction Matrices for Cross-Camera Color Constancy
Dongyoung Kim (Yonsei University), Seon Joo Kim (Yonsei University)
Image TranslationDomain AdaptationConvolutional Neural NetworkImage
🎯 What it does: CCMNet is proposed, which utilizes the pre-calibrated color correction matrix (CCM) of the camera ISP for cross-camera color constancy estimation.
CE-FAM: Concept-Based Explanation via Fusion of Activation Maps
Michihiro Kuroki (University of Tokyo), Toshihiko Yamasaki (University of Tokyo)
Explainability and InterpretabilityVision Language ModelImage
🎯 What it does: A concept-level explainable method CE-FAM is proposed, which can simultaneously identify the concepts learned by image classification models, locate their corresponding areas, and quantify their contributions to predictions.
Certifiably Optimal Anisotropic Rotation Averaging
Carl Olsson (Lund University), Christopher Zach (Chalmers University of Technology)
Pose EstimationOptimizationPoint CloudStochastic Differential Equation
🎯 What it does: This paper proposes a provably optimal rotation averaging method under anisotropic uncertainty, which explicitly incorporates the confidence information of each relative rotation into the optimization objective.
CF3: Compact and Fast 3D Feature Fields
Hyunjoon Lee (Seoul National University), Jaesik Park (Seoul National University)
SegmentationCompressionAutonomous DrivingAuto EncoderGaussian SplattingPoint Cloud
🎯 What it does: A pipeline for constructing compact feature fields based on 3D Gaussian point clouds is proposed, which combines multi-view feature enhancement, low-dimensional AutoEncoder compression, and adaptive sparsification to achieve high-quality 3D feature representation with extremely low storage requirements.
CharaConsist: Fine-Grained Consistent Character Generation
Mengyu Wang (Beijing Jiaotong University), Yunchao Wei
GenerationData SynthesisTransformerDiffusion modelImageText
🎯 What it does: A training-free consistency text-to-image generation method called CharaConsist is proposed, which maintains fine-grained consistency between characters and backgrounds across different scenes and actions.
CHARM3R: Towards Unseen Camera Height Robust Monocular 3D Detector
Abhinav Kumar (Michigan State University), Xiaoming Liu (Bosch Research North America)
Object DetectionDepth EstimationAutonomous DrivingPoint Cloud
🎯 What it does: This paper studies the robustness of monocular 3D detection in the absence of seen camera height (ego height) and proposes the CHARM3R model based on average depth estimation, significantly improving detection performance at different heights.
ChartCap: Mitigating Hallucination of Dense Chart Captioning
Junyoung Lim (Seoul National University), Gunhee Kim (Seoul National University)
Large Language ModelSupervised Fine-TuningVision Language ModelImageText
🎯 What it does: A large-scale real-world chart dataset, ChartCap (565K pairs), was constructed, and a hallucination-free dense chart description method was proposed, supplemented by the Visual Consistency Score evaluation metric.
ChartPoint: Guiding MLLMs with Grounding Reflection for Chart Reasoning
Zhengzhuo Xu (Tsinghua University), Jian Guo (Hong Kong University of Science and Technology)
TransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageMultimodalityChain-of-Thought
🎯 What it does: This paper proposes the PointCoT approach, integrating positional reflection into multimodal chain-of-thought reasoning, and constructs the ChartPoint-SFT-62k dataset, training two models, ChartPointQ2/Q2.5, which perform outstandingly on chart reasoning tasks.
ChatReID: Open-ended Interactive Person Retrieval via Hierarchical Progressive Tuning for Vision Language Models
Ke Niu (Fudan University), Xiangyang Xue (Fudan University)
RetrievalTransformerSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: ChatReID is proposed, a text-centric multimodal retrieval framework that improves fine-grained matching and reasoning in the person re-identification (Re-ID) task using large-scale instruction data and staged fine-tuning (HPT).
