CVPR 2023 Papers — Page 13
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2353 papers
Local Connectivity-Based Density Estimation for Face Clustering
Junho Shin (Chungnam National University), Yeong Jun Koh (Chungnam National University)
ClassificationRecognitionTransformerImage
🎯 What it does: A facial clustering method based on local connectivity density estimation and paired connectivity networks is proposed. First, LCENet predicts the local connectivity probability among KNN neighbors and combines it with similarity to obtain reliable density. Then, a graph is constructed using a dual-edge selection strategy based on density and similarity, and finally, PCENet is used to predict paired connectivity on the selected edges, with clustering results obtained through BFS.
Local Implicit Normalizing Flow for Arbitrary-Scale Image Super-Resolution
Jie-En Yao (National Tsing Hua University), Chun-Yi Lee (National Tsing Hua University)
RestorationSuper ResolutionFlow-based ModelImage
🎯 What it does: This paper proposes a Local Implicit Regularization Flow (LINF), which combines regularization flow with local implicit neural representations to learn image texture distributions at arbitrary scales, achieving high-quality super-resolution generation at variable scales.
Local Implicit Ray Function for Generalizable Radiance Field Representation
Xin Huang (Northwestern Polytechnical University), Qing Wang (Tencent AI Lab)
RestorationGenerationData SynthesisTransformerNeural Radiance FieldImage
🎯 What it does: A Local Implicit Ray Function (LIRF) is proposed to reconstruct the radiance field of unknown scenes, achieving multi-scale novel view synthesis.
Local-Guided Global: Paired Similarity Representation for Visual Reinforcement Learning
Hyesong Choi (Ewha Womans University), Dongbo Min (Ewha Womans University)
Representation LearningReinforcement LearningContrastive LearningImage
🎯 What it does: The PSRL method is proposed, which utilizes self-supervised pixel correspondence and action-aware transformations to jointly learn local and global visual representations, enhancing the sample efficiency of visual reinforcement learning.
Local-to-Global Registration for Bundle-Adjusting Neural Radiance Fields
Yue Chen (Xi'an Jiaotong University), Fei Wang (Xi'an Jiaotong University)
Data SynthesisPose EstimationOptimizationNeural Radiance FieldImage
🎯 What it does: A local-to-global registration based NeRF bundle adjustment method called L2G-NeRF is proposed, which can simultaneously optimize camera pose and scene representation.
Localized Semantic Feature Mixers for Efficient Pedestrian Detection in Autonomous Driving
Abdul Hannan Khan (German Research Center for Artificial Intelligence), Andreas Dengel (German Research Center for Artificial Intelligence)
Object DetectionAutonomous DrivingComputational EfficiencyKnowledge DistillationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper proposes an anchor-free pedestrian detection architecture named LSFM based on MLP, primarily addressing the challenges of low latency in high-speed pedestrian detection and the difficulties in detecting small/occluded pedestrians.
LOCATE: Localize and Transfer Object Parts for Weakly Supervised Affordance Grounding
Gen Li (University of Edinburgh), Laura Sevilla-Lara (University of Edinburgh)
Object DetectionDomain AdaptationComputational EfficiencyTransformerContrastive LearningImage
🎯 What it does: The LOCATE framework is proposed, which learns the functional use of object parts through third-person perspective character-object interaction images under weak supervision, and transfers knowledge to first-person perspective static object images to achieve localization of target functional areas.
Logical Consistency and Greater Descriptive Power for Facial Hair Attribute Learning
Haiyu Wu (University of Notre Dame), Kevin W. Bowyer (University of Notre Dame)
ClassificationRecognitionConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper constructs a new facial image dataset, FH37K, containing 17 fine-grained facial beard attributes, and trains attribute classifiers on this dataset.
Logical Implications for Visual Question Answering Consistency
Sergio Tascon-Morales (University of Bern), Raphael Sznitman (University of Bern)
ClassificationRecognitionTransformerSupervised Fine-TuningVision Language ModelImageMultimodality
🎯 What it does: A training method for visual question answering models is proposed, which directly constrains the model using logical entailment relationships to enhance the model's consistency and accuracy.
LOGO: A Long-Form Video Dataset for Group Action Quality Assessment
Shiyi Zhang (Tsinghua University), Yansong Tang (Tsinghua University)
Graph Neural NetworkVideo
🎯 What it does: A multi-person long-term artistic swimming action quality assessment dataset LOGO has been constructed, and a GOAT module based on group graph convolution and attention fusion has been proposed.
LoGoNet: Towards Accurate 3D Object Detection With Local-to-Global Cross-Modal Fusion
Xin Li (East China Normal University), Liang He (East China Normal University)
Object DetectionAutonomous DrivingTransformerImageMultimodalityPoint Cloud
🎯 What it does: This paper proposes a local-to-global cross-modal fusion network named LoGoNet, designed to fuse LiDAR point clouds and multi-camera images to enhance 3D object detection accuracy.
Long Range Pooling for 3D Large-Scale Scene Understanding
Xiang-Li Li (Tsinghua University), Shi-Min Hu (Tsinghua University)
SegmentationConvolutional Neural NetworkPoint Cloud
🎯 What it does: A Long Range Pooling (LRP) module is proposed, and LRPNet is constructed for 3D voxel scene segmentation.
