CVPR 2024 Papers — Page 5
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2716 papers
Content-Style Decoupling for Unsupervised Makeup Transfer without Generating Pseudo Ground Truth
Zhaoyang Sun (Wuhan University of Technology), Yi Rong (Wuhan University of Technology)
Image TranslationGenerationGenerative Adversarial NetworkImage
🎯 What it does: An unsupervised makeup transfer method CSD-MT is proposed, which decouples content and makeup style through frequency decomposition and trains directly on source and reference images.
ConTex-Human: Free-View Rendering of Human from a Single Image with Texture-Consistent Synthesis
Xiangjun Gao (Hong Kong University of Science and Technology), Long Quan (Hong Kong University of Science and Technology)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: A high-fidelity, texture-consistent 3D free-viewpoint rendering method for human bodies, called ConTex-Human, has been designed using a single image.
Context-Aware Integration of Language and Visual References for Natural Language Tracking
Yanyan Shao (Zhejiang University of Technology), Jiming Chen (Zhejiang University)
Object TrackingRetrievalTransformerPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes QueryNLT, an end-to-end natural language tracking framework that integrates language descriptions with visual templates, consisting of two main modules: Prompt Modulation and Target Decoding.
Context-based and Diversity-driven Specificity in Compositional Zero-Shot Learning
Yun Li (CSIRO Data61), Lina Yao (CSIRO Data61)
ClassificationRecognitionTransformerPrompt EngineeringVision Language ModelImage
🎯 What it does: A context and diversity-driven specificity learning framework (CDS-CZSL) is proposed, which enhances the accuracy of compositional zero-shot learning (CZSL) by evaluating and utilizing attribute specificity.
Context-Guided Spatio-Temporal Video Grounding
Xin Gu (University of Chinese Academy of Sciences), Libo Zhang (University of Chinese Academy of Sciences)
RecognitionObject DetectionTransformerVideoText
🎯 What it does: In the video localization task with a given text query, the authors propose the Context-Guided STVG framework (CG-STVG), which assists target localization by mining instance visual context from the video itself.
Contextrast: Contextual Contrastive Learning for Semantic Segmentation
Changki Sung (Korea Advanced Institute of Science and Technology), Hyun Myung (Korea Advanced Institute of Science and Technology)
SegmentationContrastive LearningImage
🎯 What it does: A semantic segmentation framework named Contextrast is proposed, which utilizes multi-scale features to construct representative anchors for context contrastive learning, and enhances edge details through boundary-aware negative sample sampling (BANE).
ContextSeg: Sketch Semantic Segmentation by Querying the Context with Attention
Jiawei Wang (Shandong University), Changjian Li (University of Edinburgh)
SegmentationConvolutional Neural NetworkTransformerAuto EncoderImage
🎯 What it does: This paper proposes the ContextSeg method for semantic segmentation of hand-drawn sketches, which consists of two stages: first, using an autoencoder to learn the structure and positional information of strokes, and then using a autoregressive Transformer to group and label the strokes.
Contextual Augmented Global Contrast for Multimodal Intent Recognition
Kaili Sun (Wuhan University), Huyin Zhang (Wuhan University)
RecognitionTransformerContrastive LearningVideoMultimodality
🎯 What it does: This paper proposes a Context-Aware Global Contrast Network (CAGC) based on cross-video global contrast for multimodal intent recognition, enhancing the model's ability to recognize intents through cross-video contextual learning and global contrast learning.
Continual Forgetting for Pre-trained Vision Models
Hongbo Zhao (State Key Laboratory of Multimodal Artificial Intelligence Systems, Institute of Automation, Chinese Academy of Sciences), Zhaoxiang Zhang (State Key Laboratory of Multimodal Artificial Intelligence Systems, Institute of Automation, Chinese Academy of Sciences)
RecognitionObject DetectionSafty and PrivacyTransformerSupervised Fine-TuningImage
🎯 What it does: This paper proposes the problem of 'Continual Forgetting', aiming to sequentially and gradually delete specified knowledge (such as privacy, bias, etc.) from pre-trained visual models while keeping the remaining knowledge unaffected. To achieve this, the authors designed the GS-LoRA method: adding a LoRA low-rank adaptation module to the FFN layer of the Transformer block and automatically selecting the Transformer blocks that need modification through group sparse regularization, thus enabling efficient and precise forgetting.
Continual Learning for Motion Prediction Model via Meta-Representation Learning and Optimal Memory Buffer Retention Strategy
DaeJun Kang (Korea Automotive Technology Institute), Sanmin Kim (Korea Advanced Institute of Science and Technology)
Autonomous DrivingRepresentation LearningMeta LearningConvolutional Neural NetworkReinforcement LearningTime Series
🎯 What it does: A continuous motion prediction framework CMP is proposed, combining sparse representation learning and optimal memory buffer retention strategies to address the catastrophic forgetting and continuous updating issues of motion prediction models in autonomous driving.
Continual Segmentation with Disentangled Objectness Learning and Class Recognition
Yizheng Gong (University of Liverpool), Jimin Xiao (Xi'an Jiaotong-Liverpool University)
Object DetectionSegmentationKnowledge DistillationTransformerContrastive LearningImage
🎯 What it does: This paper proposes the CoMasTRe framework, which achieves continuous segmentation by separating object-centric learning and category recognition.
Continual Self-supervised Learning: Towards Universal Multi-modal Medical Data Representation Learning
Yiwen Ye (Northwestern Polytechnical University), Yong Xia (Northwestern Polytechnical University)
Representation LearningContrastive LearningMultimodalityBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: This paper proposes a continuous self-supervised learning framework MedCoSS for constructing multimodal medical pre-training models, addressing the issues of multimodal data conflicts and model scalability.
