CVPR 2023 Papers — Page 4
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2353 papers
CHMATCH: Contrastive Hierarchical Matching and Robust Adaptive Threshold Boosted Semi-Supervised Learning
Jianlong Wu (Harbin Institute of Technology), Liqiang Nie (Shandong University)
ClassificationContrastive LearningImage
🎯 What it does: This paper proposes a semi-supervised learning framework named CHMatch, which combines instance-level prediction matching with hierarchical label-based graph-level contrastive learning, aiming to improve classification performance under few labeled samples.
CiaoSR: Continuous Implicit Attention-in-Attention Network for Arbitrary-Scale Image Super-Resolution
Jiezhang Cao (ETH Zurich), Luc Van Gool (ETH Zurich)
RestorationSuper ResolutionImage
🎯 What it does: A continuous implicit attention network called CiaoSR is proposed for arbitrary scale image super-resolution, which can be seamlessly integrated with any SR backbone network.
CiCo: Domain-Aware Sign Language Retrieval via Cross-Lingual Contrastive Learning
Yiting Cheng (Fudan University), Wenqiang Zhang (Fudan University)
RetrievalDomain AdaptationTransformerContrastive LearningVideoText
🎯 What it does: This paper proposes a sign language retrieval framework CiCo based on cross-lingual contrastive learning, which can achieve natural language query retrieval of corresponding sign language videos, as well as sign language video retrieval of corresponding text.
CIGAR: Cross-Modality Graph Reasoning for Domain Adaptive Object Detection
Yabo Liu (Harbin Institute of Technology), Yong Xu (Harbin Institute of Technology)
Object DetectionDomain AdaptationGraph Neural NetworkTransformerMultimodalityGraph
🎯 What it does: A cross-modal graph reasoning framework (CIGAR) is proposed, which integrates visual feature maps with language graphs (word embeddings) for unsupervised domain adaptation in object detection.
CIMI4D: A Large Multimodal Climbing Motion Dataset Under Human-Scene Interactions
Ming Yan (Xiamen University), Cheng Wang (Xiamen University)
GenerationPose EstimationOptimizationVideoMultimodalityPoint CloudBenchmark
🎯 What it does: This study constructed the first large-scale 3D climbing action dataset, CIMI4D, which includes 180k frames of RGB video, LiDAR point clouds, IMU measurements, and high-precision static scene scans, providing accurate human pose, global trajectory, and rock wall contact annotations. Benchmark evaluations were conducted on tasks such as 3D pose estimation, motion prediction, and generation using this dataset.
CIRCLE: Capture in Rich Contextual Environments
João Pedro Araújo (Stanford University), Karen Liu
GenerationPose EstimationTransformerVideoPoint Cloud
🎯 What it does: A system for capturing human motion in virtual reality is proposed, and based on this, the CIRCLE dataset is constructed.
CLAMP: Prompt-Based Contrastive Learning for Connecting Language and Animal Pose
Xu Zhang (University of Sydney), Dacheng Tao (University of Sydney)
Pose EstimationTransformerPrompt EngineeringVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: The CLAMP method is proposed, which combines learnable text prompts with the CLIP language model to improve animal pose estimation.
Class Adaptive Network Calibration
Bingyuan Liu (École de technologie supérieure), Ismail Ben Ayed (Universitat Pompeu Fabra)
ClassificationSegmentationOptimizationConvolutional Neural NetworkTransformerImageText
🎯 What it does: This paper proposes a Category Adaptive Label Smoothing (CALS) method based on an improved Augmented Lagrangian Multiplier (ALM) algorithm to simultaneously optimize accuracy and calibration in deep network training.
Class Attention Transfer Based Knowledge Distillation
Ziyao Guo (Xidian University), Xiaodong Lin (University of Guelph)
ClassificationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: By training the student network to align only with the class activation maps (CAM) generated by the teacher, a highly interpretable distillation method called CAT-KD is constructed.
Class Balanced Adaptive Pseudo Labeling for Federated Semi-Supervised Learning
Ming Li (East China Normal University), Yan Wang (East China Normal University)
Federated LearningImage
🎯 What it does: A framework called CBAFed is proposed for using class-balanced adaptive pseudo-labels in federated semi-supervised learning, addressing issues of non-IID, catastrophic forgetting, and class imbalance.
Class Prototypes Based Contrastive Learning for Classifying Multi-Label and Fine-Grained Educational Videos
Rohit Gupta (Center for Research in Computer Vision University of Central Florida), Mubarak Shah (Center for Research in Computer Vision University of Central Florida)
ClassificationTransformerContrastive LearningVideoTextMultimodalityAudio
🎯 What it does: This paper proposes a prototype-based contrastive learning framework for detecting and fine-grained classification of multi-label content in educational videos.
Class Relationship Embedded Learning for Source-Free Unsupervised Domain Adaptation
Yixin Zhang (University of Science and Technology of China), Weinan He (University of Science and Technology of China)
Domain AdaptationContrastive LearningImage
🎯 What it does: In the unsupervised domain adaptation scenario with no data in the source domain, knowledge transfer is achieved by utilizing the class relationships of the pre-trained source model and implementing contrastive learning through class relationship embedding;
Class-Balancing Diffusion Models
Yiming Qin (Shanghai Jiao-Tong University), Ya Zhang (Shanghai Jiao-Tong University)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: The study addresses the issue of declining generative quality when training diffusion models on class-imbalanced (long-tail) data, and proposes a Class-Balancing Diffusion Model (CBDM) based on regularization to enhance the diversity and authenticity of tail class images.
