CVPR 2025 Papers — Page 17
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2871 papers
MIMO: A Medical Vision Language Model with Visual Referring Multimodal Input and Pixel Grounding Multimodal Output
Yanyuan Chen (Peking University), Hang Li (Peking University)
RecognitionSegmentationTransformerLarge Language ModelVision Language ModelImageTextMultimodalityMagnetic Resonance Imaging
🎯 What it does: A MIMO model is proposed, achieving a unified framework for visual prompt input and pixel-level segmentation output in medical imaging for the first time.
MIMO: Controllable Character Video Synthesis with Spatial Decomposed Modeling
Yifang Men (Alibaba Group), Liefeng Bo (Alibaba Group)
Object DetectionObject TrackingGenerationData SynthesisPose EstimationDepth EstimationDiffusion modelVideo
🎯 What it does: A controllable character video synthesis framework MIMO based on spatial decomposition modeling is proposed, which can generate high-quality animated videos with controllable appearance, actions, and scenes using character images, pose sequences, or scene videos provided by users.
Mind the Gap: Confidence Discrepancy Can Guide Federated Semi-Supervised Learning Across Pseudo-Mismatch
Yijie Liu (Xiamen University), Hanzi Wang (Xiamen University)
Federated LearningContrastive LearningImage
🎯 What it does: This study investigates the issue of mismatched pseudo-labels caused by data heterogeneity in federated semi-supervised learning and proposes the SAGE method, which enhances model performance through adaptive pseudo-label correction based on confidence differences.
Mind the Gap: Detecting Black-box Adversarial Attacks in the Making through Query Update Analysis
Jeonghwan Park (Queens University Belfast), Ihsen Alouani (Queens University Belfast)
Anomaly DetectionAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a GWAD framework based on query update similarity (Delta Similarity) for real-time detection of black-box adversarial attacks.
Mind the Time: Temporally-Controlled Multi-Event Video Generation
Ziyi Wu (Snap Research), Sergey Tulyakov (Snap Research)
GenerationData SynthesisTransformerDiffusion modelVideoText
🎯 What it does: This paper presents MinT, a model capable of controlling multi-event video generation based on timestamps.
Mind the Trojan Horse: Image Prompt Adapter Enabling Scalable and Deceptive Jailbreaking
Junxi Chen (Sun Yat-Sen University), Xiaohua Xie (Sun Yat-Sen University)
GenerationAdversarial AttackTransformerPrompt EngineeringDiffusion modelImageText
🎯 What it does: This paper systematically studies and verifies the feasibility of 'hijacking attacks' on text-to-image diffusion models based on IP-Adapter using image prompts, revealing the potential harm of such attacks to the reputation of service providers and user experience.
Minding Fuzzy Regions: A Data-driven Alternating Learning Paradigm for Stable Lesion Segmentation
Lexin Fang (Shandong University), Caiming Zhang (Shandong University)
SegmentationAuto EncoderImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper proposes a data-driven alternating learning (DALE) framework to enhance the stability and accuracy of medical image lesion segmentation in ambiguous regions.
MINIMA: Modality Invariant Image Matching
Jiangwei Ren (Huazhong University of Science and Technology), Xiang Bai (Wuhan University)
RecognitionData SynthesisDomain AdaptationImageMultimodality
🎯 What it does: By constructing a unified cross-modal image matching framework MINIMA and utilizing a large data engine to automatically generate multi-modal paired data from RGB images, the 'modal gap' problem in cross-modal matching is addressed;
Minimal Interaction Seperated Tuning: A New Paradigm for Visual Adaptation
Ningyuan Tang (Nanjing University), Jianxin Wu (Nanjing University)
ClassificationRecognitionDomain AdaptationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper proposes a visual adaptation paradigm called 'Separated Tuning': running large-scale pre-trained models in the cloud and only transmitting compressed intermediate features to low-power devices, which then locally train a small side network to achieve task transfer.
Minimizing Labeled, Maximizing Unlabeled: An Image-Driven Approach for Video Instance Segmentation
Fangyun Wei (University of Sydney), Chang Xu (Peking University)
Object DetectionSegmentationTransformerImageVideoRetrieval-Augmented Generation
🎯 What it does: A video instance segmentation framework named MinMaxVIS is proposed, which can be trained with only a small number of labeled images and a large amount of unlabeled images.
Minority-Focused Text-to-Image Generation via Prompt Optimization
Soobin Um (KAIST), Jong Chul Ye (KAIST)
GenerationData SynthesisOptimizationPrompt EngineeringDiffusion modelImageText
🎯 What it does: This paper proposes a MinorityPrompt framework based on online prompt optimization to actively generate low-density (minority group) samples in text-to-image generation.
MIRE: Matched Implicit Neural Representations
Dhananjaya Jayasundara (Johns Hopkins University), Vishal M. Patel (Johns Hopkins University)
RestorationSuper ResolutionConvolutional Neural NetworkNeural Radiance FieldImage
🎯 What it does: This paper studies how to dynamically match activation functions for each input signal in implicit neural representations (INR), constructing a network structure that aligns with the characteristics of the signal.
MirrorVerse: Pushing Diffusion Models to Realistically Reflect the World
Ankit Dhiman (Indian Institute of Science), R Venkatesh Babu (Indian Institute of Science)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: Trained and deployed a diffusion model-based specular reflection generator, MirrorFusion 2.0, specifically for high-quality specular reflection synthesis.
