CVPR 2026 Papers — Page 22
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 4071 papers
META: Meta Evolution of Tool Trajectory Adaptation for Long-Video Understanding
Jing Huang (Zhejiang University), Qiang Zhu (Zhejiang University)
Meta LearningVision Language ModelVideoBenchmark
🎯 What it does: Propose the META framework, leveraging a training-agnostic tool evolution mechanism to achieve self-improvement in long video understanding.
MetaSpectra+: A Compact Broadband Metasurface Camera for Snapshot Hyperspectral+ Imaging
Yuxuan Liu (Purdue University), Qi Guo (Purdue University)
Diffusion modelAuto EncoderImageMultimodalityPhysics Related
🎯 What it does: Developed and demonstrated the MetaSpectra+ ultra-wideband multifunctional camera, enabling single-shot acquisition of high dynamic range, polarization, and hyperspectral images.
MeteorPred: A Meteorological Multimodal Large Model and Dataset for Severe Weather Event Prediction
Shuo Tang (Institute of Automation, Chinese Academy of Sciences), Shiming Xiang (Institute of Automation, Chinese Academy of Sciences)
TransformerLarge Language ModelTextMultimodalityTime SeriesSequentialBenchmarkPhysics Related
🎯 What it does: Proposed the MP-Bench large-scale multimodal disaster prediction dataset and developed the Meteorology Multimodal Large Model (MMLM) capable of directly processing 4D meteorological data
MeToM: Metadata-Guided Token Merging for Efficient Video LLMs
Zhuojie Wu (University of Queensland), Xin Yu (Adelaide University)
Computational EfficiencyTransformerLarge Language ModelVision Language ModelVideo
🎯 What it does: Propose a training-free metadata-driven visual token merging framework called MeToM, which enables spatial and temporal adaptive token merging in video large language models (VLLMs) by leveraging video bitstream metadata, significantly improving inference speed and efficiency.
MetricHMSR: Metric Human Mesh and Scene Recovery from Monocular Images
Chentao Song (Tsinghua University), Tao Yu (Tsinghua University)
Pose EstimationDepth EstimationTransformerMixture of ExpertsImageMesh
🎯 What it does: This paper proposes MetricHMSR, an end-to-end framework capable of simultaneously recovering human pose, size, global position, and scene geometry from a single monocular image.
MetroGS: Efficient and Stable Reconstruction of Geometrically Accurate High-Fidelity Large-Scale Scenes
Kehua Chen (Institute of Computing Technology, Chinese Academy of Sciences), Zhaoqi Wang (Institute of Computing Technology, Chinese Academy of Sciences)
GenerationOptimizationComputational EfficiencyGaussian SplattingImagePoint Cloud
🎯 What it does: Propose a large-scale urban scene geometry reconstruction framework named MetroGS, achieving high-precision modeling based on distributed 2D Gaussian Splatting.
MFEN: Multi-Frequency Expert Network for Visible-Infrared Person Re-ID
Xulin Li (University of Science and Technology of China), Nenghai Yu (University of Science and Technology of China)
RecognitionRetrievalConvolutional Neural NetworkMixture of ExpertsContrastive LearningImageMultimodality
🎯 What it does: Propose a multi-band expert network (MFEN) along with three techniques: random frequency augmentation (RFA) and frequency-assisted optimization (FAO), specifically designed for visible-infrared person re-identification.
MGDHand: Multi-Granularity Prior-to-Inertial Distillation Framework for Sequential 3D Hand Pose Estimation from Sparse IMUs
Xinyi Wang (Shanghai Jiao Tong University), Erwei Yin (Shanghai Jiao Tong University)
Pose EstimationKnowledge DistillationTransformerSequential
🎯 What it does: Proposes a multi-grained prior-to-inertial distillation framework called MGDHand for estimating 3D hand pose from sparse IMU data.
MHopReg: Efficient Hierarchical Multi-Hop Graph Search for Point Cloud Registration
Yue Wu (Xidian University), Wenping Ma (Xidian University)
Pose EstimationGraph Neural NetworkPoint Cloud
🎯 What it does: Designed a hierarchical multi-hop graph search framework, MHopReg, which effectively eliminates outlier correspondences in low-overlap and large-scale scenarios by leveraging SE(3)-equivariant graph encoding, cluster-balanced seed sampling, and multi-resolution multi-hop expansion, achieving precise point cloud registration.
MIBURI: Towards Expressive Interactive Gesture Synthesis
M. Hamza Mughal (Max Planck Institute for Informatics), Christian Theobalt (Max Planck Institute for Informatics)
GenerationData SynthesisTransformerAuto EncoderContrastive LearningMultimodalitySequential
🎯 What it does: MIBURI realizes real-time, expressive full-body and facial co-speech animation generation through an online, causal framework, directly utilizing the internal speech-text embeddings from the dialog system Moshi to drive animation generation.
MICo-150K: A Comprehensive Dataset Advancing Multi-Image Composition
Xinyu Wei (Hong Kong Polytechnic University), Lei Zhang (Hong Kong Polytechnic University)
GenerationData SynthesisLarge Language ModelSupervised Fine-TuningVision Language ModelDiffusion modelImageMultimodalityBenchmark
🎯 What it does: This paper studies the multi-image synthesis (MICo) task, constructs a large-scale high-quality dataset called MICo-150K, and designs an evaluation benchmark named MICo-Bench.
MICON-Bench: Benchmarking and Enhancing Multi-Image Context Image Generation in Unified Multimodal Models
Mingrui Wu (Xiamen University), Rongrong Ji (Xiamen University)
GenerationLarge Language ModelImageMultimodalityBenchmark
🎯 What it does: Propose the MICON-Bench benchmark and the Dynamic Attention Rebalancing (DAR) mechanism to evaluate and enhance the consistency and coherence of unified multimodal models (UMMs) in multi-image context generation.