Continual-MAE: Adaptive Distribution Masked Autoencoders for Continual Test-Time Adaptation
Jiaming Liu (Peking University), Shanghang Zhang (Peking University)
ClassificationSegmentationDomain AdaptationTransformerAuto EncoderImage
🎯 What it does: This study proposes the Adaptive Distribution Masked Autoencoders (ADMA) framework for continuous test-time adaptation (CTTA) of unlabeled data in the context of changing target domain distributions.
Continuous Optical Zooming: A Benchmark for Arbitrary-Scale Image Super-Resolution in Real World
Huiyuan Fu (Beijing University of Posts and Telecommunications), Huadong Ma (Beijing University of Posts and Telecommunications)
RestorationSuper ResolutionMeta LearningImageBenchmark
🎯 What it does: This paper constructs a continuous optical zoom dataset COZ and proposes the Local Mix Implicit Network (LMI) based on MLP-Mixer and meta-learning for super-resolution of images at arbitrary scales in realistic scenes.
Continuous Pose for Monocular Cameras in Neural Implicit Representation
Qi Ma (ETH Zurich), Luc Van Gool (ETH Zurich)
Pose EstimationNeural Radiance FieldSimultaneous Localization and MappingImageVideo
🎯 What it does: This paper proposes an end-to-end optimization framework that represents the pose of a monocular camera as a time-continuous neural function, integrating it with NeRF, event cameras, visual SLAM, and IMU.
Contrasting Intra-Modal and Ranking Cross-Modal Hard Negatives to Enhance Visio-Linguistic Compositional Understanding
Le Zhang (Mila - Quebec AI Institute), Aishwarya Agrawal (Mila - Quebec AI Institute)
RetrievalRepresentation LearningTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: In visual-language models like CLIP, incorporating hard negative sample enhancement improves the fine-grained correspondence between images and texts, thereby enhancing combinatorial reasoning capabilities.
Contrastive Denoising Score for Text-guided Latent Diffusion Image Editing
Hyelin Nam (Korea Advanced Institute of Science and Technology), Jong Chul Ye (Korea Advanced Institute of Science and Technology)
Image TranslationGenerationDiffusion modelScore-based ModelContrastive LearningImageText
🎯 What it does: This paper proposes the Contrastive Denoising Score (CDS), which achieves text-guided latent diffusion image editing through zero-shot training by incorporating the CUT loss into the Delta Denoising Score (DDS) framework.
Contrastive Learning for DeepFake Classification and Localization via Multi-Label Ranking
Cheng-Yao Hong (Academia Sinica), Tyng-Luh Liu (Academia Sinica)
ClassificationRecognitionConvolutional Neural NetworkTransformerContrastive LearningVideo
🎯 What it does: Unified solution for deepfake binary classification, local localization, and forgery sequence recognition
Contrastive Mean-Shift Learning for Generalized Category Discovery
Sua Choi (Pohang University of Science and Technology), Minsu Cho (Pohang University of Science and Technology)
ClassificationRecognitionContrastive LearningImage
🎯 What it does: This paper proposes a framework that combines mean shift and contrastive learning for general category discovery (GCD) on partially labeled image collections.
Contrastive Pre-Training with Multi-View Fusion for No-Reference Point Cloud Quality Assessment
Ziyu Shan (Shanghai Jiao Tong University), Shan Liu (Tencent)
RecognitionData SynthesisTransformerContrastive LearningPoint Cloud
🎯 What it does: This paper proposes a no-reference point cloud quality assessment framework called CoPA based on contrastive learning, which generates anchors by mixing local blocks of projected images and achieves high-quality predictions through semantic-guided multi-view fusion during the fine-tuning phase.
Control4D: Efficient 4D Portrait Editing with Text
Ruizhi Shao (Tsinghua University), Yebin Liu (Tsinghua University)
GenerationData SynthesisComputational EfficiencyDiffusion modelGenerative Adversarial NetworkGaussian SplattingVideo
🎯 What it does: The Control4D framework is proposed, enabling high-fidelity and temporally consistent editing of dynamic 4D portraits solely based on text instructions.
ControlRoom3D: Room Generation using Semantic Proxy Rooms
Jonas Schult (Meta GenAI), Ji Hou (Meta GenAI)
GenerationData SynthesisDepth EstimationDiffusion modelMesh
🎯 What it does: A control method called ControlRoom3D is proposed, which is based on user-specified 3D semantic proxy rooms and text descriptions to generate high-quality 3D room meshes.
ConvoFusion: Multi-Modal Conversational Diffusion for Co-Speech Gesture Synthesis
Muhammad Hamza Mughal (Max Planck Institute for Informatics), Christian Theobalt (Max Planck Institute for Informatics)
GenerationData SynthesisTransformerDiffusion modelAuto EncoderTextMultimodalityAudio
🎯 What it does: A multimodal dialogue gesture synthesis framework called ConvoFusion based on diffusion models is proposed, which can generate voice-driven full-body and hand gestures in both single and dual-person dialogues, and supports controllability of text and audio.
Convolutional Prompting meets Language Models for Continual Learning
Anurag Roy (Indian Institute of Technology Kharagpur), Abir Das (Indian Institute of Technology Kharagpur)
ClassificationRecognitionConvolutional Neural NetworkTransformerLarge Language ModelPrompt EngineeringImage
🎯 What it does: The paper proposes a continuous learning method based on convolutional prompts, ConvPrompt, which can adapt to new tasks without storing old data.
Cooperation Does Matter: Exploring Multi-Order Bilateral Relations for Audio-Visual Segmentation
Qi Yang (University of Chinese Academy of Sciences), Shiming Xiang (Chinese Academy of Sciences)
SegmentationTransformerVideoMultimodalityAudio
🎯 What it does: This paper proposes the COMBO framework, which combines pixel, modality, and temporal three-way bidirectional coupling relationships to achieve audio-visual segmentation tasks.