Class-Conditional Sharpness-Aware Minimization for Deep Long-Tailed Recognition
Zhipeng Zhou (University of Science and Technology of China), Wei Gong (University of Science and Technology of China)
ClassificationRecognitionConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A class-conditioned Sharpness-Aware Minimization (CC-SAM) two-stage training framework is proposed to enhance the generalization performance of deep long-tail recognition models.
Class-Incremental Exemplar Compression for Class-Incremental Learning
Zilin Luo (Singapore Management University), Qianru Sun (Singapore Management University)
CompressionImage
🎯 What it does: An adaptive mask compression method CIM is proposed, which utilizes CAM and learnable activation functions to compress samples of old classes in class-incremental learning;
CLIP for All Things Zero-Shot Sketch-Based Image Retrieval, Fine-Grained or Not
Aneeshan Sain (University of Surrey), Yi-Zhe Song (University of Surrey)
RetrievalTransformerVision Language ModelContrastive LearningImage
🎯 What it does: Utilizing the open-source pre-trained model of CLIP, zero-shot sketch retrieval (ZS-SBIR) and fine-grained zero-shot sketch retrieval (FG-ZS-SBIR) are achieved through visual prompt learning.
CLIP Is Also an Efficient Segmenter: A Text-Driven Approach for Weakly Supervised Semantic Segmentation
Yuqi Lin (Zhejiang University), Xiaofei He (Zhejiang University)
SegmentationComputational EfficiencyTransformerPrompt EngineeringImage
🎯 What it does: A framework called CLIP-ES is proposed, which utilizes a frozen CLIP model and a text-driven strategy to generate pseudo-masks and train segmentation models for weakly supervised semantic segmentation without training additional networks.
CLIP the Gap: A Single Domain Generalization Approach for Object Detection
Vidit Vidit (EPFL), Mathieu Salzmann (EPFL)
Object DetectionDomain AdaptationAutonomous DrivingPrompt EngineeringVision Language ModelImage
🎯 What it does: Utilize the pre-trained visual language model CLIP for semantic enhancement to train a single-domain generalization object detector.
CLIP-S4: Language-Guided Self-Supervised Semantic Segmentation
Wenbin He (Bosch Research North America), Liu Ren (Bosch Research North America)
Object TrackingSegmentationVision Language ModelContrastive LearningImageVideo
🎯 What it does: By combining self-supervised pixel contrastive learning with the CLIP vision-language model, a unified pixel representation is learned, achieving language-driven semantic segmentation without the need for pixel-level annotations.
CLIP-Sculptor: Zero-Shot Generation of High-Fidelity and Diverse Shapes From Natural Language
Aditya Sanghi, Daniel Ritchie
GenerationData SynthesisTransformerContrastive LearningPoint CloudMesh
🎯 What it does: We propose CLIP-Sculptor, a zero-shot text-to-3D shape generation model that can generate high-fidelity and diverse 3D voxel shapes without using (text, shape) pairs.
CLIP2: Contrastive Language-Image-Point Pretraining From Real-World Point Cloud Data
Yihan Zeng (Huawei Noah's Ark Lab), Hang Xu (Huawei Noah's Ark Lab)
ClassificationRecognitionObject DetectionDomain AdaptationRepresentation LearningContrastive LearningImageTextMultimodalityPoint Cloud
🎯 What it does: This paper proposes a new CLIP 2 framework that utilizes a tri-modal (text, image, point cloud) triplet proxy for contrastive learning, achieving open-world zero-shot transfer of point cloud representations.
CLIP2Protect: Protecting Facial Privacy Using Text-Guided Makeup via Adversarial Latent Search
Fahad Shamshad (Mohamed Bin Zayed University of AI), Karthik Nandakumar (Mohamed Bin Zayed University of AI)
GenerationSafty and PrivacyAdversarial AttackGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a text-guided concealment method based on a pre-trained generative model (StyleGAN) to generate 'naturalized' facial images that can mislead black-box face recognition systems without affecting user experience.
CLIP2Scene: Towards Label-Efficient 3D Scene Understanding by CLIP
Runnan Chen (University of Hong Kong), Wenping Wang (Texas A&M University)
SegmentationAutonomous DrivingTransformerContrastive LearningPoint Cloud
🎯 What it does: This paper proposes the CLIP2Scene framework, which transfers the image-text pre-training knowledge of CLIP to 3D point cloud networks, achieving unsupervised 3D semantic segmentation and label-efficient fine-tuning.
CLIPPING: Distilling CLIP-Based Models With a Student Base for Video-Language Retrieval
Renjing Pei (Huawei Noah's Ark Lab), Youliang Yan (Huawei Noah's Ark Lab)
RetrievalKnowledge DistillationConvolutional Neural NetworkRecurrent Neural NetworkContrastive LearningVideoTextMultimodality
🎯 What it does: The CLIPPING method is used to distill knowledge from a large-scale CLIP-base teacher model (Clip4clip) to a small student model during the fine-tuning phase, achieving efficient video-language retrieval.
CLIPPO: Image-and-Language Understanding From Pixels Only
Michael Tschannen (Google Research), Neil Houlsby (Google Research)
RetrievalTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: A single-tower Vision Transformer (CLIPPO) is proposed and trained, which can handle image, text, and text-image mixed tasks simultaneously using only pixel input.