Long-Tailed Visual Recognition via Self-Heterogeneous Integration With Knowledge Excavation
Yan Jin (Xiamen University), Hanzi Wang (Xiamen University)
ClassificationRecognitionKnowledge DistillationMixture of ExpertsImage
🎯 What it does: This paper proposes a long-tail visual recognition framework called SHIKE based on mixed experts, which enhances the performance of tail classes by adaptively fusing features of different depths and mining knowledge.
Long-Term Visual Localization With Mobile Sensors
Shen Yan (National University of Defense Technology), Xiaowei Zhou (Zhejiang University)
Pose EstimationRetrievalOptimizationTransformerSimultaneous Localization and MappingImageMultimodality
🎯 What it does: This paper proposes a long-term visual localization framework called SensLoc based on multi-sensor information from mobile devices and constructs a new dataset.
Look Around for Anomalies: Weakly-Supervised Anomaly Detection via Context-Motion Relational Learning
MyeongAh Cho (Yonsei University), Sangyoun Lee (Yonsei University)
Anomaly DetectionConvolutional Neural NetworkVideo
🎯 What it does: A weakly supervised video anomaly detection method is proposed, utilizing a single backbone to achieve anomaly frame detection through implicit category activation feature learning and context-motion relationship modeling.
Look Before You Match: Instance Understanding Matters in Video Object Segmentation
Junke Wang (Fudan University), Yu-Gang Jiang (Fudan University)
Object DetectionSegmentationTransformerVideo
🎯 What it does: A two-branch network (ISVOS) is proposed to achieve video object segmentation by incorporating instance segmentation query information into memory matching.
Look, Radiate, and Learn: Self-Supervised Localisation via Radio-Visual Correspondence
Mohammed Alloulah (Nokia Bell Labs), Maximilian Arnold (Nokia Bell Labs)
Object DetectionContrastive LearningMultimodality
🎯 What it does: This paper proposes a method for target localization using unsupervised wireless visual correspondence and constructs a high-resolution MaxRay dataset for the first time.
Lookahead Diffusion Probabilistic Models for Refining Mean Estimation
Guoqiang Zhang (University of Technology Sydney), W. Bastiaan Kleijn (Victoria University of Wellington)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: A Lookahead Diffusion Probabilistic Model (LA-DPM) is proposed, which improves mean estimation by extrapolating the estimates of x for two consecutive steps during the reverse sampling process, thereby enhancing sampling quality.
Looking Through the Glass: Neural Surface Reconstruction Against High Specular Reflections
Jiaxiong Qiu (Nankai University), Bo Ren (Nankai University)
RestorationGenerationNeural Radiance FieldImage
🎯 What it does: A monocular 3D surface reconstruction method for high specular reflection (HSR) scenes, NeuS-HSR, is proposed, which can recover the true surface of the target object from images taken through glass.
Low-Light Image Enhancement via Structure Modeling and Guidance
Xiaogang Xu (Zhejiang Lab), Jiangbo Lu (SmartMore Corporation)
RestorationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: A low-light image enhancement framework is proposed that simultaneously performs structural modeling and appearance enhancement, achieving clearer and more realistic enhancement effects through a structure-guided appearance enhancement module.
LP-DIF: Learning Local Pattern-Specific Deep Implicit Function for 3D Objects and Scenes
Meng Wang (Tsinghua University), Zhizhong Han (Wayne State University)
GenerationData SynthesisPoint CloudMesh
🎯 What it does: This paper proposes LP-DIF, which reconstructs detail-rich geometries by partitioning 3D shapes into local regions and learning dedicated decoders for each pattern cluster.
LSTFE-Net:Long Short-Term Feature Enhancement Network for Video Small Object Detection
Jinsheng Xiao (Wuhan University), Jiayi Ma (Guangdong University of Technology)
Object DetectionConvolutional Neural NetworkVideo
🎯 What it does: This paper proposes a Long-Short Term Feature Enhancement Network (LSTFE-Net) that improves the performance of small object detection in videos by aligning short-term frames, selecting the most informative long-term frames, and aggregating multi-scale features.
LVQAC: Lattice Vector Quantization Coupled With Spatially Adaptive Companding for Efficient Learned Image Compression
Xi Zhang (Shanghai Jiao Tong University), Xiaolin Wu (McMaster University)
CompressionConvolutional Neural NetworkImage
🎯 What it does: In the end-to-end CNN image compression framework, a new quantization module called LVQAC (a combination of lattice vector quantization and spatial adaptive A-law compression) is proposed to replace traditional uniform scalar quantization.
M6Doc: A Large-Scale Multi-Format, Multi-Type, Multi-Layout, Multi-Language, Multi-Annotation Category Dataset for Modern Document Layout Analysis
Hiuyi Cheng (South China University of Technology), Lianwen Jin (IntSig Information Co., Ltd.)
Object DetectionSegmentationTransformerImage
🎯 What it does: This paper constructs a large-scale, multi-format, multi-type, multi-layout, multi-language, and multi-annotated document layout analysis dataset M Doc 06, and proposes a Transformer-based document layout analysis model TransDLANet.
MACARONS: Mapping and Coverage Anticipation With RGB Online Self-Supervision
Antoine Guédon, Vincent Lepetit
Depth EstimationRobotic IntelligenceTransformerSimultaneous Localization and MappingPoint Cloud
🎯 What it does: This paper studies a next best view (NBV) method for self-supervised online learning that uses only RGB cameras, capable of exploring and reconstructing in unknown large-scale 3D environments simultaneously.