Missing Target-Relevant Information Prediction with World Model for Accurate Zero-Shot Composed Image Retrieval
Yuanmin Tang (Institute of Information Engineering Chinese Academy of Sciences), Qi Wu (University of Adelaide)
RetrievalContrastive LearningWorld ModelImage
🎯 What it does: A prediction-based image-to-word mapping network called PrediCIR is proposed, which is based on a world model to predict missing target visual content and map it to pseudo-words in zero-shot synthesized image retrieval.
Mitigating Ambiguities in 3D Classification with Gaussian Splatting
Ruiqi Zhang (Nanjing University), Zhan Ma (Nanjing University)
ClassificationGaussian SplattingPoint Cloud
🎯 What it does: This paper proposes a point cloud representation based on Gaussian Splatting (GS) to address the ambiguity issues in distinguishing between linear and planar surfaces, as well as transparent/reflective objects in traditional point clouds, and applies it to 3D classification tasks.
Mitigating Hallucinations in Large Vision-Language Models via DPO: On-Policy Data Hold the Key
Zhihe Yang (Chinese University of Hong Kong), Dongsheng Li (Chinese University of Hong Kong)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningTextMultimodality
🎯 What it does: A framework named OPA-DPO is proposed, utilizing expert feedback and a policy alignment method to reduce the hallucination phenomenon in large visual language models.
Mitigating Object Hallucinations in Large Vision-Language Models with Assembly of Global and Local Attention
Wenbin An (Xi'an Jiaotong University), Shijian Lu (Nanyang Technological University)
Object DetectionTransformerVision Language ModelImageMultimodality
🎯 What it does: This study investigates the object hallucination problem in large visual language models (LVLM) and proposes a training-free, pluggable decoding method—Assembly of Global and Local Attention (AGLA). This method reduces hallucinations and enhances visual understanding by combining the global features of the original image with the local features of the augmented image.
Mitigating the Human-Robot Domain Discrepancy in Visual Pre-training for Robotic Manipulation
Jiaming Zhou (Hong Kong University of Science and Technology), Junwei Liang (Hong Kong University of Science and Technology)
Domain AdaptationRobotic IntelligenceContrastive LearningVideo
🎯 What it does: This paper proposes a parameter-efficient adaptation method called HR-Align, which utilizes human-robot paired video alignment semantics to alleviate the domain gap of human visual pre-trained models in robot control.
MITracker: Multi-View Integration for Visual Object Tracking
Mengjie Xu (ShanghaiTech University), Qian Wang (ShanghaiTech University)
Object TrackingTransformerVideo
🎯 What it does: A multi-view integrated tracking framework MITracker is proposed, and a large-scale multi-view tracking dataset MVTrack is released, addressing challenges such as occlusion and target loss in traditional single-view tracking.
MixerMDM: Learnable Composition of Human Motion Diffusion Models
Pablo Ruiz-Ponce (Universidad de Alicante), José García-Rodríguez (Universitat de Barcelona)
GenerationData SynthesisTransformerDiffusion modelVideoMultimodality
🎯 What it does: This paper proposes MixerMDM, a learnable model ensemble method that combines two pre-trained text-conditioned human motion diffusion models to generate motion sequences that balance interaction and individual actions without the need for retraining.
Mixture of Submodules for Domain Adaptive Person Search
Minsu Kim (Samsung Electronics), Kwanghoon Sohn (Yonsei University)
RecognitionObject DetectionDomain AdaptationMixture of ExpertsContrastive LearningImage
🎯 What it does: To address the cross-domain person search problem, this paper proposes the MoS framework, which achieves the separation of detection and ReID through the mixing of task- and domain-specific submodules and the generation of dual-domain samples.
MLLM-as-a-Judge for Image Safety without Human Labeling
Zhenting Wang (Meta), Ankit Jain (Meta)
Safty and PrivacyTransformerLarge Language ModelVision Language ModelImageBenchmarkChain-of-Thought
🎯 What it does: By transforming the security constitution into objective rules, conducting image-rule relevance screening, extracting precondition chains, making bias token probability judgments, and employing chain reasoning, the CLUE framework was constructed to achieve unsupervised image security assessment.
MLVU: Benchmarking Multi-task Long Video Understanding
Junjie Zhou (Beijing University of Posts and Telecommunications), Zheng Liu (Beijing University of Posts and Telecommunications)
GenerationTransformerLarge Language ModelVideoMultimodalityBenchmark
🎯 What it does: This paper proposes a new long video understanding benchmark, MLVU, for systematically evaluating the multi-task performance of multimodal large language models (MLLMs) on long videos, and experiments were conducted on 23 of the latest MLLMs.
MM-OR: A Large Multimodal Operating Room Dataset for Semantic Understanding of High-Intensity Surgical Environments
Ege Özsoy (Technical University of Munich), Nassir Navab (Technical University of Munich)
Object DetectionSegmentationGenerationTransformerLarge Language ModelVision Language ModelImageVideoMultimodalityPoint CloudAudio
🎯 What it does: This paper presents the MM-OR multimodal operating room dataset and the MM2SG model for multimodal scene graph generation.
MMAR: Towards Lossless Multi-Modal Auto-Regressive Probabilistic Modeling
Jian Yang (University of Science and Technology of China), Zheng-Jun Zha (University of Science and Technology of China)
GenerationData SynthesisTransformerLarge Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: A multi-modal autoregressive probabilistic model MMAR is proposed, which jointly models continuous image tokens and text tokens to achieve a unified framework for image understanding and generation.
MMAudio: Taming Multimodal Joint Training for High-Quality Video-to-Audio Synthesis
Ho Kei Cheng (University of Illinois Urbana-Champaign), Yuki Mitsufuji (Sony Group Corporation)
GenerationData SynthesisTransformerFlow-based ModelAuto EncoderVideoTextMultimodalityAudio
🎯 What it does: A multi-modal joint training framework (MMAudio) is proposed, capable of synthesizing high-quality, video-synchronized audio given video and optional text conditions.
