M3DLayout: A Multi-Source Dataset of 3D Indoor Layouts and Structured Descriptions for 3D Generation
Yiheng Zhang (Tsinghua University), Yuwang Wang (Tsinghua University)
CodeGenerationData SynthesisLarge Language ModelDiffusion modelTextMultimodalityPoint CloudMesh
🎯 What it does: Built a large-scale multi-source 3D indoor layout dataset named M3DLayout, providing corresponding structured text descriptions for text-driven 3D scene generation.
M4-RAG: A Massive-Scale Multilingual Multi-Cultural Multimodal RAG
David Anugraha (Stanford University), Genta Indra Winata (Capital One)
CodeRetrievalTransformerVision Language ModelTextMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: Designed and implemented a large-scale multilingual, multicultural, and multimodal retrieval-augmented generation (RAG) evaluation framework, M4-RAG, to assess retrieval effectiveness for visual question answering tasks across different languages and modalities.
M4-SAM: Multi-Modal Mixture-of-Experts with Memory-Augmented SAM for RGB-D Video Salient Object Detection
Jiyuan Liu (Hangzhou Dianzi University), Zhi Liu (Shandong University)
CodeSegmentationTransformerMixture of ExpertsVideoMultimodality
🎯 What it does: A prompt-free end-to-end framework named M-SAM is proposed for the RGB-D video salient object detection task, which performs salient object segmentation based on SAM2;
MAD: Modality-Adaptive Decoding for Mitigating Cross-Modal Hallucinations in Multimodal Large Language Models
Sangyun Chung (KAIST), Yong Man Ro (KAIST)
CodeTransformerLarge Language ModelVision Language ModelContrastive LearningVideoTextMultimodalityBenchmarkAudio
🎯 What it does: Propose a training-free Modality-Adaptive Decoding (MAD) that eliminates audio-visual cross-modal hallucinations by self-assessing task-relevant modalities and dynamically weighted contrastive decoding.
🎯 What it does: In this work, we propose the MakeAnything framework, which can generate multi-step creation tutorials (such as painting, crafts, cooking, etc.) from text or a single image, and support decomposing completed works into visual step sequences;
Making the Classification Explanation Faithful to the Confidence Score
Jian-Xun Mi (Chongqing University of Posts and Telecommunications), Weisheng Li (Chongqing University of Posts and Telecommunications)
CodeClassificationExplainability and InterpretabilityConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes a black-box explainable method called MHE based on Metropolis-Hastings sampling, which generates explanation maps highly consistent with model confidence;
Making Training-Free Diffusion Segmentors Scale with the Generative Power
Benyuan Meng (Chinese Academy of Sciences), Qingming Huang (University of Chinese Academy of Sciences)
CodeSegmentationDiffusion modelImage
🎯 What it does: Proposes a method for automatically aggregating cross-attention maps and performing per-pixel reweighting, enabling training-agnostic diffusion segmenters to significantly improve segmentation performance as the generative model's capabilities increase.
🎯 What it does: Propose a context learning framework based on Mamba (SADG) for multi-task point cloud domain generalization (reconstruction, denoising, registration), maintaining cross-domain structural consistency through structure-aware serialization and hierarchical domain-aware modeling;
🎯 What it does: Insert a lightweight linear Mapper module after the text encoder to perform conditional identity mapping for target concepts, enabling precise concept elimination without modifying the main model parameters.
Bingyu Li (University of Science and Technology of China), Xuelong Li (Institute of Artificial Intelligence TeleAI China Telecom)
CodeSegmentationDepth EstimationTransformerPrompt EngineeringVision Language ModelImageTextBenchmark
🎯 What it does: Created the MARIS fine-grained ocean open-vocabulary instance segmentation benchmark, and proposed a unified framework combining the Geometric Prior Enhancement Module (GPEM) and Semantic Alignment Injection Mechanism (SAIM) to address underwater visual degradation and semantic ambiguity.
