CVPR 2024 Papers — Page 15
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2716 papers
Local-consistent Transformation Learning for Rotation-invariant Point Cloud Analysis
Yiyang Chen (South China University of Technology), Dacheng Tao (Nanyang Technological University)
ClassificationSegmentationGraph Neural NetworkContrastive LearningPoint Cloud
🎯 What it does: A point cloud rotation-invariant learning framework called LocoTrans is proposed, which utilizes a Local Consistent Reference Frame (LCRF) and a Relative Pose Recovery (RPR) module to achieve rotation-invariant feature extraction.
Localization Is All You Evaluate: Data Leakage in Online Mapping Datasets and How to Fix It
Adam Lilja (Chalmers University of Technology), Lars Hammarstrand (Chalmers University of Technology)
Object DetectionAutonomous DrivingTransformerSupervised Fine-TuningImage
🎯 What it does: This paper studies the data leakage problem in online mapping and proposes a geographically separated training/validation/testing split scheme (Near, Far), and re-evaluates mainstream methods.
Locally Adaptive Neural 3D Morphable Models
Michail Tarasiou (Imperial College London), Stefanos Zafeiriou (Imperial College London)
TransformerAuto EncoderMesh
🎯 What it does: A Local Adaptive Shape Model (LAMM) is proposed, which is an autoencoder framework that can directly manipulate 3D mesh with sparse control point displacements using a single forward pass;
LocLLM: Exploiting Generalizable Human Keypoint Localization via Large Language Model
Dongkai Wang (Peking University), Shiliang Zhang (Peking University)
Pose EstimationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageText
🎯 What it does: A keypoint localization method based on large language models, LocLLM, is proposed, which utilizes a visual encoder to extract image features and inputs the image features along with text descriptions (including keypoint types, locations, and relationships) into a pre-trained LLM for inference, outputting keypoint coordinates.
LoCoNet: Long-Short Context Network for Active Speaker Detection
Xizi Wang (Indiana University), Gedas Bertasius (University of North Carolina at Chapel Hill)
RecognitionObject DetectionConvolutional Neural NetworkTransformerVideoAudio
🎯 What it does: A proactive speaker detection model called LoCoNet is proposed, which combines long-term intrinsic speaker context with short-term cross-speaker context to address the challenges of multiple speakers and small face scenarios.
Lodge: A Coarse to Fine Diffusion Network for Long Dance Generation Guided by the Characteristic Dance Primitives
Ronghui Li (Tsinghua University), Xiu Li (Tsinghua University)
GenerationTransformerDiffusion modelVideo
🎯 What it does: A dual-stage coarse-fine diffusion network called Lodge is proposed, capable of generating extremely long 3D dance sequences in parallel based on music while maintaining global choreography rules.
Logarithmic Lenses: Exploring Log RGB Data for Image Classification
Bruce A. Maxwell (Northeastern University), Zewen Li (Northeastern University)
ClassificationConvolutional Neural NetworkImage
🎯 What it does: This study investigates the use of logarithmic RGB (log RGB) as input images, which can enhance the performance of deep networks in image classification tasks compared to traditional sRGB/linear RGB, as well as improve robustness against variations in illumination intensity and color balance.
Logit Standardization in Knowledge Distillation
Shangquan Sun (Institute of Information Engineering, Chinese Academy of Sciences), Xiaochun Cao (Shenzhen Campus of Sun Yat-sen University)
ClassificationKnowledge DistillationImage
🎯 What it does: This paper proposes a preprocessing method for Z-score normalization of logits in knowledge distillation, allowing the student model to freely adjust the range and variance of logits, thereby better learning the relationship with the teacher model.
Long-Tail Class Incremental Learning via Independent Sub-prototype Construction
Xi Wang (Xidian University), Cheng Deng (Xidian University)
ClassificationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: A long-tail category incremental learning framework based on sub-prototype space and memory space is proposed, addressing the issues of inter-class forgetting and intra-class imbalance under imbalanced data distribution.
Long-Tailed Anomaly Detection with Learnable Class Names
Chih-Hui Ho (University of California), Nuno Vasconcelos (University of California)
Anomaly DetectionTransformerVision Language ModelAuto EncoderImage
🎯 What it does: In the industrial defect detection task, a single-model unsupervised anomaly detection method for long-tail distribution, LTAD, is proposed, which can detect and locate defects in multi-class and extremely imbalanced data.
Look-Up Table Compression for Efficient Image Restoration
Yinglong Li (University of Science and Technology of China), Zhiwei Xiong (University of Science and Technology of China)
RestorationSuper ResolutionCompressionImage
🎯 What it does: This paper proposes a Look-Up Table (LUT)-based image recovery compression framework, utilizing diagonal re-indexing and non-diagonal subsampling (Diagonal-First Compression, DFC) to compress high-dimensional LUTs to smaller storage, and designs an SPF-LUT structure to enhance recovery performance.
Lookahead Exploration with Neural Radiance Representation for Continuous Vision-Language Navigation
Zihan Wang (Institute of Computing Technology, Chinese Academy of Sciences), Shuqiang Jiang (Indiana University)
Representation LearningTransformerNeural Radiance FieldContrastive LearningMultimodality
🎯 What it does: A forward exploration method based on Hierarchical Neural Radiance Representation (HNR) is proposed, which predicts future environments in continuous visual-language navigation using multi-layer semantic features, thereby enhancing navigation planning.