MMRL: Multi-Modal Representation Learning for Vision-Language Models
Yuncheng Guo (Fudan University), Xiaodong Gu (Fudan University)
Domain AdaptationRepresentation LearningTransformerVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: This paper proposes the MMRL framework, which introduces a shared learnable multimodal representation space on the pre-trained CLIP VLM, and inserts representation subwords in the high-level encoder, achieving efficient adaptation to downstream tasks with a small number of samples while maintaining generality.
MMTL-UniAD: A Unified Framework for Multimodal and Multi-Task Learning in Assistive Driving Perception
Wenzhuo Liu (Beijing Institute of Technology), Huaping Liu (Tsinghua University)
RecognitionAutonomous DrivingTransformerImageVideoMultimodality
🎯 What it does: The MMTL-UniAD framework is proposed, which can simultaneously utilize multimodal perception data (driving perspective images, in-car cameras, pose joints, etc.) for the joint recognition of four tasks: driver emotion, behavior, traffic context, and vehicle behavior.
MMVU: Measuring Expert-Level Multi-Discipline Video Understanding
Yilun Zhao, Arman Cohan
Prompt EngineeringVideoTextMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Designed and released MMVU, a multimodal video question-answering benchmark containing 3,000 expert annotations across 27 disciplines and 4 subject areas.
MNE-SLAM: Multi-Agent Neural SLAM for Mobile Robots
Tianchen Deng (Shanghai Jiao Tong University), Weidong Chen (Shanghai Jiao Tong University)
OptimizationRobotic IntelligenceSimultaneous Localization and MappingOptical FlowPoint CloudMesh
🎯 What it does: This paper proposes MNE-SLAM, a distributed neural SLAM system that supports multi-robot collaboration, capable of performing distributed mapping, camera tracking, loop closure, and subgraph fusion without sharing raw sensor data.
MobileH2R: Learning Generalizable Human to Mobile Robot Handover Exclusively from Scalable and Diverse Synthetic Data
Zifan Wang (Tsinghua University), Li Yi (Tsinghua University)
Robotic IntelligenceDiffusion modelPoint Cloud
🎯 What it does: This paper proposes the MobileH2R framework, which utilizes scalable synthetic full-body motion data and safety-friendly demonstration generation to enable mobile robots to learn human handover tasks in both simulated and real environments.
MobileMamba: Lightweight Multi-Receptive Visual Mamba Network
Haoyang He (Zhejiang University), Lei Xie (Zhejiang University)
ClassificationObject DetectionSegmentationComputational EfficiencyKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: A lightweight visual network called MobileMamba is proposed and implemented, integrating multi-scale perception and high-frequency detail extraction, suitable for downstream tasks such as classification, detection, and segmentation at high resolutions.
MobilePortrait: Real-Time One-Shot Neural Head Avatars on Mobile Devices
Jianwen Jiang (Bytedance Intelligent Creation), Tianyun Zhong (Zhejiang University)
GenerationData SynthesisPose EstimationImageVideoAudio
🎯 What it does: We propose MobilePortrait, a real-time single-image facial animation generation model suitable for mobile devices;
MODA: Motion-Drift Augmentation for Inertial Human Motion Analysis
Yinghao Wu (Xiamen University), Yipeng Qin
RecognitionPose EstimationDomain AdaptationMultimodalityTime Series
🎯 What it does: A method based on IMU drift, called MODA, is proposed and validated to improve the robustness of human action recognition and pose estimation driven by IMU.
MoDec-GS: Global-to-Local Motion Decomposition and Temporal Interval Adjustment for Compact Dynamic 3D Gaussian Splatting
Sangwoon Kwak (Electronics and Telecommunications Research Institute), Munchurl Kim (Korea Advanced Institute of Science and Technology)
Data SynthesisCompressionGaussian SplattingVideo
🎯 What it does: The MoDec-GS framework is proposed, utilizing Global-to-Local Motion Decomposition (GLMD) to achieve high-quality, low-storage view synthesis in dynamic 3D Gaussian splitting.
Model Diagnosis and Correction via Linguistic and Implicit Attribute Editing
Xuanbai Chen (AWS AI Labs), Yifan Xing (AWS AI Labs)
RecognitionImage TranslationAdversarial AttackTransformerLarge Language ModelDiffusion modelAuto EncoderGenerative Adversarial NetworkImageMultimodality
🎯 What it does: This paper proposes a framework called MDC for automatic diagnosis and correction of visual models, which can automatically discover linguistic and implicit attributes from model errors and perform causal validation through the generation of adversarial samples, ultimately enhancing model robustness with synthesized adversarial data.
Model Poisoning Attacks to Federated Learning via Multi-Round Consistency
Yueqi Xie (Hong Kong University of Science and Technology), Neil Zhenqiang Gong (Duke University)
Federated LearningAdversarial AttackImage
🎯 What it does: A model poisoning attack named PoisonedFL is proposed, which significantly improves the attack effectiveness in federated learning by enforcing consistency in malicious client updates across multiple training rounds and dynamically adjusting the attack magnitude.
Modeling Multiple Normal Action Representations for Error Detection in Procedural Tasks
Wei-Jin Huang (Sun Yat-sen University), Wei-Shi Zheng (Sun Yat-sen University)
Anomaly DetectionTransformerVideo
🎯 What it does: Proposes the AMNAR framework, which utilizes task graphs and dynamic programming to predict all valid next actions, and dynamically reconstructs normative action representations for each action to detect erroneous behaviors in process videos.