MicroFM: Physics-guided Flow Matching for Isotropic Microscopy Reconstruction
Xingzu Zhan (Carnegie Mellon University), Min Xu (Carnegie Mellon University)
RestorationFlow-based ModelBiomedical DataPhysics Related
🎯 What it does: Proposed the MicroFM framework, which achieves isotropic 3D reconstruction in optical microscopy by leveraging physical PSF prediction, implicit volume prior, and flow matching network.
Mimic Human Cognition, Master Multi-Image Reasoning: A Meta-Action Framework for Enhanced Visual Understanding
Jianghao Yin (East China Normal University), Liang He (ByteDance)
RecognitionExplainability and InterpretabilityReinforcement LearningVision Language ModelImageVideoMultimodalityBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose the CINEMA (Cognition-Inspired Meta-Action) framework, decomposing multi-graph reasoning into five structured meta-actions: Global, Focus, Hint, Think, and Answer, and enabling model self-learning through retrieval-based tree sampling and two-stage reinforcement learning.
MimiCAT: Mimic with Correspondence-Aware Cascade-Transformer for Category-Free 3D Pose Transfer
Zenghao Chai (National University of Singapore), Mohan Kankanhalli (National University of Singapore)
Image TranslationPose EstimationTransformerVision Language ModelAuto EncoderMeshSequential
🎯 What it does: A method is proposed for the category-free 3D pose transfer task, which transfers poses from a source character to a target character while preserving the target geometry and source pose features.
MimicTalker: A Multimodal Interactive and Memory-Enhanced Framework for Real-Time Dyadic 3D Head Generation
Yinuo Wang (Xi'an Jiaotong University), Fei Wang (Xi'an Jiaotong University)
GenerationTransformerLarge Language ModelMultimodalityRetrieval-Augmented GenerationAudio
🎯 What it does: Proposed the MimicTalker framework to achieve real-time dual-person 3D head generation, capable of synchronously responding to the speaker's voice and non-verbal signals to generate natural and coherent head movements.
Mind the Discriminability Trap in Source-Free Cross-domain Few-shot Learning
Zhenyu Zhang (Huazhong University of Science and Technology), Guangyao Chen (Peking University)
Domain AdaptationRepresentation LearningMeta LearningSupervised Fine-TuningVision Language ModelContrastive LearningMultimodality
🎯 What it does: Conduct in-depth research on the fine-tuning process of vision-language models (e.g., CLIP, SigLIP, PE-Core) in the source-free cross-domain few-shot learning (SFCDFSL) task;
Mind the Gap: Transferring Labels to Align Object Detection Datasets
Mikhail Kennerley (National University of Singapore), Robby T. Tan (National University of Singapore)
Object DetectionDomain AdaptationConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes the Label-Aligned Transfer (LAT) framework, which aligns pseudo labels from the source dataset with the target label space, and integrates the Privileged Proposal Generator (PPG) and Semantic Feature Fusion (SFF) to enhance detection performance on the target dataset.
Mind the Generative Details: Direct Localized Detail Preference Optimization for Video Diffusion Models
Zitong Huang (Harbin Institute of Technology), Wangmeng Zuo (Harbin Institute of Technology)
GenerationTransformerDiffusion modelAuto EncoderVideoText
🎯 What it does: Propose the LocalDPO framework, which constructs fine-grained preference pairs using negative samples generated by local mask erosion on real videos, followed by training to enhance detail quality and semantic alignment in text-to-video diffusion models.
Mind the Hitch: Dynamic Calibration and Articulated Perception for Autonomous Trucks
Morui Zhu (University of North Texas), Qing Yang (University of North Texas)
Pose EstimationAutonomous DrivingTransformerImageMultimodalityPoint Cloud
🎯 What it does: Implemented the dynamic calibration and coherent perception framework dCAP based on multi-view vision, which real-time estimates the 6-DoF pose between tractor and trailer and enhances 3D object detection.
Mind the Way You Select Negative Texts: Pursuing the Distance Consistency in OOD Detection with VLMs
Zhikang Xu (Chinese Academy of Sciences), Qingming Huang (Chinese Academy of Sciences)
Anomaly DetectionRepresentation LearningPrompt EngineeringContrastive LearningImageTextMultimodality
🎯 What it does: Propose the InterNeg framework, which leverages the cross-modal consistency distance of the CLIP vision-language model to enhance out-of-distribution (OOD) detection, systematically selecting negative texts and generating additional negative text embeddings through modal inversion.
MindDriver: Introducing Progressive Multimodal Reasoning for Autonomous Driving
Lingjun Zhang (Amap, Alibaba Group), Mu Xu (Amap, Alibaba Group)
Autonomous DrivingTransformerLarge Language ModelReinforcement LearningVision Language ModelImageTextMultimodality
🎯 What it does: Proposes MindDriver, an end-to-end autonomous driving framework that achieves text semantic understanding, future scene image imagination, and physical trajectory planning through advanced multimodal reasoning.
MindPower: Enabling Theory-of-Mind Reasoning in VLM-based Embodied Agents
Ruoxuan Zhang (Jilin University), Wen-Huang Cheng (National Taiwan University)
Robotic IntelligenceSupervised Fine-TuningReinforcement LearningVision Language ModelMultimodalityBenchmark
🎯 What it does: Developed the MindPower Benchmark and introduced a Robot-Centric ToM reasoning hierarchy to evaluate the performance of Vision-Language Models (VLMs) in robotic perception, reasoning, and action.
Minerva-Ego: Spatiotemporal Hints for Egocentric Video Understanding
Arsha Nagrani (Google DeepMind), Cordelia Schmid (Google DeepMind)
Object TrackingSegmentationTransformerLarge Language ModelPrompt EngineeringVision Language ModelVideoMultimodalityBenchmark
🎯 What it does: Created and released Minerva-Ego—a dataset containing 1,160 multi-step, cross-temporal first-person video question-answering tasks, with human-annotated spatiotemporal reasoning trajectories provided for each question.