Looking 3D: Anomaly Detection with 2D-3D Alignment
Ankan Bhunia (University of Edinburgh), Hakan Bilen (University of Edinburgh)
Anomaly DetectionTransformerContrastive LearningImagePoint Cloud
🎯 What it does: This paper proposes a conditional anomaly detection task based on the comparison between query images and reference 3D models, and constructs a large-scale BrokenChairs-180K dataset.
Looking Similar Sounding Different: Leveraging Counterfactual Cross-Modal Pairs for Audiovisual Representation Learning
Nikhil Singh (Massachusetts Institute of Technology), Mahdi Kalayeh (Netflix)
Representation LearningTransformerContrastive LearningVideoMultimodalityAudio
🎯 What it does: By using multilingual audio tracks of dubbed movies to construct pairs of 'similar scenes with different voices', we improve audio-video self-supervised contrastive learning.
Loopy-SLAM: Dense Neural SLAM with Loop Closures
Lorenzo Liso (ETH Zurich), Martin R. Oswald (ETH Zurich)
OptimizationRobotic IntelligenceSimultaneous Localization and MappingPoint Cloud
🎯 What it does: This paper proposes Loopy-SLAM, a dense RGB-D SLAM method that utilizes neural point cloud subgraphs, capable of online loop closure and global optimization of trajectories and maps.
Loose Inertial Poser: Motion Capture with IMU-attached Loose-Wear Jacket
Chengxu Zuo (Xiamen University), Yipeng Qin (Cardiff University)
Pose EstimationRecurrent Neural NetworkAuto EncoderTime Series
🎯 What it does: Proposes Loose Inertial Poser (LIP), achieving real-time pose capture using sparse IMUs on loosely fitted jackets;
LORS: Low-rank Residual Structure for Parameter-Efficient Network Stacking
Jialin Li (Tencent), Chengjie Wang (Tencent)
Object DetectionCompressionTransformerImage
🎯 What it does: A low-rank residual structure (LORS) is proposed, sharing most parameters in stacked networks while retaining a small number of private parameters, significantly reducing the model parameters.
LoS: Local Structure-Guided Stereo Matching
Kunhong Li (Sun Yat-Sen University), Yulan Guo (National University of Defense Technology)
Depth EstimationOptimizationRecurrent Neural NetworkOptical FlowImage
🎯 What it does: A stereo matching method based on local structure guidance, LoS, is proposed. It first initializes local structure information (LSI) using monocular depth priors and binocular features, and then incorporates local structure guided propagation (LSGP) for efficient disparity updates during multi-level ConvGRU iterative optimization.
LoSh: Long-Short Text Joint Prediction Network for Referring Video Object Segmentation
Linfeng Yuan (Tongji University), Qijun Chen (Tongji University)
SegmentationTransformerOptical FlowVideoTextMultimodality
🎯 What it does: Proposes the LoSh network, which jointly predicts long and short texts, achieving a balance between appearance and action information through long-short cross attention and prediction intersection loss;
LOTUS: Evasive and Resilient Backdoor Attacks through Sub-Partitioning
Siyuan Cheng (Purdue University), Xiangyu Zhang (University of Massachusetts at Amherst)
Adversarial AttackConvolutional Neural NetworkImage
🎯 What it does: A backdoor attack framework called LOTUS is proposed, which first divides the samples of the victim class into multiple subsets, injects a unique trigger for each subset, and ultimately achieves a high success rate attack on the target class.
Low-Latency Neural Stereo Streaming
Qiqi Hou (Qualcomm), Hoang Le (Qualcomm)
CompressionAutonomous DrivingAuto EncoderOptical FlowVideo
🎯 What it does: This paper proposes a low-latency stereo video stream coding method called LLSS, which utilizes a bidirectional feature displacement module to achieve parallel compression of the left and right views.
Low-power Continuous Remote Behavioral Localization with Event Cameras
Friedhelm Hamann (Technische Universitat Berlin), Guillermo Gallego (Technische Universitat Berlin)
RecognitionObject DetectionConvolutional Neural NetworkVideo
🎯 What it does: Continuous low-power monitoring of the 'ecstatic display' behavior of Chinstrap penguins in Antarctica using event cameras, and proposing a two-stage temporal action detection framework.
Low-Rank Approximation for Sparse Attention in Multi-Modal LLMs
Lin Song (Tencent AILab), Ying Shan (Tencent AILab)
OptimizationComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelTextMultimodality
🎯 What it does: Low-Rank Approximation Sparse Attention (LoRA-Sparse) is proposed in large-scale language models, which reduces the computational cost of self-attention by compressing the query-key vectors.
Low-Rank Knowledge Decomposition for Medical Foundation Models
Yuhang Zhou (Shanghai Jiao Tong University), Yanfeng Wang (Shanghai Jiao Tong University)
Computational EfficiencyKnowledge DistillationMixture of ExpertsBiomedical DataAlzheimer's Disease
🎯 What it does: A knowledge decomposition framework called LoRKD is proposed, which splits the medical foundation model into multiple low-rank expert modules for generating lightweight and switchable expert models.
Low-Rank Rescaled Vision Transformer Fine-Tuning: A Residual Design Approach
Wei Dong (University of Electronic Science and Technology of China), Yang Yang (Northwestern Polytechnical University)
ClassificationTransformerSupervised Fine-TuningImage
🎯 What it does: A residual-based low-rank reparameterization (RLRR) strategy is proposed for parameter-efficient fine-tuning of pre-trained Vision Transformers.