MarkushGrapher-2: End-to-end Multimodal Recognition of Chemical Structures
Tim Strohmeyer (IBM Research), Peter Staar (IBM Research)
CodeRecognitionDrug DiscoveryTransformerLarge Language ModelVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: Propose MarkushGrapher-2, achieving end-to-end identification of chemical structures and multimodal Markush structures by integrating image, text, and layout information;
MaskDiME: Adaptive Masked Diffusion for Precise and Efficient Visual Counterfactual Explanations
Changlu Guo (Technical University of Denmark), Morten Rieger Hannemose (Technical University of Denmark)
CodeAutonomous DrivingExplainability and InterpretabilityDiffusion modelImage
🎯 What it does: Propose MaskDiME, a training-free diffusion model framework that achieves efficient and accurate visual counterfactual explanations by adaptively applying dual masks to locally control updates during the reverse diffusion process.
🎯 What it does: Propose Masked Representation Modeling (MRM), performing random masking and reconstruction in the latent representation space of semantic segmentation as an auxiliary objective for unsupervised domain adaptation (UDA) tasks;
CodeTransformerLarge Language ModelVision Language ModelPoint Cloud
🎯 What it does: Proposes 3D-SLIM, a decoder masking strategy that can directly replace traditional causal masks, significantly enhancing LLM's reasoning capabilities in 3D scene-language understanding.
MaxMark: High-Capacity Diffusion-Native Watermarking via Robust and Invertible Latent Embedding
Xuanhang Chang (Harbin Institute of Technology), Yu Li (Zhejiang University)
CodeGenerationSafty and PrivacyDiffusion modelFlow-based ModelImageText
🎯 What it does: Designed and implemented a high-capacity, robust diffusion-native watermarking framework called MaxMark, which can embed a large amount of information in the latent space of LDMs while preserving the latent distribution through invertible neural networks.
MDS-VQA: Model-Informed Data Selection for Video Quality Assessment
Jian Zou (City University of Hong Kong), Kede Ma (Google Inc)
CodeData-Centric LearningTransformerSupervised Fine-TuningReinforcement LearningVision Language ModelContrastive LearningVideo
🎯 What it does: Proposes MDS-VQA, an active data selection framework combining model difficulty prediction with content diversity, to select challenging and diverse video samples for fine-tuning VQA models under limited annotation budgets.
🎯 What it does: Train the MeanFlow (MF) model in the high-dimensional semantic latent space of the Representation Autoencoder (RAE) to achieve single-step or few-step generation while significantly reducing training and inference costs.
🎯 What it does: Propose MeanFuser, an end-to-end multi-modal trajectory generation and adaptive reconstruction framework based on one-shot sampling, for autonomous driving planning.
Mechanisms of Object Localization in Vision-Language Models
Timothy Schaumlöffel (Goethe University Frankfurt), Gemma Roig (Goethe University Frankfurt)
CodeObject DetectionExplainability and InterpretabilityTransformerVision Language ModelImage
🎯 What it does: This paper conducts mechanism explainability experiments on two visual-language models, LLaVA-1.5 and InternVL-3.5, to investigate how models achieve object localization, revealing that localization depends on 'containerization' mechanisms, the complementarity of global and local perspectives, and the role of a small number of task-critical attention heads.
Med-CMR: A Fine-Grained Benchmark Integrating Visual Evidence and Clinical Logic for Medical Complex Multimodal Reasoning
Haozhen Gong (National University Of Singapore), Hongwei Bran Li (National University Of Singapore)
CodeTransformerLarge Language ModelVision Language ModelMultimodalityBiomedical DataBenchmark
🎯 What it does: Constructed and released the Med-CMR benchmark, providing 20,653 visual question-answer pairs covering 11 organ systems, 12 imaging modalities, and decomposing medical reasoning into 7 visual dimensions and 4 reasoning dimensions;
MedFG-VQA: Low-Frequency Memory and Graph Attention for Lightweight Medical VQA
Haowen Gu (Nanjing University of Science and Technology), Fumin Shen (University of Electronic Science and Technology of China)
CodeGraph Neural NetworkTransformerLarge Language ModelVision Language ModelMultimodalityBiomedical DataBenchmark
🎯 What it does: Propose a lightweight medical vision question answering framework MedFG-VQA, leveraging frequency domain memory fusion and graph attention to enhance global and local feature representations.
Medic-AD: Towards Medical Vision-Language Model's Clinical Intelligence
Woohyeon Park (AIDAS Laboratory, Seoul National University), Jaeyoung Do (AIDAS Laboratory, Seoul National University)
CodeAnomaly DetectionExplainability and InterpretabilityConvolutional Neural NetworkTransformerPrompt EngineeringVision Language ModelBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: Proposed MEDIC-AD, a multi-stage medical vision-language model, aimed at enhancing anomaly detection, symptom tracking, and visual interpretability capabilities.