Chimera: Improving Generalist Model with Domain-Specific Experts
Tianshuo Peng (Chinese University of Hong Kong), Xiangyu Yue
Domain AdaptationOptimizationTransformerLarge Language ModelMixture of ExpertsMultimodalityTabular
🎯 What it does: Construct a Chimera system that integrates specialized expert models with general multimodal models to achieve unified reasoning and structural extraction in areas such as charts, tables, mathematics, and documents.
CHORDS: Diffusion Sampling Accelerator with Multi-core Hierarchical ODE Solvers
Jiaqi Han (Stanford University), Stefano Ermon (Stanford University)
GenerationComputational EfficiencyDiffusion modelImageVideoOrdinary Differential Equation
🎯 What it does: A multi-core parallel diffusion sampling acceleration framework called CHORDS is proposed, utilizing a multi-core hierarchical ODE solver to achieve training-free, model-independent inference acceleration.
CHROME: Clothed Human Reconstruction with Occlusion-Resilience and Multiview-Consistency from a Single Image
Arindam Dutta, Ziyan Wu
RestorationGenerationDiffusion modelGaussian SplattingImage
🎯 What it does: This paper studies a single-view occluded human reconstruction method called CHROME, based on a multi-view diffusion model and 3D Gaussian fitting.
CIARD: Cyclic Iterative Adversarial Robustness Distillation
Liming Lu (Nanjing University of Science and Technology), Yongbin Zhou (Nanjing University of Science and Technology)
Knowledge DistillationAdversarial AttackConvolutional Neural NetworkGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: By using multi-teacher adversarial distillation, the robustness and accuracy of the teacher model are transferred to a lightweight student model, proposing the CIARD framework.
CityGS-X: A Scalable Architecture for Efficient and Geometrically Accurate Large-Scale Scene Reconstruction
Yuanyuan Gao (Northwestern Polytechnical University), Junwei Han (Shanghai Artificial Intelligence Laboratory)
Depth EstimationComputational EfficiencyGaussian SplattingPoint Cloud
🎯 What it does: This paper proposes the CityGSX architecture, which utilizes a parallelizable hybrid hierarchical 3D representation and batch multi-task rendering to achieve efficient and geometrically accurate reconstruction of large-scale scenes.
CityNav: A Large-Scale Dataset for Real-World Aerial Navigation
Jungdae Lee (Institute of Science), Nakamasa Inoue (Institute of Science)
Autonomous DrivingRobotic IntelligenceConvolutional Neural NetworkRecurrent Neural NetworkLarge Language ModelImageMultimodalityBenchmark
🎯 What it does: Introduced and released the CityNav dataset, providing a large-scale real urban 3D environment and 32,637 human demonstration trajectories for audiovisual language navigation with drones.
CL-Splats: Continual Learning of Gaussian Splatting with Local Optimization
Jan Ackermann (ETH Zurich), Songyou Peng (Google DeepMind)
OptimizationGaussian SplattingSimultaneous Localization and MappingPoint Cloud
🎯 What it does: Proposes the CL-Splats method, which continuously improves the Gaussian Splats scene representation under sparse perspectives through local updates and efficient optimization.
ClaraVid: A Holistic Scene Reconstruction Benchmark From Aerial Perspective With Delentropy-Based Complexity Profiling
Radu Beche (Technical University of Cluj-Napoca), Sergiu Nedevschi (Technical University of Cluj-Napoca)
SegmentationData SynthesisDepth EstimationNeural Radiance FieldGaussian SplattingImagePoint CloudBenchmark
🎯 What it does: A new high-resolution synthetic drone perspective dataset ClaraVid is proposed, and a differential entropy-based scene complexity assessment framework DSP is constructed on this dataset to quantify reconstruction difficulty.