Low-Res Leads the Way: Improving Generalization for Super-Resolution by Self-Supervised Learning
Haoyu Chen (Hong Kong University of Science and Technology), Lei Zhu (Hong Kong University of Science and Technology)
RestorationSuper ResolutionGenerative Adversarial NetworkImage
🎯 What it does: The 'Low-Res Leads the Way (LWay)' training framework is proposed, which combines supervised pre-training on synthetic data and self-supervised fine-tuning on real test images to enhance the adaptability of super-resolution models to real-world degradation.
Low-Resource Vision Challenges for Foundation Models
Yunhua Zhang (Leiden University), Cees G. M. Snoek (University of Amsterdam)
ClassificationRetrievalDiffusion modelImageBenchmark
🎯 What it does: This paper addresses low-resource scenarios in computer vision by constructing the LITE benchmark and studying three major challenges: data scarcity, fine-grained differences, and domain shift; it proposes three targeted baselines to improve the adaptation of visual foundation models in low-resource tasks and validates their effectiveness on three low-resource tasks.
LowRankOcc: Tensor Decomposition and Low-Rank Recovery for Vision-based 3D Semantic Occupancy Prediction
Linqing Zhao (Tsinghua University), Jiwen Lu (Tsinghua University)
SegmentationAutonomous DrivingConvolutional Neural NetworkTransformerSupervised Fine-TuningImagePoint Cloud
🎯 What it does: Proposes the LowRankOcc method, which utilizes tensor decomposition and low-rank recovery to achieve visual-based 3D semantic occupancy prediction.
LP++: A Surprisingly Strong Linear Probe for Few-Shot CLIP
Yunshi Huang (Ecole de Technologie Superieure), Ismail Ben Ayed (Ecole de Technologie Superieure)
ClassificationOptimizationComputational EfficiencyTransformerVision Language ModelContrastive LearningImage
🎯 What it does: An improved linear probe LP++ is proposed, which adapts to few-shot CLIP by learning a linear combination of visual prototypes and text embeddings.
LPSNet: End-to-End Human Pose and Shape Estimation with Lensless Imaging
Haoyang Ge (Tianjin University), Kun Li (Nanjing University)
Pose EstimationImage
🎯 What it does: This paper proposes an end-to-end LPSNet framework that can directly estimate 3D human posture and shape from lensless measurement data.
LQMFormer: Language-aware Query Mask Transformer for Referring Image Segmentation
Nisarg A. Shah (Johns Hopkins University), Vishal M. Patel (Johns Hopkins University)
Object DetectionSegmentationTransformerImageMultimodality
🎯 What it does: This paper proposes LQMFormer, which addresses the query collapse problem in Referring Image Segmentation by achieving high-quality predictions of a single mask through multimodal fusion and dynamic query generation.
LSK3DNet: Towards Effective and Efficient 3D Perception with Large Sparse Kernels
Tuo Feng (University of Technology Sydney), Yi Yang (Zhejiang University)
Object DetectionSegmentationAutonomous DrivingComputational EfficiencyConvolutional Neural NetworkSupervised Fine-TuningPoint Cloud
🎯 What it does: A 3D perception network LSK3DNet based on sparse large kernels is designed and implemented, utilizing Spatial Dynamic Sparsity (SDS) and Channel Weight Selection (CWS) to achieve efficient semantic segmentation and detection.
LTA-PCS: Learnable Task-Agnostic Point Cloud Sampling
Jiaheng Liu (Beihang University), Jinyang Guo (Institute of Automation, Chinese Academy of Sciences)
ClassificationRetrievalPoint Cloud
🎯 What it does: This paper proposes a learnable, task-agnostic point cloud sampling framework LTA-PCS, aimed at preserving the geometric and semantic information of point clouds without relying on task labels.
LTGC: Long-tail Recognition via Leveraging LLMs-driven Generated Content
Qihao Zhao (Beijing University of Chemical Technology), Jun Liu (Northwestern Polytechnical University)
RecognitionData SynthesisDomain AdaptationTransformerLarge Language ModelSupervised Fine-TuningImage
🎯 What it does: A long-tail recognition framework LTGC based on large model-generated content is proposed, utilizing LLM, LMM, and T2I to generate diverse data for tail classes, and fine-tuning the visual encoder with BalanceMix.
LTM: Lightweight Textured Mesh Extraction and Refinement of Large Unbounded Scenes for Efficient Storage and Real-time Rendering
Jaehoon Choi (University of Maryland), Johannes Kopf (Meta)
CompressionOptimizationComputational EfficiencyNeural Radiance FieldMesh
🎯 What it does: This paper proposes a method to extract a lightweight texture mesh from a large-scale unbounded scene neural SDF model and perform joint optimization of geometry and appearance through differential rendering;
LucidDreamer: Towards High-Fidelity Text-to-3D Generation via Interval Score Matching
Yixun Liang (Hong Kong University of Science and Technology), Yingcong Chen (Hong Kong University of Science and Technology)
GenerationData SynthesisDiffusion modelGaussian SplattingTextPoint Cloud
🎯 What it does: Proposes the LucidDreamer framework for generating 3D assets from text, utilizing Interval Score Matching (ISM) to replace Score Distillation Sampling (SDS), and combining it with 3D Gaussian Splatting to generate high-fidelity 3D assets.
LUWA Dataset: Learning Lithic Use-Wear Analysis on Microscopic Images
Jing Zhang (New York University), Chen Feng (New York University)
ClassificationRecognitionConvolutional Neural NetworkTransformerSupervised Fine-TuningImageMultimodality
🎯 What it does: The first and largest-scale micro-image dataset for lithic use wear analysis (LUWA) has been constructed and made publicly available, containing 23,130 images at different magnifications and perceptual modalities, and a systematic evaluation of various visual models has been conducted based on this dataset.