Modeling Thousands of Human Annotators for Generalizable Text-to-Image Person Re-identification
Jiayu Jiang (South China University of Technology), Xiangmin Xu (South China University of Technology)
RetrievalTransformerLarge Language ModelPrompt EngineeringImageText
🎯 What it does: Developed the Human Annotator Modeling (HAM) method, which utilizes multimodal large language models (MLLM) to automatically generate diverse text descriptions, thereby enhancing human retrieval performance from text to image.
ModeSeq: Taming Sparse Multimodal Motion Prediction with Sequential Mode Modeling
Zikang Zhou (City University of Hong Kong), Yu-Kai Huang (Hon Hai Research Institute)
Autonomous DrivingTransformerMultimodalityPoint Cloud
🎯 What it does: This paper proposes ModeSeq, which employs a sequential multimodal reasoning framework to predict the future trajectories of traffic participants, directly outputting sparse yet diverse and reliable multiple modes.
MODfinity: Unsupervised Domain Adaptation with Multimodal Information Flow Intertwining
Shanglin Liu (South China University of Technology), Shengfeng He (Singapore Management University)
Domain AdaptationKnowledge DistillationImageMultimodalityAudio
🎯 What it does: Proposes the MODfinity framework, which implements unsupervised multimodal domain adaptation using modality affinity measurement and modality affinity distillation to dynamically control information flow.
MoEdit: On Learning Quantity Perception for Multi-object Image Editing
Yanfeng Li (Macao Polytechnic University), Tao Tan (Macao Polytechnic University)
Image TranslationGenerationTransformerDiffusion modelImage
🎯 What it does: The MoEdit framework is proposed, achieving multi-object image editing without auxiliary tools, capable of consistently controlling the number and attributes of objects while maintaining the overall structure.
MoEE: Mixture of Emotion Experts for Audio-Driven Portrait Animation
Huaize Liu (Zhejiang Lab), Hujun Bao (Zhejiang University)
GenerationData SynthesisMixture of ExpertsDiffusion modelVideoMultimodalityAudio
🎯 What it does: Designed and implemented the MoEE (Mixture of Emotion Experts) framework for audio-driven avatar animation, supporting fine-grained emotion and expression control from multimodal inputs such as labels, text, and audio.
MoFlow: One-Step Flow Matching for Human Trajectory Forecasting via Implicit Maximum Likelihood Estimation based Distillation
Yuxiang Fu (University of British Columbia), Renjie Liao (Simon Fraser University)
GenerationOptimizationComputational EfficiencyKnowledge DistillationTransformerFlow-based ModelVideoTime Series
🎯 What it does: A single-step trajectory prediction model called MoFlow based on flow matching is proposed, and combined with IMLE to achieve distillation of the teacher model, resulting in a fast and multimodal trajectory generation method.
MoGe: Unlocking Accurate Monocular Geometry Estimation for Open-Domain Images with Optimal Training Supervision
Ruicheng Wang (University of Science and Technology of China), Jiaolong Yang (Microsoft Research)
Depth EstimationAutonomous DrivingOptimizationConvolutional Neural NetworkTransformerImagePoint Cloud
🎯 What it does: A single-image geometric estimation model MoGe is proposed, which directly predicts a reversible affine-invariant 3D point map.
Molmo and PixMo: Open Weights and Open Data for State-of-the-Art Vision-Language Models
Matt Deitke (Allen Institute for AI), Aniruddha Kembhavi (Allen Institute for AI)
GenerationData SynthesisRetrievalTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: This paper presents Molmo—a fully open visual language model, along with its weights, training code, and dataset, aiming to achieve performance comparable to closed-source models.
MoManipVLA: Transferring Vision-language-action Models for General Mobile Manipulation
Zhenyu Wu (Beijing University of Posts and Telecommunications), Haibin Yan (Beijing University of Posts and Telecommunications)
OptimizationRobotic IntelligenceTransformerReinforcement LearningVision-Language-Action ModelMultimodality
🎯 What it does: The MoManipVLA framework is proposed to transfer the pre-trained fixed-base Visual-Language-Action (VLA) model to mobile manipulation robots, achieving coordinated motion between end-effector path planning and mobile base.
Mono-InternVL: Pushing the Boundaries of Monolithic Multimodal Large Language Models with Endogenous Visual Pre-training
Gen Luo (OpenGVLab Shanghai AI Laboratory), Xizhou Zhu (Tsinghua University)
TransformerLarge Language ModelMixture of ExpertsVision Language ModelTextMultimodality
🎯 What it does: Designed and trained Mono-InternVL—a single-modal multimodal model that embeds visual experts into a pre-trained large language model (LLM) and implements a mixture of experts (MoE) structure, proposing an Endogenous Visual Pre-training (EViP) phased coarse-to-fine visual learning process;
Mono2Stereo: A Benchmark and Empirical Study for Stereo Conversion
Songsong Yu (DLUT), Huchuan Lu (DLUT)
Image TranslationGenerationDiffusion modelImageBenchmark
🎯 What it does: The Mono2Stereo dataset is proposed, and the stereo conversion task is studied based on this dataset.
Mono3DVLT: Monocular-Video-Based 3D Visual Language Tracking
Hongkai Wei (Chang'an University), Ajmal Saeed Mian (University of Western Australia)
Object DetectionObject TrackingTransformerVision Language ModelVideoTextMultimodality
🎯 What it does: This paper proposes a monocular video 3D object tracking task guided by natural language, called Mono3DVLT, and provides a complete end-to-end solution.
