CVPR 2025 Papers — Page 4
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2871 papers
Bringing CLIP to the Clinic: Dynamic Soft Labels and Negation-Aware Learning for Medical Analysis
Hanbin Ko (Seoul National University), Chang-Min Park (Seoul National University)
ClassificationRetrievalRepresentation LearningGraph Neural NetworkTransformerContrastive LearningImageTextMultimodalityBiomedical Data
🎯 What it does: This paper proposes a method that introduces clinically enhanced dynamic soft labels and negation-based hard negative samples in medical image-text pre-training, as well as graph embeddings to improve the contrastive learning performance of CLIP on medical data.
Buffer Anytime: Zero-Shot Video Depth and Normal from Image Priors
Zhengfei Kuang (Stanford University), Fujun Luan (Adobe Research)
Depth EstimationAutonomous DrivingOptical FlowVideo
🎯 What it does: Proposed the Buffer Anytime framework, achieving zero-shot estimation of depth and normal maps from unlabeled videos;
Building a Mind Palace: Structuring Environment-Grounded Semantic Graphs for Effective Long Video Analysis with LLMs
Zeyi Huang (University of Wisconsin-Madison), Miao Liu (Meta)
RecognitionObject TrackingTransformerLarge Language ModelVision Language ModelVideo
🎯 What it does: This paper proposes VideoMindPalace, a method that divides long videos into a hierarchical topological semantic graph (including human-object interactions, activity area clustering, and scene layout), and inputs this graph in JSON format into an LLM for reasoning;
Building Vision Models upon Heat Conduction
Zhaozhi Wang (University of Chinese Academy of Sciences), Qixiang Ye (University of Chinese Academy of Sciences)
ClassificationRestorationObject DetectionSegmentationComputational EfficiencyConvolutional Neural NetworkTransformerSupervised Fine-TuningImagePhysics Related
🎯 What it does: A visual representation model vHeat based on the physical principles of thermal diffusion is proposed, and a Heat Conduction Operator (HCO) is designed to achieve global information propagation using DCT/IDCT;
BWFormer: Building Wireframe Reconstruction from Airborne LiDAR Point Cloud with Transformer
Yuzhou Liu (Institute of Automation, Chinese Academy of Sciences), Shuhan Shen (Institute of Automation, Chinese Academy of Sciences)
Object DetectionGenerationData SynthesisTransformerDiffusion modelPoint Cloud
🎯 What it does: Construct an end-to-end Transformer architecture for reconstructing 3D wireframe models of buildings from aerial LiDAR point clouds.
ByTheWay: Boost Your Text-to-Video Generation Model to Higher Quality in a Training-free Way
Jiazi Bu (Shanghai AI Laboratory), Jiaqi Wang (Shanghai Innovation Institute)
GenerationData SynthesisDiffusion modelVideoText
🎯 What it does: An improved method called ByTheWay has been developed, which is training-independent, requires no additional parameters, and incurs no sampling costs. It aims to enhance the structural rationality, temporal consistency, and motion amplitude of text-to-video (T2V) generation models.
CacheQuant: Comprehensively Accelerated Diffusion Models
Xuewen Liu (Institute of Automation, Chinese Academy of Sciences), Qingyi Gu (Institute of Automation, Chinese Academy of Sciences)
GenerationCompressionComputational EfficiencyDiffusion modelImage
🎯 What it does: A training-independent method called CacheQuant is proposed, which significantly improves inference speed and compresses model size by jointly optimizing the time-layer caching and structural-layer quantization of diffusion models.
CAD-Llama: Leveraging Large Language Models for Computer-Aided Design Parametric 3D Model Generation
Jiahao Li (Fudan University), Xiangdong Zhou (Fudan University)
GenerationAI Code AssistantTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelTextMultimodality
🎯 What it does: The CAD-Llama framework is proposed, utilizing large language models (LLM) to generate parameterized CAD code, supporting the creation, editing, and completion of CAD models.
CADCrafter: Generating Computer-Aided Design Models from Unconstrained Images
Cheng Chen (Nanyang Technological University), Fayao Liu
GenerationDomain AdaptationTransformerDiffusion modelAuto EncoderImageMesh
🎯 What it does: A framework called CADCrafter is proposed to directly convert unconstrained real-world images into editable CAD command sequences.
CADDreamer: CAD Object Generation from Single-view Images
Yuan Li (University of Texas at Dallas), Xiaohu Guo (Texas A&M University)
GenerationOptimizationDiffusion modelImageMesh
🎯 What it does: This paper presents CADDreamer, which generates high-quality CAD B-rep models from a single RGB image.
CADRef: Robust Out-of-Distribution Detection via Class-Aware Decoupled Relative Feature Leveraging
Zhiwei Ling (Zhejiang University), Shuiguang Deng (Zhejiang University)
Anomaly DetectionConvolutional Neural NetworkImage
🎯 What it does: A post-hoc OOD detection method based on category-related relative feature error, CARef, is proposed, and further enhanced to a more powerful CADRef through feature decomposition and error scaling.
Calibrated Multi-Preference Optimization for Aligning Diffusion Models
Kyungmin Lee (Google DeepMind), Yinxiao Li (Google)
GenerationOptimizationReinforcement LearningDiffusion modelImageText
🎯 What it does: The Calibrated Preference Optimization (CaPO) method is proposed, which aligns and fine-tunes text-to-image diffusion models using calibrated expected win rate rewards and frontier rejection sampling.