Class Token as Proxy: Optimal Transport-assisted Proxy Learning for Weakly Supervised Semantic Segmentation
Jian Wang (University of Liverpool), Jimin Xiao (XJTLU)
SegmentationTransformerContrastive LearningImage
🎯 What it does: The OTPL framework is proposed to bridge the class label token gap through optimal transport learning agents, thereby improving CAM generation in weakly supervised semantic segmentation.
Class-Wise Federated Averaging for Efficient Personalization
Gyuejeong Lee (SAKAK Inc), Daeyoung Choi (Cyber University of Korea)
Federated LearningImage
🎯 What it does: A category-based federated averaging framework cwFedAvg is proposed, achieving class-wise aggregation for efficient personalization.
CleanPose: Category-Level Object Pose Estimation via Causal Learning and Knowledge Distillation
Xiao Lin (Tongji University), Qijun Chen (Tongji University)
Pose EstimationKnowledge DistillationImagePoint Cloud
🎯 What it does: This work proposes CleanPose, a method that utilizes causal learning and knowledge distillation to suppress data bias in category-level pose estimation.
ClearSight: Human Vision-Inspired Solutions for Event-Based Motion Deblurring
Xiaopeng Lin (Hong Kong University of Science and Technology), Bojun Cheng (Hong Kong University of Science and Technology)
RestorationSpiking Neural NetworkImageVideo
🎯 What it does: This paper proposes the Bio-Inspired Dual-Drive Hybrid Network (BDHNet), which achieves motion deblurring by integrating motion information from event cameras with color information from traditional frame cameras.
Client2Vec: Improving Federated Learning by Distribution Shifts Aware Client Indexing
Yongxin Guo (Westlake University), Tao Lin (Westlake University)
Federated LearningSafty and PrivacyContrastive LearningImageText
🎯 What it does: Proposes the Client2Vec mechanism, which generates an index vector containing distribution shift information for each client before federated learning training, and uses this index to improve client sampling, model aggregation, and local training;
Clink! Chop! Thud! - Learning Object Sounds from Real-World Interactions
Mengyu Yang (Georgia Institute of Technology), James Hays (Georgia Institute of Technology)
RecognitionObject DetectionContrastive LearningVideoMultimodalityAudio
🎯 What it does: A multimodal object perception framework is proposed, which utilizes audio and video information to jointly identify specific objects that produce sound during interactions.
CLIP-Adapted Region-to-Text Learning for Generative Open-Vocabulary Semantic Segmentation
Jiannan Ge (University of Science and Technology of China), Yongdong Zhang (University of Science and Technology of China)
SegmentationGenerationTransformerLarge Language ModelVision Language ModelImageText
🎯 What it does: A generative open vocabulary semantic segmentation framework called CRTNet is proposed, which can generate category names and descriptions for segmented areas without using a predefined vocabulary.
CLIP-GS: Unifying Vision-Language Representation with 3D Gaussian Splatting
Siyu Jiao (Beijing Jiaotong University), Yunchao Wei (Beijing Jiaotong University)
RetrievalRepresentation LearningTransformerContrastive LearningGaussian SplattingMultimodalityPoint Cloud
🎯 What it does: Proposes the CLIP-GS framework, utilizing 3D Gaussian Scatter (3DGS) for 3D representation learning, and aligning it with CLIP's visual-text representation to achieve a unified multimodal representation;
CLIPer: Hierarchically Improving Spatial Representation of CLIP for Open-Vocabulary Semantic Segmentation
Lin Sun (Tianjin University), Yanwei Pang (Tianjin University)
SegmentationTransformerVision Language ModelDiffusion modelImage
🎯 What it does: Utilizing pre-trained CLIP for unsupervised open vocabulary semantic segmentation and enhancing its spatial representation through a hierarchical approach.
CLIPSym: Delving into Symmetry Detection with CLIP
Tinghan Yang (Purdue University), Raymond A. Yeh (Purdue University)
RecognitionTransformerPrompt EngineeringVision Language ModelImage
🎯 What it does: This paper proposes the CLIPSym framework, which utilizes a pre-trained CLIP vision-language model for detecting reflection and rotation symmetry, and generates symmetry heatmaps through a rotation equivariant decoder.