CloSET: Modeling Clothed Humans on Continuous Surface With Explicit Template Decomposition
Hongwen Zhang (Tsinghua University), Yebin Liu (Tsinghua University)
GenerationPose EstimationPoint Cloud
🎯 What it does: Proposes the CloSET method, which represents clothing using point clouds, decomposing garment deformation into explicit templates and pose-related wrinkles, and learns features on continuous surfaces for end-to-end training.
CLOTH4D: A Dataset for Clothed Human Reconstruction
Xingxing Zou (Hong Kong Polytechnic University), Waikeung Wong (Hong Kong Polytechnic University)
MeshBenchmark
🎯 What it does: A large-scale high-quality 4D clothing dataset CLOTH4D is proposed, and this dataset is used to evaluate and improve existing garment-human reconstruction methods.
Clothed Human Performance Capture With a Double-Layer Neural Radiance Fields
Kangkan Wang (Nanjing University of Science and Technology), Jian Yang (Nanjing University of Science and Technology)
GenerationPose EstimationNeural Radiance FieldVideo
🎯 What it does: This paper proposes a dual-layer neural radiance field (NeRF) framework for simultaneously capturing the body motion of clothed humans and the motion of clothing, enabling novel view synthesis.
Cloud-Device Collaborative Adaptation to Continual Changing Environments in the Real-World
Yulu Gan (Peking University), Shanghang Zhang (Peking University)
Object DetectionDomain AdaptationKnowledge DistillationImage
🎯 What it does: A cloud-device collaborative continuous adaptation framework has been designed and implemented: by using uncertainty-guided sampling (UGS) to select the most environmentally diverse samples for upload to the cloud, the cloud utilizes a large model to train visual prompts and transfers generalization capabilities to the lightweight model on the device through teacher-student joint optimization (U-VPA), achieving continuous domain adaptation.
Clover: Towards a Unified Video-Language Alignment and Fusion Model
Jingjia Huang (ByteDance Inc), Rongrong Ji (Xiamen University)
RetrievalTransformerVision Language ModelContrastive LearningVideoTextMultimodality
🎯 What it does: A unified video-language pre-training model called Clover has been designed and trained, achieving efficient end-to-end inference for video retrieval and video question answering tasks.
CNVid-3.5M: Build, Filter, and Pre-Train the Large-Scale Public Chinese Video-Text Dataset
Tian Gan (Shandong University), Qingpei Guo (Ant Group)
RetrievalTransformerVision Language ModelVideoTextMultimodality
🎯 What it does: This paper constructs and publicly releases CNVid-3.5M, which contains 3.5 million Chinese video-text pairs, and conducts pixel-level pre-training and evaluation on this dataset.
Co-Salient Object Detection With Uncertainty-Aware Group Exchange-Masking
Yang Wu (Nanjing University of Information Science and Technology), Dong Liu
Object DetectionSegmentationTransformerAuto EncoderImage
🎯 What it does: This paper proposes a Co-Salient Object Detection (CoSOD) learning framework based on Group Exchange-Masking (GEM) and designs a dual-branch feature extraction module (Latent Variable Generator Branch and CoSOD Transformer Branch). By exchanging images of different categories and masking labels, the model's robustness to irrelevant images is enhanced. Meanwhile, a Conditional Variational Autoencoder (CVAE) is utilized to generate global uncertainty features, and a Transformer is employed to capture intra-group consistency information, ultimately outputting high-quality co-salient object segmentation results.
Co-SLAM: Joint Coordinate and Sparse Parametric Encodings for Neural Real-Time SLAM
Hengyi Wang (University College London), Lourdes Agapito (University College London)
OptimizationRobotic IntelligenceSimultaneous Localization and MappingPoint Cloud
🎯 What it does: Developed a real-time neural RGB-D SLAM system called Co-SLAM, which combines coordinates and sparse parameter encoding to achieve efficient tracking and mapping;
Co-Speech Gesture Synthesis by Reinforcement Learning With Contrastive Pre-Trained Rewards
Mingyang Sun (Dalian University of Technology), Jianye Hao (Huawei Cloud Computing Technologies Co., Ltd)
GenerationData SynthesisTransformerReinforcement LearningContrastive LearningMultimodalityAudio
🎯 What it does: A reinforcement learning-based co-speech gesture synthesis framework called RACER is proposed, which generates gesture sequences that are synchronized and coherent with speech.
Co-Training 2L Submodels for Visual Recognition
Hugo Touvron (Meta AI), Hervé Jégou (Sorbonne University)
ClassificationRecognitionSegmentationKnowledge DistillationConvolutional Neural NetworkTransformerSupervised Fine-TuningImage
🎯 What it does: A co-training method for sub-models (Cosub) is proposed, which does not require an external teacher and uses a single parameter set, by randomly sampling sub-networks from the same network during training and allowing them to provide soft labels to each other as a form of regularization.
Coaching a Teachable Student
Jimuyang Zhang (Boston University), Eshed Ohn-Bar (Boston University)
Autonomous DrivingKnowledge DistillationTransformerImage
🎯 What it does: A knowledge distillation-based framework (CaT) is proposed, which guides sensor-motion (image-based) students in learning autonomous driving strategies through a privileged teacher (BEV perspective).