CALICO: Part-Focused Semantic Co-Segmentation with Large Vision-Language Models
Kiet A. Nguyen (University of Illinois), Ismini Lourentzou (University of Illinois)
Object DetectionSegmentationTransformerLarge Language ModelVision Language ModelImageMultimodality
🎯 What it does: A multi-image part-level semantic co-segmentation method CALICO based on large-scale visual language models is proposed, and a new task - part-focused semantic co-segmentation - is defined.
Camera Resection from Known Line Pencils and a Radially Distorted Scanline
Juan C. Dibene (Stevens Institute of Technology), Enrique Dunn (Stevens Institute of Technology)
Pose EstimationSimultaneous Localization and MappingImage
🎯 What it does: This paper proposes a framework for absolute camera pose estimation based on a single radial distortion scanning line, including 6-point minimal solutions, 7-point unique solutions, and 8+ linear solutions;
CamFreeDiff: Camera-free Image to Panorama Generation with Diffusion Model
Xiaoding Yuan (Johns Hopkins University), Peng Wang (ByteDance)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: The CamFreeDiff model is proposed, which utilizes diffusion models and homography transformations to predict and can generate 360° panoramic images from a single image without camera parameters.
Camouflage Anything: Learning to Hide using Controlled Out-painting and Representation Engineering
Biplab Das (International Institute of Information Technology Bangalore), Viswanath Gopalakrishnan (International Institute of Information Technology Bangalore)
SegmentationGenerationDiffusion modelImage
🎯 What it does: Proposes the Camouflage Anything method, which generates high-quality camouflage images through controlled painting and representation engineering, and introduces a new CamOT metric and LoRA fine-tuning of BiRefNet for camouflage target segmentation.
CamPoint: Boosting Point Cloud Segmentation with Virtual Camera
Jianhui Zhang (University of Chinese Academy of Sciences), Bonan Li (University of Chinese Academy of Sciences)
ClassificationSegmentationPoint Cloud
🎯 What it does: Utilize virtual cameras to generate camera visibility features, thereby enhancing point cloud segmentation performance.
CaMuViD: Calibration-Free Multi-View Detection
Amir Etefaghi Daryani (University of Florida), Henry Medeiros (University of Florida)
Object DetectionConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a calibration-free multi-view pedestrian detection framework called CaMuViD.
Can Generative Video Models Help Pose Estimation?
Ruojin Cai (Google), Ricardo Martin-Brualla (Google)
GenerationPose EstimationPrompt EngineeringImageVideo
🎯 What it does: This study investigates how to use pre-trained generative video models to interpolate intermediate frames between two nearly non-overlapping images and input these interpolated frames along with the original images into a camera pose estimator to enhance pose estimation accuracy.
Can Large Vision-Language Models Correct Semantic Grounding Errors By Themselves?
Yuan-Hong Liao (University of Toronto), David Acuna (NVIDIA)
SegmentationTransformerVision Language ModelImageMultimodality
🎯 What it does: This paper explores the self-correction ability of large visual language models (VLM) in semantic localization tasks and proposes an iterative self-correction framework based on binary feedback.
Can Machines Understand Composition? Dataset and Benchmark for Photographic Image Composition Embedding and Understanding
Zhaoran Zhao (Beijing University of Posts and Telecommunications), Wenhao Guo (Beijing University of Posts and Telecommunications)
RecognitionObject DetectionGraph Neural NetworkTransformerPrompt EngineeringImageBenchmark
🎯 What it does: A large-scale photographic image composition dataset, PICD, has been constructed, and benchmark tasks have been designed based on this dataset to evaluate the embedding and understanding of photographic composition.
Can Text-to-Video Generation help Video-Language Alignment?
Luca Zanella (University of Trento), Elisa Ricci (Fondazione Bruno Kessler)
GenerationData SynthesisRetrievalTransformerLarge Language ModelVision Language ModelVideoText
🎯 What it does: The research enhances the learning effectiveness of video-language alignment models by utilizing synthetic videos generated by text-to-video generators.
Can't Slow Me Down: Learning Robust and Hardware-Adaptive Object Detectors against Latency Attacks for Edge Devices
Tianyi Wang (Zhejiang University), Jiming Chen (Zhejiang University)
Object DetectionAutonomous DrivingComputational EfficiencyAdversarial AttackImage
🎯 What it does: A hardware-adaptive background attention adversarial training defense method against latency attacks on NMS-based object detectors for edge devices is proposed.
CAP-Net: A Unified Network for 6D Pose and Size Estimation of Categorical Articulated Parts from a Single RGB-D Image
Jingshun Huang (Fudan University), Yi Zhu (Huawei)
Object DetectionSegmentationPose EstimationImagePoint Cloud
🎯 What it does: This paper proposes CAP-Net, a unified network capable of simultaneously estimating the 6D pose and size of category-level articulated object parts from a single RGB-D image.
CAP4D: Creating Animatable 4D Portrait Avatars with Morphable Multi-View Diffusion Models
Felix Taubner (University of Toronto), David B. Lindell (University of Toronto)
GenerationData SynthesisDiffusion modelGaussian SplattingImageVideo
🎯 What it does: Generate and render in real-time animated 4D avatars constructed from any number of reference images (1–100).
CARE Transformer: Mobile-Friendly Linear Visual Transformer via Decoupled Dual Interaction
Yuan Zhou (Nanyang Technological University), Hanwang Zhang (Nanyang Technological University)
RecognitionObject DetectionSegmentationComputational EfficiencyTransformerImage
🎯 What it does: Designed the CARE mechanism to achieve decoupled interaction of linear attention and local features in a visual Transformer.