MAESTER: Masked Autoencoder Guided Segmentation at Pixel Resolution for Accurate, Self-Supervised Subcellular Structure Recognition
Ronald Xie (University of Toronto), Bo Wang (Vector Institute)
RecognitionSegmentationTransformerAuto EncoderImage
🎯 What it does: The study proposes a self-supervised segmentation method called MAESTER, which utilizes Vision Transformer for pixel-level segmentation of subcellular structures in electron microscopy images of cells.
MAGE: MAsked Generative Encoder To Unify Representation Learning and Image Synthesis
Tianhong Li (Massachusetts Institute of Technology), Dilip Krishnan (Google)
GenerationRepresentation LearningTransformerGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: The MAGE framework is proposed, unifying image generation and self-supervised representation learning, using variable mask ratios for pre-training semantic tokens generated by VQGAN;
Magic3D: High-Resolution Text-to-3D Content Creation
Chen-Hsuan Lin (NVIDIA Corporation), Tsung-Yi Lin (NVIDIA Corporation)
GenerationDiffusion modelTextMeshStochastic Differential Equation
🎯 What it does: A two-stage coarse-fine optimization framework is utilized to generate high-resolution, detail-rich 3D mesh models from text prompts.
MagicNet: Semi-Supervised Multi-Organ Segmentation via Magic-Cube Partition and Recovery
Duowen Chen (East China Normal University), Yan Wang (East China Normal University)
SegmentationConvolutional Neural NetworkImageComputed Tomography
🎯 What it does: This paper proposes MagicNet, a semi-supervised multi-organ CT segmentation method based on a teacher-student framework. The method enhances the segmentation quality of small organs by partitioning 3D CT voxels into N³ small cubes (magic-cubes), performing partition-recovery data augmentation across and within images, and utilizing cube-level local representations to fuse pseudo-labels within the same image branch.
MagicPony: Learning Articulated 3D Animals in the Wild
Shangzhe Wu (Visual Geometry Group, University of Oxford), Andrea Vedaldi (Visual Geometry Group, University of Oxford)
GenerationPose EstimationTransformerContrastive LearningImageMesh
🎯 What it does: This paper proposes the MagicPony method, which utilizes single-view images of wildlife to learn to predict the 3D shape, pose, texture, and lighting of animals from a single image, and can achieve animation and relighting during testing.
MAGVIT: Masked Generative Video Transformer
Lijun Yu (Carnegie Mellon University), Lu Jiang (Google Research)
GenerationData SynthesisTransformerAuto EncoderGenerative Adversarial NetworkVideo
🎯 What it does: A multi-task video generation model MAGVIT is proposed, achieving high-quality, efficient, and scalable generation and editing of videos through a 3D vector quantizer and a mask generator.
MAGVLT: Masked Generative Vision-and-Language Transformer
Sungwoong Kim (Korea University), Jongmin Kim (Kakao Brain)
GenerationData SynthesisTransformerVision Language ModelGenerative Adversarial NetworkImageTextMultimodality
🎯 What it does: This paper proposes a unified generative visual-language Transformer (MAGVLT) that can generate text from images, generate images from text, and even generate images and text simultaneously within a single model; it employs a non-autoregressive masking prediction mechanism to achieve parallel decoding and iterative optimization.
MAIR: Multi-View Attention Inverse Rendering With 3D Spatially-Varying Lighting Estimation
JunYong Choi, Junghyun Cho
Convolutional Neural NetworkImage
🎯 What it does: This paper explores a specific problem in the field of computer vision and proposes a new solution.
Make Landscape Flatter in Differentially Private Federated Learning
Yifan Shi (Tsinghua University), Dacheng Tao (JD Explore Academy)
OptimizationFederated LearningSafty and PrivacyImage
🎯 What it does: This paper introduces the SAM optimizer in federated learning to enhance model performance under client-level differential privacy (DP) protection.
Make-a-Story: Visual Memory Conditioned Consistent Story Generation
Tanzila Rahman (University of British Columbia), Leonid Sigal (University of British Columbia)
GenerationDiffusion modelImageText
🎯 What it does: Implementing visual story generation in text narratives
Making Vision Transformers Efficient From a Token Sparsification View
Shuning Chang (National University of Singapore), Mike Zheng Shou
ClassificationRecognitionObject DetectionComputational EfficiencyTransformerImageVideo
🎯 What it does: Proposes Semantic Token ViT (STViT), which generates a small number of semantic centers through self-attention clustering to replace the original image/video tokens, significantly reducing computational load while maintaining or improving accuracy;
MaLP: Manipulation Localization Using a Proactive Scheme
Vishal Asnani (Michigan State University), Xiaoming Liu (Michigan State University)
ClassificationAnomaly DetectionConvolutional Neural NetworkTransformerGenerative Adversarial NetworkImageBenchmark
🎯 What it does: This paper proposes an active image manipulation localization method called MaLP, which uses learned templates to encrypt real images, and then performs forgery detection and pixel-level localization on the encrypted images through a two-branch network.
MammalNet: A Large-Scale Video Benchmark for Mammal Recognition and Behavior Understanding
Jun Chen (King Abdullah University of Science and Technology), Mohamed Elhoseiny (King Abdullah University of Science and Technology)
ClassificationRecognitionObject DetectionConvolutional Neural NetworkTransformerSupervised Fine-TuningVideoBenchmark
🎯 What it does: A large-scale video benchmark called MammalNet is proposed, covering 173 species of mammals and 12 types of higher-order behaviors, and three challenge tasks are designed: standard classification, low-shot combination recognition, and behavior detection.