MiniCPM-V 4.5: Cooking Efficient MLLMs via Architecture, Data, and Training Recipe
Tianyu Yu (Tsinghua University), Maosong Sun (Tsinghua University)
Computational EfficiencyReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningVision Language ModelMultimodality
🎯 What it does: MiniCPM-V 4.5 is an 8B parameter multimodal large language model designed to improve training and inference efficiency.
Minimal Constraint Relaxation for Multiview Autocalibration
Norio Kosaka (National Institute of Informatics), Tomas Pajdla (CIIRC, CTU in Prague)
Pose EstimationOptimizationImage
🎯 What it does: This paper studies constraint relaxation on the tri-view Kruppa equations, systematically enumerating and analyzing all possible minimal relaxation subsets, and proposes a Global-Best minimal relaxation scheme to achieve efficient and stable self-calibration;
Mining Attribute Subspaces for Efficient Fine-tuning of 3D Foundation Models
Yu Jiang (The University of Texas at Austin), Qixing Huang (The University of Texas at Austin)
Computational EfficiencyData-Centric LearningSupervised Fine-TuningImageMesh
🎯 What it does: This paper addresses LoRA fine-tuning for 3D foundational models by proposing a method to extract subspaces for attributes such as texture, geometry, camera, and lighting through the generation of controlled synthetic data, integrating them into a low-rank LoRA base to achieve efficient and interpretable fine-tuning.
Mining Instance-Centric Vision-Language Contexts for Human-Object Interaction Detection
Soo Won Seo (Seoul National University), Jun Won Choi (Seoul National University)
Object DetectionTransformerVision Language ModelImageText
🎯 What it does: The paper proposes an Instance-centric Context Mining Network (InCoM-Net), which achieves more fine-grained human-object interaction reasoning by extracting instance-centric multi-level contextual information from vision-language models such as CLIP and fusing it with DETR detector features.
Mirai: Autoregressive Visual Generation Needs Foresight
Yonghao Yu (University of Tokyo), Toshihiko Yamasaki (University of Tokyo)
GenerationTransformerImage
🎯 What it does: Propose the Mirai training framework, injecting future information into visual autoregressive generative models to enhance global consistency and convergence speed.
Mirror Illusion Art
Xiaopei Zhu (Tsinghua University), Xiaolin Hu (Tsinghua University)
GenerationData SynthesisOptimizationNeural Radiance FieldImageMesh
🎯 What it does: Automatically generate a 3D mirror illusion art model that displays different patterns from front and mirror views.
Missing No More: Dictionary-Guided Cross-Modal Image Fusion under Missing Infrared
Yafei Zhang (Kunming University of Science and Technology), Yu Liu (Kunming University of Science and Technology)
RestorationConvolutional Neural NetworkLarge Language ModelImageMultimodality
🎯 What it does: Propose a cross-modal image fusion framework under missing infrared conditions, utilizing a shared convolutional dictionary to perform infrared feature reasoning and fusion in the coefficient domain;
Mistake Attribution: Fine-Grained Mistake Understanding in Egocentric Videos
Yayuan Li (University of Michigan), Jason J. Corso (University of Michigan)
RecognitionTransformerVision Language ModelVideoMultimodality
🎯 What it does: Propose the MATT task to achieve the localization of semantic, temporal, and spatial attributes of erroneous actions in first-person videos.
Mitigating Error Amplification in Fast Adversarial Training
Mengnan Zhao (Anhui University), Geyong Min (Anhui University)
ClassificationAdversarial AttackImage
🎯 What it does: This study proposes a distribution-aware dynamic guidance (DDG) strategy to enhance Fast Adversarial Training (FAT), aiming to reduce catastrophic overfitting and alleviate the robustness-accuracy trade-off.
Mitigating Instance Entanglement in Instance-Dependent Partial Label Learning
Rui Zhao (Xi'an Jiaotong University), Bo Dong (Xi'an Jiaotong University)
ClassificationDiffusion modelContrastive LearningImage
🎯 What it does: This paper proposes the Class-specific Augmentation based Disentanglement (CAD) framework to address the instance entanglement problem in instance-dependent partial label learning (ID-PLL).
Mitigating Multimodal Hallucinations via Gradient-based Self-Reflection
Shan Wang (NVIDIA), Jose M. Alvarez (NVIDIA)
Explainability and InterpretabilityComputational EfficiencyTransformerLarge Language ModelVision Language ModelContrastive LearningMultimodalityBenchmark
🎯 What it does: Proposes a gradient-based self-reflective method called GACD to reduce hallucinatory outputs in multimodal large language models during inference, enhancing the credibility and quality of visual information generation.
Mitigating Objectness Bias and Region-to-Text Misalignment for Open-Vocabulary Panoptic Segmentation
Nikolay Kormushev (University of Ljubljana), Matej Kristan (University of Ljubljana)
SegmentationConvolutional Neural NetworkTransformerSupervised Fine-TuningVision Language ModelContrastive LearningImageText
🎯 What it does: Propose a framework named OVRCOAT for open-vocabulary panoptic segmentation, enhancing the robustness of mask selection and classification through two-step improvements.
Mitigating Simplicity Bias in OOD Detection through Object Co-occurrence Analysis
Boyang Dai (University of Hong Kong), Yizhou Yu (University of Hong Kong)
Anomaly DetectionTransformerContrastive LearningImage
🎯 What it does: Proposes an OOD detection framework based on object co-occurrence (OCO), leveraging Slot Attention to achieve decoupled representations of objects in images, partitioning test samples through object co-occurrence patterns, and employing targeted scoring to implement 'divide and conquer,' thereby alleviating the simplicity bias of traditional models.
Mitigating The Distribution Shift of Diffusion-based Dataset Distillation
Yue Xu (Shanghai Jiao Tong University), Yong-Lu Li (Shanghai Jiao Tong University)
Domain AdaptationKnowledge DistillationTransformerDiffusion modelScore-based ModelImage
🎯 What it does: This paper proposes a two-stage method to alleviate distribution shift in dataset distillation (DD) based on diffusion models. The first stage introduces L1 sparse regularization during diffusion model training, restricting the learned latent space to retain only the most important and transferable features, resulting in a more concise generative prior. The second stage employs collaborative guided sampling (CGS) during sampling, suppressing diversity collapse and distribution drift caused by low-capacity sampling through synchronized denoising and two regularizations (DPP diversity regularization and mean matching regularization).