Closed-Loop Transfer for Weakly-supervised Affordance Grounding
Jiajin Tang, Sibei Yang
Object DetectionSegmentationDomain AdaptationKnowledge DistillationTransformerImageVideo
🎯 What it does: Achieving interactive object functional localization under weak supervision, this paper proposes LoopTrans, a closed-loop knowledge transfer framework that enables mutual learning between exocentric and egocentric images through Shared Activation Maps (SCAM) and denoising distillation.
CLOT: Closed Loop Optimal Transport for Unsupervised Action Segmentation
Elena Bueno-Benito (Institut de Robótica y Informática Industrial), Mariella Dimiccoli (Institut de Robótica y Informática Industrial)
SegmentationOptimizationVideo
🎯 What it does: This paper studies a new framework for unsupervised action segmentation called CLOT, which optimizes frame and segment representations through closed-loop optimal transport.
CMAD: Correlation-Aware and Modalities-Aware Distillation for Multimodal Sentiment Analysis with Missing Modalities
Yan Zhuang (University of Electronic Science and Technology of China), Fuji Ren (University of Electronic Science and Technology of China)
Knowledge DistillationRepresentation LearningTransformerTextMultimodality
🎯 What it does: The CMAD framework is proposed, which achieves unified representation in the task of missing multimodal emotion analysis through teacher-student knowledge distillation, and designs two modules: CAFD for sample-level feature and relevance alignment, and MAR for modality-aware regularization.
CMB-ML: A Cosmic Microwave Background Dataset for the Oldest Possible Computer Vision Task
James Amato (University of Texas at Dallas), Nicholas Ruozzi (University of Texas at Dallas)
RestorationSegmentationConvolutional Neural NetworkImagePhysics Related
🎯 What it does: Proposed the CMB-ML framework and dataset, achieving a complete simulation-modeling-evaluation pipeline for CMB signal cleaning tasks.
CMT: A Cascade MAR with Topology Predictor for Multimodal Conditional CAD Generation
Jianyu Wu (Shanghai Artificial Intelligence Laboratory), Shixiang Tang (Shanghai Artificial Intelligence Laboratory)
GenerationData SynthesisTransformerDiffusion modelTextMultimodalityPoint CloudMesh
🎯 What it does: A multi-modal CAD generation framework named CMT is proposed, which can accept text, point clouds, and multi-view images as input. It uses a cascaded Masked Autoregressive Network (MAR) to first generate edges and then surfaces, and directly outputs edge-surface associations through a topology predictor, thereby generating complete and topologically correct B-Rep CAD models in continuous space.
CNS-Bench: Benchmarking Image Classifier Robustness Under Continuous Nuisance Shifts
Olaf Dünkel (Max Planck Institute for Informatics), Adam Kortylewski (Max Planck Institute for Informatics)
ClassificationDomain AdaptationAnomaly DetectionConvolutional Neural NetworkTransformerDiffusion modelContrastive LearningImageBenchmark
🎯 What it does: Proposes the CNS-Bench benchmark, utilizing Stable Diffusion and LoRA sliders to achieve continuously controllable realistic noise shifts, assessing the robustness of image classifiers under OOD conditions.
Co-Painter: Fine-Grained Controllable Image Stylization via Implicit Decoupling and Adaptive Injection
Bowen Fu (Northwestern Polytechnical University), Lei Zhang (Northwestern Polytechnical University)
Image TranslationGenerationDiffusion modelImageText
🎯 What it does: CO-PAINTER is proposed, which enables precise control over image styles (brush strokes, colors, content) through fine-grained decoupling and gated feature injection.