CODA-Prompt: COntinual Decomposed Attention-Based Prompting for Rehearsal-Free Continual Learning
James Seale Smith (Georgia Institute of Technology), Zsolt Kira (Georgia Institute of Technology)
TransformerPrompt EngineeringImage
🎯 What it does: This paper proposes CODA-Prompt, a prompt mechanism based on decomposable attention for example-free continual learning.
CodeTalker: Speech-Driven 3D Facial Animation With Discrete Motion Prior
Jinbo Xing (Chinese University of Hong Kong), Tien-Tsin Wong (Chinese University of Hong Kong)
GenerationTransformerAuto EncoderMeshAudio
🎯 What it does: This paper presents CodeTalker, which utilizes discrete motion priors for speech-driven 3D facial animation generation.
Collaboration Helps Camera Overtake LiDAR in 3D Detection
Yue Hu (Shanghai Jiao Tong University), Yanfeng Wang (Shanghai AI Laboratory)
Object DetectionAutonomous DrivingPoint Cloud
🎯 What it does: The CoCa3D framework is proposed, achieving more accurate 3D detection using multi-agent collaboration with only camera input.
Collaborative Diffusion for Multi-Modal Face Generation and Editing
Ziqi Huang (Nanyang Technological University), Ziwei Liu (Nanyang Technological University)
GenerationData SynthesisDiffusion modelImageTextMultimodality
🎯 What it does: A Collaborative Diffusion framework is proposed, which achieves multi-modal (text, mask, etc.) face generation and editing through a pre-trained unimodal diffusion model without the need for retraining.
Collaborative Noisy Label Cleaner: Learning Scene-Aware Trailers for Multi-Modal Highlight Detection in Movies
Bei Gan (Tencent YouTu Lab), Bo Ren (Tencent YouTu Lab)
Object DetectionData-Centric LearningTransformerContrastive LearningVideoMultimodality
🎯 What it does: This paper proposes a Collaborative Noisy Label Cleaner (CLC) framework that utilizes the noisy labels present in movie trailers to learn movie highlight detection.
Collaborative Static and Dynamic Vision-Language Streams for Spatio-Temporal Video Grounding
Zihang Lin (Sun Yat-sen University), Wei-Shi Zheng (Tencent)
RecognitionObject DetectionTransformerVision Language ModelVideoText
🎯 What it does: This paper proposes a two-stream visual-language framework for the task of Spatio-Temporal Video Grounding, where the static stream focuses on the appearance of objects within frames, and the dynamic stream focuses on motion information across frames. A cross-stream collaboration module is implemented to achieve complementary information transfer between the two streams, thereby locating the spatial bounding box and temporal segment of the target object.
Collecting Cross-Modal Presence-Absence Evidence for Weakly-Supervised Audio-Visual Event Perception
Junyu Gao (Chinese Academy of Sciences), Changsheng Xu (Chinese Academy of Sciences)
RecognitionAnomaly DetectionVideoMultimodalityAudio
🎯 What it does: A weakly supervised cross-modal presence-absence evidence learning framework CMPAE is proposed for audio-video event perception, utilizing the Presence-Absence Evidence Collector and Joint-modal Mutual Learning to achieve joint modeling of presence evidence from a single modality and absence evidence from a complementary modality.
Color Backdoor: A Robust Poisoning Attack in Color Space
Wenbo Jiang (University of Electronic Science and Technology of China), Tianwei Zhang (Nanyang Technological University)
OptimizationAdversarial AttackImage
🎯 What it does: A backdoor attack based on unified color space displacement is proposed, and particle swarm optimization (PSO) is used to find the optimal trigger in a black-box scenario;
Combining Implicit-Explicit View Correlation for Light Field Semantic Segmentation
Ruixuan Cong (Beihang University), Hao Sheng (Beihang University)
SegmentationConvolutional Neural NetworkTransformerImage
🎯 What it does: A LF-IENet network is proposed, utilizing the implicit and explicit perspective correlation of light fields for semantic segmentation.
CoMFormer: Continual Learning in Semantic and Panoptic Segmentation
Fabio Cermelli (Politecnico di Torino), Arthur Douillard (Sorbonne Universite)
SegmentationKnowledge DistillationTransformerImageBenchmark
🎯 What it does: The CoMFormer model is proposed, which can continuously learn and simultaneously handle semantic segmentation and panoptic segmentation tasks without retraining.
Command-Driven Articulated Object Understanding and Manipulation
Ruihang Chu (Chinese University of Hong Kong), Jiaya Jia (Chinese University of Hong Kong)
Robotic IntelligencePoint Cloud
🎯 What it does: This paper proposes a user-instruction-based joint manipulation framework called Cart, which can identify and manipulate movable components with only a single point cloud, achieving predictions of joint types, axes, displacements/angles, and actual motion control.
Common Pets in 3D: Dynamic New-View Synthesis of Real-Life Deformable Categories
Samarth Sinha, David Novotny
GenerationData SynthesisSuper ResolutionTransformerNeural Radiance FieldOptical FlowVideoBenchmark
🎯 What it does: A new perspective synthesis method called Tracker-NeRF is proposed for reconstructing non-rigid pets (cats and dogs) 3D/4D models from very few viewpoint videos, and a large dataset named 'Common Pets in 3D (CoP3D)' is created.
Compacting Binary Neural Networks by Sparse Kernel Selection
Yikai Wang (Tsinghua University), Anbang Yao (Intel Labs China)
CompressionComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: Proposes Sparse Kernel Selection (Sparks), which further compresses the model and accelerates inference by selecting sub-codebooks in binary networks.