MedKCO: Medical Vision-Language Pretraining via Knowledge-Driven Cognitive Orchestration
Chenran Zhang (Southeast University), Yi Zhou (Nanjing University of Science and Technology)
CodeRetrievalRepresentation LearningTransformerVision Language ModelContrastive LearningTextMultimodalityBiomedical Data
🎯 What it does: Propose a medical vision-language pre-training framework called MedKCO, which integrates two-level knowledge-driven curriculum learning (label hierarchy and description hierarchy) along with an adaptive asymmetric contrastive loss to enhance the model's alignment and representation learning of medical images and text.
MedLoc-R1: Performance-Aware Curriculum Reward Scheduling for GRPO-Based Medical Visual Grounding
Guangjing Yang (Beijing University of Posts and Telecommunications), Qicheng Lao (Beijing University of Posts and Telecommunications)
CodeObject DetectionReinforcement LearningVision Language ModelBiomedical Data
🎯 What it does: Propose MedLoc-R1, a performance-aware reward scheduling framework designed to enhance GRPO training in medical vision localization, addressing the issue of sparse rewards.
MedTVT-R1: A Multimodal LLM Empowering Medical Reasoning and Diagnosis
Yuting Zhang (Hong Kong University of Science & Technology), Kaishun Wu (Hong Kong University of Science & Technology)
CodeTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningImageTextMultimodalityTabularBiomedical DataElectrocardiogramChain-of-Thought
🎯 What it does: Propose MedTVT-R1, a multimodal large language model capable of simultaneously processing electrocardiograms (ECG), chest X-rays, and laboratory tables, supporting long-text medical reasoning and diagnosis for multiple diseases.
🎯 What it does: Propose a lightweight forward memory framework called MemFlow, which uses a frozen deep backbone network for feature extraction. It achieves feature-to-label memory encoding, distributed storage, and retrieval through randomly connected neurons and spiking propagation. It also supports memory reinforcement on unlabeled data to enable fast domain adaptation.
🎯 What it does: Proposes the MASKED EDGE PREDICTION (MEMO) framework, achieving human-like clear edge detection through masked training and confidence ranking inference.
🎯 What it does: To address the significant memory consumption of large Diffusion Transformers (DiT) during personalized fine-tuning, this paper proposes the DiTBlockSkip scheme, combining dynamic patch sampling with block skipping and precomputed residual features, significantly reducing training memory usage while maintaining generation quality.
🎯 What it does: Proposes Masking Dual Path Distillation (MDPD), achieving memory-efficient transfer learning by using a lightweight side network and main network to distill knowledge mutually during training; after training, only the main network is retained for inference to avoid inference overhead from the side network.
🎯 What it does: Propose MeshSplatting, which performs end-to-end optimization on connected, fully opaque triangular meshes using differentiable rendering to achieve high-quality real-time novel view synthesis.
🎯 What it does: Constructed a training-free, cross-subject, and cross-scanner fMRI visual decoding framework called BrainCoDec, which infers encoding models for each voxel using hierarchical context learning, then decodes image embeddings through functional inversion with multi-voxel context.
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)
CodeTransformerLarge 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
Mind the Discriminability Trap in Source-Free Cross-domain Few-shot Learning
Zhenyu Zhang (Huazhong University of Science and Technology), Guangyao Chen (Peking University)
CodeDomain 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;
Mining Instance-Centric Vision-Language Contexts for Human-Object Interaction Detection
Soo Won Seo (Seoul National University), Jun Won Choi (Seoul National University)
CodeObject 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.
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)
CodeRestorationConvolutional 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;
🎯 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).
🎯 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.
CodeImage 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.
🎯 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)
CodeSegmentationTransformerVision 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
MMBench-GUI: A Unified Hierarchical Evaluation Framework for Multi-Platform GUI Agents
Xuehui Wang (Shanghai Jiao Tong University), Wenhai Wang (Tsinghua University)
CodeOptimizationComputational 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;
MMDIR: Multimodal Instruction-Driven Framework for Mixed-Degradation Document Image Restoration
Heng Li (Harbin Institute of Technology), Qingcai Chen (Harbin Institute of Technology)
CodeRestorationTransformerLarge 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.