MixerCSeg: An Efficient Mixer Architecture for Crack Segmentation via Decoupled Mamba Attention
Zilong Zhao (Shandong University), Feng Guo (Shandong University)
SegmentationConvolutional Neural NetworkTransformerImageBenchmark
🎯 What it does: Propose a hybrid architecture named MixerCSeg for pixel-level segmentation of road cracks;
MixFlow Training: Alleviating Exposure Bias with Slowed Interpolation Mixture
Hui Li (Fudan University), Jingdong Wang (Baidu)
GenerationDiffusion modelFlow-based ModelRectified FlowImageTextOrdinary Differential Equation
🎯 What it does: Proposed a training method called MixFlow, which utilizes a 'slowed time step' corresponding to high-noise interpolation (slowed interpolation mixture) sampled during training, thereby reducing the training-test discrepancy (exposure bias) in diffusion models.
Mixture of Prototypes for Test-time Adaptive Segmentation
Guangrui Li (University Of Technology Sydney), Yongxin Ge (University Of Technology Sydney)
SegmentationDomain AdaptationTransformerMixture of ExpertsImage
🎯 What it does: This paper proposes a test-time adaptive semantic segmentation framework based on Mixture of Experts (MoE), which generates diverse experts by clustering source domain prototypes and dynamically weights them during inference to adapt to the test distribution.
Mixture of States: Routing Token-Level Dynamics for Multimodal Generation
Haozhe Liu (King Abdullah University of Science and Technology), Jürgen Schmidhuber (King Abdullah University of Science and Technology)
GenerationTransformerDiffusion modelImageTextMultimodality
🎯 What it does: Propose the Mixture of States (MoS) framework, achieving adaptive sparse state interaction in cross-modal diffusion models through a learnable token-level router.
Mixture of Style Experts for Diverse Image Stylization
Shihao Zhu (Nankai University), Qibin Hou (Nankai University)
Image TranslationGenerationTransformerMixture of ExpertsVision Language ModelDiffusion modelContrastive LearningImage
🎯 What it does: Proposes StyleExpert, an image style transfer framework based on Mixture of Experts (MoE) and pre-trained style encoders, capable of handling diverse styles at multiple semantic levels.
Mixture-of-Experts based Feature Decoupling for Open Vocabulary Scene Graph Generation
Yiming Li (Hefei University of Technology), Bing-Kun Bao (Hefei University of Technology)
GenerationMixture of ExpertsVision Language ModelImageTextMultimodality
🎯 What it does: This paper studies Open-Vocabulary Scene Graph Generation (OVSGG), proposing the MoE-FD framework. It utilizes Mixture-of-Experts to perform fine-grained decoupling of object and relation features, combined with iterative cross-attention to achieve bidirectional semantic interaction between objects and relations, ultimately aligning visual features with semantic labels.
MLLM-HWSI: A Multimodal Large Language Model for Hierarchical Whole Slide Image Understanding
Basit Alawode (Khalifa University of Science and Technology), Sajid Javed (Khalifa University of Science and Technology)
ClassificationRetrievalExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelContrastive LearningMultimodalityBiomedical DataBenchmark
🎯 What it does: This paper proposes a multi-scale hierarchical large language model, MLLM-HWSI, for multimodal understanding and diagnostic reasoning of whole slide images (WSI).
MLLMSplat: A 2D MLLM-Powered Framework for 3D Gaussian Splatting Understanding, Generation, and Editing
Jingqiao Xiu (University of Hong Kong), Dong Xu (University of Hong Kong)
GenerationData SynthesisTransformerLarge Language ModelVision Language ModelDiffusion modelRectified FlowAuto EncoderGaussian SplattingImageTextMultimodality
🎯 What it does: Proposes the MLLMSplat framework, extending 2D multimodal large language models (MLLM) to high-level understanding, generation, and editing in 3D Gaussian Splatting (3DGS) scenarios.
MM-ACT: Learn from Multimodal Parallel Generation to Act
Haotian Liang (University of Science and Technology of China), Ping Luo (University of Hong Kong)
GenerationRobotic IntelligenceTransformerVision-Language-Action ModelDiffusion modelImageTextMultimodality
🎯 What it does: Propose a unified Vision-Language-Action (VLA) model MM-ACT, which can simultaneously generate text, images, and robotic actions within a shared discrete token space, achieving low-latency execution through parallel decoding.
MM-OVSeg: Multimodal Optical-SAR Fusion for Open-Vocabulary Segmentation in Remote Sensing
Yimin Wei (University of Tokyo), Naoto Yokoya (University of Tokyo)
SegmentationTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Proposed a multi-modal optical-SAR fusion framework named MM-OVSeg for achieving open-vocabulary segmentation (OVS) under adverse weather conditions such as clouds and fog
MM-ReCoder: Advancing Chart-to-Code Generation with Reinforcement Learning and Self-Correction
Zitian Tang (Brown University), Davide Modolo (Amazon AGI)
GenerationAI Code AssistantTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: Based on a multimodal large language model, we propose MM-ReCoder, which can self-correct and generate executable drawing code during multi-round interactions.
MM-SeR: Multimodal Self-Refinement for Lightweight Image Captioning
Junha Song (KAIST), Jaegul Choo (KAIST)
GenerationComputational EfficiencyTransformerLarge Language ModelVision Language ModelImageVideoTextMultimodality
🎯 What it does: This paper proposes a lightweight image description model based on OPT-125M and designs a multimodal self-refinement framework called MM-SeR, using two-stage generation to improve description quality.