Complementary Intrinsics From Neural Radiance Fields and CNNs for Outdoor Scene Relighting
Siqi Yang (Peking University), Boxin Shi (Peking University)
RestorationGenerationConvolutional Neural NetworkNeural Radiance FieldImage
🎯 What it does: The paper proposes an outdoor scene relighting method that combines NeRF and CNN, using NeRF to generate pseudo-labels for training CNN, achieving physically interpretable intrinsic decomposition and relighting from a single image.
Complete 3D Human Reconstruction From a Single Incomplete Image
Junying Wang (University of Southern California), Ulrich Neumann (University of Southern California)
GenerationPose EstimationConvolutional Neural NetworkGenerative Adversarial NetworkImageMesh
🎯 What it does: Reconstruct complete 3D human geometry and texture from a single occluded human image.
Complete-to-Partial 4D Distillation for Self-Supervised Point Cloud Sequence Representation Learning
Zhuoyang Zhang (Tsinghua University), Li Yi (Tsinghua University)
RecognitionSegmentationKnowledge DistillationRepresentation LearningTransformerContrastive LearningVideoPoint Cloud
🎯 What it does: A complete-part 4D knowledge distillation method is proposed, using a teacher-student framework to learn point cloud sequence representations.
CompletionFormer: Depth Completion With Convolutions and Vision Transformers
Youmin Zhang (University of Bologna), Stefano Mattoccia (University of Bologna)
RestorationDepth EstimationConvolutional Neural NetworkTransformerImage
🎯 What it does: This study focuses on the depth completion task for sparse depth and RGB images, proposing a single-branch network that integrates convolutional attention with visual Transformers to achieve unified modeling of local and global features.
Complexity-Guided Slimmable Decoder for Efficient Deep Video Compression
Zhihao Hu (Beihang University), Dong Xu (University of Hong Kong)
CompressionConvolutional Neural NetworkVideo
🎯 What it does: This paper proposes a Complexity-Guided Scalable Decoder (cgSlimDecoder) and Skip Adaptive Entropy Coding (SaEC) to achieve efficient video decoding with a single decoder under various complexity constraints.
Compositor: Bottom-Up Clustering and Compositing for Robust Part and Object Segmentation
Ju He (Johns Hopkins University), Alan L. Yuille
Object DetectionSegmentationConvolutional Neural NetworkTransformerImage
🎯 What it does: A low-level clustering and synthesis framework called Compositor is proposed, which can simultaneously achieve semantic segmentation of objects and their parts.
Comprehensive and Delicate: An Efficient Transformer for Image Restoration
Haiyu Zhao (Sichuan University), Xi Peng (Sichuan University)
RestorationTransformerImage
🎯 What it does: An efficient image restoration Transformer (CODE) is proposed, which first aggregates features through superpixel-level global attention and then returns global information to pixel-level through a dual adaptive module to complete the image restoration task.
Compressing Volumetric Radiance Fields to 1 MB
Lingzhi Li (Alibaba Group), Liefeng Bo (Alibaba Group)
CompressionNeural Radiance FieldPoint Cloud
🎯 What it does: A compression framework named VQRF is proposed, which compresses voxel grid-based volumetric radiance field models to about 1 MB, maintaining almost no visual quality loss.
Compression-Aware Video Super-Resolution
Yingwei Wang (Dalian University of Technology), Yu-Wing Tai (The Hong Kong University of Science and Technology)
RestorationSuper ResolutionCompressionRecurrent Neural NetworkOptical FlowVideo
🎯 What it does: A compressed sensing video super-resolution model CAVSR is proposed, which can adapt to input videos at different compression levels, achieving high-quality super-resolution of compressed videos.
Computational Flash Photography Through Intrinsics
Sepideh Sarajian Maralan (Simon Fraser University), Yagiz Aksoy
Image TranslationGenerationData SynthesisConvolutional Neural NetworkImage
🎯 What it does: Proposes a flash illumination decomposition and generation method based on intrinsic image decomposition.
Computationally Budgeted Continual Learning: What Does Matter?
Ameya Prabhu (University of Oxford), Adel Bibi
ClassificationComputational EfficiencyKnowledge DistillationConvolutional Neural NetworkImageBenchmark
🎯 What it does: This paper constructs a large-scale dataset benchmark under the premise of allowing a limited computational budget at each time step, systematically experiments with various common continual learning methods (sampling strategies, distillation loss, FC layer calibration, etc.), and demonstrates that the simplest experience replay + uniform/class-balanced sampling baseline method can outperform most advanced methods under computational constraints.
Conditional Generation of Audio From Video via Foley Analogies
Yuexi Du (University of Michigan), Andrew Owens (University of Michigan)
GenerationData SynthesisTransformerGenerative Adversarial NetworkVideoMultimodalityAudio
🎯 What it does: Generate audio for silent videos and specify the 'sound style' of the audio through user-provided audio and video examples, achieving conditional Foley generation.
Conditional Image-to-Video Generation With Latent Flow Diffusion Models
Haomiao Ni (Pennsylvania State University), Martin Renqiang Min (NEC Laboratories America)
GenerationData SynthesisDiffusion modelAuto EncoderOptical FlowVideo
🎯 What it does: Proposes a conditional image-to-video generation framework based on a latent optical flow diffusion model.
