ECCV 2024 Papers — Page 14
European Conference on Computer Vision · 2387 papers
Make a Cheap Scaling: A Self-Cascade Diffusion Model for Higher-Resolution Adaptation
Lanqing Guo (Nanyang Technological University), Bihan Wen (Nanyang Technological University)
GenerationSuper ResolutionDiffusion modelImageVideo
🎯 What it does: This paper proposes a self-cascade diffusion model (Self-Cascade Diffusion Model), which can quickly migrate low-resolution diffusion models to higher resolutions, achieving high-quality image and video generation.
Make a Strong Teacher with Label Assistance: A Novel Knowledge Distillation Approach for Semantic Segmentation
Shoumeng Qiu (Fudan University), Jian Pu (Fudan University)
SegmentationKnowledge DistillationImage
🎯 What it does: Introduce noisy labels as auxiliary information for the teacher model in semantic segmentation knowledge distillation, construct a lightweight teacher model, and enhance the student model's performance through distillation.
Make Your ViT-based Multi-view 3D Detectors Faster via Token Compression
Dingyuan Zhang (Huazhong University of Science and Technology), Xiang Bai (Huazhong University of Science and Technology)
Object DetectionAutonomous DrivingComputational EfficiencyTransformerImagePoint Cloud
🎯 What it does: Propose the ToC3D method for multi-view 3D detection, which compresses tokens in the ViT backbone to improve inference speed;
Make-Your-3D: Fast and Consistent Subject-Driven 3D Content Generation
Fangfu Liu (Tsinghua University), Yueqi Duan (Tsinghua University)
GenerationData SynthesisTransformerPrompt EngineeringVision Language ModelDiffusion modelNeural Radiance FieldGaussian SplattingImageTextMesh
🎯 What it does: Generate personalized 3D content consistent with text instructions within 5 minutes based on a single image.
Making Large Language Models Better Planners with Reasoning-Decision Alignment
Zhijian Huang (Shenzhen Campus of Sun Yat-sen University), Xiaodan Liang (Meituan Inc.)
Autonomous DrivingTransformerLarge Language ModelSupervised Fine-TuningVision-Language-Action ModelMultimodalityChain-of-Thought
🎯 What it does: Proposes an end-to-end autonomous driving planning framework named RDA-Driver, which utilizes large language models to achieve scene understanding, reasoning, and trajectory planning.
Mamba-ND: Selective State Space Modeling for Multi-Dimensional Data
Shufan Li (University of California, Los Angeles), Harkanwar Singh (University of California, Los Angeles)
ClassificationRecognitionSegmentationImageVideoBiomedical DataComputed Tomography
🎯 What it does: Propose Mamba-ND, a lightweight architecture that extends the 1D state space model Mamba to arbitrary multidimensional data by alternating sequence orders between layers;
MambaIR: A Simple Baseline for Image Restoration with State-Space Model
Hang Guo (Tsinghua University), Shu-Tao Xia (Tsinghua University)
RestorationSuper ResolutionImage
🎯 What it does: Proposed and implemented MambaIR, a simple baseline based on the Mamba structured state space model for image restoration tasks.
ManiGaussian: Dynamic Gaussian Splatting for Multi-task Robotic Manipulation
Guanxing Lu (Tsinghua University), Yansong Tang (Tsinghua University)
Robotic IntelligenceDiffusion modelGaussian SplattingWorld ModelTextMultimodality
🎯 What it does: Proposes a robot multi-task manipulation framework called ManiGaussian based on dynamic Gaussian projection, which leverages future scene reconstruction to explore spatiotemporal dynamics of the scene, achieving efficient manipulation under language conditions.
MANIKIN: Biomechanically Accurate Neural Inverse Kinematics for Human Motion Estimation
Jiaxi Jiang (ETH Zürich), Christian Holz (ETH Zürich)
Pose EstimationMesh
🎯 What it does: Proposed MANIKIN, a neural inverse kinematics model based on biomechanical constraints, which infers full-body pose from sparse head and hand poses;
MAP-ADAPT: Real-Time Quality-Adaptive Semantic 3D Maps
Jianhao Zheng (Stanford University), Iro Armeni (Stanford University)
SegmentationComputational EfficiencyRepresentation LearningConvolutional Neural NetworkSimultaneous Localization and MappingImageMesh
🎯 What it does: Propose a real-time quality-adaptive semantic 3D map construction method (MAP-ADAPT), which generates a single map using RGB-D images and automatically assigns different resolutions under varying semantic categories and geometric complexity.
MapDistill: Boosting Efficient Camera-based HD Map Construction via Camera-LiDAR Fusion Model Distillation
Xiaoshuai Hao (Samsung R&D Institute), Jing Zhang (University of Sydney)
Autonomous DrivingComputational EfficiencyKnowledge DistillationRepresentation LearningImagePoint Cloud
🎯 What it does: Propose the MapDistill method, using knowledge distillation to transfer knowledge from the camera-LiDAR fusion model to a lightweight camera model, achieving efficient high-definition map construction.
MapTracker: Tracking with Strided Memory Fusion for Consistent Vector HD Mapping
Jiacheng Chen (Simon Fraser University), Yasutaka Furukawa (Simon Fraser University)
Autonomous DrivingTransformerSimultaneous Localization and MappingVideoPoint Cloud
🎯 What it does: This paper proposes a vector HD map construction framework called MapTracker, which is based on the fusion of tracking and sparse memory, aiming to achieve temporally consistent high-precision road element maps;
MarineInst: A Foundation Model for Marine Image Analysis with Instance Visual Description
Ziqiang Zheng (Hong Kong University of Science and Technology), Sai-Kit Yeung (Hong Kong University of Science and Technology)
RecognitionSegmentationTransformerVision Language ModelImage
🎯 What it does: This paper develops MarineInst, a foundation model for the marine domain capable of performing instance segmentation and instance-level semantic description in both automatic or interactive modes.