MMBench-GUI: A Unified Hierarchical Evaluation Framework for Multi-Platform GUI Agents
Xuehui Wang (Shanghai Jiao Tong University), Wenhai Wang (Tsinghua University)
OptimizationComputational EfficiencyLarge Language ModelAgentic AIVision-Language-Action ModelMultimodalityBenchmark
🎯 What it does: Propose MMBench-GUI, a cross-platform hierarchical GUI agent evaluation framework covering Windows, macOS, Linux, iOS, Android, and Web, divided into four layers: content understanding, element localization, single-application automation, and multi-application collaboration;
MMCP-GEN: A Modality-Extensible Diffusion Language Model for Conditional Protein Sequence Generation
Zeyu An (Hong Kong Polytechnic University), Shujun Wang (Hong Kong Polytechnic University)
GenerationTransformerLarge Language ModelDiffusion modelContrastive LearningMultimodalityBiomedical Data
🎯 What it does: Propose MMCP-GEN, a diffusion language model supporting multi-modal and multi-condition control for de novo protein sequence design.
MMDIR: Multimodal Instruction-Driven Framework for Mixed-Degradation Document Image Restoration
Heng Li (Harbin Institute of Technology), Qingcai Chen (Harbin Institute of Technology)
RestorationTransformerLarge Language ModelVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: Proposes MMDIR, a multimodal instruction-driven document image restoration framework that can automatically identify degradation types and remove them under mixed degradation conditions.
MMFace-DiT: A Dual-Stream Diffusion Transformer for High-Fidelity Multimodal Face Generation
Bharath Krishnamurthy (University of North Texas), Ajita Rattani
GenerationData SynthesisTransformerVision Language ModelDiffusion modelRectified FlowAuto EncoderImageTextMultimodality
🎯 What it does: Proposed a unified dual-stream diffusion Transformer, MMFace-DiT, for generating high-fidelity facial images from multimodal inputs such as text, semantic masks, or sketches.
MMGait: Towards Multi-Modal Gait Recognition
Chenye Wang (Beijing Normal University), Yongzhen Huang (Beijing Normal University)
RecognitionContrastive LearningMultimodalityBenchmark
🎯 What it does: Established the MMGait multimodal gait recognition benchmark, proposed a unified Omni multimodal gait recognition task and the OmniGait model;
MMLandmarks: a Cross-View Instance-Level Benchmark for Geo-Spatial Understanding
Oskar Kristoffersen (Technical University of Denmark), Dim P. Papadopoulos (Technical University of Denmark)
RetrievalVision Language ModelContrastive LearningImageTextMultimodalityBenchmark
🎯 What it does: Proposed the MMLANDMARKS cross-perspective instance-level multimodal geospatial benchmark and trained a CLIP-based multimodal contrastive learning model.
MMR-AD: A Large-Scale Multimodal Dataset for Benchmarking General Anomaly Detection with Multimodal Large Language Models
Xincheng Yao (Shanghai Jiao Tong University), Chongyang Zhang (Shanghai Jiao Tong University)
Anomaly DetectionTransformerLarge Language ModelReinforcement LearningPrompt EngineeringVision Language ModelImageMultimodalityBenchmark
🎯 What it does: Proposed the MMR-AD large-scale multimodal reasoning-based industrial anomaly detection dataset, and constructed the Anomaly-R1 baseline model based on it;
MMSD3.0: A Multi-Image Benchmark for Real-World Multimodal Sarcasm Detection
Haochen Zhao, Haoliang Zhang (Institute of Information Engineering, Chinese Academy of Sciences)
ClassificationTransformerSupervised Fine-TuningVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: Proposed a new multi-graph sarcasm detection dataset, MMSD3.0, and the corresponding cross-graph reasoning model CIRM, addressing the semantic and emotional association problems of sarcasm in multi-graph scenarios.
MMTIT-Bench: A Multilingual and Multi-Scenario Benchmark with Cognition-Perception-Reasoning Guided Text-Image Machine Translation
Gengluo Li (Institute of Information Engineering, Chinese Academy of Sciences), Yu Zhou (Institute of Information Engineering, Chinese Academy of Sciences)
Image TranslationExplainability and InterpretabilityPrompt EngineeringVision Language ModelMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Proposed the MMTIT-Bench multilingual multi-scenario text-image translation benchmark and introduced the CPR-Trans translation data paradigm based on cognitive-perception-reasoning, significantly enhancing the translation accuracy and interpretability of VLLM in multilingual and multi-scenario contexts.
MMVIP: A Visible-infrared Paired Dataset for Multi-weather Marine Vision
Yunpeng Yin (Guangdong University of Technology), Lianglun Cheng (Guangdong University of Technology)
Image TranslationObject DetectionImageVideoMultimodalityBenchmark
🎯 What it does: Constructed MMVIP, a multi-weather ocean visual dataset covering seven weather conditions, 128,100 visible-infrared aligned images, and 50 video clips, providing complete registration, fusion, detection, and cross-modal translation benchmarks.
mmWaveFlow: Unified Enhancement and Generation of mmWave Human Point Clouds
Chang Su (Institute of Software, Chinese Academy of Sciences), Zhi Wang (Institute of Software, Chinese Academy of Sciences)
GenerationData SynthesisFlow-based ModelAuto EncoderPoint Cloud
🎯 What it does: Unified implementation of enhancement and generation for millimeter-wave human point clouds, constructing the mmWaveFlow framework;
Mobile-VTON: High-Fidelity On-Device Virtual Try-On
Zhenchen Wan, Mingming Gong
Image TranslationKnowledge DistillationConvolutional Neural NetworkDiffusion modelScore-based ModelGenerative Adversarial NetworkImage
🎯 What it does: Proposes MOBILE‑VTON, a high-fidelity virtual try-on system that can operate completely offline on mobile devices;
MoBind: Motion Binding for Fine-Grained IMU-Video Pose Alignment
Duc Duy Nguyen (Adelaide University), Minh Hoai (Adelaide University)
Pose EstimationRetrievalConvolutional Neural NetworkTransformerContrastive LearningVideoMultimodalityTime Series
🎯 What it does: Design and implement the MoBind framework to learn joint representations between inertial measurement unit (IMU) signals and 2D skeleton motion sequences extracted from videos, supporting cross-modal retrieval, temporal synchronization, subject and body part localization, and action recognition.