🎯 What it does: Established the MMGait multimodal gait recognition benchmark, proposed a unified Omni multimodal gait recognition task and the OmniGait model;
CodeAnomaly 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;
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)
CodeComputational 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).
MoEActok: A MoE-based Action Tokenizer for Vision-Language-Action Models
Chunpu Xu (Hong Kong Polytechnic University), Yao Mu (Shanghai Jiao Tong University)
CodeRobotic 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)
CodeAnomaly 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).
CodeClassificationKnowledge 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.
Monet: Reasoning in Latent Visual Space Beyond Image and Language
Qixun Wang (Peking University), Yisen Wang (Peking University)
CodeRepresentation 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)
CodeSegmentationDepth 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.
MooCap: A Multi-View Benchmark for Cow-Object-Human Interaction and Behavior Dynamics
Ian Noronha (Purdue University), Upinder Kaur (Purdue University)
CodeClassificationPose 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.
🎯 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)
CodeTransformerSupervised 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.
🎯 What it does: Propose the MOS framework to address the modality gap problem in cross-modal ship ReID between optical and synthetic aperture radar (SAR) modalities
Mostly Text, Smart Visuals: Asymmetric Text-Visual Pruning for Large Vision-Language Models
Sijie Li (University of Sheffield), Jungong Han (Tsinghua University)
CodeComputational EfficiencyVision Language ModelMultimodalityBenchmark
🎯 What it does: Proposes a weight quantization method for large vision-language models (LVLMs) called ATV-Pruning, aiming to achieve high sparsity ratios while maintaining performance.
🎯 What it does: Propose the MotionEdit dataset and benchmark, focusing on modifying actions and interactions in images according to text instructions while preserving identity and scene consistency.
🎯 What it does: Propose MotionHiFlow, a hierarchical flow matching framework that generates text-driven 3D human motions layer by layer from coarse to fine temporal scales;
🎯 What it does: Propose a world model-based MRI contrast enhancement dynamics simulation framework, MRI CEKWorld, to generate continuous time-point contrast-enhanced images from single non-contrast MRI scans, avoiding the need for contrast agent injection.
MSPT: Efficient Large-Scale Physical Modeling via Parallelized Multi-Scale Attention
Pedro M. P. Curvo (University of Amsterdam), Maksim Zhdanov (University of Amsterdam)
CodeComputational EfficiencyTransformerMeshBenchmarkPhysics Related
🎯 What it does: Proposed a multi-scale patch Transformer (MSPT) for large-scale physical modeling, achieving parallel processing of local self-attention and global pooling through spherical tree partitioning.
MSRL: Scaling Generative Multimodal Reward Modeling via Multi-Stage Reinforcement Learning
Chenglong Wang (Northeastern University), Tong Xiao (Northeastern University)
CodeKnowledge DistillationReinforcement Learning from Human FeedbackReinforcement LearningVideoTextMultimodalityChain-of-Thought
🎯 What it does: Proposes a multi-stage reinforcement learning framework, MSRL, which first trains a reward model on large-scale text preference data using RL, and then transfers it to multimodal tasks through caption-based RL and cross-modal knowledge distillation, significantly enhancing the performance of multimodal reward models.
MuCo: Multi-turn Contrastive Learning for Multimodal Embedding Model
Geonmo Gu (NAVER AI Lab), Dongyoon Han (NAVER AI Lab)
CodeComputational EfficiencyRepresentation LearningData-Centric LearningSupervised Fine-TuningPrompt EngineeringVision Language ModelContrastive LearningMultimodalityBenchmark
🎯 What it does: This paper proposes a multi-round contrastive learning framework, MuCo, which processes multiple query-target pairs from the same image in one go using a conversational structure, enhancing the training efficiency and effectiveness of multi-modal embeddings.
🎯 What it does: Propose the MMRest model, addressing the overlapping sentiment centers problem in multimodal sentiment analysis through clustering and multi-metric learning.
🎯 What it does: Propose a multi-modal test-time adaptation method based on adaptive probability Gaussian calibration (AdaPGC), which explicitly models class-conditional distributions and continuously updates statistics during inference, supplemented by adaptive contrastive imbalance correction to address distribution asymmetry caused by single-modal distortions.