CoralSCOP: Segment any COral Image on this Planet
Ziqiang Zheng (Hong Kong University of Science and Technology), Sai-Kit Yeung (Hong Kong University of Science and Technology)
SegmentationTransformerPrompt EngineeringImage
🎯 What it does: A basic segmentation model for coral reef images, CoralSCOP, has been developed, capable of achieving automatic or semi-automatic dense coral segmentation under various interaction modes such as no prompt, single point, or single point + box.
CORE-MPI: Consistency Object Removal with Embedding MultiPlane Image
Donggeun Yoon (Chungnam National University), Donghyeon Cho (Hanyang University)
RestorationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: A method for object removal based on multi-plane images (MPI) called CORE-MPI is proposed, which utilizes embedded images to achieve online interaction and completes object masking and restoration on the embedded images.
CORES: Convolutional Response-based Score for Out-of-distribution Detection
Keke Tang (Guangzhou University), Zhihong Tian (Guangzhou University)
Anomaly DetectionConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a method for out-of-distribution (OOD) detection based on convolutional kernel responses, called CORES, which can determine whether the input belongs to the training distribution without requiring ID data.
Correcting Diffusion Generation through Resampling
Yujian Liu (University of California Santa Barbara), Shiyu Chang (University of California Santa Barbara)
Object DetectionGenerationDiffusion modelImageText
🎯 What it does: A sampling framework based on particle filtering is proposed to correct errors of missing objects and low image quality when generating images with diffusion models, mainly by reducing the gap between the generated distribution and the real distribution.
Correlation-aware Coarse-to-fine MLPs for Deformable Medical Image Registration
Mingyuan Meng (University of Sydney), Jinman Kim (University of Sydney)
Image TranslationSegmentationConvolutional Neural NetworkImageBiomedical DataMagnetic Resonance ImagingAlzheimer's Disease
🎯 What it does: This paper proposes a coarse-fine level co-registration network for medical image deformation based on a multi-layer perceptron (CorrMLP), and conducts unsupervised training and evaluation on brain MRI and cardiac MRI.
Correlation-Decoupled Knowledge Distillation for Multimodal Sentiment Analysis with Incomplete Modalities
Mingcheng Li (Fudan University), Lihua Zhang (Fudan University)
Knowledge DistillationRepresentation LearningTransformerContrastive LearningMultimodality
🎯 What it does: The CorrKD framework is proposed to address the issue of incomplete modalities in multimodal emotion analysis through teacher-student distillation, enhancing model robustness.
Correspondence-Free Non-Rigid Point Set Registration Using Unsupervised Clustering Analysis
Mingyang Zhao (Chinese Academy of Sciences), Dong-Ming Yan (Chinese Academy of Sciences)
OptimizationImagePoint CloudMagnetic Resonance ImagingComputed Tomography
🎯 What it does: A correspondence-free non-rigid point set registration framework is proposed, transforming the registration problem into unsupervised clustering, and a closed-form solution is provided.
CorrMatch: Label Propagation via Correlation Matching for Semi-Supervised Semantic Segmentation
Boyuan Sun (Nankai University), Qibin Hou (Nankai University)
SegmentationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes CorrMatch, a semi-supervised semantic segmentation method that achieves pixel and region propagation through correlation matching, utilizing the correlation information of unlabeled images to generate more accurate pseudo-labels.
CosalPure: Learning Concept from Group Images for Robust Co-Saliency Detection
Jiayi Zhu (East China Normal University), Geguang Pu (East China Normal University)
RestorationObject DetectionDiffusion modelImage
🎯 What it does: By learning common concepts in a group of images and using these concepts to guide diffusion models for image purification, the robustness of co-occurring object detection against adversarial attacks and common distortions is enhanced.
CoSeR: Bridging Image and Language for Cognitive Super-Resolution
Haoze Sun (Tsinghua University), Yujiu Yang (Tsinghua University)
RestorationSuper ResolutionDiffusion modelImageText
🎯 What it does: A cognitive super-resolution framework called CoSeR is proposed for low-resolution images. It first extracts cognitive embeddings, then uses these embeddings to generate reference images that are semantically consistent and texture-similar. Subsequently, the LR image, cognitive embeddings, and reference images are simultaneously injected into a diffusion-based SR network to achieve high-quality super-resolution.
CosmicMan: A Text-to-Image Foundation Model for Humans
Shikai Li (Shanghai AI Laboratory), Wayne Wu (Shanghai AI Laboratory)
GenerationData SynthesisDiffusion modelImageText
🎯 What it does: A human-oriented text-to-image generation foundational model called CosmicMan has been constructed, along with a sustainable data production process named Annotate Anyone.
COTR: Compact Occupancy TRansformer for Vision-based 3D Occupancy Prediction
Qihang Ma (East China Normal University), Yuan Xie (East China Normal University)
Object DetectionSegmentationAutonomous DrivingTransformerImagePoint CloudBenchmark
🎯 What it does: Proposes the Compact Occupancy Transformer (COTR), achieving vision-based 3D occupancy prediction, constructing a compact 3D OCC representation, and enhancing semantic recognition.
Countering Personalized Text-to-Image Generation with Influence Watermarks
Hanwen Liu (Peking University), Yadong Mu (Peking University)
GenerationData SynthesisCompressionDiffusion modelImage
🎯 What it does: An influence watermarking method named InMark is proposed, making personal images unlearnable when fine-tuning diffusion models, thereby preventing unauthorized text-to-image generation.