Manipulating Transfer Learning for Property Inference
Yulong Tian (Nanjing University), David Evans (University of Virginia)
Domain AdaptationAdversarial AttackImage
🎯 What it does: This paper studies how, in the context of transfer learning, if the upstream model is maliciously manipulated, it can perform attribute inference attacks on the downstream model, revealing whether the training set contains samples of specific individuals or groups.
MAP: Multimodal Uncertainty-Aware Vision-Language Pre-Training Model
Yatai Ji (Tsinghua University), Yujiu Yang (Tencent)
RecognitionRetrievalTransformerVision Language ModelImageTextMultimodality
🎯 What it does: A multi-modal uncertainty-aware pre-training framework MAP is proposed, which models visual and linguistic representations as multivariate Gaussian distributions using a Probability Distribution Encoder (PDE), and designs three distribution-based pre-training tasks.
MaPLe: Multi-Modal Prompt Learning
Muhammad Uzair Khattak (Mohamed bin Zayed University of AI), Fahad Shahbaz Khan (Linköping University)
ClassificationRecognitionDomain AdaptationTransformerPrompt EngineeringVision Language ModelImageMultimodality
🎯 What it does: Proposed and implemented a method for simultaneous prompt learning for both the visual and language branches of CLIP (MaPLe), achieving efficient fine-tuning for downstream visual recognition tasks.
Mapping Degeneration Meets Label Evolution: Learning Infrared Small Target Detection With Single Point Supervision
Xinyi Ying (National University of Defense Technology), Shilin Zhou (National University of Defense Technology)
Object DetectionConvolutional Neural NetworkImage
🎯 What it does: A weakly supervised infrared small target detection framework called LESPS is proposed, which can gradually evolve single-point labels into pixel-level target masks through intermediate predictions of the network.
Marching-Primitives: Shape Abstraction From Signed Distance Function
Weixiao Liu (National University of Singapore), Gregory S. Chirikjian (Johns Hopkins University)
GenerationOptimizationExplainability and InterpretabilityPoint CloudMesh
🎯 What it does: An algorithm is proposed to directly extract geometric primitives (such as superquadrics) from voxelized SDF, called Marching-Primitives;
MarginMatch: Improving Semi-Supervised Learning with Pseudo-Margins
Tiberiu Sosea (University of Illinois Chicago), Cornelia Caragea (University of Illinois Chicago)
ClassificationDomain AdaptationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: Proposes the MarginMatch method, which combines consistency regularization and pseudo-labeling, and utilizes the dynamic changes of pseudo-margins during training to filter high-quality pseudo-labels to enhance semi-supervised learning effectiveness.
Markerless Camera-to-Robot Pose Estimation via Self-Supervised Sim-to-Real Transfer
Jingpei Lu (University of California, San Diego), Michael C. Yip (University of California, San Diego)
Pose EstimationRobotic IntelligenceImage
🎯 What it does: An end-to-end CtRNet is proposed for unmarked camera-to-robot pose estimation, enhancing performance through self-supervised sim-to-real training.
MARLIN: Masked Autoencoder for Facial Video Representation LearnINg
Zhixi Cai (Monash University), Munawar Hayat (Monash University)
RecognitionRepresentation LearningTransformerAuto EncoderGenerative Adversarial NetworkVideo
🎯 What it does: Self-supervised facial video representation learning is based on reconstructing densely occluded facial regions to learn general facial features;
MarS3D: A Plug-and-Play Motion-Aware Model for Semantic Segmentation on Multi-Scan 3D Point Clouds
Jiahui Liu (University of Hong Kong), Xiaojuan Qi (University of Hong Kong)
SegmentationAutonomous DrivingConvolutional Neural NetworkPoint Cloud
🎯 What it does: In the task of multi-scan 3D point cloud semantic segmentation, a pluggable MarS3D module is proposed, which can add multi-scan perception capabilities to existing single-scan models and achieve joint prediction of semantic categories and motion states.
Mask DINO: Towards a Unified Transformer-Based Framework for Object Detection and Segmentation
Feng Li (Hong Kong University of Science and Technology), Heung-Yeung Shum (Hong Kong University of Science and Technology)
Object DetectionSegmentationTransformerImage
🎯 What it does: Mask DINO achieves a unified framework for object detection and image segmentation (instance, panoptic, semantic) by adding a mask prediction branch to the DINO detection framework.
Mask-Free OVIS: Open-Vocabulary Instance Segmentation Without Manual Mask Annotations
Vibashan VS (Salesforce Research), Ran Xu (Salesforce Research)
Object DetectionSegmentationVision Language ModelImage
🎯 What it does: An open-source vocabulary instance segmentation method is proposed that does not require manual mask annotations, training Mask-RCNN through the generation of pseudo-masks.
Mask-Free Video Instance Segmentation
Lei Ke (ETH Zurich), Fisher Yu (ETH Zurich)
Object DetectionObject TrackingSegmentationTransformerVideo
🎯 What it does: A mask-free annotation method for video instance segmentation called MaskFreeVIS is proposed.
Mask-Guided Matting in the Wild
Kwanyong Park (Korea Advanced Institute of Science and Technology), Joon-Young Lee (Adobe Research)
RestorationSegmentationConvolutional Neural NetworkImageBenchmark
🎯 What it does: This paper proposes a mask-guided image matting method for outdoor scenes that can generate high-quality alpha mattes with only a rough mask provided.