CaricatureBooth: Data-Free Interactive Caricature Generation in a Photo Booth
Zhiyu Qu (University of Surrey), Yi-Zhe Song (University of Surrey)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: Using TPS to generate cartoon training samples, combined with Bézier curve interaction and the diffusion model maintained by InstantID, to achieve identity-preserving and interactive cartoon portrait generation.
CARL: A Framework for Equivariant Image Registration
Hastings Greer (University of North Carolina at Chapel Hill), Marc Niethammer (University of North Carolina at Chapel Hill)
Convolutional Neural NetworkDiffusion modelImageBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: A multi-step isometric registration framework called CARL has been developed, utilizing coordinate attention to achieve isometry for independent translations of input images, while maintaining overall isometry in subsequent refinement layers.
CarPlanner: Consistent Auto-regressive Trajectory Planning for Large-Scale Reinforcement Learning in Autonomous Driving
Dongkun Zhang (Zhejiang University), Yue Wang (Zhejiang University)
Autonomous DrivingTransformerReinforcement LearningPoint Cloud
🎯 What it does: This paper presents CarPlanner, a consistency autoregressive trajectory planner that utilizes reinforcement learning to generate multimodal trajectories, addressing the training efficiency and performance issues of large-scale real driving tasks.
CASAGPT: Cuboid Arrangement and Scene Assembly for Interior Design
Weitao Feng (University of Science and Technology of China), Wenbo Zhou (University of Science and Technology of China)
GenerationData SynthesisTransformerLarge Language ModelPoint Cloud
🎯 What it does: The CASAGPT model is proposed, which generates indoor scenes in an autoregressive manner, using cube primitives to significantly reduce object intersections.
CASP: Compression of Large Multimodal Models Based on Attention Sparsity
Mohsen Gholami (Huawei Technologies Canada), Yong Zhang (Huawei Technologies Canada)
CompressionTransformerLarge Language ModelImageVideoTextMultimodality
🎯 What it does: A very low-bit compression method CASP based on attention sparsity is proposed, which compresses large multimodal models using low-rank decomposition and quantization.
CASP: Consistency-aware Audio-induced Saliency Prediction Model for Omnidirectional Video
Zhaolin Wan (Harbin Institute of Technology), Debin Zhao (Harbin Institute of Technology)
RecognitionSegmentationData SynthesisTransformerVideoMultimodalityAudio
🎯 What it does: A CASP model is proposed and experimentally validated for the task of audio-induced saliency prediction in panoramic videos.
CAT4D: Create Anything in 4D with Multi-View Video Diffusion Models
Rundi Wu (Google DeepMind), Aleksander Holynski
GenerationData SynthesisDiffusion modelGaussian SplattingVideoStochastic Differential Equation
🎯 What it does: Proposes the CAT4D method, which utilizes a multi-view video diffusion model to convert monocular videos into multi-view videos, and then achieves 4D scene creation through deformable 3D Gaussian optimization.
CATANet: Efficient Content-Aware Token Aggregation for Lightweight Image Super-Resolution
Xin Liu (Nanjing University), Gangshan Wu (Nanjing University)
RestorationSuper ResolutionTransformerImage
🎯 What it does: A lightweight super-resolution network called CATANet is proposed, which achieves long-distance information interaction through content-aware token aggregation.
Category-Agnostic Neural Object Rigging
Guangzhao He (Stanford University), Jiajun Wu (Stanford University)
Object DetectionGenerationPose EstimationTransformerPoint CloudMesh
🎯 What it does: This paper studies an unsupervised, category-independent neural object assembly method (CANOR) that can learn a low-dimensional pose space from dynamic 3D object point clouds and achieve intuitive control over poses through editable sparse blobs.
Causal Composition Diffusion Model for Closed-loop Traffic Generation
Haohong Lin (Carnegie Mellon University), Hongge Chen (Cruise LLC)
GenerationAutonomous DrivingDiffusion modelTime Series
🎯 What it does: This paper studies a causal combination diffusion model called CCDiff, designed for generating closed-loop traffic scenarios while balancing controllability and realism.
CAV-MAE Sync: Improving Contrastive Audio-Visual Mask Autoencoders via Fine-Grained Alignment
Edson Araujo (Goethe University of Frankfurt), Hilde Kuehne (Tuebingen AI Center University of Tuebingen)
ClassificationSegmentationRetrievalTransformerAuto EncoderContrastive LearningMultimodalityAudio
🎯 What it does: Proposes CAV-MAE Sync, which improves audio-visual self-supervised learning through fine-grained audio-visual alignment, separation of contrastive and reconstruction objectives, and the introduction of global and register tokens.
CCIN: Compositional Conflict Identification and Neutralization for Composed Image Retrieval
Likai Tian (Wuhan University), Xuelong Li (Northwestern Polytechnical University)
RetrievalTransformerLarge Language ModelContrastive LearningImage
🎯 What it does: This paper proposes a method to improve image retrieval accuracy by first identifying attribute conflicts between the reference image and modification instructions, and then neutralizing these conflicts through a dual-instruction mechanism.
CDI: Copyrighted Data Identification in Diffusion Models
Jan Dubiński (Warsaw University of Technology), Adam Dziedzic (CISPA Helmholtz Center for Information Security)
RecognitionData-Centric LearningTransformerDiffusion modelImage
🎯 What it does: This paper proposes a copyright data identification framework called CDI based on dataset inference, which is used to determine whether a Diffusion model has utilized a specific copyrighted dataset.