Coupled Laplacian Eigenmaps for Locally-Aware 3D Rigid Point Cloud Matching
Matteo Bastico (Mines Paris), David Ryckelynck
Pose EstimationAnomaly DetectionPoint Cloud
🎯 What it does: The Coupled Laplacian is proposed to generate an aligned Laplacian feature space by adding cross edges on registered point clouds, achieving unsupervised 3D rigid point cloud matching for local geometric details, and applying it to bone side estimation and industrial 3D anomaly detection.
CPGA: Coding Priors-Guided Aggregation Network for Compressed Video Quality Enhancement
Qiang Zhu (University of Electronic Science and Technology of China), Shuyuan Zhu (Kuaishou Technology)
RestorationCompressionConvolutional Neural NetworkOptical FlowVideo
🎯 What it does: This paper proposes a coding prior-based aggregation network (CPGA) and a corresponding coding prior dataset (VCP) to enhance the quality of compressed videos.
CPLIP: Zero-Shot Learning for Histopathology with Comprehensive Vision-Language Alignment
Sajid Javed (Khalifa University of Science and Technology), Mohammed Bennamoun (University of the Western Australia)
ClassificationSegmentationRetrievalTransformerLarge Language ModelVision Language ModelContrastive LearningImageTextMultimodalityBiomedical Data
🎯 What it does: An unsupervised pathological visual language pre-training framework, CPLIP, has been developed, which utilizes a large number of unlabeled pathological images and a custom pathological dictionary to generate diverse text prompts. It achieves alignment between images and text through multi-to-multi aligned MIL-NCE contrastive learning, thereby enhancing zero-shot classification and segmentation performance.
CPP-Net: Embracing Multi-Scale Feature Fusion into Deep Unfolding CP-PPA Network for Compressive Sensing
Zhen Guo (Northwestern Polytechnical University), Hongping Gan (Northwestern Polytechnical University)
RestorationCompressionConvolutional Neural NetworkTransformerImageMagnetic Resonance Imaging
🎯 What it does: This paper proposes a deep unfolding network called CPP-Net based on CP-PPA for image compressed sensing reconstruction, enhancing detail recovery through multi-scale feature fusion (DPFB) and iterative stage fusion (IFS).
CPR-Coach: Recognizing Composite Error Actions based on Single-class Training
Shunli Wang (Fudan University), Lihua Zhang (Fudan University)
RecognitionConvolutional Neural NetworkOptical FlowVideoMultimodality
🎯 What it does: This paper first constructs a vision-based CPR-COACH dataset, which includes 13 types of single error actions and 74 types of combined error actions, and proposes a human cognition-inspired ImagineNet framework for recognizing composite error actions in a single-class training and multi-class testing scenario.
CPR: Retrieval Augmented Generation for Copyright Protection
Aditya Golatkar (AWS AI Labs), Stefano Soatto (AWS AI Labs)
GenerationRetrievalSafty and PrivacyComputational EfficiencyTransformerDiffusion modelImageRetrieval-Augmented Generation
🎯 What it does: Proposed and implemented the CPR (Copy-Protected Retrieval) method, achieving copyright protection and privacy security in retrieval-augmented generation with diffusion models.
CricaVPR: Cross-image Correlation-aware Representation Learning for Visual Place Recognition
Feng Lu (Tsinghua University), Chun Yuan (University of Chinese Academy of Sciences)
RecognitionRepresentation LearningTransformerContrastive LearningImage
🎯 What it does: This study investigates the global feature representation for visual place recognition and proposes a cross-image correlation-aware method.
CRKD: Enhanced Camera-Radar Object Detection with Cross-modality Knowledge Distillation
Lingjun Zhao (University of Michigan), Katherine A. Skinner (University of Michigan)
Object DetectionAutonomous DrivingKnowledge DistillationMultimodalityPoint Cloud
🎯 What it does: A cross-modal knowledge distillation framework CRKD is proposed, which distills a LiDAR-Camera prior fusion model into a Camera-Radar model to improve 3D object detection performance.
CroSel: Cross Selection of Confident Pseudo Labels for Partial-Label Learning
Shiyu Tian (Chongqing University), Lei Feng (Singapore University of Technology and Design)
ClassificationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes a cross-selection method called CroSel, which uses historical predictions to identify true labels in partial label learning and enhances training effectiveness through co-mix consistency regularization.
Cross Initialization for Face Personalization of Text-to-Image Models
Lianyu Pang (Sun Yat-sen University), Xudong Mao (Hong Kong Polytechnic University)
GenerationOptimizationDiffusion modelImageText
🎯 What it does: This paper studies a facial personalization method for text-to-image diffusion models, proposing cross-initialization and regularization strategies to enhance reconstruction quality and editability, while significantly accelerating the optimization process.
Cross-Dimension Affinity Distillation for 3D EM Neuron Segmentation
Xiaoyu Liu (University of Science and Technology of China), Zhiwei Xiong (University of Science and Technology of China)
SegmentationKnowledge DistillationConvolutional Neural NetworkSupervised Fine-TuningBiomedical Data
🎯 What it does: A lightweight 2D convolutional network is used to generate a 3D affinity map, enabling efficient segmentation of neurons in electron microscopy (EM) volumes.
Cross-Domain Few-Shot Segmentation via Iterative Support-Query Correspondence Mining
Jiahao Nie (Nanyang Technological University), Shijian Lu (Nanyang Technological University)
SegmentationDomain AdaptationMeta LearningConvolutional Neural NetworkImage
🎯 What it does: This paper studies cross-domain few-shot semantic segmentation (CD-FSS) and proposes an Iterative Few-Shot Adapter (IFA) to maximize the utilization of the limited supervisory signals from support samples, addressing the dual challenges of cross-domain transfer and overfitting.