M&M VTO: Multi-Garment Virtual Try-On and Editing
Luyang Zhu (Google Research), Ira Kemelmacher-Shlizerman (University of Washington)
GenerationData SynthesisDiffusion modelImageText
🎯 What it does: A multi-clothing virtual fitting and editing system M&M VTO is proposed, which allows trying on multiple garments on a single portrait and adjusting the clothing layout based on textual descriptions.
M3-UDA: A New Benchmark for Unsupervised Domain Adaptive Fetal Cardiac Structure Detection
Bin Pu (Hong Kong University of Science and Technology), Xiaomeng Li (Hong Kong University of Science and Technology)
Object DetectionDomain AdaptationConvolutional Neural NetworkGraph Neural NetworkImageBiomedical DataUltrasoundBenchmark
🎯 What it does: This paper proposes a new fetal heart structure detection framework called M3-UDA and constructs a cross-center FCS dataset to address the challenge of fetal ultrasound heart structure detection under unsupervised domain adaptation.
MA-LMM: Memory-Augmented Large Multimodal Model for Long-Term Video Understanding
Bo He (University of Maryland), Ser-Nam Lim (University of Central Florida)
ClassificationRecognitionRetrievalCompressionTransformerLarge Language ModelVision Language ModelVideoTextMultimodality
🎯 What it does: MA-LMM is proposed, which enhances the performance of multimodal large models in long video understanding through online frame-by-frame processing and the use of a long-term memory bank.
MACE: Mass Concept Erasure in Diffusion Models
Shilin Lu (Nanyang Technological University), Adams Wai-Kin Kong (Nanyang Technological University)
GenerationData SynthesisDiffusion modelImageText
🎯 What it does: The MACE framework is proposed to achieve large-scale concept erasure in text-to-image diffusion models, capable of erasing up to 100 concepts at once.
MADTP: Multimodal Alignment-Guided Dynamic Token Pruning for Accelerating Vision-Language Transformer
Jianjian Cao (Fudan University), Tao Chen (Fudan University)
RetrievalCompressionComputational EfficiencyTransformerVision Language ModelImageMultimodality
🎯 What it does: Designed and implemented the MADTP framework, which performs dynamic token pruning guided by multi-modal alignment for the visual-language Transformer, significantly reducing the model's computational load and GFLOPs.
MAFA: Managing False Negatives for Vision-Language Pre-training
Jaeseok Byun (Seoul National University), Taesup Moon (Seoul National University)
RecognitionRetrievalTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: This paper addresses the issue of false negative samples caused by the many-to-many correspondence between images and texts in visual-language pre-training. It proposes the MAFA method, which utilizes Efficient Connection Mining (ECM) to convert false negative samples into positive samples, and incorporates label smoothing (S-ITC) into the contrastive loss to mitigate the negative impact of false negatives on learning.
MaGGIe: Masked Guided Gradual Human Instance Matting
Chuong Huynh (University of Maryland), Joon-Young Lee (Adobe)
Object DetectionSegmentationTransformerImageVideo
🎯 What it does: This paper proposes an instance-guided progressive human instance masking network named MaGGIe, which can predict multiple instance alpha mattes in images and videos in one go.
Magic Tokens: Select Diverse Tokens for Multi-modal Object Re-Identification
Pingping Zhang (Dalian University of Technology), Huchuan Lu (Dalian University of Technology)
RecognitionRetrievalTransformerImageMultimodality
🎯 What it does: The EDITOR framework is proposed, which significantly enhances the feature representation and robustness of multi-modal object ReID through spatial-frequency token selection and hierarchical mask aggregation on Vision Transformer.
MagicAnimate: Temporally Consistent Human Image Animation using Diffusion Model
Zhongcong Xu (National University of Singapore), Mike Zheng Shou
GenerationData SynthesisDiffusion modelImageVideo
🎯 What it does: This paper presents MagicAnimate, a portrait animation framework based on diffusion models that can generate cross-frame consistent, identity-preserving, and high-quality videos.
MAGICK: A Large-scale Captioned Dataset from Matting Generated Images using Chroma Keying
Ryan D. Burgert (Stony Brook University), Michael S. Ryoo (Stony Brook University)
SegmentationGenerationData SynthesisDiffusion modelImageStochastic Differential Equation
🎯 What it does: This paper presents the MAGICK dataset, which contains 150,000 generated object images along with their high-quality alpha masks, and demonstrates the use of this dataset to train a control network for alpha-to-RGB generation.
Make Me a BNN: A Simple Strategy for Estimating Bayesian Uncertainty from Pre-trained Models
Gianni Franchi (Institut Polytechnique de Paris), Angela Yao (National University of Singapore)
ClassificationAnomaly DetectionImage
🎯 What it does: This paper proposes ABNN, which transforms a pre-trained DNN into a network capable of producing Bayesian posteriors by inserting Bayesian noise into the normalization layers and performing a small amount of fine-tuning, achieving posterior uncertainty estimation.
Make Pixels Dance: High-Dynamic Video Generation
Yan Zeng (ByteDance Research), Hang Li (ByteDance Research)
GenerationData SynthesisDiffusion modelVideoText
🎯 What it does: Introducing PixelDance, a diffusion model for video generation that utilizes text, the first frame image, and the last frame image as conditions.