MaRINeR: Enhancing Novel Views by Matching Rendered Images with Nearby References
Lukas Bösiger (ETH Zurich), Zuria Bauer (ETH Zurich)
Image TranslationImage HarmonizationRestorationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: Propose MaRINeR, a novel view rendering method that post-processes noise/geometric defects in 3D reconstructions by leveraging nearby reference images;
Markov Knowledge Distillation: Make Nasty Teachers trained by Self-undermining Knowledge Distillation Fully Distillable
En-hui Yang (University of Waterloo), Linfeng Ye (University of Waterloo)
Knowledge DistillationImage
🎯 What it does: This paper proposes a new knowledge distillation method called Markov Knowledge Distillation (MKD), demonstrating its ability to make previously 'nasty' (malicious) teacher models fully distillable, while significantly enhancing student model performance even with normal teachers.
MARs: Multi-view Attention Regularizations for Patch-based Feature Recognition of Space Terrain
Timothy Chase Jr (University at Buffalo), Karthik Dantu (University at Buffalo)
RecognitionRepresentation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: Research on single-frame description method for aerospace terrain surface features based on metric learning
MART: MultiscAle Relational Transformer Networks for Multi-agent Trajectory Prediction
Seongju Lee (Gwangju Institute of Science and Technology), Kyoobin Lee (Gwangju Institute of Science and Technology)
Autonomous DrivingComputational EfficiencyTransformerTime Series
🎯 What it does: Propose a multi-scale relational Transformer network called MART for multi-agent trajectory prediction.
MarvelOVD: Marrying Object Recognition and Vision-Language Models for Robust Open-Vocabulary Object Detection
Kuo Wang (Sun Yat Sen University), Guanbin Li (Sun Yat Sen University)
RecognitionObject DetectionConvolutional Neural NetworkTransformerPrompt EngineeringVision Language ModelImageText
🎯 What it does: This paper proposes the MarvelOVD framework, which leverages collaboration between an object detector and a vision-language model (VLM) to generate high-quality pseudo labels, and enhances open-vocabulary object detection (OVD) performance through online pseudo label mining and adaptive proposal weighting during training.
Mask as Supervision: Leveraging Unified Mask Information for Unsupervised 3D Pose Estimation
Yuchen Yang (Fudan University), Xiao Sun (Shanghai Artificial Intelligence Laboratory)
Pose EstimationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes an unsupervised 3D human pose estimation framework based on masks, which achieves direct prediction from a single image to 3D keypoints by using foreground masks as supervision signals.
Mask2Map: Vectorized HD Map Construction Using Bird's Eye View Segmentation Masks
Sehwan Choi (Hanyang University), Hongjae Shin (Hanyang University)
Autonomous DrivingTransformerImage
🎯 What it does: Propose Mask2Map, an end-to-end online HD map construction framework that generates instance-level queries through BEV segmentation masks and predicts vectorized map instances.
Masked Angle-Aware Autoencoder for Remote Sensing Images
Zhihao Li (Xidian University), Licheng Jiao (Xidian University)
ClassificationObject DetectionSegmentationRepresentation LearningTransformerAuto EncoderContrastive LearningImage
🎯 What it does: Designed an angle-aware pre-training framework MA3E based on Masked Autoencoder, which learns rotation-invariant features on remote sensing images using rotated cropping and OT loss.
Masked Generative Video-to-Audio Transformers with Enhanced Synchronicity
Santiago Pascual (Dolby Laboratories), Joan Serrà (Dolby Laboratories)
GenerationData SynthesisTransformerDiffusion modelAuto EncoderVideoMultimodalityAudio
🎯 What it does: Propose the MaskVAT model, which utilizes a high-quality audio encoder and a mask generation Transformer to generate full-bandwidth, semantically matching, and video-synchronized audio.
Masked Motion Prediction with Semantic Contrast for Point Cloud Sequence Learning
yuehui han, Jin Xie (Nanjing University)
ClassificationRepresentation LearningTransformerAuto EncoderContrastive LearningPoint Cloud
🎯 What it does: This paper proposes a self-supervised point cloud sequence representation learning framework called M2PSC based on Masked Autoencoder, integrating three pre-training tasks: motion trajectory prediction, semantic contrast, and appearance reconstruction.
Masked Video and Body-worn IMU Autoencoder for Egocentric Action Recognition
Mingfang Zhang (University of Tokyo), Yoichi Sato (University of Tokyo)
ClassificationRecognitionGraph Neural NetworkTransformerAuto EncoderContrastive LearningVideoMultimodalityTime Series
🎯 What it does: This paper proposes an end-to-end method for action recognition by fusing first-person view videos with body-worn IMU signals.
MasterWeaver: Taming Editability and Face Identity for Personalized Text-to-Image Generation
Yuxiang Wei (Harbin Institute of Technology), Wangmeng Zuo (Harbin Institute of Technology)
GenerationTransformerDiffusion modelImageText
🎯 What it does: Proposed a no-tuning method called MasterWeaver for generating personalized images with high identity fidelity and flexible editability.