Cross-spectral Gated-RGB Stereo Depth Estimation
Samuel Brucker (Torc Robotics), Felix Heide (Princeton University)
Depth EstimationTransformerImagePoint Cloud
🎯 What it does: A cross-spectral stereo depth estimation framework is proposed, combining a gated camera with a high-resolution RCCB camera, utilizing multi-view and time-gated information to achieve high-precision depth inference at long distances.
Cross-view and Cross-pose Completion for 3D Human Understanding
Matthieu Armando, Grégory Rogez
ClassificationRestorationPose EstimationTransformerSupervised Fine-TuningImageVideoMesh
🎯 What it does: This paper learns a general human representation through cross-view and cross-pose self-supervised pre-training, and fine-tunes it on tasks such as human mesh recovery, hand mesh recovery, pose estimation, DensePose, and gesture/grasp classification.
CrossKD: Cross-Head Knowledge Distillation for Object Detection
Jiabao Wang (Nankai University), Qibin Hou (Nankai University)
Object DetectionKnowledge DistillationImage
🎯 What it does: A new knowledge distillation method called CrossKD is proposed, where the intermediate features of the student detection head are sent to the teacher detection head to generate cross predictions. These cross predictions are then aligned with the teacher's predictions, alleviating target conflicts in prediction simulation and enhancing the student's detection performance.
CrossMAE: Cross-Modality Masked Autoencoders for Region-Aware Audio-Visual Pre-Training
Yuxin Guo (University of Chinese Academy of Sciences), Yun Zheng (Alibaba Group)
ClassificationRetrievalTransformerAuto EncoderContrastive LearningMultimodalityAudio
🎯 What it does: Proposes CrossMAE, a cross-modal pre-training model with a dual-layer masked autoencoder for audio-visual tasks.
CrowdDiff: Multi-hypothesis Crowd Density Estimation using Diffusion Models
Yasiru Ranasinghe (Johns Hopkins University), Vishal M. Patel (Johns Hopkins University)
Object DetectionGenerationDiffusion modelImageBenchmark
🎯 What it does: This paper proposes CrowdDiff, which utilizes a conditional diffusion model to generate narrow kernel crowd density maps and achieves counting through threshold detection.
CSTA: CNN-based Spatiotemporal Attention for Video Summarization
Jaewon Son (Sungkyunkwan University), Kwangsu Kim (Sungkyunkwan University)
Computational EfficiencyConvolutional Neural NetworkVideoBenchmark
🎯 What it does: A spatiotemporal attention model based on CNN, CSTA, is proposed, which concatenates frame features into images, utilizes 2D convolution to generate attention, and predicts important frames for video summarization.
Curriculum Point Prompting for Weakly-Supervised Referring Image Segmentation
Qiyuan Dai (ShanghaiTech University), Sibei Yang (ShanghaiTech University)
Object DetectionSegmentationTransformerPrompt EngineeringContrastive LearningImage
🎯 What it does: The paper proposes a weakly supervised referential image segmentation framework called Point Prompt Learning (PPT), which utilizes a multi-source curriculum learning strategy to achieve semantic alignment and fine segmentation using frozen CLIP and SAM.
CURSOR: Scalable Mixed-Order Hypergraph Matching with CUR Decomposition
Qixuan Zheng (City University of Hong Kong), Hong Yan (City University of Hong Kong)
OptimizationGraph
🎯 What it does: This paper proposes the CURSOR framework, which achieves scalable mixed-order hypergraph matching through CUR decomposition and designs an efficient matching algorithm based on PRL.
CurveCloudNet: Processing Point Clouds with 1D Structure
Colton Stearns (Stanford University), Leonidas J. Guibas (Stanford University)
SegmentationAutonomous DrivingConvolutional Neural NetworkPoint Cloud
🎯 What it does: Proposes CurveCloudNet, which segments point clouds into curve clouds based on laser scanning trajectories to construct a novel backbone network.
Customization Assistant for Text-to-Image Generation
Yufan Zhou (Adobe Research), Tong Sun (Adobe Research)
GenerationData SynthesisKnowledge DistillationTransformerLarge Language ModelDiffusion modelImageText
🎯 What it does: This paper proposes a parameter-free custom assistant (CAFE) based on large language models and diffusion models, capable of creative generation from a single user image within 2-5 seconds, and handling vague prompts in conversation while providing textual explanations.
Customize your NeRF: Adaptive Source Driven 3D Scene Editing via Local-Global Iterative Training
Runze He, Si Liu
GenerationData SynthesisDiffusion modelNeural Radiance FieldImage
🎯 What it does: The CustomNeRF framework is proposed to achieve local foreground editing of 3D scenes driven by text or reference images.
CustomListener: Text-guided Responsive Interaction for User-friendly Listening Head Generation
Xi Liu (Meituan), Pengfei Yan (Meituan)
GenerationTransformerLarge Language ModelDiffusion modelVideoTextAudio
🎯 What it does: A framework named CustomListener is proposed, which utilizes free-text priors to achieve user-oriented listening head generation. It can generate non-verbal responses with personalized attributes (identity, personality, behavior habits) based on the speaker's audio, semantics, and emotions in face-to-face conversations.
CuVLER: Enhanced Unsupervised Object Discoveries through Exhaustive Self-Supervised Transformers
Shahaf Arica (Technion Israel Institute of Technology), Shlomi Laufer (Technion Israel Institute of Technology)
Object DetectionSegmentationDomain AdaptationTransformerContrastive LearningImage
🎯 What it does: Proposes two unsupervised object detection and segmentation frameworks: VoteCut and CuVLER;
CVT-xRF: Contrastive In-Voxel Transformer for 3D Consistent Radiance Fields from Sparse Inputs
Yingji Zhong (Hong Kong University of Science and Technology), Dan Xu (Hong Kong University of Science and Technology)
TransformerNeural Radiance FieldContrastive LearningPoint Cloud
🎯 What it does: Under sparse view conditions, a voxel-based Contrastive Transformer (CVT) framework is proposed, which enhances the quality of the NeRF's sparse input light field by mutually constraining ray points and surrounding points within the voxel.