Certified Human Trajectory Prediction
Mohammadhossein Bahari (École Polytechnique Fédérale de Lausanne), Alexandre Alahi (École Polytechnique Fédérale de Lausanne)
Object TrackingAutonomous DrivingDiffusion modelAuto EncoderMultimodalityTime SeriesBenchmark
🎯 What it does: A robust certification framework for trajectory prediction based on random smoothing is proposed, introducing an unconditional diffusion denoiser to enhance prediction accuracy and confidence boundaries.
CGMatch: A Different Perspective of Semi-supervised Learning
Bo Cheng (Jilin University), Lan Du (Monash University)
ClassificationOptimizationSupervised Fine-TuningImage
🎯 What it does: This paper proposes a new semi-supervised learning framework called CGMatch, focusing on model training in scenarios with few labeled samples.
CH3Depth: Efficient and Flexible Depth Foundation Model with Flow Matching
Jiaqi Li (Huazhong University of Science and Technology), Zhiguo Cao (Huazhong University of Science and Technology)
Depth EstimationConvolutional Neural NetworkFlow-based ModelOptical FlowImageVideo
🎯 What it does: This paper proposes CH Depth, an efficient, fine-grained, and temporally consistent depth estimation baseline model based on flow matching.
Chain of Attack: On the Robustness of Vision-Language Models Against Transfer-Based Adversarial Attacks
Peng Xie (Hong Kong University of Science and Technology), Kani Chen (Hong Kong University of Science and Technology)
Adversarial AttackTransformerLarge Language ModelVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Evaluate the robustness of VLM in black-box transfer adversarial attacks, propose the Chain of Attack (CoA) method to generate adversarial images, and introduce LLM-based ASR automatic evaluation metrics.
Chain of Semantics Programming in 3D Gaussian Splatting Representation for 3D Vision Grounding
Jiaxin Shi (Inner Mongolia University), Zhi Weng (Inner Mongolia University)
Large Language ModelVision Language ModelGaussian SplattingPoint Cloud
🎯 What it does: This paper proposes a zero-shot neural symbolic framework based on 3D Gaussian splatting representation, achieving multi-step spatial reasoning through chained semantic programming to complete 3D visual grounding.
ChainHOI: Joint-based Kinematic Chain Modeling for Human-Object Interaction Generation
Ling-An Zeng (Sun Yat-sen University), Wei-Shi Zheng (Sun Yat-sen University)
GenerationData SynthesisPose EstimationGraph Neural NetworkTransformerDiffusion modelVideoText
🎯 What it does: This paper proposes ChainHOI, a text-driven human-object interaction (HOI) generation method based on diffusion models, which can generate high-quality, logically consistent, and realistic full-body motion sequences based on text descriptions and geometric information of target objects.
Change3D: Revisiting Change Detection and Captioning from A Video Modeling Perspective
Duowang Zhu (Wuhan University), Zhenfeng Shao (Wuhan University)
GenerationConvolutional Neural NetworkTransformerImageVideo
🎯 What it does: Treating dual-temporal remote sensing images as short videos, inserting learnable perceptual frames, and using a video encoder for change detection and change description generation.
Channel Consistency Prior and Self-Reconstruction Strategy Based Unsupervised Image Deraining
Guanglu Dong (Sichuan University), Chao Ren (Beijing Jiaotong University)
Image TranslationRestorationGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes an unsupervised single image de-raining framework called CSUD, which eliminates the dependence on real rain-clean paired data through pseudo-pair generation and self-reconstruction strategies.
Channel-wise Noise Scheduled Diffusion for Inverse Rendering in Indoor Scenes
JunYong Choi (Korea Institute of Science and Technology), Junghyun Cho (Korea Institute of Science and Technology)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: A diffusion model-based inverse rendering framework is proposed, utilizing channel-level noise scheduling to train two models separately (PDM for diversity and SDM for accuracy), achieving the simultaneous prediction of multi-modal results such as geometry, material, and lighting from a single RGB image.
Chapter-Llama: Efficient Chaptering in Hour-Long Videos with LLMs
Lucas Ventura (University Gustave Eiffel), Gül Varol (University Gustave Eiffel)
TransformerLarge Language ModelSupervised Fine-TuningVideoTextAudio
🎯 What it does: This study investigates how to automatically generate chapter divisions and corresponding titles for long videos (on the order of 1 hour), proposing the Chapter-Llama framework based on large language models.
Charm: The Missing Piece in ViT Fine-Tuning for Image Aesthetic Assessment
Fatemeh Behrad (KU Leuven University), Johan Wagemans (KU Leuven University)
ClassificationComputational EfficiencyTransformerSupervised Fine-TuningImage
🎯 What it does: A new tokenization method called Charm is proposed, which retains the composition, high definition, aspect ratio, and multi-scale information of images without modifying the pre-trained ViT structure, in order to enhance the performance of image aesthetic assessment and image quality assessment.
Chat-based Person Retrieval via Dialogue-Refined Cross-Modal Alignment
Yang Bai (Wuhan University), Mang Ye (Wuhan University)
RetrievalTransformerLarge Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: A framework for dialogue-oriented person retrieval (ChatPR) is proposed and implemented, which refines the retrieval query through interactive multi-turn dialogue and constructs the first ChatPedes dataset using dialogue data generated by large language models;
Chat2SVG: Vector Graphics Generation with Large Language Models and Image Diffusion Models
Ronghuan Wu, Jing Liao
GenerationTransformerLarge Language ModelDiffusion modelImageText
🎯 What it does: A text-to-SVG generation system (Chat2SVG) has been developed, enabling the generation of structured, editable high-quality SVG graphics from natural language prompts.