Match-Stereo-Videos: Bidirectional Alignment for Consistent Dynamic Stereo Matching
Junpeng Jing (Imperial College London), Krystian Mikolajczyk (Imperial College London)
Depth EstimationConvolutional Neural NetworkRecurrent Neural NetworkOptical FlowVideo
🎯 What it does: A dynamic stereo video matching framework named BiDAStereo based on bidirectional alignment is proposed, enabling the generation of temporally consistent and high-accuracy disparity maps in dynamic scenes.
MathVerse: Does Your Multi-modal LLM Truly See the Diagrams in Visual Math Problems?
Renrui Zhang (CUHK MMLab), Hongsheng Li (Shanghai AI Laboratory)
TransformerLarge Language ModelVision Language ModelMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Constructed a new visual math problem benchmark called MathVerse to fairly evaluate the ability of multimodal large language models (MLLMs) in graphical mathematical reasoning; and proposed a Chain-of-Thought (CoT) evaluation strategy based on GPT-4 to perform fine-grained assessment of the model's intermediate reasoning quality.
MaxFusion: Plug&Play Multi-Modal Generation in Text-to-Image Diffusion Models
Nithin Gopalakrishnan Nair (Johns Hopkins University), Vishal Patel
GenerationDiffusion modelImageTextMultimodality
🎯 What it does: Propose a no-retraining multimodal generation method called MaxFusion, which leverages intermediate features from existing text-to-image diffusion models for fusion, enabling image generation under multiple conditions.
MaxMI: A Maximal Mutual Information Criterion for Manipulation Concept Discovery
Pei Zhou (University of Hong Kong), Yanchao Yang (University of Hong Kong)
Robotic IntelligenceTransformerSequential
🎯 What it does: This paper proposes a self-supervised framework for discovering key states (manipulation concepts), which identifies important physical states in demonstration trajectories using the MaxMI criterion and employs these states to train concept-guided manipulation strategies.
MC-PanDA: Mask Confidence for Panoptic Domain Adaptation
Ivan Martinović (University of Zagreb), Siniša Šegvić (University of Zagreb)
SegmentationDomain AdaptationTransformerImage
🎯 What it does: This paper proposes MC-PanDA, a domain adaptive panoptic segmentation method based on masked transformers, which significantly improves the model's robustness to noise in source domain pseudo labels through mask-level confidence regulation and confidence-based point sampling.
McGrids: Monte Carlo-Driven Adaptive Grids for Iso-Surface Extraction
Daxuan Ren (Nanyang Technological University), Jianfei Cai (Nanyang Technological University)
OptimizationComputational EfficiencyMesh
🎯 What it does: An adaptive grid (McGrids) implementation based on Monte Carlo sampling is proposed for isosurface extraction from neural implicit representations.
MedRAT: Unpaired Medical Report Generation via Auxiliary Tasks
Elad Hirsch (Technion Israel Institute of Technology), Ayellet Tal (Technion Israel Institute of Technology)
GenerationTransformerAuto EncoderContrastive LearningImageTextBiomedical Data
🎯 What it does: Proposed a model called MedRAT that generates medical reports from unpaired image and report data, achieving full report generation from X-ray images through autoencoding and multimodal alignment.
Meerkat: Audio-Visual Large Language Model for Grounding in Space and Time
Sanjoy Chowdhury (University of Maryland, College Park), Dinesh Manocha (University of Maryland, College Park)
TransformerLarge Language ModelSupervised Fine-TuningVision Language ModelContrastive LearningImageMultimodalityBenchmarkAudio
🎯 What it does: Developed Meerkat, a multimodal large language model capable of achieving fine-grained audio-visual alignment in both spatial and temporal dimensions, and proposed the AVFIT dataset along with the MeerkatBench unified evaluation framework.
MegaScenes: Scene-Level View Synthesis at Scale
Joseph Tung (Cornell University), Noah Snavely (Stanford University)
GenerationData SynthesisDepth EstimationDiffusion modelSimultaneous Localization and MappingImageBenchmark
🎯 What it does: Propose the MegaScenes dataset and use it to fine-tune single-view novel view synthesis models, enhancing scene-level perspective synthesis performance.
MemBN: Robust Test-Time Adaptation via Batch Norm with Statistics Memory
Juwon Kang (GENGENAI), Suha Kwak (Pohang University of Science and Technology)
Domain AdaptationImage
🎯 What it does: Proposes a memory-based batch normalization (MemBN) method that achieves robust test-time adaptation by maintaining the latest test batch statistics in each batch normalization layer.
Memory-Efficient Fine-Tuning for Quantized Diffusion Model
Hyogon Ryu (Korea Advanced Institute of Science and Technology), Hyunjung Shim (Korea Advanced Institute of Science and Technology)
GenerationComputational EfficiencySupervised Fine-TuningDiffusion modelMultimodality
🎯 What it does: This paper addresses the challenge of fine-tuning trillion-parameter quantized diffusion models and proposes TuneQDM, a memory-efficient fine-tuning method.
Merging and Splitting Diffusion Paths for Semantically Coherent Panoramas
Fabio Quattrini (University of Modena and Reggio Emilia), Rita Cucchiara (University of Modena and Reggio Emilia)
GenerationData SynthesisDiffusion modelImageText
🎯 What it does: Designed and implemented the MAD (Merge-Attend-Diffuse) operator to achieve multi-view interaction in the attention layer of pre-trained diffusion models, thereby generating panoramic images with both visual and semantic consistency.