CyberDemo: Augmenting Simulated Human Demonstration for Real-World Dexterous Manipulation
Jun Wang (University of California San Diego), Xiaolong Wang (University of California San Diego)
Robotic IntelligenceTransformerSupervised Fine-TuningMultimodality
🎯 What it does: We propose CyberDemo: a complete pipeline for learning fine operations such as hand grasping and rotation in the real world from human demonstrations in simulation. This pipeline first completes human remote demonstrations using low-cost cameras in the SAPIEN simulator, then enhances the demonstrations in multiple dimensions (viewpoint, lighting, texture, object shape, object and arm posture) within the simulator. It utilizes automatic curriculum learning and action chunking Transformers to perform supervised learning on the enhanced data, and finally fine-tunes with only a small amount (about 3 minutes) of real robot demonstrations for direct deployment on real robots.
CycleINR: Cycle Implicit Neural Representation for Arbitrary-Scale Volumetric Super-Resolution of Medical Data
Wei Fang (Alibaba Group), Minfeng Xu (Alibaba Group)
SegmentationSuper ResolutionBiomedical DataComputed Tomography
🎯 What it does: This paper proposes CycleINR, which utilizes implicit neural representations to achieve arbitrary scale medical volume super-resolution, and suppresses over-smoothing and noise inconsistency in generated slices through cyclic consistency loss.
Cyclic Learning for Binaural Audio Generation and Localization
Zhaojian Li (Northwestern Polytechnical University), Yuan Yuan (Northwestern Polytechnical University)
GenerationData SynthesisConvolutional Neural NetworkDiffusion modelVideoAudio
🎯 What it does: This paper proposes a cyclic learning framework that combines visual sound source localization with monaural audio to generate binaural audio, utilizing monaural waveforms and visual information of sound source objects to achieve binaural audio upmixing.
D^4: Dataset Distillation via Disentangled Diffusion Model
Duo Su (University of Chinese Academy of Sciences), Bowen Tang (Chinese Academy of Sciences)
Data SynthesisKnowledge DistillationDiffusion modelImage
🎯 What it does: A method for architecture-independent dataset distillation using a decoupled diffusion model is proposed, which directly extracts class prototypes from the original data and synthesizes a high-quality synthetic dataset through a latent diffusion model.
D3still: Decoupled Differential Distillation for Asymmetric Image Retrieval
Yi Xie (South China University of Technology), Shengfeng He (Singapore Management University)
RetrievalKnowledge DistillationImage
🎯 What it does: A decoupled differential distillation framework (D3still) is proposed, which enhances the ranking consistency and performance of lightweight query networks in heterogeneous image retrieval by decomposing the similarity differential matrix in the atlas domain into feature knowledge, asynchronous similarity differential knowledge, and synchronous similarity differential knowledge.
D3T: Distinctive Dual-Domain Teacher Zigzagging Across RGB-Thermal Gap for Domain-Adaptive Object Detection
Dinh Phat Do (Ajou University), Wonjun Hwang (Ajou University)
Object DetectionDomain AdaptationKnowledge DistillationImageMultimodality
🎯 What it does: The D3T framework is proposed, which uses two specialized teacher models (RGB and thermal imaging) and a zigzag learning strategy to transfer the visible light domain detection model to the thermal imaging domain, achieving unsupervised domain adaptation.
DanceCamera3D: 3D Camera Movement Synthesis with Music and Dance
Zixuan Wang (Tsinghua University), Jiebo Luo (University of Rochester)
GenerationData SynthesisTransformerDiffusion modelVideoMultimodality
🎯 What it does: This paper presents the DanceCamera3D model, which can generate 3D camera movements based on music and dance actions.
Dancing with Still Images: Video Distillation via Static-Dynamic Disentanglement
Ziyu Wang (Shanghai Jiao Tong University), Yong-Lu Li (Shanghai Jiao Tong University)
Knowledge DistillationVideo
🎯 What it does: A dataset distillation method for video data is proposed, which first learns static information through image distillation and then compensates for dynamic information using a learnable dynamic memory network, achieving efficient video distillation.
DAP: A Dynamic Adversarial Patch for Evading Person Detectors
Amira Guesmi (New York University), Muhammad Shafique (New York University)
Object DetectionAdversarial AttackImageVideo
🎯 What it does: This paper proposes a Dynamic Adversarial Patch (DAP) that directly optimizes patches in pixel space, combining similarity loss and total variation loss, ensuring that the patch remains natural and realistic after being printed on a T-shirt while effectively deceiving detection systems.
DaReNeRF: Direction-aware Representation for Dynamic Scenes
Ange Lou (United Imaging Intelligence), Ziyan Wu (United Imaging Intelligence)
RestorationGenerationData SynthesisCompressionNeural Radiance FieldVideo
🎯 What it does: This paper proposes a direction-aware representation called DaReNeRF based on dual-tree complex wavelet transform for efficient modeling and re-rendering of complex dynamic scenes.
DART: Implicit Doppler Tomography for Radar Novel View Synthesis
Tianshu Huang (Carnegie Mellon University), Anthony Rowe (Bosch Research)
GenerationData SynthesisNeural Radiance FieldPoint Cloud
🎯 What it does: A DART method based on implicit radar radiation fields has been developed to synthesize new perspective radar scans from the range-Doppler images of millimeter-wave radar.