ChatGarment: Garment Estimation, Generation and Editing via Large Language Models
Siyuan Bian (Max Planck Institute for Intelligent Systems), Yao Feng (Meshcapade)
GenerationTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodalityChain-of-Thought
🎯 What it does: This paper presents ChatGarment, a system that utilizes visual language models to automatically estimate, generate, and edit 3D garment sewing patterns based on images or text.
ChatGen: Automatic Text-to-Image Generation From FreeStyle Chatting
Chengyou Jia (Xi'an Jiaotong University), Minnan Luo (Xi'an Jiaotong University)
GenerationData SynthesisTransformerLarge Language ModelPrompt EngineeringTextMultimodalityBenchmark
🎯 What it does: An automated text-to-image (Automatic T2I) system called ChatGen is proposed, which can automatically generate prompts based on free chat input, select suitable text-to-image generation models, and configure parameters.
ChatHuman: Chatting about 3D Humans with Tools
Jing Lin (Max Planck Institute for Intelligent Systems), Michael J. Black (Max Planck Institute for Intelligent Systems)
GenerationPose EstimationRetrievalTransformerLarge Language ModelAgentic AIMultimodalityRetrieval-Augmented Generation
🎯 What it does: A proxy ChatHuman based on large language models is proposed, capable of natural language interaction and completing complex 3D human tasks by calling and integrating various 3D human-related tools (such as pose estimation, shape measurement, emotion recognition, etc.).
Cheb-GR: Rethinking K-nearest Neighbor Search in Re-ranking for Person Re-identification
Jinxi Yang (Wuhan University), Mang Ye (Wuhan University)
RecognitionRetrievalGraph Neural NetworkImage
🎯 What it does: The Cheb-GR method is proposed, utilizing adaptive neighbor search guided by the Chebyshev theorem and graph convolution rearrangement, eliminating the reliance on k-nearest neighbor search in traditional rearrangement.
Chebyshev Attention Depth Permutation Texture Network with Latent Texture Attribute Loss
Ravishankar Evani (Nanyang Technological University), Shangbo Mao (Nanyang Technological University)
ClassificationRecognitionConvolutional Neural NetworkImage
🎯 What it does: A texture recognition framework combining Chebyshev Attention Depth Permutation Texture Network (CAPTN) and Stochastic Local Texture Masking (SLTM) is proposed, and end-to-end training is achieved through Latent Texture Attribute Loss.
CheckManual: A New Challenge and Benchmark for Manual-based Appliance Manipulation
Yuxing Long (Peking University), Hao Dong (Peking University)
Robotic IntelligenceTransformerLarge Language ModelTextBenchmark
🎯 What it does: Proposed the CheckManual benchmark and ManualPlan planning model, supporting learning and executing appliance operations from multi-page manuals.
CheXwhatsApp: A Dataset for Exploring Challenges in the Diagnosis of Chest X-rays through Mobile Devices
Mariamma Antony (Indian Institute of Science), Chiranjib Bhattacharyya (Indian Institute of Science)
ClassificationObject DetectionCompressionExplainability and InterpretabilityConvolutional Neural NetworkImage
🎯 What it does: The CheXwhatsApp dataset has been constructed and made publicly available, containing 141,804 pairs of original and WhatsApp-compressed chest X-ray images. Three quantification metrics, PIP, OLS, and LI, are proposed to evaluate the prediction, interpretability, and localization instability caused by compression.
CheXWorld: Exploring Image World Modeling for Radiograph Representation Learning
Yang Yue (Tsinghua University), Gao Huang (Tsinghua University)
ClassificationSegmentationRepresentation LearningTransformerContrastive LearningWorld ModelImageBiomedical Data
🎯 What it does: This paper proposes CheXWorld, a self-supervised world modeling framework for chest X-ray images, aimed at learning anatomical structures, layouts, and domain transformation knowledge.
CholecTrack20: A Multi-Perspective Tracking Dataset for Surgical Tools
Chinedu Innocent Nwoye (University of Strasbourg), Nicolas Padoy (University of Strasbourg)
Object DetectionObject TrackingVideoBenchmark
🎯 What it does: Created and publicly released the CholecTrack20 multi-view surgical tool tracking dataset, which includes multi-class and multi-tool tracking annotations for 20 complete cholecystectomy surgery videos.
Circumventing Shortcuts in Audio-visual Deepfake Detection Datasets with Unsupervised Learning
Stefan Smeu (Bitdefender), Elisabeta Oneata (Bitdefender)
ClassificationAnomaly DetectionContrastive LearningVideoMultimodalityAudio
🎯 What it does: This paper reveals the existence of a 'leading silence' short-term silence bias in audio-video deepfake datasets and proposes an unsupervised learning framework called AVH-Align, which utilizes only real samples. It enhances detection robustness by aligning AV-HuBERT self-supervised features, avoiding reliance on dataset shortcuts like brief silences.
CityWalker: Learning Embodied Urban Navigation from Web-Scale Videos
Xinhao Liu (New York University), Chen Feng (New York University)
Autonomous DrivingRobotic IntelligenceTransformerSupervised Fine-TuningSimultaneous Localization and MappingVideo
🎯 What it does: By collecting and utilizing 2000 hours of urban walking and driving videos, action pseudo-labels are generated using visual odometry to train a Transformer-based navigation model, achieving adaptive navigation for pedestrians and vehicles in urban environments.