Make-It-Vivid: Dressing Your Animatable Biped Cartoon Characters from Text
Junshu Tang (Shanghai Jiao Tong University), Lizhuang Ma (Shanghai Jiao Tong University)
GenerationData SynthesisLarge Language ModelDiffusion modelGenerative Adversarial NetworkImageTextMesh
🎯 What it does: Proposes the Make-It-Vivid framework for automatic generation and animation of UV textures for 3D biped cartoon characters based on text instructions.
Make-Your-Anchor: A Diffusion-based 2D Avatar Generation Framework
Ziyao Huang (Institute of Computing Technology Chinese Academy of Sciences), Tong-Yee Lee (National Cheng Kung University)
GenerationData SynthesisPose EstimationDiffusion modelVideoMesh
🎯 What it does: A 2D human anchor video generation framework based on diffusion models, called 'Make-Your-Anchor', has been developed, which can train realistic anchor videos with complete body movements and facial expressions using just one minute of video.
Makeup Prior Models for 3D Facial Makeup Estimation and Applications
Xingchao Yang (CyberAgent), Yoshihiro Kanamori (University of Tsukuba)
RestorationGenerationGenerative Adversarial NetworkImage
🎯 What it does: Two prior models for makeup are proposed (PCA linear model and StyleGAN2 generative model), and corresponding estimation networks are constructed to achieve robust estimation of 3D facial makeup and its applications in reconstruction, editing, transfer, and interpolation.
Making Vision Transformers Truly Shift-Equivariant
Renan A. Rojas-Gomez (University of Illinois at Urbana-Champaign), Raymond A. Yeh (Purdue University)
ClassificationSegmentationTransformerImage
🎯 What it does: A Vision Transformer module that achieves complete cyclic translation invariance (adaptive patching, attention, image merging, relative position encoding) is proposed, constructing a truly cyclic translation invariant ViT.
Making Visual Sense of Oracle Bones for You and Me
Runqi Qiao (Beijing University of Posts and Telecommunications), Honggang Zhang (Beijing University of Posts and Telecommunications)
GenerationData SynthesisTransformerLarge Language ModelVision Language ModelDiffusion modelImageText
🎯 What it does: This paper proposes a visual guide system based on generative AI to help the public understand the relationship between oracle bone characters and their semantics, and designs a quantitative evaluation metric called TransOV.
ManiFPT: Defining and Analyzing Fingerprints of Generative Models
Hae Jin Song (University of Southern California), Wael AbdAlmageed (Clemson University)
RecognitionGenerationData SynthesisConvolutional Neural NetworkScore-based ModelFlow-based ModelGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes and formally defines the 'artifact' (image-level trace) and 'fingerprint' (model-level trace) of generative models, and provides algorithms for estimating them from limited samples; subsequently, it achieves multi-class attribution of generative models through these traces, obtaining high recognition accuracy across multiple datasets.
ManipLLM: Embodied Multimodal Large Language Model for Object-Centric Robotic Manipulation
Xiaoqi Li (Peking University), Hao Dong (Peking University)
Robotic IntelligenceTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodalityChain-of-Thought
🎯 What it does: By fine-tuning the adapters of multimodal large language models (MLLMs), it is enabled to predict object contact points, hand claw directions, and perform robot grasping and manipulation at the object center from RGB images and text prompts.
MANUS: Markerless Grasp Capture using Articulated 3D Gaussians
Chandradeep Pokhariya (Indian Institute of Information Technology Hyderabad), Srinath Sridhar (Brown University)
Pose EstimationRobotic IntelligenceGaussian SplattingVideoPoint Cloud
🎯 What it does: The MANUS method is proposed, which uses a deformable 3D Gaussian representation to achieve markerless hand-object interaction capture.
Map-Relative Pose Regression for Visual Re-Localization
Shuai Chen (Niantic), Eric Brachmann
Pose EstimationConvolutional Neural NetworkTransformerImage
🎯 What it does: A visual re-localization method that combines scene-specific map representation with a general Transformer pose regression network is proposed.
MAP: MAsk-Pruning for Source-Free Model Intellectual Property Protection
Boyang Peng (Tongji University), Changjun Jiang (Tongji University)
ClassificationDomain AdaptationSafty and PrivacyKnowledge DistillationConvolutional Neural NetworkTransformerImage
🎯 What it does: A framework called MAP (Mask-Pruning) is proposed to achieve model intellectual property protection in source-free and data-free scenarios by learning binary masks to prune the model, thereby weakening its generalization ability to unauthorized domains.
MAPLM: A Real-World Large-Scale Vision-Language Benchmark for Map and Traffic Scene Understanding
Xu Cao (University of Illinois Urbana-Champaign), Chao Zheng (SambaNova Systems)
Autonomous DrivingTransformerLarge Language ModelVision Language ModelContrastive LearningImageTextMultimodalityPoint CloudBenchmark
🎯 What it does: This paper constructs the MAPLM, a large-scale real traffic scene visual-language dataset, which includes panoramic 2D images, 3D LiDAR point clouds, and high-precision map texts, and releases the MAPLM-QA question-answering benchmark on it;
MAPSeg: Unified Unsupervised Domain Adaptation for Heterogeneous Medical Image Segmentation Based on 3D Masked Autoencoding and Pseudo-Labeling
Xuzhe Zhang (Columbia University), Yun Wang (Duke University)
SegmentationDomain AdaptationAuto EncoderImageBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: This paper proposes a unified unsupervised domain adaptation framework MAPSeg, capable of voxel-level segmentation under different medical imaging domain shifts (cross-sequence, cross-site, cross-age, cross-modality).