Data Poisoning based Backdoor Attacks to Contrastive Learning
Jinghuai Zhang (University of California), Neil Zhenqiang Gong (Duke University)
Representation LearningAdversarial AttackData-Centric LearningContrastive LearningImageMultimodality
🎯 What it does: This study investigates backdoor attacks based on data poisoning in contrastive learning and proposes a new attack method.
Data Valuation and Detections in Federated Learning
Wenqian Li (National University of Singapore), Yan Pang (National University of Singapore)
Anomaly DetectionFederated LearningSafty and PrivacyComputational EfficiencyImage
🎯 What it does: Proposes the FedBary method, which utilizes the Wasserstein barycenter to evaluate client contributions and detect noisy data in federated learning without sharing raw data.
Data-Efficient Multimodal Fusion on a Single GPU
Noël Vouitsis (Layer 6 AI), Maksims Volkovs (Layer 6 AI)
RetrievalData-Centric LearningTransformerContrastive LearningImageTextMultimodalityAudio
🎯 What it does: We propose FuseMix, a multimodal data augmentation scheme that performs mixing in the latent space of a pretrained unimodal encoder, achieving efficient multimodal alignment using a frozen unimodal encoder and a lightweight fusion adapter.
Data-Efficient Unsupervised Interpolation Without Any Intermediate Frame for 4D Medical Images
JungEun Kim (Korea Advanced Institute of Science and Technology), Eunho Yang (Korea Advanced Institute of Science and Technology)
RestorationData SynthesisConvolutional Neural NetworkOptical FlowImageBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: An unsupervised 4D medical image interpolation framework called UVI-Net is proposed, which can synthesize any intermediate frames along the time axis using only the starting and ending frames.
Data-Free Quantization via Pseudo-label Filtering
Chunxiao Fan (Hefei University of Technology), Meng Wang (Hefei University of Technology)
CompressionKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: A quantization method is proposed for scenarios without original training data: first, the reliability of synthetic data is assessed through self-entropy, dividing it into high and low reliability samples; high reliability samples use primary pseudo-labels, while low reliability samples use auxiliary pseudo-labels; subsequently, knowledge distillation (KL + cross-entropy + MSE) is employed to train the quantization network.
DAVE - A Detect-and-Verify Paradigm for Low-Shot Counting
Jer Pelhan (University of Ljubljana), Matej Kristan (University of Ljubljana)
Object DetectionTransformerContrastive LearningImage
🎯 What it does: A low-sample counting method called DAVE is proposed, which combines density estimation and detection approaches by first generating high-recall detection candidate boxes and then verifying to eliminate false detections, achieving both counting and localization capabilities.
Day-Night Cross-domain Vehicle Re-identification
Hongchao Li (Anhui Normal University), Yonglong Luo (Anhui Normal University)
RecognitionDomain AdaptationAutonomous DrivingConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a framework called DNDM for Day-Night Cross-Domain Vehicle ReID, addressing issues such as glare from headlights at night, low-light environments, and inter-domain differences.
De-confounded Data-free Knowledge Distillation for Handling Distribution Shifts
Yuzheng Wang (Fudan University), Lizhe Qi (Fudan University)
Domain AdaptationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: The KDCI (Knowledge Distillation Causal Intervention) framework is proposed, which eliminates distribution shift in Data-Free Knowledge Distillation (DFKD) through causal inference, enhancing the performance of the student model.
De-Diffusion Makes Text a Strong Cross-Modal Interface
Chen Wei (Google DeepMind), Jiahui Yu (Johns Hopkins University)
GenerationData SynthesisLarge Language ModelDiffusion modelAuto EncoderImageTextMultimodality
🎯 What it does: A self-encoder named De-Diffusion is proposed, which encodes input images into text ('scrambled caption') and then decodes them using a pre-trained text-to-image diffusion model to achieve high-quality image reconstruction. The generated text serves as a cross-modal interface for multimodal tasks.
DEADiff: An Efficient Stylization Diffusion Model with Disentangled Representations
Tianhao Qi, Yongdong Zhang
Image TranslationGenerationDiffusion modelImage
🎯 What it does: Developed the DEADiff model, which utilizes decoupled representation and decoupled conditioning mechanisms to achieve style transfer from reference images to generated images while maintaining text control.
Decentralized Directed Collaboration for Personalized Federated Learning
Yingqi Liu (Nanjing University of Science and Technology), Li Shen (JD Explore Academy)
Federated LearningConvolutional Neural NetworkImageStochastic Differential Equation
🎯 What it does: A decentralized personalized federated learning framework DFedPGP based on directed graphs is proposed, utilizing gradient pushing and partial model personalization to achieve efficient communication and better personalization performance.
Deciphering 'What' and 'Where' Visual Pathways from Spectral Clustering of Layer-Distributed Neural Representations
Xiao Zhang (University of Chicago), Michael Maire (University of Chicago)
SegmentationTransformerDiffusion modelImage
🎯 What it does: A global activation analysis method based on spectral clustering is proposed, which simultaneously utilizes the attention features from all layers of pre-trained visual models (such as Stable Diffusion, CLIP, DINO, etc.) to extract two information pathways: 'what' (semantics) and 'where' (spatial), thereby achieving unsupervised semantic segmentation and instance segmentation.
Decompose-and-Compose: A Compositional Approach to Mitigating Spurious Correlation
Fahimeh Hosseini Noohdani (Sharif University of Technology), Mahdieh Soleymani Baghshah (Sharif University of Technology)
Domain AdaptationExplainability and InterpretabilityConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A Decompose-and-Compose (DaC) method is proposed, which utilizes the interpretability of standard ERM models (xGradCAM) to first identify the causal and non-causal parts of images, and then combines parts from different images to create new samples, thereby eliminating the correlation between non-causal features and labels, thus enhancing the model's robustness under distribution shifts.