Monocular and Generalizable Gaussian Talking Head Animation
Shengjie Gong (South China University of Technology), Zhuoman Liu (Hong Kong Polytechnic University)
GenerationDepth EstimationGaussian SplattingVideo
🎯 What it does: A general 3D Gaussian scattering head animation framework MGGTalk based on monocular data is proposed, which can generate high-quality, multi-view animations of unseen identities without the need for personalized retraining.
MonoDGP: Monocular 3D Object Detection with Decoupled-Query and Geometry-Error Priors
Fanqi Pu (Shenzhen International Graduate School Tsinghua University), Wenming Yang (Shenzhen International Graduate School Tsinghua University)
Object DetectionDepth EstimationAutonomous DrivingTransformerPoint Cloud
🎯 What it does: A monocular 3D object detection framework called MonoDGP based on Transformer is proposed, which utilizes geometric error priors, decoupled queries, and a region segmentation head to achieve more accurate depth estimation and 3D object localization.
MonoInstance: Enhancing Monocular Priors via Multi-view Instance Alignment for Neural Rendering and Reconstruction
Wenyuan Zhang (Tsinghua University), Zhizhong Han (Wayne State University)
RestorationSegmentationDepth EstimationNeural Radiance FieldGaussian SplattingPoint Cloud
🎯 What it does: Estimate monocular depth uncertainty through multi-view instance alignment, improving the monocular prior for neural rendering and reconstruction.
MonoPlace3D: Learning 3D-Aware Object Placement for 3D Monocular Detection
Rishubh Parihar (Indian Institute of Science Bangalore), R. Venkatesh Babu (Indian Institute of Science Bangalore)
Object DetectionData SynthesisAutonomous DrivingDiffusion modelAuto EncoderImage
🎯 What it does: This paper proposes MonoPlace3D, which learns feasible 3D object placement distributions from a single road surface image and combines it with ControlNet-based realistic rendering to generate scene-aware 3D data augmentation.
MonoSplat: Generalizable 3D Gaussian Splatting from Monocular Depth Foundation Models
Yifan Liu (Chinese University of Hong Kong), Yixuan Yuan (Chinese University of Hong Kong)
RestorationDepth EstimationTransformerGaussian SplattingImage
🎯 What it does: This paper proposes MonoSplat, a method for generalizable 3D Gaussian dispersion reconstruction that utilizes knowledge from a pre-trained monocular depth model.
MonoTAKD: Teaching Assistant Knowledge Distillation for Monocular 3D Object Detection
Hou-I Liu (National Yang Ming Chiao Tung University), Wen-Huang Cheng (National Taiwan Normal University)
Object DetectionAutonomous DrivingKnowledge DistillationPoint Cloud
🎯 What it does: A teaching assistant knowledge distillation framework for monocular 3D detection, MonoTAKD, is proposed, which achieves efficient knowledge transfer under the same modality using a camera-based teaching assistant model and extracts unique 3D spatial information from LiDAR through residual distillation.
MonSter: Marry Monodepth to Stereo Unleashes Power
Junda Cheng (Huazhong University of Science and Technology), Xin Yang (Huazhong University of Science and Technology)
Depth EstimationOptimizationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a dual-branch structure called MonSter, which combines monocular depth estimation with stereo matching to form an iterative complementary enhancement process.
Morpheus: Text-Driven 3D Gaussian Splat Shape and Color Stylization
Jamie Wynn (Niantic), Mohamed Sayed (Niantic)
GenerationData SynthesisDiffusion modelGaussian SplattingImagePoint Cloud
🎯 What it does: This paper proposes a text prompt-based 3D Gaussian Splatting scene stylization method that can simultaneously control the transformation of appearance and geometric shape.
MOS-Attack: A Scalable Multi-objective Adversarial Attack Framework
Ping Guo (City University of Hong Kong), Zhenkun Wang (Southern University of Science and Technology)
OptimizationAdversarial AttackImage
🎯 What it does: This paper proposes a multi-objective attack framework called MOS-Attack, which generates stronger adversarial samples by simultaneously optimizing various surrogate loss functions.
MOS: Modeling Object-Scene Associations in Generalized Category Discovery
Zhengyuan Peng (Shanghai Jiao Tong University), Lizhuang Ma (Shanghai Jiao Tong University)
ClassificationSegmentationTransformerContrastive LearningImage
🎯 What it does: This paper proposes a framework called MOS for General Category Discovery (GCD) that utilizes scene information to address the misjudgment problem caused by the confusion between scenes and objects.
Mosaic of Modalities: A Comprehensive Benchmark for Multimodal Graph Learning
Jing Zhu (University of Michigan), Danai Koutra (University of Michigan)
Graph Neural NetworkContrastive LearningImageTextMultimodalityGraphBenchmark
🎯 What it does: This paper presents MM-GRAPH, a multimodal graph learning benchmark that includes text and visual features, covering seven real-world graph datasets of different scales for node/edge prediction and knowledge graph completion.
Mosaic3D: Foundation Dataset and Model for Open-Vocabulary 3D Segmentation
Junha Lee (NVIDIA), Chris Choy (NVIDIA)
SegmentationData SynthesisVision Language ModelContrastive LearningTextPoint Cloud
🎯 What it does: An automated data generation pipeline was developed to construct the Mosaic3D‑5.6M large-scale 3D mask-text dataset, and based on this, the Mosaic3D foundational model was trained to achieve 3D semantic and instance segmentation of open vocabulary.
MoSca: Dynamic Gaussian Fusion from Casual Videos via 4D Motion Scaffolds
Jiahui Lei (University of Pennsylvania), Kostas Daniilidis (University of Pennsylvania)
GenerationData SynthesisDepth EstimationGaussian SplattingOptical FlowVideo
🎯 What it does: This paper presents a fully automated four-dimensional reconstruction system called MoSca, which can reconstruct and render dynamic scenes from unposed monocular casual videos.