CO2-Net: A Physics-Informed Spatio-Temporal Model for Global Surface CO2 Reconstruction
Hao Zheng (Shanghai Jiaotong University), Shiyu Liang (Shanghai Jiaotong University)
TransformerTime SeriesPhysics Related
🎯 What it does: This paper proposes CO‑2Net, a physics-informed spatiotemporal model designed to reconstruct global surface CO2 concentrations from sparse observations and meteorological auxiliary variables, avoiding reliance on large-scale prior data.
CoA-VLA: Improving Vision-Language-Action Models via Visual-Text Chain-of-Affordance
Jinming Li (Shanghai University), Feifei Feng (Midea Group)
Robotic IntelligenceTransformerVision-Language-Action ModelDiffusion modelMultimodality
🎯 What it does: The CoA-VLA model is proposed, embedding four types of affordance—object, grasping, space, and motion—into a visual language model in a chain reasoning format, enhancing the robot's action planning based on visual language.
CObL: Toward Zero-Shot Ordinal Layering without User Prompting
Aneel Damaraju (Harvard University), Todd Zickler (Harvard University)
RestorationSegmentationGenerationDiffusion modelImage
🎯 What it does: This paper proposes a model named CObL, which can infer multiple layers of objects arranged in occlusion order from a single image in a zero-shot manner without requiring user prompts and without prior knowledge of the number of objects, and complete their full (occluded) filling;
CoDa-4DGS: Dynamic Gaussian Splatting with Context and Deformation Awareness for Autonomous Driving
Rui Song (Fraunhofer Institute for Transportation and Infrastructure Systems), Alois Knoll (Technical University of Munich)
Autonomous DrivingGaussian SplattingPoint Cloud
🎯 What it does: This paper proposes CoDa-4DGS, a 4D Gaussian projection method that combines context awareness and deformation awareness for dynamic rendering in autonomous driving scenarios.
CODA: Repurposing Continuous VAEs for Discrete Tokenization
Zeyu Liu, Gao Huang
GenerationCompressionAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: The CODA framework decouples the compression capability of pre-trained continuous VAE from discretization, directly transforming continuous VAE into an efficient discrete tokenizer.
CODE-CL: Conceptor-Based Gradient Projection for Deep Continual Learning
Marco P. E. Apolinario (Purdue University), Kaushik Roy (Purdue University)
ClassificationOptimizationImage
🎯 What it does: A gradient projection method based on Conceptor, CODE-CL, is proposed to simultaneously suppress catastrophic forgetting and enhance forward knowledge transfer in continual learning.
CogCM: Cognition-Inspired Contextual Modeling for Audio-Visual Speech Enhancement
Feixiang Wang (Institute of Computing Technology Chinese Academy of Sciences), Xilin Chen (Institute of Computing Technology Chinese Academy of Sciences)
RestorationGenerative Adversarial NetworkVideoAudio
🎯 What it does: A hierarchical context modeling framework called CogCM, inspired by cognitive science, is proposed for audio and video speech enhancement, integrating semantic context and signal context.
CogNav: Cognitive Process Modeling for Object Goal Navigation with LLMs
Yihan Cao (National University of Defense Technology), Kai Xu (Peking University)
Robotic IntelligenceTransformerLarge Language ModelPrompt EngineeringVision Language ModelMultimodalityBenchmark
🎯 What it does: Proposes the CogNav framework, which uses large language models to simulate the object goal navigation process, constructs an online heterogeneous cognitive map, and drives navigation through a fine-grained state machine;
CoHD: A Counting-Aware Hierarchical Decoding Framework for Generalized Referring Expression Segmentation
Zhuoyan Luo (Tsinghua University), Yujiu Yang (Tsinghua University)
Object DetectionSegmentationTransformerImageText
🎯 What it does: A counting-aware hierarchical decoding framework CoHD is proposed for general referential expression segmentation.