MarkovGen: Structured Prediction for Efficient Text-to-Image Generation
Sadeep Jayasumana (Google Research), Sanjiv Kumar (Google Research)
GenerationData SynthesisTransformerGenerative Adversarial NetworkImageText
🎯 What it does: Utilize a lightweight Markov Random Field (MRF) model to replace the later sampling steps of Muse, achieving high-quality text-to-image generation.
MART: Masked Affective RepresenTation Learning via Masked Temporal Distribution Distillation
Zhicheng Zhang (Nankai University), Jufeng Yang (Nankai University)
Representation LearningContrastive LearningVideo
🎯 What it does: A mask-based self-supervised learning framework called MART is proposed, which extracts and verifies emotional cues through an emotional dictionary, generates emotion-guided occlusions, and restores temporal emotional distributions to enhance video emotion analysis performance.
MAS: Multi-view Ancestral Sampling for 3D Motion Generation Using 2D Diffusion
Roy Kapon (Tel Aviv University), Amit H. Bermano (Tel Aviv University)
GenerationData SynthesisPose EstimationTransformerDiffusion modelVideo
🎯 What it does: This paper proposes Multi-view Ancestral Sampling (MAS), which generates a complete three-dimensional motion sequence by performing ancestral sampling on multiple 2D motion sequences from a trained 2D diffusion model simultaneously, ensuring multi-view consistency at each step through triangulation and reprojection.
Mask Grounding for Referring Image Segmentation
Yong Xien Chng (Tsinghua University), Gao Huang (Tsinghua University)
Object DetectionSegmentationTransformerVision Language ModelImageText
🎯 What it does: Proposes a Mask Grounding auxiliary task and constructs the MagNet network to enhance fine-grained visual-text correspondence in referring image segmentation (RIS).
Mask4Align: Aligned Entity Prompting with Color Masks for Multi-Entity Localization Problems
Haoquan Zhang (South China University of Technology), Huaidong Zhang (South China University of Technology)
Object DetectionSegmentationTransformerPrompt EngineeringVision Language ModelImageMultimodality
🎯 What it does: This paper studies the multi-entity localization problem in Visual Question Answering (VQA) and proposes the Mask4Align method, which overlays colored masks on images and generates corresponding colored candidate answers, utilizing a pre-trained visual language model to achieve unsupervised multi-entity localization.
MaskCLR: Attention-Guided Contrastive Learning for Robust Action Representation Learning
Mohamed Abdelfattah (Ecole Polytechnique Federale de Lausanne), Alexandre Alahi (Ecole Polytechnique Federale de Lausanne)
RecognitionRepresentation LearningTransformerContrastive LearningVideo
🎯 What it does: The MaskCLR method is proposed, which enhances the robustness and generalization performance of skeleton action recognition through attention-guided probabilistic masking and multi-layer contrastive learning.
MaskClustering: View Consensus based Mask Graph Clustering for Open-Vocabulary 3D Instance Segmentation
Mi Yan (Peking University), He Wang (Peking University)
Object DetectionSegmentationGraph Neural NetworkVision Language ModelPoint Cloud
🎯 What it does: A global mask graph clustering method based on multi-view consistency is proposed, which utilizes the consensus relationships of 2D masks across multiple perspectives to aggregate them into 3D instances and generate open vocabulary semantic features for each instance, with the entire process requiring no training.
Masked and Shuffled Blind Spot Denoising for Real-World Images
Hamadi Chihaoui (University of Bern), Paolo Favaro (University of Bern)
RestorationImage
🎯 What it does: This paper proposes a denoising method based on single-image self-supervision called MASH, which combines blind spot networks, dynamic occlusion ratios, and local pixel shuffling to denoise real images without the need for clean images or training sets.
Masked AutoDecoder is Effective Multi-Task Vision Generalist
Han Qiu (Nanyang Technological University), Shijian Lu (Nanyang Technological University)
Object DetectionSegmentationGenerationPose EstimationTransformerAuto EncoderImageMultimodality
🎯 What it does: A parallel decoding framework based on Masked AutoDecoder is designed, capable of uniformly handling various visual tasks such as object detection, instance segmentation, keypoint detection, and image captioning.
Masked Autoencoders for Microscopy are Scalable Learners of Cellular Biology
Oren Kraus (Recursion), Berton Earnshaw (Recursion)
Representation LearningTransformerAuto EncoderImageBiomedical Data
🎯 What it does: Train a self-supervised Masked Autoencoder (MAE) to learn cellular phenotype representations from millions of microscopy images, and evaluate its effectiveness in biological relationship inference and morphological feature prediction.
Masked Spatial Propagation Network for Sparsity-Adaptive Depth Refinement
Jinyoung Jun (Korea University), Chang-Su Kim (Korea University)
RestorationDepth EstimationTransformerImage
🎯 What it does: A Sparse Depth Refinement (SDR) framework is proposed, which uses a Masked Spatial Propagation Network (MSPN) to progressively propagate sparse depth information, completing the refinement and filling of monocular depth maps.
MaskINT: Video Editing via Interpolative Non-autoregressive Masked Transformers
Haoyu Ma (University of California, Irvine), Xiaohui Xie (Meta)
GenerationData SynthesisTransformerDiffusion modelVideo
🎯 What it does: This paper presents MaskINT, a two-stage video editing framework that first uses a pre-trained text-image diffusion model for zero-shot editing of keyframes, and then employs a non-autoregressive masked generative Transformer for structure-guided frame interpolation.