CL-LoRA: Continual Low-Rank Adaptation for Rehearsal-Free Class-Incremental Learning
Jiangpeng He (Massachusetts Institute of Technology), Fengqing Zhu (Purdue University)
Knowledge DistillationRepresentation LearningTransformerImage
🎯 What it does: A dual-adapter architecture (CL-LoRA) is designed for class-incremental learning without rehearsal, integrating shared low-rank adapters with task-specific low-rank adapters to achieve knowledge sharing and task-specific learning.
CL-MoE: Enhancing Multimodal Large Language Model with Dual Momentum Mixture-of-Experts for Continual Visual Question Answering
Tianyu Huai (East China Normal University), Liang He (East China Normal University)
TransformerLarge Language ModelMixture of ExpertsVision Language ModelMultimodality
🎯 What it does: A dual momentum mixture of experts (CL-MoE) framework based on a multimodal large language model is proposed for continuous visual question answering tasks.
Classic Video Denoising in a Machine Learning World: Robust, Fast, and Controllable
Xin Jin (Nankai University), Chongyi Li (Nankai University)
RestorationConvolutional Neural NetworkOptical FlowVideo
🎯 What it does: This paper proposes a differentiable denoising pipeline that combines traditional video denoising techniques with deep learning, using neural networks to predict noise parameters for automatic analysis and removal of video noise.
Classifier-Free Guidance Inside the Attraction Basin May Cause Memorization
Anubhav Jain (New York University), Yuki Mitsufuji (Sony Group Corporation)
GenerationDiffusion modelImage
🎯 What it does: This study investigates the phenomenon of memorization in diffusion models and proposes techniques to reduce memorization by detecting and avoiding attractor basins in the diffusion trajectory, using dynamic or static transition points and Opposite Guidance.
Classifier-guided CLIP Distillation for Unsupervised Multi-label Classification
Dongseob Kim (Samsung Electronics), Hyunjung Shim (KAIST)
ClassificationKnowledge DistillationContrastive LearningImage
🎯 What it does: This paper proposes an unsupervised multi-label classification method called Classifier-guided CLIP Distillation (CCD), which utilizes pseudo-labels from CLIP and combines classifier-guided local views and debiasing techniques to train multi-label recognition models without manual annotation.
Classifier-to-Bias: Toward Unsupervised Automatic Bias Detection for Visual Classifiers
Quentin Guimard (University of Trento), Elisa Ricci (University of Trento)
ClassificationRetrievalLarge Language ModelImageTextRetrieval-Augmented Generation
🎯 What it does: An unsupervised framework named CLASSIFIER-TO-BIAS (C2B) is proposed, which can automatically identify the biases of visual classifiers solely based on the textual description of the given classification task and a pre-trained model.
CleanDIFT: Diffusion Features without Noise
Nick Stracke (LMU Munich), Björn Ommer (LMU Munich)
ClassificationSegmentationDepth EstimationDiffusion modelImage
🎯 What it does: This paper proposes the CleanDIFT method, which utilizes lightweight unsupervised fine-tuning to directly extract noise-agnostic and time-step-agnostic features from clean images using a pre-trained diffusion model.
ClearSight: Visual Signal Enhancement for Object Hallucination Mitigation in Multimodal Large Language Models
Hao Yin (University of Science and Technology of China), Zilei Wang (University of Science and Technology of China)
Object DetectionGenerationTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: A training-free and tool-free visual enhancement fusion method (VAF) is proposed, which enhances the attention to visual signals in multimodal large language models by adjusting the mid-layer attention weights, thereby reducing the generation of false objects.
ClimbingCap: Multi-Modal Dataset and Method for Rock Climbing in World Coordinate
Ming Yan (Xiamen University), Cheng Wang (Xiamen University)
Pose EstimationTransformerSupervised Fine-TuningMultimodalityPoint Cloud
🎯 What it does: A global climbing motion recovery method based on RGB+LiDAR is proposed, and the largest climbing dataset AscendMotion is released.
CLIP is Almost All You Need: Towards Parameter-Efficient Scene Text Retrieval without OCR
Xugong Qin (Nanjing University of Science and Technology), Pengwen Dai (Chinese Academy of Sciences)
RetrievalTransformerVision Language ModelContrastive LearningImageText
🎯 What it does: This paper proposes a scene text retrieval method called CAYN that is completely independent of OCR, utilizing CLIP for image-text retrieval and re-ranking.
CLIP is Strong Enough to Fight Back: Test-time Counterattacks towards Zero-shot Adversarial Robustness of CLIP
Songlong Xing (University of Trento), Nicu Sebe (University of Trento)
ClassificationAdversarial AttackTransformerContrastive LearningImage
🎯 What it does: This study investigates the robustness of CLIP under adversarial attacks and proposes a counterattack method during testing that utilizes a pretrained visual encoder to enhance zero-shot robustness against adversarial attacks.
CLIP Under the Microscope: A Fine-Grained Analysis of Multi-Object Representation
Reza Abbasi (Sharif University of Technology), Mahdieh Soleymani Baghshah (Sharif University of Technology)
RetrievalTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: This paper systematically studies the bias of the CLIP model's image and text encoders in multi-object scenes and conducts a quantitative evaluation using the designed ComCO dataset.