Decomposing Disease Descriptions for Enhanced Pathology Detection: A Multi-Aspect Vision-Language Pre-training Framework
Vu Minh Hieu Phan (Australian Institute for Machine Learning, The University of Adelaide), Johan W. Verjans (Australian Institute for Machine Learning, The University of Adelaide)
ClassificationRecognitionTransformerLarge Language ModelVision Language ModelContrastive LearningImageMultimodalityBiomedical DataComputed Tomography
🎯 What it does: A multi-faceted visual-language pre-training framework MAVL is proposed, which decomposes disease descriptions into visual features and performs multi-faceted matching.
DeconfuseTrack: Dealing with Confusion for Multi-Object Tracking
Cheng Huang (Huazhong University of Science and Technology), Yuhao Wei (Huazhong University of Science and Technology)
Object TrackingVideo
🎯 What it does: DeconfuseTrack is proposed, which combines Decomposed Data Association (DDA) and Occlusion-aware NMS (ONMS) to significantly reduce confusion and mismatches in multi-object tracking.
DeCoTR: Enhancing Depth Completion with 2D and 3D Attentions
Yunxiao Shi (Qualcomm AI Research), Fatih Porikli (Qualcomm AI Research)
RestorationDepth EstimationTransformerPoint Cloud
🎯 What it does: The DeCoTR framework is proposed, which first improves the initial depth of S2D-TR using a 2D Transformer to obtain more accurate results, then elevates 2D features to 3D point clouds and employs a 3D Transformer for cross-attention learning, supplemented by point cloud normalization and low-resolution global attention, ultimately achieving high-precision depth completion.
Decoupled Pseudo-labeling for Semi-Supervised Monocular 3D Object Detection
Jiacheng Zhang (Sun Yat-sen University), Guanbin Li (Sun Yat-sen University)
Object DetectionAutonomous DrivingKnowledge DistillationImage
🎯 What it does: This paper proposes a decoupled pseudo-label generation and gradient projection framework (DPL) for semi-supervised monocular 3D object detection, achieving pseudo-label training through a teacher-student model.
Decoupling Static and Hierarchical Motion Perception for Referring Video Segmentation
Shuting He (Nanyang Technological University), Henghui Ding (Fudan University)
Object DetectionSegmentationTransformerContrastive LearningVideoText
🎯 What it does: This paper proposes a framework that decomposes the reference video segmentation task into static perception and hierarchical motion perception, and utilizes contrastive learning to enhance the ability to distinguish similar moving objects.
Deep Equilibrium Diffusion Restoration with Parallel Sampling
Jiezhang Cao (ETH Zurich), Luc Van Gool (KU Leuven)
RestorationDiffusion modelImage
🎯 What it does: A zero-shot diffusion model image restoration method based on Deep Equilibrium (DEQ) is designed, which allows for parallel sampling and improves image quality and controls the generation direction through initialization optimization.
Deep Generative Model based Rate-Distortion for Image Downscaling Assessment
Yuanbang Liang (Cardiff University), Yipeng Qin (Cardiff University)
GenerationSuper ResolutionCompressionFlow-based ModelGenerative Adversarial NetworkImage
🎯 What it does: A new image downsampling evaluation metric IDA-RD based on rate-distortion theory is proposed, and blind random super-resolution is implemented through deep generative models to measure the extent to which downsampling algorithms retain high-resolution information.
Deep Imbalanced Regression via Hierarchical Classification Adjustment
Haipeng Xiong (National University of Singapore), Angela Yao (National University of Singapore)
Object DetectionDepth EstimationKnowledge DistillationImage
🎯 What it does: To address the issue of sample imbalance in visual regression tasks, a Hierarchical Classification Adjustment (HCA) framework is proposed and implemented, utilizing a combination of multi-level fine and coarse classifiers to reduce quantization error and improve regression accuracy.
Deep Single Image Camera Calibration by Heatmap Regression to Recover Fisheye Images Under Manhattan World Assumption
Nobuhiko Wakai (Panasonic Holdings Corporation), Takayoshi Yamashita (Chubu University)
RestorationPose EstimationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A learning-based single image camera calibration method is proposed, which uses heatmap regression to simultaneously estimate rotation (pan, tilt, roll), focal length, and distortion under the Manhattan world assumption, thereby recovering fisheye images.
Deep Video Inverse Tone Mapping Based on Temporal Clues
Yuyao Ye (Peking University), Ronggang Wang (Peking University)
RestorationRecurrent Neural NetworkTransformerOptical FlowVideo
🎯 What it does: A novel pipeline for Inverse Tone Mapping using video temporal information is proposed, which can recover HDR video from a single frame LDR video and address the issues of texture loss in overexposed areas and temporal inconsistency.
Deep-TROJ: An Inference Stage Trojan Insertion Algorithm through Efficient Weight Replacement Attack
Sabbir Ahmed (Binghamton University), Adnan Siraj Rakin (New Jersey Institute of Technology)
TransformerImage
🎯 What it does: This paper addresses deep neural network models that inject malicious backdoors through memory page table bit flips during the inference phase.
DeepCache: Accelerating Diffusion Models for Free
Xinyin Ma (National University of Singapore), Xinchao Wang (National University of Singapore)
GenerationComputational EfficiencyDiffusion modelImage
🎯 What it does: A training-free cache acceleration scheme called DeepCache is proposed, which utilizes the temporal redundancy of high-level features in continuous denoising steps of diffusion models. It caches features on the skip connections of U-Net to reduce redundant computations, thereby accelerating inference.