Merlin: Empowering Multimodal LLMs with Foresight Minds
En Yu (Huazhong University Of Science And Technology), Wenbing Tao (Huazhong University Of Science And Technology)
Object TrackingTransformerLarge Language ModelVision Language ModelImageVideoTextMultimodalityChain-of-Thought
🎯 What it does: This paper designs and implements Merlin, a multimodal large language model capable of predicting future trajectories and performing inference based on single images or multi-frame videos.
MERLiN: Single-Shot Material Estimation and Relighting for Photometric Stereo
Ashish Tiwari (Indian Institute of Technology Gandhinagar), Shanmuganathan Raman (Indian Institute of Technology Gandhinagar)
Image TranslationGenerationConvolutional Neural NetworkImagePhysics Related
🎯 What it does: This paper proposes the MERLiN network, which can jointly estimate svBRDF parameters and achieve physically accurate relighting from a single image.
Mesh2NeRF: Direct Mesh Supervision for Neural Radiance Field Representation and Generation
Yujin Chen (Technical University of Munich), Matthias Niessner (Intel Labs)
GenerationDiffusion modelNeural Radiance FieldMesh
🎯 What it does: Propose the Mesh2NeRF method, which can directly parse a complete light radiance field from a textured 3D mesh and use it as a 3D supervisory signal for NeRF training and 3D generation tasks.
MeshAvatar: Learning High-quality Triangular Human Avatars from Multi-view Videos
Yushuo Chen (Tsinghua University), Yebin Liu (Tsinghua University)
GenerationPose EstimationConvolutional Neural NetworkVideoMesh
🎯 What it does: End-to-end learning of high-quality triangular human avatars from multi-view videos, resulting in editable and re-lightable explicit meshes and implicit material fields.
MeshFeat: Multi-Resolution Features for Neural Fields on Meshes
Mihir Mahajan (Technical University of Munich), Daniel Cremers (Technical University of Munich)
RestorationGenerationComputational EfficiencyNeural Radiance FieldImageMesh
🎯 What it does: Propose MeshFeat, a multi-resolution parameterized feature encoding for meshes, used in neural field representations;
MeshSegmenter: Zero-Shot Mesh Segmentation via Texture Synthesis
Ziming Zhong (Shanghaitech University), Shenghua Gao (Shanghaitech University)
SegmentationVision Language ModelDiffusion modelImageMesh
🎯 What it does: Propose the MeshSegmenter framework to achieve zero-shot grid semantic segmentation based on texture synthesis, by migrating 2D segmentation models to 3D grids and combining multi-view voting to complete unsupervised semantic segmentation.
MeshVPR: Citywide Visual Place Recognition Using 3D Meshes
Gabriele Berton (Politecnico di Torino), Carlo Masone (Politecnico di Torino)
RecognitionData SynthesisRetrievalDomain AdaptationSupervised Fine-TuningImageMesh
🎯 What it does: Propose MeshVPR, which utilizes synthetic images rendered from 3D meshes for city-level visual place recognition, and bridges the gap between real and synthetic domains through lightweight feature alignment.
MesonGS: Post-training Compression of 3D Gaussians via Efficient Attribute Transformation
Shuzhao Xie (Tsinghua University), Zhi Wang (Tsinghua University)
CompressionNeural Radiance FieldGaussian SplattingPoint Cloud
🎯 What it does: Proposed MesonGS, a codec for post-training compression of 3D Gaussians, which first prunes unimportant Gaussians through importance evaluation considering both view-dependent and view-independent factors, then compresses attributes using Euler angles instead of quaternions, RAHT transform, block quantization, and vector quantization, and finally employs LZ77 encoding with fine-grained quantization and finetuning to restore visual quality.
Meta-optimized Angular Margin Contrastive Framework for Video-Language Representation Learning
Thong Thanh Nguyen (National University of Singapore), Anh Tuan Luu (VinAI Research)
Representation LearningMeta LearningLarge Language ModelContrastive LearningVideoTextMultimodality
🎯 What it does: Propose a contrastive learning framework based on subtracted angular margin, and enhance the quality and generalization performance of video-language representation learning through a sample weighting function parameterized by MLP and text-enhanced data generated by large vision-language models.
Meta-Prompting for Automating Zero-shot Visual Recognition with LLMs
Muhammad Jehanzeb Mirza (TU Graz), Horst Possegger (TU Graz)
ClassificationRecognitionLarge Language ModelPrompt EngineeringVision Language ModelImageText
🎯 What it does: For zero-shot visual recognition, the MPVR framework is proposed, which automatically generates category-specific VLM prompts through meta-prompts, constructing a diverse set of prompts to achieve zero-shot classification.
MetaAT: Active Testing for Label-Efficient Evaluation of Dense Recognition Tasks
Sanbao Su (University of Connecticut), Liu Ren (Bosch Research North America)
Object DetectionSegmentationTransformerImage
🎯 What it does: This paper proposes MetaAT, an active testing method based on Vision Transformer, for evaluating the true loss of semantic segmentation and object detection models under limited annotation budgets.
MetaAug: Meta-Data Augmentation for Post-Training Quantization
Cuong Van Pham (Monash University), Thanh-Toan Do (Monash University)
OptimizationComputational EfficiencyMeta LearningConvolutional Neural NetworkImage
🎯 What it does: Designed and implemented a meta-learning based calibration data augmentation method to address overfitting in post-training quantization.