MoST: Efficient Monarch Sparse Tuning for 3D Representation Learning
Xu Han (Huazhong University of Science and Technology), Xianzhi Li (Huazhong University of Science and Technology)
Representation LearningPoint Cloud
🎯 What it does: This paper proposes a parameter-efficient fine-tuning method called MoST based on reparameterization, which replaces the dense weights of the full model with a sparse Point Monarch matrix specifically designed for 3D point clouds, allowing for fine-tuning of only a small number of parameters without increasing inference overhead.
MotiF: Making Text Count in Image Animation with Motion Focal Loss
Shijie Wang (Meta), Xi Yin
GenerationData SynthesisDiffusion modelOptical FlowVideoTextBenchmark
🎯 What it does: This paper studies the Text-guided Image Animation (TI2V) task and proposes a Motion-Focused Loss (MotiF) based on optical flow heatmaps, allowing the model to focus on high-motion areas during training, thereby improving text consistency and motion quality.
Motion Modes: What Could Happen Next?
Karran Pandey (University of Toronto), Paul Guerrero (Adobe Research)
GenerationOptimizationDiffusion modelImageVideo
🎯 What it does: In a single static image, a pre-trained image-to-video diffusion model is utilized to discover and generate diverse motions focused on the target object by guiding energy during the inference phase.
Motion Prompting: Controlling Video Generation with Motion Trajectories
Daniel Geng (Google DeepMind), Deqing Sun (Google DeepMind)
SegmentationGenerationData SynthesisDiffusion modelVideo
🎯 What it does: A Motion Prompt framework based on point trajectories is proposed, training ControlNet on a pre-trained video diffusion model to achieve various video motion control and editing using sparse or dense trajectories.
Motion-Grounded Video Reasoning: Understanding and Perceiving Motion at Pixel Level
Andong Deng (University of Central Florida), Chen Chen (University of Texas at Dallas)
Object DetectionSegmentationTransformerLarge Language ModelVision Language ModelVideo
🎯 What it does: A new task of Motion-Grounded Video Reasoning is proposed, which requires the model to generate pixel-level masks for target objects in videos based on motion-related questions, and a large-scale dataset GROUNDMORE is constructed for this task.
MotionBench: Benchmarking and Improving Fine-grained Video Motion Understanding for Vision Language Models
Wenyi Hong (Tsinghua University), Jie Tang (Tsinghua University)
RecognitionCompressionComputational EfficiencyTransformerLarge Language ModelVision Language ModelVideoMultimodalityBenchmark
🎯 What it does: Proposes the MotionBench benchmark to evaluate fine-grained motion understanding in video VLMs and conducts a systematic evaluation of existing models.
MotionMap: Representing Multimodality in Human Pose Forecasting
Reyhaneh Hosseininejad (Ecole Polytechnique Federale de Lausanne), Alexandre Alahi (Ecole Polytechnique Federale de Lausanne)
Pose EstimationRecurrent Neural NetworkDiffusion modelAuto EncoderMultimodalityTime Series
🎯 What it does: This paper proposes MotionMap, a multimodal human pose prediction framework based on 2D heatmaps, which can automatically capture different future motion patterns, provide confidence assessments, and support controllable predictions and uncertainty analysis.
MotionPro: A Precise Motion Controller for Image-to-Video Generation
Zhongwei Zhang (University of Science and Technology of China), Tao Mei (HiDream.ai Inc.)
GenerationData SynthesisDiffusion modelOptical FlowImageVideoBenchmark
🎯 What it does: A precise motion control framework named MotionPro has been developed, which achieves fine-grained and object-level motion control from images to videos by utilizing region-level trajectories and motion masks.
MotionPRO: Exploring the Role of Pressure in Human MoCap and Beyond
Shenghao Ren (Nanjing University), Xun Cao (Tsinghua University)
Pose EstimationRobotic IntelligenceVideoMultimodality
🎯 What it does: A large-scale multimodal human motion capture dataset called MotionPRO has been constructed, and a method for pose and trajectory estimation called FRAPPE, which integrates pressure and RGB data, has been proposed.
Motions as Queries: One-Stage Multi-Person Holistic Human Motion Capture
Kenkun Liu (Alibaba Group), Xiaoguang Han (Chinese University of Hong Kong)
Object TrackingPose EstimationTransformerVideo
🎯 What it does: A single-stage multi-person full-body motion capture system is proposed, capable of predicting the motion trajectories of all individuals directly from a monocular video in one go.
MotionStone: Decoupled Motion Intensity Modulation with Diffusion Transformer for Image-to-Video Generation
Shuwei Shi (University of Tokyo), Yinqiang Zheng (Zhejiang University)
GenerationData SynthesisTransformerDiffusion modelContrastive LearningVideo
🎯 What it does: This paper proposes a motion-to-video generation model called MotionStone based on a diffusion transformer, which utilizes a decoupled motion intensity estimator to achieve precise control of object and camera motion.
Move-in-2D: 2D-Conditioned Human Motion Generation
Hsin-Ping Huang (Adobe Research), Zhan Xu (Adobe Research)
GenerationData SynthesisTransformerVision Language ModelDiffusion modelVideoTextMultimodality
🎯 What it does: A conditional diffusion model based on 2D scene images and text prompts is proposed to generate human motion sequences that can be projected into the scene.
MoVE-KD: Knowledge Distillation for VLMs with Mixture of Visual Encoders
Jiajun Cao (Peking University), Shanghang Zhang (Peking University)
Knowledge DistillationRepresentation LearningTransformerMixture of ExpertsVision Language ModelMultimodality
🎯 What it does: In large-scale visual language models, knowledge distillation integrates the unique advantages of multiple visual encoders into a single encoder, achieving more efficient visual representation.