Mocap-2-to-3: Multi-view Lifting for Monocular Motion Recovery with 2D Pretraining
Zhumei Wang, Siyuan Huang
Pose EstimationDepth EstimationTransformerDiffusion modelImage
🎯 What it does: Develop a multi-view enhancement framework called Mocap-2-to-3 that lifts monocular 2D keypoints into absolute 3D motion coordinates.
MoCapAnything: Unified 3D Motion Capture for Arbitrary Skeletons from Monocular Videos
Kehong Gong (Huawei International Pte Ltd), Mingyuan Zhang (Huawei Central Media Technology Institute)
Pose EstimationGraph Neural NetworkTransformerVideoMultimodalityMeshBenchmark
🎯 What it does: Propose a unified 3D motion capture framework called MoCapAnything, which can directly generate asset-specific animated rotations from monocular videos and arbitrary 3D assets, achieving cross-category and cross-skeleton motion capture and retargeting.
MoCha: End-to-End Video Character Replacement without Structural Guidance
Zhengbo Xu (HUJING Digital Media & Entertainment Group), Jing Li (Huazhong University of Science and Technology)
Object DetectionSegmentationGenerationData SynthesisReinforcement Learning from Human FeedbackTransformerSupervised Fine-TuningDiffusion modelRectified FlowAuto EncoderVideoMultimodality
🎯 What it does: Propose MoCha, an end-to-end video role replacement framework that can perform identity replacement using only a single-frame mask while preserving the original video's background and motion.
MoCoDiff: A Controllable Autoregressive Diffusion Model for Expressive Motion Generation
Wenfeng Song (Beijing Information Science and Technology University), Aimin Hao (Beihang University)
GenerationData SynthesisConvolutional Neural NetworkVision Language ModelDiffusion modelTime SeriesSequential
🎯 What it does: Propose a controllable autoregressive diffusion model called MoCoDiff for generating long-term, expressive-style human motions.
MoD-DPO: Towards Mitigating Cross-modal Hallucinations in Omni LLMs using Modality Decoupled Preference Optimization
Ashutosh Chaubey (University of Southern California), Mohammad Soleymani (University of Southern California)
OptimizationMultimodality
🎯 What it does: By performing preference optimization on multimodal large language models, we propose Modality-Decoupled Direct Preference Optimization (MoD-DPO) and incorporate language prior debiasing (LPD) to reduce cross-modal hallucinations.
Model Merging in the Essential Subspace
Longhua Li (Southeast University), Xin Geng (Southeast University)
ClassificationRepresentation LearningTransformerImageBenchmark
🎯 What it does: This paper proposes the ESM framework for model merging in the functional subspace, aiming to integrate multi-task fine-tuned models and significantly reduce task interference.
Modeling Cross-vision Synergy for Unified Large Vision Model
Shengqiong Wu (National University of Singapore), Tat-Seng Chua (National University of Singapore)
Knowledge DistillationRepresentation LearningTransformerMixture of ExpertsVision Language ModelImageVideo
🎯 What it does: Designed and implemented PolyV, a unified large-scale visual model capable of achieving collaborative reasoning across three visual modalities: images, videos, and 3D.
Modeling Spatiotemporal Neural Frames for High Resolution Brain Dynamic
Wanying Qu (Fudan University), Yanwei Fu (Fudan University)
RestorationGenerationData SynthesisTransformerDiffusion modelScore-based ModelAuto EncoderMultimodalityBiomedical DataMagnetic Resonance Imaging
🎯 What it does: Propose a diffusion Transformer framework conditioned on EEG signals to reconstruct high-resolution fMRI dynamic sequences and support training-free reconstruction of intermediate frames;
Modeling the Brain's Grammar: ROI-Guided fMRI Pretraining for Transferable and Interpretable Vision Decoding
Yulong Liu (Hong Kong University of Science and Technology), Yike Guo (Hong Kong University of Science and Technology)
RetrievalExplainability and InterpretabilityRepresentation LearningDiffusion modelAuto EncoderImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: Propose a ROI-guided pre-training framework for fMRI called ROITok, which encodes brain activity using ROI-level tokens and learns cross-subject transferable visual representations through sparse ROI context fusion.
Modeling the Visual Ambiguity of Human Sketches
Yang Zhou (Zhejiang University), Shengfeng He (Singapore Management University)
RetrievalTransformerContrastive LearningImageText
🎯 What it does: This paper proposes a metric called AmbiScore to measure the visual ambiguity of human sketch-image pairs, and based on this, designs the DisAmb framework, which reduces the negative impact of ambiguity on zero-shot sketch-based image retrieval (ZS-SBIR) performance through two mechanisms: elastic matching and purification matching.
Models as Lego Builders: Assembling Malice from Benign Blocks via Semantic Blueprints
Chenxi Li (Dail Tech), Xiande Huang (Dail Tech)
Adversarial AttackLarge Language ModelPrompt EngineeringVision Language ModelImageText
🎯 What it does: Propose a black-box jailbreak method called StructAttack that exploits semantic slot filling vulnerabilities, decomposing malicious requests into seemingly harmless slot types and inducing LVLMs to reassemble and generate harmful content through structured visual prompts.
MoDES: Accelerating Mixture-of-Experts Multimodal Large Language Models via Dynamic Expert Skipping
Yushi Huang (Hong Kong University of Science and Technology), Jun Zhang (Hong Kong University of Science and Technology)
Computational EfficiencyMixture of ExpertsMultimodality
🎯 What it does: Proposed a training-agnostic dynamic expert skipping framework called MoDES to accelerate the inference of multimodal mixture-of-experts large language models (MLLMs).