Mask3D: Pre-Training 2D Vision Transformers by Learning Masked 3D Priors
Ji Hou (Meta Reality Labs), Matthias Nießner (Technical University of Munich)
Object DetectionSegmentationTransformerContrastive LearningImage
🎯 What it does: Mask3D injects 3D priors into the 2D Vision Transformer through a self-supervised masking reconstruction task on single-view RGB-D images.
MaskCLIP: Masked Self-Distillation Advances Contrastive Language-Image Pretraining
Xiaoyi Dong (University of Science and Technology of China), Nenghai Yu (Xiamen University)
Object DetectionSegmentationRetrievalKnowledge DistillationRepresentation LearningTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: This paper proposes MaskCLIP, which combines mask self-distillation and visual-language contrastive learning for pre-training, enhancing the transferability of visual models.
MaskCon: Masked Contrastive Learning for Coarse-Labelled Dataset
Chen Feng (Queen Mary University of London), Ioannis Patras (Queen Mary University of London)
RetrievalRepresentation LearningContrastive LearningImage
🎯 What it does: In this study, the authors propose a contrastive learning framework named MaskCon, which generates soft labels using coarse-grained labels and inter-sample relationships to learn fine-grained representations.
Masked and Adaptive Transformer for Exemplar Based Image Translation
Chang Jiang (Hangzhou Dianzi University), Gang Xu (Hangzhou Dianzi University)
Image TranslationGenerationTransformerGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: A sample-driven image translation framework named MATEBIT is proposed, which utilizes Masked and Adaptive Transformer to learn cross-domain correspondences and achieves local and global style control through contrastive style learning and a U-Net decoder.
Masked Auto-Encoders Meet Generative Adversarial Networks and Beyond
Zhengcong Fei (Meituan), Xiaolin Wei (Meituan)
Object DetectionSegmentationGenerationRepresentation LearningTransformerAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: A GAN-based Masked AutoEncoder pre-training framework, GAN-MAE, has been designed and implemented. It reconstructs masked image patches using a generator and distinguishes authenticity with a discriminator to enhance the effectiveness of self-supervised visual representation learning.
Masked Autoencoders Enable Efficient Knowledge Distillers
Yutong Bai (Johns Hopkins University), Cihang Xie (University of California Santa Cruz)
Computational EfficiencyKnowledge DistillationTransformerAuto EncoderImage
🎯 What it does: This paper proposes the DMAE framework, which utilizes a teacher model pre-trained with MAE to achieve efficient knowledge distillation during the self-supervised phase by aligning intermediate features.
Masked Autoencoding Does Not Help Natural Language Supervision at Scale
Floris Weers (Apple), Tom Gunter (Apple)
TransformerAuto EncoderContrastive LearningImageTextMultimodality
🎯 What it does: The study combines self-supervised masked autoencoders with natural language supervised CLIP in large-scale image-text pretraining, proposing the MAE-CLIP model and evaluating its performance.
Masked Image Modeling With Local Multi-Scale Reconstruction
Haoqing Wang (Peking University), Kai Han (Huawei Noah's Ark Lab)
Object DetectionSegmentationRepresentation LearningTransformerContrastive LearningImage
🎯 What it does: This paper proposes a Local Multi-scale Reconstruction (LocalMIM) framework to improve representation learning in Masked Image Modeling (MIM).
Masked Image Training for Generalizable Deep Image Denoising
Haoyu Chen (Hong Kong University of Science and Technology), Lei Zhu (Hong Kong University of Science and Technology)
RestorationTransformerImage
🎯 What it does: A mask training strategy is proposed to enhance the generalization ability of deep image denoising models to non-training noise.
Masked Images Are Counterfactual Samples for Robust Fine-Tuning
Yao Xiao (Sun Yat-sen University), Liang Lin (Sun Yat-sen University)
ClassificationDomain AdaptationKnowledge DistillationTransformerSupervised Fine-TuningImage
🎯 What it does: Using CAM-based masked images as counterfactual samples, combined with feature distillation from a pre-trained model, to fine-tune large models like CLIP, aiming to enhance OOD robustness while maintaining ID performance.
Masked Jigsaw Puzzle: A Versatile Position Embedding for Vision Transformers
Bin Ren (University of Trento), Wei Wang (Beijing Jiaotong University)
ClassificationSafty and PrivacyTransformerImage
🎯 What it does: This paper studies the positional embeddings of visual Transformers, demonstrating their explicit learning of two-dimensional spatial relationships and leading to privacy leakage, and proposes the Masked Jigsaw Puzzle (MJP) positional embedding scheme.
Masked Motion Encoding for Self-Supervised Video Representation Learning
Xinyu Sun (South China University of Technology), Chuang Gan (UMass Amherst)
Representation LearningTransformerOptical FlowVideo
🎯 What it does: This paper proposes a self-supervised video representation learning method that utilizes Masked Motion Encoding (MME) to learn video features by recovering motion trajectories.
Masked Representation Learning for Domain Generalized Stereo Matching
Zhibo Rao (Nanchang Hangkong University), Xing Li (Northwestern Polytechnical University)
Depth EstimationDomain AdaptationRepresentation LearningConvolutional Neural NetworkAuto EncoderImage
🎯 What it does: A pseudo-multi-task framework based on occlusion representation learning is designed, where the left image is randomly occluded and input into the network along with the complete right image. An image reconstruction task is added during the training phase to enhance the generalization performance of cross-domain stereo matching.