Movie Weaver: Tuning-Free Multi-Concept Video Personalization with Anchored Prompts
Feng Liang (University of Texas at Austin), Diana Marculescu (University of Texas at Austin)
GenerationData SynthesisPrompt EngineeringDiffusion modelVideoText
🎯 What it does: This paper presents Movie Weaver, a parameter-free multi-concept video personalization model that can combine various reference images such as faces, bodies, and animals with text prompts to generate high-quality videos.
MovieBench: A Hierarchical Movie Level Dataset for Long Video Generation
Weijia Wu (National University of Singapore), Mike Zheng Shou (Zhejiang University)
GenerationData SynthesisTransformerLarge Language ModelVision Language ModelVideoMultimodalityBenchmarkAudio
🎯 What it does: This paper proposes MovieBench, a hierarchical movie-level dataset aimed at long video generation, and reevaluates existing video generation models based on this dataset.
MOVIS: Enhancing Multi-Object Novel View Synthesis for Indoor Scenes
Ruijie Lu (Peking University), Siyuan Huang (BIGAI)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: Improved multi-object indoor scene new perspective synthesis model, enhancing structural perception and cross-view consistency.
MP-GUI: Modality Perception with MLLMs for GUI Understanding
Ziwei Wang (Zhejiang University), Jiajun Bu (Ant Group)
TransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: A multimodal large language model specifically designed for graphical user interfaces (GUI) called MP-GUI has been developed. It extracts screen information through three types of perceivers (graphic, text, spatial) and dynamically fuses these features based on task semantics using a Fusion Gate, achieving comprehensive understanding and application of GUIs.
MP-SfM: Monocular Surface Priors for Robust Structure-from-Motion
Zador Pataki (ETH Zurich), Marc Pollefeys (ETH Zurich)
Depth EstimationOptimizationImage
🎯 What it does: An improved incremental structured light reconstruction method based on monocular depth and normal vector priors is proposed, significantly enhancing the reliability of SfM in low overlap, low disparity, and highly symmetric scenes.
MPDrive: Improving Spatial Understanding with Marker-Based Prompt Learning for Autonomous Driving
Zhiyuan Zhang (South China University of Technology), Shuangping Huang
Autonomous DrivingPrompt EngineeringMultimodality
🎯 What it does: Designed and implemented the MPDrive framework, which enhances spatial understanding in autonomous driving visual question answering by converting object space coordinates into textual visual markers.
Mr. DETR: Instructive Multi-Route Training for Detection Transformers
Chang-Bin Zhang (Visual AI Lab, University of Hong Kong), Kai Han (Meituan Inc.)
Object DetectionSegmentationTransformerImage
🎯 What it does: This paper proposes a Multi-route Training mechanism for DETR-like detectors, which simultaneously uses one main route for one-to-one matching and one or two auxiliary routes for one-to-many matching during the training phase. It employs Instructive Self-Attention to guide the auxiliary routes in learning one-to-many targets, while directly using the main route during the inference phase without additional inference costs.
MTADiffusion: Mask Text Alignment Diffusion Model for Object Inpainting
Jun Huang (Meitu Inc), Xiaolin Hu (Tsinghua University)
RestorationGenerationDiffusion modelImage
🎯 What it does: Proposes MTADiffusion, a mask-text alignment-based diffusion model for object filling.
Multi-focal Conditioned Latent Diffusion for Person Image Synthesis
Jiaqi Liu (University of Trento), Nicu Sebe (Ocean University of China)
GenerationData SynthesisPose EstimationDiffusion modelImageBenchmark
🎯 What it does: This paper proposes a Multi-Focal Conditional Latent Diffusion Model (MCLD) for pose-guided portrait generation, enhancing detail and identity preservation by separating high-frequency areas such as facial and clothing textures as independent conditions.
Multi-Granularity Class Prototype Topology Distillation for Class-Incremental Source-Free Unsupervised Domain Adaptation
Peihua Deng (Hangzhou Dianzi University), Liang Li (Institute of Computing Technology, Chinese Academy of Sciences)
Domain AdaptationKnowledge DistillationTransformerContrastive LearningImage
🎯 What it does: An algorithm GROTO is proposed for the class-incremental source-free unsupervised domain adaptation (CI-SFUDA) problem, which can continuously transfer the pre-trained source model to the target domain and maintain old class knowledge without source samples and as target classes gradually appear.
Multi-Group Proportional Representations for Text-to-Image Models
Sangwon Jung (Seoul National University), Flavio P. Calmon (Harvard University)
GenerationOptimizationDiffusion modelImageText
🎯 What it does: This paper proposes and evaluates the Multi-Group Proportional Representation (MPR) metric to quantify the representation fairness of text-to-image models across multiple intersecting groups, using it as a fine-tuning objective.
Multi-Label Prototype Visual Spatial Search for Weakly Supervised Semantic Segmentation
Songsong Duan (Xidian University), Nannan Wang (Xidian University)
SegmentationTransformerContrastive LearningImage
🎯 What it does: A multi-label prototype visual space search (MuP-VSS) framework is designed, which generates multi-label prototypes through global embedding and the Prototype Embedding Module (PEM), and optimizes the prototypes using three contrastive losses (CCP, CIP, P2P) to produce high-quality pseudo-pixel labels, addressing the classification bias problem in weakly supervised semantic segmentation.