MaskPLAN: Masked Generative Layout Planning from Partial Input
Hang Zhang (ETH Zurich), Benjamin Dillenburger (ETH Zurich)
GenerationData SynthesisGraph Neural NetworkTransformerAuto EncoderImage
🎯 What it does: MaskPLAN proposes a generative layout planning model based on a graph-structured Dynamic Masked Autoencoder (GDMAE), which can automatically complete a full floor plan based on partial user input and supports multi-attribute interaction.
Matching 2D Images in 3D: Metric Relative Pose from Metric Correspondences
Axel Barroso-Laguna (Niantic), Eric Brachmann (University of Oxford)
Pose EstimationDepth EstimationConvolutional Neural NetworkReinforcement LearningImage
🎯 What it does: MicKey is an end-to-end neural network that predicts 3D metric keypoints and their descriptors from a single image, enabling the matching of 2D images in 3D camera space and recovering the scale-accurate relative pose of the camera.
Matching Anything by Segmenting Anything
Siyuan Li (ETH Zurich INSAIT), Fisher Yu (ETH Zurich INSAIT)
Object DetectionObject TrackingSegmentationContrastive LearningImageVideo
🎯 What it does: On unsupervised static images, dense instance masks are generated using the Segment Anything Model (SAM), and two views are produced through strong data augmentation. Instance-level association features are learned using contrastive learning, and then this feature mapper (MASA Adapter) is injected into any detection/separation base model, enabling zero-shot multi-object tracking in any domain.
MatchU: Matching Unseen Objects for 6D Pose Estimation from RGB-D Images
Junwen Huang (Technical University of Munich), Benjamin Busam (Technical University of Munich)
Pose EstimationTransformerImageMultimodality
🎯 What it does: The MatchU framework is proposed, utilizing RGB-D fusion and rotation-invariant 3D descriptors to achieve 6D pose estimation on unseen objects.
Material Palette: Extraction of Materials from a Single Image
Ivan Lopes (Inria), Raoul de Charette (Inria)
GenerationDomain AdaptationConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: Extract material textures and SVBRDF information usable for PBR rendering from a single real-world image.
MatFuse: Controllable Material Generation with Diffusion Models
Giuseppe Vecchio (University of Catania), Concetto Spampinato (University of Catania)
GenerationData SynthesisDiffusion modelGenerative Adversarial NetworkImageMultimodality
🎯 What it does: A material generation framework called MatFuse based on diffusion models is proposed, which can generate high-quality SVBRDF materials through multimodal conditions such as text, images, color palettes, and sketches, and supports material editing.
MatSynth: A Modern PBR Materials Dataset
Giuseppe Vecchio (University of Catania), Valentin Deschaintre (Adobe Research)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: A large-scale, high-quality 4K PBR material dataset called MatSynth has been created and used to evaluate material capture and generation methods.
MaxQ: Multi-Axis Query for N:M Sparsity Network
Jingyang Xiang (Zhejiang University), Yong Liu (Zhejiang University)
ClassificationObject DetectionSegmentationConvolutional Neural NetworkImage
🎯 What it does: A dynamic soft masking method called MaxQ based on multi-axis queries is proposed to identify and enhance important weights in N:M sparse networks during the training phase.
MCD: Diverse Large-Scale Multi-Campus Dataset for Robot Perception
Thien-Minh Nguyen (Nanyang Technological University), Noel Blunder (Technische Universität Hamburg)
SegmentationAutonomous DrivingOptimizationRobotic IntelligenceSimultaneous Localization and MappingMultimodalityPoint CloudBenchmark
🎯 What it does: A multimodal large-scale dataset MCD has been proposed and released across three universities in Eurasia, covering traditional and low-cost NRE LiDAR, IMU, cameras, ultra-wideband, and other sensors, and for the first time providing 29 classes of semantic annotations and continuous temporal high-precision ground truth for NRE point clouds.
MCNet: Rethinking the Core Ingredients for Accurate and Efficient Homography Estimation
Haokai Zhu (Zhejiang University), Hui-Liang Shen (Zhejiang University)
OptimizationComputational EfficiencyConvolutional Neural NetworkImageVideoMultimodality
🎯 What it does: A multi-scale correlation search network (MCNet) is proposed for high-precision and efficient single-frame homography estimation.
MCPNet: An Interpretable Classifier via Multi-Level Concept Prototypes
Bor-Shiun Wang (National Yang Ming Chiao Tung University), Wei-Chen Chiu (National Yang Ming Chiao Tung University)
ClassificationExplainability and InterpretabilityContrastive LearningImage
🎯 What it does: MCPNet is proposed, a model for interpretable classification achieved through multi-layer concept prototypes;
MeaCap: Memory-Augmented Zero-shot Image Captioning
Zequn Zeng (Xidian University), Zhengjue Wang (Xidian University)
GenerationRetrievalTransformerVision Language ModelImageTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes a memory-enhanced zero-shot image description framework called MeaCap, which retrieves and filters key concepts from external text memory, and then generates descriptions through a keyword-to-sentence language model, supporting both training-free and text-only training modes.
Mean-Shift Feature Transformer
Takumi Kobayashi (National Institute of Advanced Industrial Science and Technology)
ClassificationSegmentationTransformerGaussian SplattingImage
🎯 What it does: A feature transformation module (MSF-transformer) based on mean-shift updates is proposed to replace the existing Transformer module; PROBE projection and efficient grouped projection are introduced to further compress parameters.