🎯 What it does: Propose a multi-scale Patch Transformer (MPDiT), which uses large Patches in the early network stages to capture global information and then upsamples to small Patches in later stages to refine local details, thereby significantly reducing computational costs while maintaining generation quality.
Multi-Scale Gaussian-Language Map for Zero-shot Embodied Navigation and Reasoning
Sixian Zhang (State Key Laboratory of AI Safety, Institute of Computing Technology, Chinese Academy of Sciences), Shuqiang Jiang (University of Chinese Academy of Sciences)
Multi-speaker Attention Alignment for Multimodal Social Interaction
Liangyang Ouyang (University of Tokyo), Yoichi Sato (University of Tokyo)
CodeTransformerLarge Language ModelVision Language ModelMultimodality
🎯 What it does: Propose a multi-speaker attention alignment method to enhance the performance of multi-modal large language models in multi-speaker social interaction tasks.
🎯 What it does: Propose a multi-perspective crowd tracking model based on Transformer, named MVTrackTrans, and collect and annotate two large-scale real-world multi-perspective tracking datasets, MVCrowdTrack and CityTrack.
MultiModalPFN: Extending Prior-Data Fitted Networks for Multimodal Tabular Learning
Wall Kim (Samsung Electronics), Hanul Kim (Seoul National University of Science and Technology)
CodeClassificationRepresentation LearningTransformerVision Language ModelGenerative Adversarial NetworkImageTextMultimodalityTabularElectronic Health RecordsFinance Related
🎯 What it does: Proposed a multimodal TabPFN (MMPFN) framework capable of uniformly processing three different modalities of data: tables, images, and text.
🎯 What it does: Proposed a multi-view masked image modeling (MuM) framework that extends MAE to any number of images from the same scene, using a ViT-L encoder-decoder for 3D visual pre-training.
NEC-Diff: Noise-Robust Event-RAW Complementary Diffusion for Seeing Motion in Extreme Darkness
Haoyue Liu (Huazhong University of Science and Technology), Luxin Yan (Huazhong University of Science and Technology)
CodeRestorationDiffusion modelMultimodality
🎯 What it does: Designed and implemented a diffusion-model-based event-RAW composite imaging framework, NEC-Diff, for high-quality image reconstruction in extremely dark environments.
Neighbor-Aware Localized Concept Erasure in Text-to-Image Diffusion Models
Zhuan Shi (McGill University), Golnoosh Farnadi (McGill University)
CodeGenerationSafty and PrivacyConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: Propose a training-free, three-stage neighborhood-aware local concept erasure framework called NLCE, which can precisely remove specified concepts in text-to-image diffusion models while preserving the visual expression of adjacent concepts.
🎯 What it does: Proposes the NeighborMAE framework, which leverages adjacent Earth observation images to jointly complete the reconstruction task of the Masked AutoEncoder, thereby learning spatial dependencies.
Neu-PiG: Neural Preconditioned Grids for Fast Dynamic Surface Reconstruction on Long Sequences
Julian Kaltheuner, Reinhard Klein
CodeRestorationRepresentation LearningPoint Cloud
🎯 What it does: This paper proposes Neu-PiG, a fast dynamic surface reconstruction method based on preconditioned multi-scale latent grids, which achieves spatiotemporally consistent high-precision 3D reconstruction from long sequences of unstructured point clouds.
🎯 What it does: Proposed a test-time adaptation method called NCTTA based on Neural Collapse, extending NC for the first time at the sample level, revealing the alignment collapse between sample features and classifier weights (NC3+) and correcting pseudo-label errors using a hybrid objective.
🎯 What it does: A hybrid framework based on neural fields, NFH-SEM, achieves high-fidelity 3D surface reconstruction of microstructures using multi-view and quadrant backscattered electron (BSE) detector images.
NeuroSeg Meets DINOv3: Transferring 2D Self-Supervised Visual Priors to 3D Neuron Segmentation via DINOv3 Initialization
Yik San Cheng (University of Sydney), Weidong Cai (University of Sydney)
CodeSegmentationRepresentation LearningConvolutional Neural NetworkTransformerContrastive LearningBiomedical Data
🎯 What it does: Propose the NeurINO model, by migrating the 2D self-supervised visual prior DINOv3 to a 3D convolutional network, achieving efficient neuronal segmentation and reconstruction.