MetaCap: Meta-learning Priors from Multi-View Imagery for Sparse-view Human Performance Capture and Rendering
Guoxing Sun (Max Planck Institute for Informatics), Marc Habermann (Max Planck Institute for Informatics)
RestorationGenerationPose EstimationMeta LearningNeural Radiance FieldImageMesh
🎯 What it does: Propose the MetaCap method, which achieves high-quality human geometry and appearance recovery and free-viewpoint rendering under sparse or single-view inputs by pre-training implicit network weights through meta-learning.
MetaWeather: Few-Shot Weather-Degraded Image Restoration
Youngrae Kim (Korea Advanced Institute of Science and Technology), Dongman Lee (Korea Advanced Institute of Science and Technology)
RestorationMeta LearningConvolutional Neural NetworkTransformerImage
🎯 What it does: Propose MetaWeather, a weather denoising model that performs few-shot adaptation using a small number of examples, enabling image restoration under any unseen weather conditions.
MEVG : Multi-event Video Generation with Text-to-Video Models
Gyeongrok Oh (Korea University), Sangpil Kim (Korea University)
GenerationData SynthesisLarge Language ModelPrompt EngineeringDiffusion modelVideoText
🎯 What it does: Generate multi-event videos without requiring additional training or video data.
Mew: Multiplexed Immunofluorescence Image Analysis through an Efficient Multiplex Network
Sukwon Yun (University of North Carolina at Chapel Hill), Tianlong Chen (University of North Carolina at Chapel Hill)
ClassificationGraph Neural NetworkImageBiomedical Data
🎯 What it does: Construct a dual-layer multi-network (Voronoi geometry layer and cell type homogeneity layer), and use an expandable GNN combined with an attention mechanism to perform patient-level multi-fluorescence image phenotype prediction.
MICDrop: Masking Image and Depth Features via Complementary Dropout for Domain-Adaptive Semantic Segmentation
Linyan Yang (TU Munich), Luc Van Gool (ETH Zurich)
SegmentationDomain AdaptationTransformerImageBenchmark
🎯 What it does: Propose the MICDrop framework, which enhances unsupervised domain adaptation semantic segmentation performance by adopting complementary dropout and cross-modal feature fusion on RGB and depth features.
MIGS: Multi-Identity Gaussian Splatting via Tensor Decomposition
Aggelina Chatziagapi (Stony Brook University), Dimitris Samaras (Stony Brook University)
GenerationGaussian SplattingVideo
🎯 What it does: Learn a unified neural representation using monocular video, which can be shared across multiple human identities and enable real-time animation for multiple identities.
milliFlow: Scene Flow Estimation on mmWave Radar Point Cloud for Human Motion Sensing
Fangqiang Ding (University of Edinburgh), Chris Xiaoxuan Lu (University College London)
RecognitionSegmentationPose EstimationRecurrent Neural NetworkOptical FlowMultimodalityPoint Cloud
🎯 What it does: Propose milliFlow, which uses deep learning to estimate scene flow from millimeter-wave radar point clouds, providing fine-grained motion information for human motion perception and enhancing downstream tasks such as activity recognition, human parsing, and part tracking through this scene flow.
Mind the Interference: Retaining Pre-trained Knowledge in Parameter Efficient Continual Learning of Vision-Language Models
Longxiang Tang (Tsinghua University), Jiaya Jia (Hong Kong University)
Computational EfficiencyKnowledge DistillationTransformerSupervised Fine-TuningContrastive LearningMultimodality
🎯 What it does: Proposes the DIKI framework on visual-language models (e.g., CLIP) to address domain-class incremental learning (DCIL), leveraging residual knowledge integration to preserve pre-trained knowledge and significantly reduce forward forgetting.
MinD-3D: Reconstruct High-quality 3D objects in Human Brain
Jianxiong Gao (Fudan University), Yanwei Fu (Fudan University)
GenerationTransformerDiffusion modelVideoMeshBiomedical DataMagnetic Resonance Imaging
🎯 What it does: Studied the task of reconstructing 3D visual objects from fMRI signals
Mini-Splatting: Representing Scenes with a Constrained Number of Gaussians
Guangchi Fang (Hong Kong Polytechnic University), Bing Wang (Hong Kong Polytechnic University)
OptimizationComputational EfficiencyRepresentation LearningNeural Radiance FieldGaussian SplattingImage
🎯 What it does: Propose Mini-Splatting, which significantly reduces the number of Gaussians while maintaining rendering quality through Gaussian densification and simplification algorithms, achieving efficient scene representation.
Minimalist Vision with Freeform Pixels
Jeremy Klotz (Columbia University), Shree Nayar
OptimizationComputational EfficiencyImageVideo
🎯 What it does: This paper designs and implements a simplified camera using freeform pixels for lightweight visual tasks, such as indoor space monitoring, room illumination measurement, and traffic flow estimation.
MirrorGaussian: Reflecting 3D Gaussians for Reconstructing Mirror Reflections
Jiayue Liu (Tsinghua Shenzhen International Graduate School), Chun Yuan (Huawei Noah's Ark Lab)
GenerationGaussian SplattingSimultaneous Localization and MappingImagePoint Cloud
🎯 What it does: Proposes the MirrorGaussian method, leveraging 3D Gaussian Splatting and a dual rendering strategy to achieve high-quality real-time reconstruction and rendering of scenes containing mirrors, while supporting scene editing (e.g., inserting new mirrors and objects).
Mismatch Quest: Visual and Textual Feedback for Image-Text Misalignment
Brian Gordon (Tel Aviv University), Idan Szpektor (Google Research)
RetrievalAnomaly DetectionExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: Developed a multi-task model capable of automatically providing textual explanations and visual bounding box annotations when images and text are mismatched, and constructed a large-scale automatically generated training set (TV-Feedback) and a human-annotated evaluation set (SeeTRUE-Feedback).