Conditional Text Image Generation With Diffusion Models
Yuanzhi Zhu (Alibaba DAMO Academy), Cong Yao (Alibaba DAMO Academy)
GenerationDiffusion modelImageText
🎯 What it does: A conditional text-image generation framework based on diffusion models, CTIG-DM, is proposed, which can control the generated text images through three conditions: image, text, and style.
Confidence-Aware Personalized Federated Learning via Variational Expectation Maximization
Junyi Zhu (KU Leuven), Matthew B. Blaschko (KU Leuven)
Federated LearningImage
🎯 What it does: A personalized federated learning framework based on hierarchical Bayesian models and variational expectation maximization (pFedVEM) is proposed, which adaptively computes confidence for aggregation and regularization.
Conflict-Based Cross-View Consistency for Semi-Supervised Semantic Segmentation
Zicheng Wang (University of Sydney), Luping Zhou (University of Sydney)
SegmentationAutonomous DrivingConvolutional Neural NetworkImage
🎯 What it does: A conflict-based cross-view consistency (CCVC) semi-supervised semantic segmentation method based on co-training is proposed;
Conjugate Product Graphs for Globally Optimal 2D-3D Shape Matching
Paul Roetzer (University of Bonn), Florian Bernard (University of Bonn)
OptimizationGraph Neural NetworkMesh
🎯 What it does: This paper proposes a globally optimal 2D-3D shape matching method based on conjugate product graphs, achieving continuous and unbiased matching from 2D contours to 3D meshes by introducing higher-order costs in the graph to implement local rigidity constraints.
Connecting the Dots: Floorplan Reconstruction Using Two-Level Queries
Yuanwen Yue (ETH Zurich), Francis Engelmann (ETH Zurich)
Object DetectionSegmentationTransformerPoint Cloud
🎯 What it does: This paper proposes a single-stage Transformer model called RoomFormer, which can directly predict the polygons of all rooms, room types, and structural elements such as doors and windows from the top-down density maps obtained from 3D point cloud projections.
Connecting Vision and Language With Video Localized Narratives
Paul Voigtlaender (Google Research), Vittorio Ferrari (Google Research)
RecognitionObject DetectionRetrievalTransformerVision Language ModelVideoMultimodalityBenchmark
🎯 What it does: Through the 'Video Localized Narratives' protocol, multimodal annotations were performed on 20k general domain videos, generating dense annotations corresponding to each word and spatial-temporal trajectory, and based on this, VNG and VideoQA benchmarks were created.
ConQueR: Query Contrast Voxel-DETR for 3D Object Detection
Benjin Zhu (Chinese University of Hong Kong), Hongsheng Li (Chinese University of Hong Kong)
Object DetectionAutonomous DrivingTransformerContrastive LearningPoint Cloud
🎯 What it does: A sparse 3D object detection framework called ConQueR based on DETR is proposed, achieving high-quality sparse predictions by introducing a Query Contrast mechanism.
Consistent Direct Time-of-Flight Video Depth Super-Resolution
Zhanghao Sun (Stanford University), Rakesh Ranjan (Meta Platforms)
Depth EstimationSuper ResolutionConvolutional Neural NetworkOptical FlowVideo
🎯 What it does: This paper proposes a multi-frame direct time-of-flight (dToF) video depth super-resolution method (DVSR and HVSR), which enhances the depth quality of low-resolution dToF videos by utilizing multi-frame correlation and dToF histogram information.
Consistent View Synthesis With Pose-Guided Diffusion Models
Hung-Yu Tseng (Meta), Johannes Kopf (Meta)
GenerationData SynthesisDiffusion modelOptical FlowImageVideo
🎯 What it does: A pose-guided diffusion model is proposed to generate long-range consistent perspective videos from a single image.
Consistent-Teacher: Towards Reducing Inconsistent Pseudo-Targets in Semi-Supervised Object Detection
Xinjiang Wang (SenseTime Research), Wayne Zhang (SenseTime Research)
Object DetectionConvolutional Neural NetworkGaussian SplattingImage
🎯 What it does: A semi-supervised object detection framework named Consistent-Teacher is proposed to address the overfitting problem caused by inconsistent pseudo-labels.
Constrained Evolutionary Diffusion Filter for Monocular Endoscope Tracking
Xiongbiao Luo (Xiamen University)
Object TrackingPose EstimationDepth EstimationSimultaneous Localization and MappingVideoBiomedical DataComputed Tomography
🎯 What it does: This paper proposes a 'Constrained Evolutionary Diffusion Filter (CEDF)' for monocular endoscopic 3D tracking, combining spatial state constraints with evolutionary random diffusion to address particle degradation and impoverishment issues.
ConStruct-VL: Data-Free Continual Structured VL Concepts Learning
James Seale Smith (MIT-IBM Watson AI Lab), Leonid Karlinsky (MIT-IBM Watson AI Lab)
Representation LearningAdversarial AttackData-Centric LearningTransformerVision Language ModelGenerative Adversarial NetworkMultimodalityBenchmark
🎯 What it does: Proposes the ConStruct-VL benchmark and a data-free continual learning method to enhance the learning and memory capabilities of multimodal VL models in structural concepts (attributes, relationships, states).
Constructing Deep Spiking Neural Networks From Artificial Neural Networks With Knowledge Distillation
Qi Xu (Dalin University of Technology), Gang Pan (Zhejiang University)
ClassificationKnowledge DistillationSpiking Neural NetworkImage
🎯 What it does: Using knowledge distillation, a pre-trained ANN is employed as a teacher model to train a deep SNN capable of achieving high-accuracy classification in just 4 time steps.