MODIX: A Training-Free Multimodal Information-Driven Positional Index Scaling for Vision-Language Models
Ruoxiang Huang (Peking University), Zhen Yuan (Peking University)
TransformerVision Language ModelMultimodality
🎯 What it does: Propose MODIX, a training-agnostic method that dynamically adjusts the RoPE position encoding stride in multi-modal Transformers during inference, aiming to allocate finer-grained position encodings based on information density to enhance the multi-modal reasoning capability of vision-language models.
ModularAgent: A Task-Aware Modular Framework for Joint Optimization of Multimodal Large Language Models and World Models
Yu-Wei Zhan (Tsinghua University), Wenwu Zhu (Tsinghua University)
OptimizationRobotic IntelligenceTransformerLarge Language ModelReinforcement LearningMixture of ExpertsWorld ModelTextSequential
🎯 What it does: Designed and implemented ModularAgent, a task-aware modular framework that achieves bidirectional coupling between multimodal large language models and world models, enhancing the performance of embodied agents in multi-task and cross-environment scenarios.
MoE-GRPO: Optimizing Mixture-of-Experts via Reinforcement Learning in Vision-Language Models
Dohwan Ko (Korea University), Hyunwoo J. Kim (KAIST)
OptimizationReinforcement LearningMixture of ExpertsVision Language ModelMultimodality
🎯 What it does: This paper proposes the MoE-GRPO framework, which optimizes expert routing in vision-language models through reinforcement learning to achieve more efficient and diverse expert selection.
MoEActok: A MoE-based Action Tokenizer for Vision-Language-Action Models
Chunpu Xu (Hong Kong Polytechnic University), Yao Mu (Shanghai Jiao Tong University)
Robotic IntelligenceLarge Language ModelMixture of ExpertsVision-Language-Action ModelAuto EncoderMultimodality
🎯 What it does: Proposed a Mixture-of-Experts (MoE)-based action tokenizer, MoEActok, which can decompose continuous control signals into discrete action tokens aligned with language models and is directly utilized within the Vision-Language-Action (VLA) framework;
MoECLIP: Patch-Specialized Experts for Zero-shot Anomaly Detection
Jun Yeong Park (Yonsei University), Yu Rang Park (Yonsei University)
Anomaly DetectionTransformerMixture of ExpertsVision Language ModelContrastive LearningImageBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: Propose MoECLIP——integrating a Mixture-of-Experts module into CLIP to achieve dynamic adaptation at the image patch level for Zero-Shot Anomaly Detection (ZSAD).
MOFA-VTON: More Fashion Possibilities with Fine-Grained Adaptations in Virtual Try-On
Xiaoyu Han (Harbin Institute Of Technology), Shengping Zhang (Harbin Institute Of Technology)
Image TranslationSegmentationGenerationPose EstimationConvolutional Neural NetworkTransformerDiffusion modelImage
🎯 What it does: Propose a user-interactive virtual try-on framework called MOFA-VTON, which allows users to draw curve sketches to control the layout of upper and lower clothing, achieving diverse clothing effects.
MOGeo: Beyond One-to-One Cross-View Object Geo-localization
Bo Lv (Shenzhen University), Yingying Zhu (Shenzhen University)
Object DetectionConvolutional Neural NetworkContrastive LearningImageBenchmark
🎯 What it does: Propose the cross-view multi-object geolocation task (CVMOGL) and implement an end-to-end model MOGeo
MoLingo: Motion-Language Alignment for Text-to-Human Motion Generation
Yannan He (University of Tübingen), Gerard Pons-Moll (University of Tübingen)
GenerationTransformerLarge Language ModelVision-Language-Action ModelDiffusion modelRectified FlowAuto EncoderMultimodality
🎯 What it does: Proposed MoLingo, a text-to-motion generation model based on a continuous latent space, capable of generating realistic and semantically coherent 3D human motions from textual descriptions.
Molmo2: Open Weights and Data for Vision-Language Models with Video Understanding and Grounding
Christopher Clark (Allen Institute for AI), Ranjay Krishna (Allen Institute for AI)
TransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageVideoTextMultimodality
🎯 What it does: This paper proposes Molmo2, a fully open-weight and open-data visual language model that supports single-image, multi-image, and video inputs, and can output free text and pixel-level localization.
Momentum Memory for Knowledge Distillation in Computational Pathology
Yongxin Guo (Wake Forest University), Metin N. Gurcan (Wake Forest University)
ClassificationKnowledge DistillationGraph Neural NetworkMultimodalityBiomedical Data
🎯 What it does: Improved cross-modal knowledge distillation using a momentum memory mechanism, constructing a framework named MoMKD that enables injecting genomic information into a single histology model during training with only a one-time global dictionary, allowing multiple biomarker predictions during inference with only H&E slices.
MOMO: Mars Orbital MOdel Foundation Model for Mars Orbital Applications
Mirali Purohit (Arizona State University), Hannah Kerner (Arizona State University)
ClassificationSegmentationAuto EncoderImageBenchmark
🎯 What it does: A multi-sensor foundation model named MOMO was constructed and trained for Mars orbit remote sensing tasks, achieving a unified representation across different resolution sensors (HiRISE, CTX, THEMIS) through model fusion.
Monet: Reasoning in Latent Visual Space Beyond Image and Language
Qixun Wang (Peking University), Yisen Wang (Peking University)
Representation LearningLarge Language ModelSupervised Fine-TuningReinforcement LearningVision Language ModelMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Training multimodal large language models (MLLM) for reasoning in a latent visual space, utilizing continuous latent embeddings as intermediate visual thinking.
Monocular Open Vocabulary Occupancy Prediction for Indoor Scenes
Changqing Zhou (Hong Kong University of Science and Technology Guangzhou), Changhao Chen (Hong Kong University of Science and Technology Guangzhou)
SegmentationDepth EstimationRepresentation LearningVision Language ModelGaussian SplattingTextPoint Cloud
🎯 What it does: This paper proposes a monocular open-source vocabulary occupancy prediction framework that jointly models the geometry and semantics of indoor scenes using 3D language-embedded Gaussian representations.