MedBN: Robust Test-Time Adaptation against Malicious Test Samples
Hyejin Park (Pohang University of Science and Technology), Jungseul Ok (Pohang University of Science and Technology)
ClassificationSegmentationDomain AdaptationAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: This paper addresses the vulnerability of Test-Time Adaptation (TTA) methods when subjected to adversarial sample attacks, proposing a Median-based Batch Normalization (MedBN) that is seamlessly integrated into various existing TTA frameworks to enhance robustness against data poisoning attacks.
MedM2G: Unifying Medical Multi-Modal Generation via Cross-Guided Diffusion with Visual Invariant
Chenlu Zhan (Zhejiang University), Jian Wu (Zhejiang University)
GenerationData SynthesisDiffusion modelContrastive LearningMultimodalityBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: A unified medical multimodal generation framework MedM2G is proposed, capable of performing various cross-modal generation tasks such as text↔image, CT↔MRI, and CT↔X-ray within the same model.
MeLFusion: Synthesizing Music from Image and Language Cues using Diffusion Models
Sanjoy Chowdhury (University of Maryland), Dinesh Manocha (University of Maryland)
GenerationData SynthesisDiffusion modelImageTextMultimodalityBenchmarkAudio
🎯 What it does: Designed and implemented a latent diffusion model MELFUSION that simultaneously utilizes image and text conditions to generate music that conforms to visual contexts, proposing a visual synapse mechanism;
MemFlow: Optical Flow Estimation and Prediction with Memory
Qiaole Dong (Fudan University), Yanwei Fu (Fudan University)
Recurrent Neural NetworkTransformerOptical FlowVideo
🎯 What it does: This paper proposes a real-time optical flow estimation and prediction framework called MemFlow, which utilizes a memory module to aggregate historical motion information and updates it in real-time for each frame.
MemoNav: Working Memory Model for Visual Navigation
Hongxin Li (University of Chinese Academy of Sciences), Zhaoxiang Zhang (University of Chinese Academy of Sciences)
Graph Neural NetworkTransformerReinforcement LearningImage
🎯 What it does: This paper proposes MemoNav, a working memory model for image target navigation that combines short-term memory (STM), long-term memory (LTM), and working memory (WM) in three scenarios.
Memory-based Adapters for Online 3D Scene Perception
Xiuwei Xu (Tsinghua University), Jiwen Lu (Tsinghua University)
Object DetectionSegmentationConvolutional Neural NetworkVideoPoint Cloud
🎯 What it does: A memory-based adapter framework is proposed to convert existing offline 3D scene perception models into models that can process RGB-D video streams in real-time online.
Memory-Scalable and Simplified Functional Map Learning
Robin Magnet (Polytechnic Institute of Paris), Maks Ovsjanikov (Polytechnic Institute of Paris)
Diffusion modelPoint CloudMeshBenchmark
🎯 What it does: The paper proposes a memory-scalable and simplified functional mapping learning framework that can directly use soft point mappings during training without storing dense matrices.
MemSAM: Taming Segment Anything Model for Echocardiography Video Segmentation
Xiaolong Deng (Shenzhen University), Jing Qin (Hong Kong Polytechnic University)
SegmentationConvolutional Neural NetworkVideoBiomedical DataUltrasound
🎯 What it does: This paper proposes a heart ultrasound video segmentation framework called MemSAM based on the Segment Anything Model.
MESA: Matching Everything by Segmenting Anything
Yesheng Zhang (Shanghai Jiao Tong University), Xu Zhao (Shanghai Jiao Tong University)
Object DetectionSegmentationPose EstimationConvolutional Neural NetworkGraph Neural NetworkSupervised Fine-TuningContrastive LearningImage
🎯 What it does: This paper proposes the MESA method, which constructs an Area Graph using the segmentation results from SAM and solves precise area matching through a graph model to reduce redundant calculations in feature matching, thereby improving the accuracy of subsequent point matching.
MeshGPT: Generating Triangle Meshes with Decoder-Only Transformers
Yawar Siddiqui (Technical University of Munich), Matthias Nießner (Technical University of Munich)
GenerationData SynthesisTransformerSupervised Fine-TuningMesh
🎯 What it does: Automatically generate triangular meshes by autoregressively generating 3D models in the form of triangle sequences directly through a Transformer model;
MeshPose: Unifying DensePose and 3D Body Mesh Reconstruction
Eric-Tuan Le, Iasonas Kokkinos (Snap Inc.)
Pose EstimationMesh
🎯 What it does: The MeshPose method is proposed, combining DensePose with human mesh recovery to simultaneously optimize pixel-level reprojection and 3D mesh quality.
Meta-Point Learning and Refining for Category-Agnostic Pose Estimation
Junjie Chen (Jiangxi University of Finance and Economics), Li Niu (Shanghai Jiao Tong University)
Pose EstimationMeta LearningTransformerImage
🎯 What it does: A category-independent pose estimation framework based on meta-point learning and refinement is proposed, capable of predicting key points of any category with only a small number of labeled support images.
MetaCloak: Preventing Unauthorized Subject-driven Text-to-image Diffusion-based Synthesis via Meta-learning
Yixin Liu (Lehigh University), Lichao Sun (Lehigh University)
GenerationData SynthesisSafty and PrivacyAdversarial AttackMeta LearningDiffusion modelImage
🎯 What it does: Proposes the MetaCloak method, which prevents unauthorized personalized text-to-image diffusion model generation by adding robust adversarial perturbations to user photos;
MFP: Making Full Use of Probability Maps for Interactive Image Segmentation
Chaewon Lee, Chang-Su Kim
RecognitionSegmentationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a new computer vision method aimed at improving the accuracy of image recognition.