Masked Scene Contrast: A Scalable Framework for Unsupervised 3D Representation Learning
Xiaoyang Wu (University of Hong Kong), Hengshuang Zhao (University of Hong Kong)
SegmentationRepresentation LearningTransformerContrastive LearningPoint Cloud
🎯 What it does: This paper proposes an unsupervised 3D representation learning framework MSC based on scene-level point clouds, combining contrastive learning and masked point modeling for pre-training.
Masked Video Distillation: Rethinking Masked Feature Modeling for Self-Supervised Video Representation Learning
Rui Wang (Fudan University), Yu-Gang Jiang (Fudan University)
Knowledge DistillationRepresentation LearningTransformerVideo
🎯 What it does: A Masked Video Distillation (MVD) framework is proposed, which first uses MAE/VideoMAE pre-trained image or video models as teachers, and then uses the high-level features output by these teachers as the target for mask feature prediction to train a student Vision Transformer to learn more semantically meaningful spatiotemporal representations.
Masked Wavelet Representation for Compact Neural Radiance Fields
Daniel Rho (AI2XL KT), Eunbyung Park (Sungkyunkwan University)
Data SynthesisCompressionNeural Radiance FieldMesh
🎯 What it does: This paper proposes a scheme for sparsifying and compressing grid-based neural fields using multi-level wavelet transforms and learnable masks, balancing rendering quality and storage efficiency.
MaskSketch: Unpaired Structure-Guided Masked Image Generation
Dina Bashkirova (Boston University), Irfan Essa (Georgia Institute of Technology)
GenerationData SynthesisTransformerContrastive LearningImage
🎯 What it does: Developed MaskSketch, a method for unpaired sketch-to-photo generation that utilizes a pre-trained MaskGIT and structural guidance through self-attention mapping.
Master: Meta Style Transformer for Controllable Zero-Shot and Few-Shot Artistic Style Transfer
Hao Tang (Peking University), Xinchao Wang (National University of Singapore)
Image TranslationMeta LearningTransformerReinforcement LearningImageText
🎯 What it does: A Transformer architecture named Meta Style Transformer (Master) is proposed for controllable zero-shot and few-shot artistic style transfer.
Matching Is Not Enough: A Two-Stage Framework for Category-Agnostic Pose Estimation
Min Shi (Huazhong University of Science and Technology), Zhiguo Cao (Huazhong University of Science and Technology)
Pose EstimationTransformerImageBenchmark
🎯 What it does: A two-stage framework for category-agnostic pose estimation is proposed, where a Transformer encoder is first used to obtain similarity cues, and then a decoder refines the keypoint locations.
MCF: Mutual Correction Framework for Semi-Supervised Medical Image Segmentation
Yongchao Wang (Chongqing University of Posts and Telecommunications), Xinbo Gao (Chongqing University of Posts and Telecommunications)
SegmentationConvolutional Neural NetworkImageBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: A mutual correction framework MCF is proposed for semi-supervised medical image segmentation, utilizing two different structured sub-networks to correct deviations from each other.
MD-VQA: Multi-Dimensional Quality Assessment for UGC Live Videos
Zicheng Zhang (Shanghai Jiao Tong University), Guangtao Zhai (Shanghai Jiao Tong University)
CompressionConvolutional Neural NetworkVideo
🎯 What it does: This paper first constructs the largest and most diverse video database, TaoLive, which includes compressed distorted videos, and proposes a no-reference multidimensional video quality assessment model MD-VQA based on semantic, distortion, and motion three-dimensional features;
MDL-NAS: A Joint Multi-Domain Learning Framework for Vision Transformer
Shiguang Wang (University of Electronic Science and Technology of China), Haijun Liu (Chongqing University)
ClassificationObject DetectionSegmentationPose EstimationNeural Architecture SearchTransformerImage
🎯 What it does: MDL-NAS is proposed, a Transformer framework that jointly handles multi-domain and multi-tasking, capable of efficiently sharing parameters within a single super network and automatically searching for optimal structures for different visual tasks (classification, detection, segmentation);
MDQE: Mining Discriminative Query Embeddings To Segment Occluded Instances on Challenging Videos
Minghan Li (Hong Kong Polytechnic University), Lei Zhang (Hong Kong Polytechnic University)
Object DetectionSegmentationTransformerContrastive LearningVideo
🎯 What it does: The MDQE method is proposed for the video instance segmentation task, which improves the segmentation performance of occluded instances by incorporating spatial-temporal priors during the query initialization phase and applying contrastive learning to the query embeddings.
MED-VT: Multiscale Encoder-Decoder Video Transformer With Application To Object Segmentation
Rezaul Karim (York University), Mennatullah Siam (York University)
Object DetectionSegmentationTransformerVideo
🎯 What it does: A unified multi-scale encoding-decoding video Transformer (MED-VT) is proposed for automatic video object segmentation and actor/action segmentation, achieving spatiotemporal consistent pixel-level predictions through multi-scale queries and multi-to-multi label propagation.
MEDIC: Remove Model Backdoors via Importance Driven Cloning
Qiuling Xu (Purdue University), Xiangyu Zhang (Purdue University)
Anomaly DetectionAutonomous DrivingConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A method called Importance-Driven Cloning (MEDIC) is proposed, which trains a model from scratch using a small number of clean samples to replicate the benign functions of the original model and remove backdoors.