CLIP-driven Coarse-to-fine Semantic Guidance for Fine-grained Open-set Semi-supervised Learning
Xiaokun Li (Beijing Jiaotong University), Qingji Guan (Beijing Jiaotong University)
ClassificationRecognitionTransformerVision Language ModelContrastive LearningImage
🎯 What it does: A coarse-fine layered semantic guidance framework based on CLIP (CFSG-CLIP) is designed, which first filters local visual features related to categories through a semantic filtering module, and then further refines the features in the fine-grained branch through a visual semantic injection strategy to achieve efficient differentiation between fine-grained IDs and OOD samples.
CLOC: Contrastive Learning for Ordinal Classification with Multi-Margin N-pair Loss
Dileepa Pitawela (University of Adelaide), Hsiang-Ting Chen (University of Adelaide)
ClassificationExplainability and InterpretabilityConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes an ordinal classification method CLOC based on multi-margin N-pair contrastive learning, which utilizes learnable margins to preserve category order and reduce critical decision errors.
Closed-Loop Supervised Fine-Tuning of Tokenized Traffic Models
Zhejun Zhang (NVIDIA Research), Marco Pavone (Stanford University)
Autonomous DrivingSupervised Fine-TuningPoint Cloud
🎯 What it does: A closed-loop supervised fine-tuning method named CAT-K is proposed to improve traffic simulation models based on next-token prediction, reducing the covariance shift problem between training and deployment.
Closest Neighbors are Harmful for Lightweight Masked Auto-encoders
Jian Meng (Cornell University), Jae-sun Seo (Cornell University)
Representation LearningTransformerAuto EncoderImage
🎯 What it does: A self-supervised pre-training method for lightweight visual Transformers, NoR-MAE, is proposed. By incorporating the concept of nearest neighbor patches (CNP) into the Masked Autoencoder and performing semantic rebounding, it addresses the issue of semantic entanglement that arises in lightweight models during reconstruction tasks.
CMMLoc: Advancing Text-to-PointCloud Localization with Cauchy-Mixture-Model Based Framework
Yanlong Xu (University of Science and Technology of China), Xun Yang (University of Science and Technology of China)
Pose EstimationRetrievalAutonomous DrivingTransformerContrastive LearningTextPoint Cloud
🎯 What it does: A text-to-point cloud localization framework called CMMLoc based on the Cauchy mixture model is proposed, addressing the issue of partial correlation between text descriptions and point clouds.
Co-op: Correspondence-based Novel Object Pose Estimation
Sungphill Moon (NAVER LABS), Sangwook Kim (NAVER LABS)
Pose EstimationTransformerImage
🎯 What it does: A Co-op framework is proposed to achieve 6DoF pose estimation of novel objects from a single RGB image using only the CAD model of the target object, without any additional fine-tuning.
Co-Speech Gesture Video Generation with Implicit Motion-Audio Entanglement
Xinjie Li (Pennsylvania State University), Jing Xiao (PingAn Technology)
GenerationData SynthesisPose EstimationTransformerDiffusion modelVideoAudio
🎯 What it does: Proposes the Implicit Motion-Audio Entanglement (IMAE) method, which generates realistic upper-body pose videos directly from audio without using additional inputs.
CO-SPY: Combining Semantic and Pixel Features to Detect Synthetic Images by AI
Siyuan Cheng (Purdue University), Vikash Sehwag (Sony AI)
Data SynthesisAnomaly DetectionAuto EncoderContrastive LearningImageBenchmark
🎯 What it does: Proposes the CO-Spy framework, which combines semantic features and pixel-level traces to detect AI-generated images.
CoA: Towards Real Image Dehazing via Compression-and-Adaptation
Long Ma, Zhuo Su
RestorationDomain AdaptationOptimizationComputational EfficiencyKnowledge DistillationContrastive LearningImage
🎯 What it does: A dual-stage learning framework based on compression and adaptation (CoA) is proposed, which performs adaptation through dual-layer optimization in the real domain after compressing the model in the synthetic domain;
COAP: Memory-Efficient Training with Correlation-Aware Gradient Projection
Jinqi Xiao (ByteDance Inc.), Bo Yuan (ByteDance Inc.)
GenerationOptimizationComputational EfficiencyDiffusion modelImageMultimodality
🎯 What it does: This paper proposes a low-rank gradient projection COAP method that significantly reduces the memory usage of the optimizer while maintaining training performance.
Coarse Correspondences Boost Spatial-Temporal Reasoning in Multimodal Language Model
Benlin Liu (Tencent), Ranjay Krishna (University of Washington)
Object DetectionObject TrackingSegmentationTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageVideoMultimodality
🎯 What it does: By visually prompting coarse-grained correspondences in 2D images, the spatial-temporal reasoning ability of multimodal large language models is enhanced.
COB-GS: Clear Object Boundaries in 3DGS Segmentation Based on Boundary-Adaptive Gaussian Splitting
Jiaxin Zhang (Harbin Institute of Technology), Xianming Liu (Harbin Institute of Technology)
SegmentationOptimizationGaussian SplattingPoint Cloud
🎯 What it does: This paper addresses the issue of boundary ambiguity in 3D Gaussian Splatting (3DGS) for object segmentation and proposes a joint optimization method for semantics and texture segmentation—COB-GS.
COBRA: COmBinatorial Retrieval Augmentation for Few-Shot Adaptation
Arnav M. Das (University of Washington), Jeff Bilmes (University of Washington)
RetrievalDomain AdaptationContrastive LearningImage
🎯 What it does: This paper proposes COBRA, a retrieval enhancement method based on Combinatorial Mutual Information (CMI), designed to select diverse and relevant samples from large auxiliary datasets (such as LAION-2B) in few-shot learning to improve model performance.