Missing Modality Prediction for Unpaired Multimodal Learning via Joint Embedding of Unimodal Models
Taesup Kim (Seoul National University), Donggeun Kim (Seoul National University)
ClassificationRepresentation LearningTransformerSupervised Fine-TuningPrompt EngineeringContrastive LearningMultimodalityBenchmark
🎯 What it does: This study addresses the problem of missing modalities in unpaired multimodal data by proposing a parameter-efficient fine-tuning and read-only prompt learning framework based on pre-trained unimodal encoders. The framework can predict missing modality embeddings during inference and fuse them with existing modalities to enhance multimodal classification performance in scenarios with missing modalities.
Mitigating Background Shift in Class-Incremental Semantic Segmentation
Gilhan Park (Sungkyunkwan University), Jae-Pil Heo (Sungkyunkwan University)
SegmentationKnowledge DistillationTransformerImage
🎯 What it does: Propose a background class separation framework that reduces catastrophic forgetting caused by background drift in class-incremental semantic segmentation through selective pseudo-labels, adaptive feature distillation, label-guided output distillation, and orthogonal objectives.
Mitigating Perspective Distortion-induced Shape Ambiguity in Image Crops
Aditya Prakash (University of Illinois Urbana Champaign), Saurabh Gupta (University of Illinois Urbana Champaign)
Object DetectionPose EstimationDepth EstimationConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes a camera intrinsic parameter-based position information encoding (KPE), which is embedded into a deep learning model with image cropping input to reduce shape ambiguity caused by perspective distortion.
MixDQ: Memory-Efficient Few-Step Text-to-Image Diffusion Models with Metric-Decoupled Mixed Precision Quantization
Tianchen Zhao (Tsinghua University), Yu Wang (Tsinghua University)
GenerationDiffusion modelImageText
🎯 What it does: For few-step text-to-image diffusion models, we propose a hybrid precision post-training quantization method called MixDQ, which includes BOS-aware text embedding quantization, metric-decoupled sensitivity analysis, and integer programming bit-width allocation.
Mixture of Efficient Diffusion Experts Through Automatic Interval and Sub-Network Selection
Alireza Ganjdanesh (University of Maryland College Park), Heng Huang (University of Maryland College Park)
GenerationComputational EfficiencyMixture of ExpertsDiffusion modelImage
🎯 What it does: Prune the pre-trained Latent Diffusion Model into a Mixture of Experts (MoE) model, significantly reducing sampling costs through automatic interval partitioning and subnetwork selection.
ML-SemReg: Boosting Point Cloud Registration with Multi-level Semantic Consistency
Shaocheng Yan (Wuhan University), Jiayuan Li (Wuhan University)
Pose EstimationAutonomous DrivingPoint Cloud
🎯 What it does: Propose the ML-SemReg framework, which enhances the matching quality of point cloud registration through multi-layer semantic consistency
MLPHand: Real Time Multi-View 3D Hand Reconstruction via MLP Modeling
Jian Yang (Chinese Academy of Science), Zhaoxin Fan (Beihang University)
Pose EstimationComputational EfficiencyImageMesh
🎯 What it does: Propose a real-time multi-view single-hand 3D reconstruction method called MLPHand, utilizing a lightweight Skeleton2Mesh model and multi-view geometric feature fusion to achieve high frame rate inference.
MM-SafetyBench: A Benchmark for Safety Evaluation of Multimodal Large Language Models
Xin Liu (East China Normal University), Yu Qiao (Shanghai AI Laboratory)
Safty and PrivacyTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: This study proposes MM-SafetyBench, a benchmark for evaluating the safety of multimodal large language models (LLMs), constructing 5,040 text-image pairs across 13 categories of harmful scenarios, and demonstrating the vulnerability of multimodal LLMs through visual prompt attacks.
MMBENCH: Is Your Multi-Modal Model an All-around Player?
Yuan Liu (Shanghai AI Laboratory), Dahua Lin (Shanghai AI Laboratory)
Large Language ModelVision Language ModelMultimodalityBenchmark
🎯 What it does: Designed and constructed MMBench, a bilingual multimodal model benchmark consisting of over 3,000 multiple-choice questions, to systematically evaluate the perception and reasoning capabilities of vision-language models.
MMEarth: Exploring Multi-Modal Pretext Tasks For Geospatial Representation Learning
Vishal Nedungadi (University of Copenhagen), Nico Lang (University of Copenhagen)
ClassificationSegmentationRepresentation LearningConvolutional Neural NetworkAuto EncoderMultimodality
🎯 What it does: Investigated the effectiveness of multimodal pre-training tasks in geospatial representation learning, proposed MP-MAE, and constructed the MMEarth dataset.
MMVR: Millimeter-wave Multi-View Radar Dataset and Benchmark for Indoor Perception
Mohammad Mahbubur Rahman (Mitsubishi Electric Research Laboratories), Petros Boufounos (Mitsubishi Electric Research Laboratories)
Object DetectionSegmentationPose EstimationConvolutional Neural NetworkTransformerImageBenchmark
🎯 What it does: This study collects and publicly releases a millimeter-wave multi-view radar (MMVR) dataset containing two scene protocols, automatically generating detection, pose, and segmentation annotations using a synchronized RGB-D camera, and conducts benchmark evaluations for three categories of indoor perception tasks on this dataset.