MonoSAOD: Monocular 3D Object Detection with Sparsely Annotated Label
Junyoung Jung (Kyung Hee University), Jung Uk Kim (Kyung Hee University)
Object DetectionAutonomous DrivingTransformerImagePoint Cloud
🎯 What it does: Under sparse annotation conditions, training for monocular 3D detection is conducted using dual modules: Road-Aware Patch Augmentation (RAPA) and Prototype Filtering (PBF).
MonoVLM: Monocular 3D Visual Grounding with Vision Language Models
Huaizhi Qu (University of North Carolina at Chapel Hill), Tianlong Chen (University of North Carolina at Chapel Hill)
Object DetectionTransformerReinforcement LearningVision Language ModelMultimodality
🎯 What it does: This paper proposes MonoVLM, a three-stage training framework that leverages Vision-Language Models (VLMs) for monocular 3D visual localization, i.e., given an RGB image and a natural language description, predicting the 3D bounding box of an object.
MooCap: A Multi-View Benchmark for Cow-Object-Human Interaction and Behavior Dynamics
Ian Noronha (Purdue University), Upinder Kaur (Purdue University)
ClassificationPose EstimationGraph Neural NetworkTransformerVideoMultimodalityBenchmarkAgriculture Related
🎯 What it does: Proposed the multi-perspective animal behavior benchmark MooCap, and constructed a detailed dataset containing multi-camera synchronized videos, fine-grained action annotations, and pose keypoints.
MOON2.0: Dynamic Modality-balanced Multimodal Representation Learning for E-commerce Product Understanding
Zhanheng Nie (Alibaba Group), Bo Zheng (Alibaba Group)
ClassificationRetrievalMixture of ExpertsVision Language ModelMultimodalityBenchmark
🎯 What it does: Proposed MOON2.0, a dynamic multi-modal balanced product understanding framework capable of simultaneously processing images, text, and their mixed queries.
More Natural, More Real: Object-aware Gaussian Splatting for 3D Visual Decoding from Human Brain
Haodong Jing (Xi'an Jiaotong University), Nanning Zheng (Xi'an Jiaotong University)
Vision Language ModelContrastive LearningGaussian SplattingOptical FlowBiomedical DataMagnetic Resonance Imaging
🎯 What it does: Propose the BrainGS framework, which directly applies 3D Gaussian Splatting to decode high-fidelity 3D object reconstruction from fMRI/EEG brain signals.
More Than Meets the Eye: A Unified Image Fusion Framework via Semantic-Pixel Entropy Trade-off for Zero-Shot Generalization
Xiaowen Liu (People's Public Security University of China), Xu Dong (People's Public Security University of China)
Convolutional Neural NetworkImageMultimodality
🎯 What it does: This paper proposes a unified image fusion framework called DECC, which can achieve zero-shot generalization to multiple fusion tasks with only infrared-visible image pairs for training.
More than the Sum: Panorama-Language Models for Adverse Omni-Scenes
Weijia Fan (Karlsruhe Institute of Technology), Rainer Stiefelhagen (Karlsruhe Institute of Technology)
TransformerSupervised Fine-TuningVision Language ModelMultimodalityBenchmark
🎯 What it does: This paper proposes an audio-visual language model (PLM) for panoramic images and constructs a large-scale panoramic visual question answering dataset, PanoVQA, aiming to achieve global understanding and reasoning of 360° environments.
MORE-STEM: Long-Short MemOry REcall and Spatio-TEmporal Consistency Model for Query-Driven 3D/4D Point Cloud Segmentation
Chade Li (Chinese Academy of Sciences), Yihong Wu (Chinese Academy of Sciences)
SegmentationTransformerLarge Language ModelContrastive LearningPoint CloudBenchmark
🎯 What it does: Proposed a unified framework named MORE-STEM for query-driven 3D/4D point cloud segmentation, integrating cross-frame text-visual alignment, spatiotemporal consistency modeling, and long-short-term memory recall;
MoRE: 3D Visual Geometry Reconstruction Meets Mixture-of-Experts
Jingnan Gao (Shanghai Jiao Tong University), Yichao Yan (Shanghai Jiao Tong University)
Pose EstimationDepth EstimationTransformerMixture of ExpertsImagePoint Cloud
🎯 What it does: Propose MoRE, a large-scale 3D vision geometry reconstruction foundation model based on Mixture-of-Experts, capable of directly predicting multiple geometric quantities such as point clouds, depth, camera pose, and surface normals from unposed images.
MoRe: Motion-aware Feed-forward 4D Reconstruction Transformer
Juntong Fang (Tsinghua University), Yu-Shen Liu (Tsinghua University)
Pose EstimationDepth EstimationTransformerVideo
🎯 What it does: Propose MoRe, a forward 4D reconstruction network based on Transformer, which can simultaneously estimate camera pose, depth maps, point maps, and dynamic object masks from monocular video streams, and supports streaming inference.
MoReGen: Multi-Agent Motion-Reasoning Engine for Code-based Text-to-Video Synthesis
Xiangyu Bai, Sarah Ostadabbas
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningAgentic AIVideoTextBenchmarkPhysics Related
🎯 What it does: Designed and implemented the MoReGen framework, which collaborates multi-agent LLMs, physics simulators, and renderers to directly generate executable simulation code from natural language instructions and render videos adhering to Newton's laws of motion.
MoRel: Long-Range Flicker-Free 4D Motion Modeling via Anchor Relay-based Bidirectioanl Blending with Hierarchical Densification
Sangwoon Kwak (ETRI), Jihyong Oh (Chung-Ang University)
GenerationOptimizationGaussian SplattingVideo
🎯 What it does: Proposes the MoRel framework, utilizing Anchor Relay-based Bidirectional Blending (ARBB) and Feature-variance-guided Hierarchical Densification (FHD) for artifact-free, memory-efficient dynamic scene reconstruction of long-term 4D Gaussian Splatting.