Megahertz Light Steering Without Moving Parts
Adithya Pediredla (Dartmouth College), Ioannis Gkioulekas (Carnegie Mellon University)
Point CloudUltrasound
🎯 What it does: A light path control technology based on ultrasonic shaping has been designed and implemented, enabling optical scanning at MHz rates without moving parts, with prototype experiments conducted in projectors and LiDAR systems.
MEGANE: Morphable Eyeglass and Avatar Network
Junxuan Li (Australian National University), Jason Saragih (Meta Reality Labs Research)
GenerationData SynthesisAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: A method is proposed to reconstruct, exchange, and relight 3D deformable models of glasses and faces in a multi-view environment.
MELTR: Meta Loss Transformer for Learning To Fine-Tune Video Foundation Models
Dohwan Ko (Korea University), Hyunwoo J. Kim (Korea University)
RetrievalMeta LearningTransformerVideoMultimodality
🎯 What it does: A Meta Loss Transformer (MELTR) module is proposed, which can automatically and non-linearly fuse the loss functions of multiple auxiliary tasks into a unified loss, helping video foundation models to fine-tune more effectively.
MeMaHand: Exploiting Mesh-Mano Interaction for Single Image Two-Hand Reconstruction
Congyi Wang (ByteDance), Shilei Wen (ByteDance)
Pose EstimationGraph Neural NetworkTransformerImageMesh
🎯 What it does: We propose MeMaHand, which can simultaneously predict the 3D mesh vertex positions of both hands and MANO parameters from a single RGB image, achieving two-hand reconstruction in a single forward pass.
Memory-Friendly Scalable Super-Resolution via Rewinding Lottery Ticket Hypothesis
Jin Lin (Xiamen University), Zongze Wu (Shenzhen University)
RestorationSuper ResolutionCompressionKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: This paper studies a memory-friendly scalable super-resolution framework MSSR, which utilizes the Lottery Ticket Hypothesis (LTH) and weight resetting to achieve gradual compression and expansion of the network, resulting in multi-scale sub-networks.
Meta Architecture for Point Cloud Analysis
Haojia Lin (Xiamen University), Rongrong Ji (Xiamen University)
ClassificationSegmentationGraph Neural NetworkTransformerPoint Cloud
🎯 What it does: A unified Meta architecture called PointMeta is proposed, which abstracts the building blocks of existing point cloud networks into four meta-functions: neighbor update, aggregation, point update, and position embedding, and based on this, an efficient PointMetaBase module is introduced.
Meta Compositional Referring Expression Segmentation
Li Xu (Singapore University of Technology and Design), Jun Liu (Singapore University of Technology and Design)
SegmentationMeta LearningTransformerVision Language ModelImage
🎯 What it does: This paper proposes a meta-learning-based framework called MCRES to enhance the generalization ability of referential expression segmentation models when dealing with new combinations of concepts.
Meta Omnium: A Benchmark for General-Purpose Learning-To-Learn
Ondrej Bohdal (University of Edinburgh), Timothy Hospedales (Samsung AI Center)
ClassificationSegmentationPose EstimationHyperparameter SearchMeta LearningConvolutional Neural NetworkGaussian SplattingImageBenchmark
🎯 What it does: This paper proposes the Meta Omnium benchmark—a multi-task few-shot meta-learning benchmark that integrates four major visual tasks: classification, semantic segmentation, keypoint localization, and regression. It provides a unified data partition, evaluation metrics, HPO strategies, and baseline models.
Meta-Causal Learning for Single Domain Generalization
Jin Chen (Beijing Institute of Technology), Jiebo Luo (University of Rochester)
Domain AdaptationMeta LearningImage
🎯 What it does: This paper addresses the single-source domain generalization problem by proposing a three-stage learning framework called simulate-analyze-reduce, and implements a meta-causal learning method based on counterfactual causal inference.
Meta-Explore: Exploratory Hierarchical Vision-and-Language Navigation Using Scene Object Spectrum Grounding
Minyoung Hwang (Seoul National University), Songhwai Oh (Seoul National University)
Robotic IntelligenceTransformerReinforcement LearningVision Language ModelMultimodality
🎯 What it does: Meta-Explore is proposed, a hierarchical visual-language navigation framework that constructs a topological map during the exploration phase and utilizes the Object Spectrum (SOS) to find unvisited local targets to correct erroneous paths during the utilization phase.
Meta-Learning With a Geometry-Adaptive Preconditioner
Suhyun Kang (Seoul National University), Wonjong Rhee (Seoul National University)
OptimizationMeta LearningImage
🎯 What it does: This paper studies preconditioned gradient descent under the MAML framework, proposing the Geometry-Adaptive Preconditioner (GAP) and its low-computation approximation, Approximate GAP. These methods can adaptively utilize task-specific and path-dependent preconditioner matrices in the inner loop while satisfying Riemannian metrics, thereby enhancing meta-learning performance.
Meta-Personalizing Vision-Language Models To Find Named Instances in Video
Chun-Hsiao Yeh (University of California), Simon Jenni (Adobe Research)
Object DetectionRetrievalMeta LearningTransformerVision Language ModelContrastive LearningVideoText
🎯 What it does: A meta-personalized visual-language model is proposed, which automatically learns representations of user-specified instances in videos and retrieves those instances through natural language queries.