COIN: Confidence Score-Guided Distillation for Annotation-Free Cell Segmentation
Sanghyun Jo (Seoul National University), Kyungsu Kim (Seoul National University)
SegmentationKnowledge DistillationImage
🎯 What it does: We propose COIN, a three-step unsupervised cell instance segmentation framework that first enhances pixel-level detection using unsupervised semantic segmentation + optimal transport (OT), then generates pseudo ground truth (GT) and evaluates instance confidence using the Segment Anything Model (SAM), and finally recursively self-distills to expand high-confidence instances.
Collaborative Instance Object Navigation: Leveraging Uncertainty-Awareness to Minimize Human-Agent Dialogues
Francesco Taioli (Polytechnic of Turin), Yiming Wang (Fondazione Bruno Kessler)
Robotic IntelligenceTransformerLarge Language ModelVision Language ModelTextBenchmark
🎯 What it does: This paper proposes a collaborative instance target navigation (CoIN) task that actively resolves instance target visual ambiguities through natural language dialogue between humans and robots in unknown environments.
CoLMDriver: LLM-based Negotiation Benefits Cooperative Autonomous Driving
Changxing Liu (Shanghai Jiao Tong University), Siheng Chen (Shanghai Jiao Tong University)
Autonomous DrivingTransformerLarge Language ModelVision Language ModelTextBenchmark
🎯 What it does: This paper proposes CoLMDriver, a complete LLM-driven collaborative driving system that combines multi-turn language negotiation and intent-guided waypoint generation, and introduces the InterDrive interactive driving benchmark.
Color Matching Using Hypernetwork-Based Kolmogorov-Arnold Networks
Artem Nikonorov (Samara National Research University), Radu Timofte (University of Wurzburg)
Image TranslationOptimizationGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes the cmKAN framework for mapping the colors of source images to target color spaces, supporting three scenarios: supervised, unsupervised, and paired optimization.
Colors See Colors Ignore: Clothes Changing ReID with Color Disentanglement
Priyank Pathak (University of Central Florida), Yogesh S. Rawat (University of Central Florida)
RecognitionRetrievalTransformerImageVideo
🎯 What it does: Proposes a clothing change re-identification method called CSCI that only uses RGB without external annotations or models, utilizing color information to decouple identity features;
CoMatch: Dynamic Covisibility-Aware Transformer for Bilateral Subpixel-Level Semi-Dense Image Matching
Zizhuo Li (Wuhan University), Jiayi Ma (Wuhan University)
Pose EstimationDepth EstimationTransformerImage
🎯 What it does: A semi-dense image matcher called CoMatch is designed, which combines a dynamic visibility-aware Transformer and a bidirectional sub-pixel refinement module to achieve high-precision and high-efficiency image matching.
CombatVLA: An Efficient Vision-Language-Action Model for Combat Tasks in 3D Action Role-Playing Games
Peng Chen (Alibaba Group), Bo Zheng (Alibaba Group)
Vision-Language-Action ModelContrastive LearningMultimodalityBenchmark
🎯 What it does: A 3B-scale visual language action model CombatVLA was developed, trained with Action of Thought (AoT) data and integrated into an action execution framework, capable of achieving real-time efficient decision-making in 3D ARPG combat.
Combinative Matching for Geometric Shape Assembly
Nahyuk Lee (POSTECH), Minsu Cho (RLWRLD)
Object DetectionOptimizationContrastive LearningMesh
🎯 What it does: A combinative matching method is proposed, which achieves high-precision matching and alignment of interlocking components in geometric shape assembly by learning three features: local direction, surface shape, and occupied volume.
COME: Dual Structure-Semantic Learning with Collaborative MoE for Universal Lesion Detection Across Heterogeneous Ultrasound Datasets
Lingyu Chen (Nanjing University of Aeronautics and Astronautics), Fang Chen (Nanjing University of Aeronautics and Astronautics)
Object DetectionMixture of ExpertsImageMultimodalityBiomedical DataUltrasound
🎯 What it does: This paper studies a cross-homogeneous ultrasound imaging multi-dataset general lesion detection framework based on dual-structure semantic learning and Collaborative Mixture of Experts (COME).