MO-EMT-NAS: Multi-Objective Continuous Transfer of Architectural Knowledge Between Tasks from Different Datasets
PENG LIAO (East China University of Science and Technology), Wenli Du (East China University of Science and Technology)
Neural Architecture SearchImageBiomedical Data
🎯 What it does: Propose a multi-objective multi-task evolutionary NAS framework called MO-EMT-NAS, which can share and transfer architectural knowledge between multi-tasks across different datasets while optimizing model accuracy and size.
MoAI: Mixture of All Intelligence for Large Language and Vision Models
Byung-Kwan Lee (Korea Advanced Institute of Science and Technology), Yong Man Ro (Korea Advanced Institute of Science and Technology)
CompressionTransformerLarge Language ModelMixture of ExpertsVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes a novel large language and vision model called MoAI, which utilizes auxiliary visual information generated by external computer vision models (segmentation, detection, scene graph, OCR). After being compressed by MoAI-Compressor, the auxiliary visual information is fused with visual features and language features through MoAI-Mixer (based on Mixture of Experts) to enhance performance on zero-shot vision-language tasks.
MobileDiffusion: Instant Text-to-Image Generation on Mobile Devices
Yang Zhao (University at Buffalo), Tingbo Hou
GenerationConvolutional Neural NetworkTransformerDiffusion modelAuto EncoderGenerative Adversarial NetworkImageTextMultimodality
🎯 What it does: This paper proposes MobileDiffusion, an ultra-lightweight text-to-image diffusion model tailored for mobile devices, achieving instant generation through dual optimization of architecture and sampling.
MobileNetV4: Universal Models for the Mobile Ecosystem
Danfeng Qin (Google), Andrew Howard (Google)
ClassificationObject DetectionComputational EfficiencyKnowledge DistillationNeural Architecture SearchConvolutional Neural NetworkTransformerReinforcement LearningImage
🎯 What it does: Proposes the MobileNetV4 architecture to achieve a general-purpose, efficient model for mobile ecosystems.
Möbius Transform for Mitigating Perspective Distortions in Representation Learning
Prakash Chandra Chhipa (Luleå Tekniska Universitet), Mubarak Shah (University of Central Florida)
Object DetectionData SynthesisRepresentation LearningConvolutional Neural NetworkTransformerContrastive LearningImage
🎯 What it does: The study synthesizes perspective distortions via Möbius transformation, proposes the MPD method, and evaluates model robustness on the ImageNet-PD dataset.
MOD-UV: Learning Mobile Object Detectors from Unlabeled Videos
Yihong Sun (Cornell University), Bharath Hariharan (Cornell University)
Object DetectionConvolutional Neural NetworkVideo
🎯 What it does: Leverage motion information in unlabeled videos to self-learn a moving object detector
Modality Translation for Object Detection Adaptation without forgetting prior knowledge
Heitor Rapela Medeiros (ETS Montreal), Marco Pedersoli (ETS Montreal)
Image TranslationObject DetectionDomain AdaptationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: Propose ModTr, which converts IR input into RGB-like images through a residual translation network, enabling pre-trained RGB object detectors to directly process new modalities while retaining their original weights.
Model Breadcrumbs: Scaling Multi-Task Model Merging with Sparse Masks
MohammadReza Davari (Concordia University), Eugene Belilovsky (Concordia University)
TransformerImageText
🎯 What it does: Propose the Model Breadcrumbs method, which merges multiple fine-tuned models based on the same base model through sparse masks to construct a multi-task model without requiring further training.
Model Stock: All we need is just a few fine-tuned models
Dong-Hwan Jang (NAVER AI Lab), Dongyoon Han (NAVER AI Lab)
ClassificationKnowledge DistillationConvolutional Neural NetworkSupervised Fine-TuningMixture of ExpertsImage
🎯 What it does: Proposed the Model Stock method, which constructs geometric relationships between a few (e.g., two) fine-tuned models and pre-trained models in the weight space, obtaining fusion weights close to the center to achieve high-quality models;
Modeling and Driving Human Body Soundfields through Acoustic Primitives
Chao Huang (University of Rochester), Alexander Richard (Codec Avatars Lab, Meta)
GenerationMultimodalityAudio
🎯 What it does: The study proposes a 3D human sound field rendering framework based on acoustic primitives, which can predict multiple low-order acoustic primitives from human poses and head-mounted microphone audio, and real-time render full spatial audio.
Modeling Label Correlations with Latent Context for Multi-Label Recognition
Zhaomin Chen (Key Laboratory of Intelligent Informatics for Safety and Emergency of Zhejiang Province, Wenzhou University), Guodao Zhang (Hangzhou Dianzi University)
RecognitionTransformerImage
🎯 What it does: This paper proposes a multi-label image recognition framework that models label correlations by leveraging potential contextual information, comprising three modules: multi-layer feature fusion, cross-attention embedding, and label correlation capture;
Modelling Competitive Behaviors in Autonomous Driving Under Generative World Model
Guanren Qiao (Chinese University of Hong Kong Shenzhen), Rongxiao Qu (Chinese University of Hong Kong Shenzhen)
Autonomous DrivingReinforcement LearningWorld ModelSequential
🎯 What it does: This paper studies the use of coarse correlated equilibrium (CCE) from game theory to model competitive behaviors of vehicles in autonomous driving scenarios under generative world models, thereby generating safety-critical events.