CocoER: Aligning Multi-Level Feature by Competition and Coordination for Emotion Recognition
Xuli Shen (Fudan University), Xiangyang Xue (Fudan University)
RecognitionTransformerContrastive LearningImageMultimodality
🎯 What it does: A multi-layer visual feature alignment framework called CocoER is proposed, which is based on competitive and collaborative mechanisms. It extracts head, body, and background features through cross-layer attention and generates pseudo-labels via a vocabulary information alignment module to guide feature refinement.
CoCoGaussian: Leveraging Circle of Confusion for Gaussian Splatting from Defocused Images
Jungho Lee (Yonsei University), Sangyoun Lee (Yonsei University)
RestorationGenerationDepth EstimationNeural Radiance FieldGaussian SplattingImage
🎯 What it does: A 3D Gaussian blur method based on Circle of Confusion (CoC) is proposed, which can complete 3D scene reconstruction and new viewpoint rendering using only defocused images.
Code-as-Monitor: Constraint-aware Visual Programming for Reactive and Proactive Robotic Failure Detection
Enshen Zhou (Beihang University), He Wang (Peking University)
Anomaly DetectionRobotic IntelligenceTransformerLarge Language ModelVision Language ModelImageVideo
🎯 What it does: Proposes the Code-as-Monitor framework, unifying open-set reactive and proactive fault detection into a spatiotemporal constraint satisfaction problem, utilizing visual language models to generate monitoring code for real-time detection.
CoE: Chain-of-Explanation via Automatic Visual Concept Circuit Description and Polysemanticity Quantification
Wenlong Yu (Tianjin University), Qinghua Hu (Tianjin University)
Explainability and InterpretabilityConvolutional Neural NetworkTransformerLarge Language ModelVision Language ModelImage
🎯 What it does: A deep visual model interpretability framework based on automated concept decoding, separation, ambiguity quantification, and chain-based explanation is proposed.
Coeff-Tuning: A Graph Filter Subspace View for Tuning Attention-Based Large Models
Zichen Miao (Purdue University), Qiang Qiu (Purdue University)
ClassificationRecognitionGraph Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningImageTextMultimodality
🎯 What it does: A parameter-efficient fine-tuning method called Coeff-Tuning is proposed, which enhances the expressive power of large models by learning a small number of subspace coefficients to recombine the filter subspaces of multi-head attention.
Coherent 3D Portrait Video Reconstruction via Triplane Fusion
Shengze Wang (UNC Chapel Hill), Koki Nagano (NVIDIA)
RestorationGenerationData SynthesisConvolutional Neural NetworkGenerative Adversarial NetworkVideo
🎯 What it does: This paper proposes a single-view 3D portrait video reconstruction method based on triplane fusion, achieving time-consistent, realistic, and dynamically expressive 3D portrait video generation.
ColabSfM: Collaborative Structure-from-Motion by Point Cloud Registration
Johan Edstedt (Linköping University), Alberto Jaenal (Ericsson Research)
TransformerPoint Cloud
🎯 What it does: A collaborative SfM (ColabSfM) framework is proposed, which achieves alignment of different SfM maps through point cloud registration without sharing visual descriptors.
Collaborative Decoding Makes Visual Auto-Regressive Modeling Efficient
Zigeng Chen (National University of Singapore), Xinchao Wang (National University of Singapore)
GenerationComputational EfficiencyKnowledge DistillationTransformerSupervised Fine-TuningImage
🎯 What it does: This paper proposes Collaborative Decoding (CoDe), which splits the coarse and fine scale generation tasks of visual autoregressive models into a large model responsible for low-frequency global sketches and a small model responsible for high-frequency details, significantly improving inference speed and memory utilization.
Collaborative Tree Search for Enhancing Embodied Multi-Agent Collaboration
Lizheng Zu (Nanyang Technological University), Pan Zhou (Harbin Institute of Technology)
Robotic IntelligenceReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningAgentic AIMultimodality
🎯 What it does: The CoTS framework is proposed, utilizing a modified Monte Carlo Tree Search and LLM reward function to enhance multi-agent collaborative planning and execution in embodied environments.
CoLLM: A Large Language Model for Composed Image Retrieval
Chuong Huynh (University of Maryland), Abhinav Shrivastava (University of Maryland)
RetrievalTransformerLarge Language ModelVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: In the absence of labeled CIR triplets, the CoLLM framework is proposed, utilizing image-caption pairs to generate synthetic triplets through Slerp and text interpolation, combined with LLM to generate composite query embeddings, completing zero-shot or self-supervised CIR training.
Color Alignment in Diffusion
Ka Chun Shum (Hong Kong University of Science and Technology), Sai-Kit Yeung (Hong Kong University of Science and Technology)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: This paper proposes a method for achieving fine-grained color conditional generation in diffusion models, capable of generating images that are highly consistent with given color patterns (color values and proportions) while maintaining diversity and quality of content.
CoMapGS: Covisibility Map-based Gaussian Splatting for Sparse Novel View Synthesis
Youngkyoon Jang (Huawei Noah's Ark Lab), Eduardo Pérez-Pellitero (Huawei Noah's Ark Lab)
GenerationData SynthesisGaussian SplattingPoint Cloud
🎯 What it does: The CoMapGS framework is proposed, which significantly improves the geometric and texture quality in sparse view synthesis through pixel-level visibility maps guiding point cloud initialization and adaptive weighted supervision.