🎯 What it does: Propose the NexusFlow framework to address multi-task learning problems under partially supervised settings, with different tasks and domain differences.
🎯 What it does: Proposed a NeRF-guided 3D Gaussian scattering segmentation framework, NG-GS, capable of refining object boundaries within a 3D Gaussian model
🎯 What it does: Propose a model named NI-Tex that can generate high-quality PBR textures even when there are non-isometric and topological differences between the image and target geometry.
NimbusGS: Unified 3D Scene Reconstruction under Hybrid Weather
Yanying Li (Ocean University of China), Yong Du (Ocean University of China)
CodeRestorationGaussian SplattingImage
🎯 What it does: Propose NimbusGS, a unified framework that recovers high-quality 3D scenes when multi-view inputs are affected by various or mixed adverse weather conditions (fog, rain, snow).
NitroGen: An Open Foundation Model for Generalist Gaming Agents
Loïc Magne (Nvidia), Linxi Fan (Nvidia)
CodeReinforcement Learning from Human FeedbackTransformerSupervised Fine-TuningVision-Language-Action ModelDiffusion modelFlow-based ModelVideoBenchmark
🎯 What it does: Trained a general-purpose game agent named NITROGEN using 40,000 hours of publicly available game videos with player action annotations for behavior cloning, and provided a multi-game, multi-task evaluation suite.
🎯 What it does: Propose an unsupervised, online video stabilization framework that integrates a classic three-stage pipeline (motion estimation, motion propagation, motion compensation) with multithreaded buffering, achieving real-time stabilization without requiring paired stabilized/unstabilized video data.
No Need For Real Anomaly: MLLM Empowered Zero-Shot Video Anomaly Detection
Zunkai Dai (Beijing University of Posts and Telecommunications), Yuanyuan Qiao (Beijing University of Posts and Telecommunications)
CodeAnomaly DetectionTransformerLarge Language ModelVision Language ModelVideoMultimodality
🎯 What it does: Designed an end-to-end zero-shot video anomaly detection framework, LAVIDA, which leverages multi-modal large language models (MLLM) and pseudo anomaly samplers to achieve cross-scenario anomaly detection without requiring training on real abnormal videos.
Nonparametric Deep Fine-grained Clustering with Low-Rank Guided Vision-Language Model
Xulun Ye (Ningbo University), Kun Zhou (Shenzhen University)
CodeOptimizationRepresentation LearningVision Language ModelContrastive LearningMultimodality
🎯 What it does: Propose a non-parametric deep fine-grained clustering framework based on a low-rank guided vision-language model (VLM), which can simultaneously learn discriminative features and dynamically infer the number of clusters under unlabeled and unknown class number scenarios.
🎯 What it does: To address the resource constraints of edge devices, the NuWa method is proposed, which can quickly derive lightweight models tailored to specific categories from pre-trained Vision Transformers (ViT) without requiring post-training fine-tuning.
🎯 What it does: Proposes a cross-category generalizable object re-identification paradigm OG-ReID and constructs the MGOR framework to achieve target-domain-free adaptation for universal identity representation learning.
🎯 What it does: Proposes OccAny, an uncalibrated, domain-agnostic urban 3D occupancy prediction model capable of performing zero-shot inference on sequences, monocular, or panoramic images while generating high-quality occupancy and segmentation features.
🎯 What it does: Propose a monocular video occlusion human motion capture method based on frequency domain denoising prior, modeling occluded motion as a wavelet coefficient selection problem.
🎯 What it does: This paper introduces the OccuFly dataset, the first to establish a 3D semantic scene benchmark from an aerial perspective using a camera.
Omni IIE Bench: Benchmarking the Practical Capabilities of Image Editing Models
Yujia Yang (University of Chinese Academy of Sciences), Hongzhu Yi (Tencent)
CodeMultimodalityBenchmark
🎯 What it does: Propose the Omni IIE Bench benchmark to evaluate the single-round consistency and multi-round coordination capabilities of instruction-driven image editing models, ensuring data quality through rigorous manual screening.
🎯 What it does: This paper proposes a unified one-time 3D editing framework called Omni-3DEdit, capable of performing multiple editing tasks such as object deletion, addition, and appearance modification in a multi-view latent space.