Content-Aware Token Sharing for Efficient Semantic Segmentation With Vision Transformers
Chenyang Lu (Eindhoven University of Technology), Gijs Dubbelman (Eindhoven University of Technology)
SegmentationComputational EfficiencyTransformerImage
🎯 What it does: This paper proposes a Content-Aware Token Sharing (CTS) method, which allows adjacent image patches of the same semantic class to share a token, thereby reducing the computational load of Vision Transformer (ViT) in semantic segmentation tasks.
Context De-Confounded Emotion Recognition
Dingkang Yang (Fudan University), Lihua Zhang (Fudan University)
RecognitionConvolutional Neural NetworkImage
🎯 What it does: A context debiasing module CCIM based on causal intervention is proposed to eliminate contextual bias in emotion recognition, which can be seamlessly integrated into existing models to enhance performance.
Context-Aware Alignment and Mutual Masking for 3D-Language Pre-Training
Zhao Jin (Sichuan University), Yinjie Lei (Sun Yat-sen University)
RecognitionObject DetectionSegmentationRecurrent Neural NetworkTransformerContrastive LearningTextMultimodalityPoint Cloud
🎯 What it does: This paper proposes a unified pre-training framework for 3D point clouds and text, achieving fine-grained interaction between point cloud and language features through context-aware spatial semantic alignment and mutual masking modeling, and transferring the pre-trained model to multi-tasks such as 3D visual localization, dense description, and question answering.
Context-Aware Pretraining for Efficient Blind Image Decomposition
Chao Wang (Baidu Inc.), Yi Yang (National University of Singapore)
RestorationComputational EfficiencyTransformerSupervised Fine-TuningImage
🎯 What it does: A context-aware pre-training framework (CP) is proposed to achieve blind image decomposition tasks that remove various mixed noises in a one-time manner.
Context-Aware Relative Object Queries To Unify Video Instance and Panoptic Segmentation
Anwesa Choudhuri (University of Illinois), Alexander G. Schwing (University of Illinois)
Object DetectionObject TrackingSegmentationTransformerVideo
🎯 What it does: A context-aware relative object query method is proposed, achieving an online, post-processing-free unified framework in video instance segmentation, video panoptic segmentation, and multi-object tracking segmentation.
Context-Based Trit-Plane Coding for Progressive Image Compression
Seungmin Jeon (Korea University), Chang-Su Kim (Korea University)
CompressionConvolutional Neural NetworkAuto EncoderImageVideo
🎯 What it does: A progressive image compression algorithm CTC based on Trit-Plane coding is proposed, utilizing context models to achieve more efficient bitrate control and image reconstruction.
Continual Detection Transformer for Incremental Object Detection
Yaoyao Liu (Max Planck Institute for Informatics), Christian Rupprecht (University of Oxford)
Object DetectionKnowledge DistillationTransformerImage
🎯 What it does: An incremental learning framework CL-DETR is designed for the Transformer-based object detector, addressing the poor performance of directly applying knowledge distillation and sample replay on Deformable DETR/UP-DETR.
Continual Semantic Segmentation With Automatic Memory Sample Selection
Lanyun Zhu (Singapore University of Technology and Design), Jun Liu (Singapore University of Technology and Design)
SegmentationConvolutional Neural NetworkReinforcement LearningImage
🎯 What it does: A reinforcement learning-based automated memory sample selection mechanism is designed to improve the performance of continuous semantic segmentation.
Continuous Intermediate Token Learning With Implicit Motion Manifold for Keyframe Based Motion Interpolation
Clinton A. Mo (University of Sydney), Zhiyong Wang (Meta Reality Labs)
GenerationPose EstimationTransformerVideo
🎯 What it does: This paper proposes a three-stage Transformer framework for continuous motion interpolation based on sparse keyframes.
Continuous Landmark Detection With 3D Queries
Prashanth Chandran (Disney Research), Derek Bradley (Disney Research)
RecognitionObject DetectionPose EstimationConvolutional Neural NetworkTransformerImage
🎯 What it does: A network has been designed and implemented that can predict a continuous and infinite number of facial landmark points during inference through 3D query points, supporting arbitrary landmark layouts, dense landmarks, and off-surface landmarks.
Continuous Pseudo-Label Rectified Domain Adaptive Semantic Segmentation With Implicit Neural Representations
Rui Gong (ETH Zurich), Luc Van Gool (ETH Zurich)
SegmentationDomain AdaptationImage
🎯 What it does: A pseudo-label correction framework based on continuous implicit neural representation (Continuous RMM) is proposed for unsupervised domain adaptation semantic segmentation, which can enhance segmentation performance through self-training in the absence of labels in the target domain.
Continuous Sign Language Recognition With Correlation Network
Lianyu Hu (Tianjin University), Wei Feng (Tianjin University)
RecognitionConvolutional Neural NetworkRecurrent Neural NetworkVideo
🎯 What it does: This paper proposes the CorrNet network, which explicitly captures body trajectories in continuous sign language videos through correlation and recognition modules, thereby enhancing continuous sign language recognition performance.