Meta-Tuning Loss Functions and Data Augmentation for Few-Shot Object Detection
Berkan Demirel (Middle East Technical University), Ramazan Gokberk Cinbis (HAVELSAN Inc.)
Object DetectionMeta LearningConvolutional Neural NetworkReinforcement LearningImage
🎯 What it does: A Meta-Tuning framework for few-shot object detection is proposed, which automatically adjusts the loss function and data augmentation parameters through meta-learning to enhance the performance of fine-tuning models.
MetaCLUE: Towards Comprehensive Visual Metaphors Research
Arjun R. Akula (Google), Varun Jampani (Google)
ClassificationGenerationRetrievalTransformerDiffusion modelContrastive LearningImage
🎯 What it does: This paper proposes the MetaCLUE framework, systematically constructing four types of tasks for visual metaphors (classification, localization, understanding, and generation), and collects high-quality annotations on advertising images, including main concepts, secondary concepts, relationships, and bounding boxes.
Metadata-Based RAW Reconstruction via Implicit Neural Functions
Leyi Li (Zhejiang University), Qinmin Yang (Zhejiang University)
RestorationImage
🎯 What it does: This paper proposes a RAW reconstruction method based on implicit neural functions (INF), utilizing the sparse metadata (sampled RAW pixels) embedded during capture and the corresponding sRGB values to construct a conditional mapping from coordinates to RAW values, thereby self-supervising the recovery of the original RAW image.
MetaFusion: Infrared and Visible Image Fusion via Meta-Feature Embedding From Object Detection
Wenda Zhao (Dalian University of Technology), Huchuan Lu (Dalian University of Technology)
Image TranslationObject DetectionMeta LearningConvolutional Neural NetworkImageMultimodality
🎯 What it does: A framework for joint learning of infrared-visible image fusion and object detection, called MetaFusion, is proposed, which achieves mutual promotion of the two tasks through meta-feature embedding.
MetaMix: Towards Corruption-Robust Continual Learning With Temporally Self-Adaptive Data Transformation
Zhenyi Wang (State University of New York at Buffalo), Mingchen Gao (State University of New York at Buffalo)
ClassificationMeta LearningRecurrent Neural NetworkImage
🎯 What it does: This paper proposes MetaMix, an adaptive data augmentation framework for continual learning, aimed at enhancing the model's robustness against unseen corruptions.
MetaPortrait: Identity-Preserving Talking Head Generation With Fast Personalized Adaptation
Bowen Zhang (University of Science and Technology of China), Fang Wen (Microsoft)
GenerationSuper ResolutionMeta LearningGenerative Adversarial NetworkImageVideo
🎯 What it does: A complete framework has been developed that can generate high-quality talking head avatars from a single source image and driving video while maintaining identity without distortion.
MetaViewer: Towards a Unified Multi-View Representation
Ren Wang (Shandong University), Yilong Yin (Shandong University)
Representation LearningMeta LearningAuto EncoderContrastive LearningImage
🎯 What it does: This paper proposes MetaViewer, a unified multi-view representation learning framework based on meta-learning, which automatically integrates and eliminates view-specific redundant information through a unified-to-specific two-layer optimization process.
MethaneMapper: Spectral Absorption Aware Hyperspectral Transformer for Methane Detection
Satish Kumar (University of California Santa Barbara), B S Manjunath (University of California Santa Barbara)
Object DetectionSegmentationConvolutional Neural NetworkTransformerImage
🎯 What it does: A methane plume detection network called MethaneMapper based on hyperspectral Transformer has been designed and implemented, capable of end-to-end detection and quantification of methane emissions, and a large methane hotspot dataset MHS has been released.
METransformer: Radiology Report Generation by Transformer With Multiple Learnable Expert Tokens
Zhanyu Wang (University of Sydney), Luping Zhou (University of Sydney)
GenerationTransformerMixture of ExpertsImageTextElectronic Health Records
🎯 What it does: The METransformer model is proposed, which implements a multi-expert joint diagnosis framework for radiology report generation by introducing multiple learnable Expert Tokens in the Transformer encoder and decoder.
MHPL: Minimum Happy Points Learning for Active Source Free Domain Adaptation
Fan Wang (Shandong University), Yilong Yin (Shandong University)
ClassificationDomain AdaptationConvolutional Neural NetworkImage
🎯 What it does: In the absence of available source data, an active source-free domain adaptation method MHPL is proposed, specifically selecting and utilizing the 'minimum happy points' (Neighbor-Chaotic, Individual-Different, Source-Dissimilar) to enhance target domain performance.
MIANet: Aggregating Unbiased Instance and General Information for Few-Shot Semantic Segmentation
Yong Yang (South China University of Technology), Tianlin Huang (South China University of Technology)
SegmentationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This study proposes a multi-information aggregation network called MIANet, which generates general category prototypes through word vectors and combines unbiased instance information to achieve few-shot semantic segmentation.
MIC: Masked Image Consistency for Context-Enhanced Domain Adaptation
Lukas Hoyer (ETH Zurich), Luc Van Gool (ETH Zurich)
ClassificationObject DetectionSegmentationDomain AdaptationKnowledge DistillationImage
🎯 What it does: This paper proposes a Masked Image Consistency (MIC) module that enhances the learning of contextual relationships in the target domain by randomly masking patches in target domain images and ensuring the network remains consistent with the pseudo-labels of the complete images, thereby improving unsupervised domain adaptation performance.