Communication-Efficient Multi-Vehicle Collaborative Semantic Segmentation via Sparse 3D Gaussian Sharing
Tianyu Hong (Tianjin University), Tie Qiu (Qinghai Minzu University)
SegmentationCompressionAutonomous DrivingComputational EfficiencyGaussian SplattingImage
🎯 What it does: A communication-efficient multi-vehicle collaborative semantic segmentation framework GSCOOP based on sparse 3D Gaussian sharing has been developed, which can generate discrete 3D Gaussian representations from multi-view images and achieve low-bandwidth communication through selection and compression.
CoMoGaussian: Continuous Motion-Aware Gaussian Splatting from Motion-Blurred Images
Jungho Lee (Yonsei University), Sangyoun Lee (Yonsei University)
RestorationGaussian SplattingImageOrdinary Differential Equation
🎯 What it does: Recovering clear 3D scenes from blurred images caused by camera motion
CoMPaSS: Enhancing Spatial Understanding in Text-to-Image Diffusion Models
Gaoyang Zhang (Zhejiang University), Xinguo Liu (vivo)
GenerationData SynthesisSupervised Fine-TuningDiffusion modelImageText
🎯 What it does: To address the shortcomings of text-to-image diffusion models in generating spatial relationships, the CoMPaSS framework is proposed, significantly enhancing spatial understanding capabilities.
CompCap: Improving Multimodal Large Language Models with Composite Captions
Xiaohui Chen (Meta), Baosheng He (Meta)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningImageTextMultimodality
🎯 What it does: Proposes the CompCap framework, which utilizes LLM and automated tools to generate six types of synthetic images and their high-quality, detailed titles;
Competitive Distillation: A Simple Learning Strategy for Improving Visual Classification
Daqian Shi (Queen Mary University of London), Cédric M John (Queen Mary University of London)
ClassificationKnowledge DistillationImage
🎯 What it does: Proposes a Competitive Distillation strategy that dynamically selects the best-performing network as the teacher to enhance the performance of visual classification models.
CompleteMe: Reference-based Human Image Completion
Yu-Ju Tsai (University of California), Ming-Hsuan Yang (University of California)
Image TranslationRestorationConvolutional Neural NetworkDiffusion modelImageMultimodalityBenchmark
🎯 What it does: A human image completion framework called CompleteMe is proposed, which utilizes a dual U-Net structure and Region-focused Attention to complete occluded portraits, maintaining pose consistency while preserving details from the reference image.
Completing 3D Partial Assemblies with View-Consistent 2D-3D Correspondence
Weihao Wang (Tongji University), Bin He (Tongji University)
Object DetectionPose EstimationConvolutional Neural NetworkTransformerContrastive LearningImagePoint Cloud
🎯 What it does: A 3D assembly completion framework based on a single view image is proposed, achieving complete recovery of partial assemblies through missing-guided feature fusion and self-supervised viewpoint alignment.
Compression of 3D Gaussian Splatting with Optimized Feature Planes and Standard Video Codecs
Soonbin Lee (Fraunhofer Heinrich Hertz Institute), Cornelius Hellge (Fraunhofer Heinrich Hertz Institute)
CompressionGaussian SplattingPoint Cloud
🎯 What it does: This work proposes a 3D Gaussian projection compression framework based on tri-plane feature representation and utilizing standard video codecs (HEVC/FFmpeg), capable of compressing three-dimensional Gaussian raw attributes to under 10MB while maintaining almost no distortion.
Compression-Aware One-Step Diffusion Model for JPEG Artifact Removal
Jinpei Guo (Carnegie Mellon University), Yulun Zhang (Shanghai Jiao Tong University)
RestorationCompressionDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: A first-order diffusion model CODiff is proposed, which uses a compressed sensing visual embedder CaVE to extract JPEG compression priors, and based on this, performs artifact removal and reconstruction on low-quality images.