Multi-Layer Visual Feature Fusion in Multimodal LLMs: Methods, Analysis, and Best Practices
Junyan Lin (Ningbo Key Laboratory of Spatial Intelligence and Digital Derivative), Xiaoyu Shen (Ningbo Key Laboratory of Spatial Intelligence and Digital Derivative)
TransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: This paper systematically studies the selection and fusion strategies of multi-layer visual features in multimodal large language models and proposes an external direct fusion method.
Multi-Modal Aerial-Ground Cross-View Place Recognition with Neural ODEs
Sijie Wang (Nanyang Technological University), Wee Peng Tay (Nanyang Technological University)
RecognitionRetrievalAutonomous DrivingImageMultimodalityPoint CloudOrdinary Differential Equation
🎯 What it does: The AGPlace model is proposed to achieve multi-modal ground (image + point cloud) queries and cross-view location recognition in aerial perspectives (satellite images or road maps).
Multi-modal Contrastive Learning with Negative Sampling Calibration for Phenotypic Drug Discovery
Jiahua Rao (Sun Yat-sen University), Yuedong Yang (Sun Yat-sen University)
Drug DiscoveryGraph Neural NetworkMixture of ExpertsContrastive LearningMultimodalityBiomedical Data
🎯 What it does: The MINER framework is proposed, which achieves contrastive learning for weakly paired and missing drug phenotype data through negative sampling calibration and multimodal fusion, generating more robust molecular phenotype representations.
Multi-Modal Contrastive Masked Autoencoders: A Two-Stage Progressive Pre-training Approach for RGBD Datasets
Muhammad Abdullah Jamal (Intuitive Surgical), Omid Mohareri (Intuitive Surgical)
SegmentationRepresentation LearningTransformerDiffusion modelAuto EncoderContrastive LearningImageMultimodality
🎯 What it does: A multi-modal contrastive masked autoencoder based on a two-stage progressive pre-training approach is proposed for representation learning on RGB-D datasets.
Multi-modal Knowledge Distillation-based Human Trajectory Forecasting
Jaewoo Jeong (KAIST), Kuk-Jin Yoon (KAIST)
Knowledge DistillationTransformerVision Language ModelTextMultimodality
🎯 What it does: This paper proposes a multi-modal knowledge distillation framework to distill richer motion intention knowledge from a teacher model trained with complete modalities (trajectory, 3D pose, and text) to a student model that uses only trajectories, thereby improving the accuracy of pedestrian trajectory prediction.
Multi-modal Medical Diagnosis via Large-small Model Collaboration
Wanyi Chen (Zhejiang University), Haishuai Wang (Zhejiang University)
ClassificationAnomaly DetectionMixture of ExpertsContrastive LearningMultimodalityBiomedical DataElectronic Health Records
🎯 What it does: The AdaCoMed framework is proposed, achieving collaborative learning between single-modal large models and multi-modal small models;
Multi-Modal Synergistic Implicit Image Enhancement for Efficient Optical Flow Estimation
Weichen Dai (Hangzhou Dianzi University), Wanzeng Kong (Hangzhou Dianzi University)
RestorationConvolutional Neural NetworkRecurrent Neural NetworkTransformerOptical FlowImageMultimodality
🎯 What it does: This paper proposes a method for optical flow estimation based on multimodal collaborative implicit image enhancement, which improves optical flow accuracy in low-light and noisy environments.
Multi-modal Topology-embedded Graph Learning for Spatially Resolved Genes Prediction from Pathology Images with Prior Gene Similarity Information
Hang Shi (Nanjing University of Aeronautics and Astronautics), Wei Shao (Nanjing University of Aeronautics and Astronautics)
Graph Neural NetworkImageBiomedical Data
🎯 What it does: This paper proposes a spatial transcriptomics gene expression prediction method based on pathological images, utilizing multimodal graph embedding to learn spatial representations of different image features and ultimately output gene expression levels.
Multi-modal Vision Pre-training for Medical Image Analysis
Shaohao Rui (Shanghai Jiao Tong University), Xiaosong Wang (Shanghai AI Laboratory)
ClassificationSegmentationConvolutional Neural NetworkTransformerContrastive LearningMultimodalityBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A self-supervised pre-training framework for multimodal brain MRI, BrainMVP, is proposed, utilizing cross-modal reconstruction, modality inductive distillation, and modality-aware contrastive learning to learn cross-modal representations.
Multi-party Collaborative Attention Control for Image Customization
Han Yang (Institute of Computing Technology), Yongjun Xu (Institute of Computing Technology)
Object DetectionGenerationTransformerDiffusion modelImageText
🎯 What it does: This paper proposes MCA-Ctrl, a parameter-free framework that achieves high-quality image customization under text or image conditions, capable of simultaneously performing three types of tasks: subject generation, subject replacement, and subject addition.
Multi-Resolution Pathology-Language Pre-training Model with Text-Guided Visual Representation
Shahad Albastaki (Khalifa University of Science and Technology), Arif Mahmood
ClassificationSegmentationRepresentation LearningTransformerVision Language ModelContrastive LearningImageTextMultimodalityBiomedical Data
🎯 What it does: A multi-resolution pathology-language pre-training model MR-PLIP was developed, utilizing whole slide images (WSI) to perform aligned training between images and automatically generated text descriptions at four magnification levels: 5×, 10×, 20×, and 40×.
Multi-Scale Neighborhood Occupancy Masked Autoencoder for Self-Supervised Learning in LiDAR Point Clouds
Mohamed Abdelsamad (Bosch Center for AI), Abhinav Valada (University of Freiburg)
Object DetectionSegmentationAutonomous DrivingTransformerAuto EncoderPoint Cloud
🎯 What it does: We propose NOMAE, a multi-scale neighborhood occupancy mask autoencoder for unsupervised pre-training of LiDAR point clouds.