ContraNeRF: Generalizable Neural Radiance Fields for Synthetic-to-Real Novel View Synthesis via Contrastive Learning
Hao Yang (Peking University), Liwei Wang (Peking University)
Data SynthesisDomain AdaptationNeural Radiance FieldContrastive LearningPoint Cloud
🎯 What it does: A generalizable NeRF model called ContraNeRF is proposed, which can generalize from synthetic data to real-world scenarios by utilizing geometric alignment contrastive learning and cross-view attention to achieve multi-view feature consistency.
Contrastive Grouping With Transformer for Referring Image Segmentation
Jiajin Tang (ShanghaiTech University), Sibei Yang (ShanghaiTech University)
Object DetectionSegmentationTransformerContrastive LearningImageMultimodality
🎯 What it does: An end-to-end Mask classification framework called CGFormer is proposed, which uses learnable query tokens to semantically align target expressions and output segmentation masks.
Contrastive Mean Teacher for Domain Adaptive Object Detectors
Shengcao Cao (University of Illinois at Urbana-Champaign), Yu-Xiong Wang (University of Illinois at Urbana-Champaign)
Object DetectionDomain AdaptationContrastive LearningImage
🎯 What it does: A framework that integrates contrastive learning with mean teacher self-supervised learning is proposed for unsupervised domain adaptation in object detection.
Contrastive Semi-Supervised Learning for Underwater Image Restoration via Reliable Bank
Shirui Huang (Xidian University), Yunsong Li (Xidian University)
RestorationContrastive LearningImage
🎯 What it does: Proposes the Semi-UIR semi-supervised underwater image restoration framework, utilizing unlabeled data to enhance model generalization capabilities.
Controllable Light Diffusion for Portraits
David Futschik (Google Research), Rohit Pandey (Google Research)
RestorationData SynthesisConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: Diffuse the light of a single portrait photo, soften shadows and highlights, and improve lighting quality.
Controllable Mesh Generation Through Sparse Latent Point Diffusion Models
Zhaoyang Lyu (Shanghai AI Laboratory), Bo Dai (Shanghai AI Laboratory)
GenerationData SynthesisDiffusion modelAuto EncoderPoint CloudMesh
🎯 What it does: A method for mesh generation is proposed through a Sparse Latent Point Diffusion Model (SLIDE), which first converts the mesh into a point cloud and then generates and decodes latent points to obtain high-quality meshes.
ConvNeXt V2: Co-Designing and Scaling ConvNets With Masked Autoencoders
Sanghyun Woo (KAIST), Saining Xie (Meta AI)
Object DetectionSegmentationConvolutional Neural NetworkAuto EncoderImage
🎯 What it does: This paper proposes the ConvNeXt V2 network architecture and designs a fully convolutional sparse Mask-AutoEncoder (FCMAE) for self-supervised pre-training, addressing the mismatch between traditional ConvNets and MAE; it achieves performance improvements on multiple tasks such as ImageNet, COCO, and ADE20K.
ConZIC: Controllable Zero-Shot Image Captioning by Sampling-Based Polishing
Zequn Zeng (Xidian University), Zhengjue Wang (Xidian University)
GenerationTransformerVision Language ModelContrastive LearningImageText
🎯 What it does: A sampling-based non-autoregressive model, Gibbs-BERT, combined with CLIP, proposes a zero-shot image description framework called ConZIC, which can generate diverse and controllable image descriptions under unsupervised training conditions.
Cooperation or Competition: Avoiding Player Domination for Multi-Target Robustness via Adaptive Budgets
Yimu Wang (University of Waterloo), Hongyang Zhang (University of Waterloo)
OptimizationAdversarial AttackImage
🎯 What it does: This study investigates the multi-objective robustness problem, revealing the player dominance phenomenon that occurs in existing MAX/MSD adversarial training, and proposes the AdaptiveBudget framework for dynamically adjusting the attack budget to enhance the overall robustness of multi-attack models.
CORA: Adapting CLIP for Open-Vocabulary Detection With Region Prompting and Anchor Pre-Matching
Xiaoshi Wu (Chinese University of Hong Kong), Hongsheng Li (Chinese University of Hong Kong)
Object DetectionTransformerPrompt EngineeringVision Language ModelContrastive LearningImage
🎯 What it does: A CLIP-based open vocabulary word detection framework called CORA is proposed, which achieves open vocabulary detection without additional image-text data through region prompts and anchor pre-matching.
CoralStyleCLIP: Co-Optimized Region and Layer Selection for Image Editing
Ambareesh Revanur (Adobe), Deepak Pai (Adobe)
Image TranslationGenerationConvolutional Neural NetworkGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: A method called Co-optimized Region And Layer Selection (CORAL) is proposed, which utilizes CLIP-driven multi-layer attention fusion and region selection to achieve high-fidelity text-driven image editing.
Coreset Sampling From Open-Set for Fine-Grained Self-Supervised Learning
Sungnyun Kim (KAIST), Se-Young Yun (KAIST)
ClassificationObject DetectionSegmentationRepresentation LearningContrastive LearningImage
🎯 What it does: This paper proposes the OpenSSL task, which utilizes unlabeled open sets for self-supervised pre-training on a fine-grained target dataset with limited annotations, and presents a core sample selection algorithm called SimCore based on submodular properties.
Correlational Image Modeling for Self-Supervised Visual Pre-Training
Wei Li (Nanyang Technological University), Chen Change Loy (Nanyang Technological University)
Object DetectionSegmentationRepresentation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A self-supervised visual pre-training method called Correlation Image Modeling (CIM) is proposed, which learns general representations by randomly cropping and predicting the correlation map between the cropped patches and the original image.