MoE-DiffIR: Task-customized Diffusion Priors for Universal Compressed Image Restoration
Yulin Ren (University of Science and Technology of China), Zhibo Chen (University of Science and Technology of China)
RestorationPrompt EngineeringMixture of ExpertsVision Language ModelDiffusion modelImageBenchmark
🎯 What it does: Propose MoE-DiffIR, a general-purpose compressed image restoration framework based on Stable Diffusion, which dynamically customizes priors for different compression tasks through Mixture-of-Experts prompts and activates SD's cross-modal priors using a visual-to-text adapter;
MoEAD: A Parameter-efficient Model for Multi-class Anomaly Detection
Shiyuan Meng (Zhejiang University), Shibo He (Zhejiang University)
Anomaly DetectionTransformerMixture of ExpertsImage
🎯 What it does: Propose a parameter-efficient ViT-style model MoEAD for unsupervised unified multi-class anomaly detection, adopting Mixture of Experts (MoE) technology to share parameters in the Transformer decoder and achieve finer-grained expert selection through SMoE layers, combined with learnable query embeddings, neighborhood mask attention, and auxiliary loss to balance expert load, ultimately achieving detection and localization of anomalous images;
MOFA-Video: Controllable Image Animation via Generative Motion Field Adaptions in Frozen Image-to-Video Diffusion Model
Muyao Niu (University of Tokyo), Yinqiang Zheng (Tencent AI Lab)
GenerationData SynthesisDiffusion modelOptical FlowImageVideoMultimodality
🎯 What it does: Propose MOFA-Video, which utilizes a frozen Stable Video Diffusion model and multi-domain motion field adapters (MOFA-Adapters) to achieve controllable image animation based on sparse control signals (e.g., hand-drawn trajectories, facial key points, etc.);
MoMA: Multimodal LLM Adapter for Fast Personalized Image Generation
Kunpeng Song (ByteDance), Xiao Yang (ByteDance)
GenerationTransformerLarge Language ModelPrompt EngineeringVision Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: MoMA proposes a zero-shot tuning, open-source personalized image generation framework based on a multi-modal LLM adapter, achieving high-quality personalized generation in Stable Diffusion through cross-attention and self-attention mechanisms using a single reference image and text prompts.
Momentum Auxiliary Network for Supervised Local Learning
Junhao Su (Southeast University), Chenyang Si (Nanyang Technological University)
ClassificationConvolutional Neural NetworkImage
🎯 What it does: Designed and verified the Momentum Auxiliary Network (MAN) to enhance inter-block information exchange in supervised local learning, thereby improving model accuracy and reducing memory consumption.
Mono-ViFI: A Unified Learning Framework for Self-supervised Single- and Multi-frame Monocular Depth Estimation
Jinfeng Liu (vivo Mobile Communication Co., Ltd), Jinwei Chen (vivo Mobile Communication Co., Ltd)
Depth EstimationConvolutional Neural NetworkOptical FlowVideoBenchmark
🎯 What it does: Propose a unified self-supervised single-frame and multi-frame monocular depth estimation framework called Mono-ViFI, which utilizes video frame interpolation (VFI) to achieve temporal domain data augmentation and designs a VFI-assisted multi-frame feature alignment and fusion module; on this basis, spatial affine transformation enhancement and triple depth consistency loss are added for mutual distillation and scale self-consistency.
Monocular Occupancy Prediction for Scalable Indoor Scenes
Hongxiao Yu (University of Chinese Academy of Sciences), Zhaoxiang Zhang (University of Chinese Academy of Sciences)
SegmentationDepth EstimationConvolutional Neural NetworkImageBenchmark
🎯 What it does: Propose the ISO model to predict 3D occupied voxels in indoor scenes using monocular images.
MonoTTA: Fully Test-Time Adaptation for Monocular 3D Object Detection
Hongbin Lin (FNii-Shenzhen), Zhen Li (CUHK-Shenzhen)
Object DetectionDomain AdaptationAutonomous DrivingImage
🎯 What it does: Proposed a new monocular 3D object detection method called MonoTTA, aiming to address issues caused by data distribution shifts during testing through fully test-time adaptation (Fully Test-Time Adaptation).
MonoWAD: Weather-Adaptive Diffusion Model for Robust Monocular 3D Object Detection
Youngmin Oh (Kyung Hee University), Jung Uk Kim (Kyung Hee University)
Object DetectionAutonomous DrivingTransformerDiffusion modelImage
🎯 What it does: Propose MonoWAD, a monocular 3D object detection framework that maintains robustness under various weather conditions
MONTAGE: Monitoring Training for Attribution of Generative Diffusion Models
Jonathan Brokman (Fujitsu Research of Europe), Hisashi Kojima (Fujitsu Limited)
GenerationExplainability and InterpretabilityTransformerVision Language ModelDiffusion modelImage
🎯 What it does: Monitor the generation process of custom diffusion models, construct a data attribution table, and train a distance metric learning model to predict the contribution of training samples on unmonitored generated images.
Motion and Structure from Event-based Normal Flow
Zhongyang Ren, Yi Zhou (Hunan University)
Pose EstimationDepth EstimationOptical Flow
🎯 What it does: Investigated how to utilize the normal flow of event cameras to estimate camera motion and scene geometry.
Motion Aware Event Representation-driven Image Deblurring
Zhijing Sun (University of Science and Technology of China), Zheng-Jun Zha (University of Science and Technology of China)
RestorationConvolutional Neural NetworkRecurrent Neural NetworkImageMultimodality
🎯 What it does: This paper proposes a deviation accumulation event representation method based on event cameras, and designs an end-to-end image deblurring network that integrates a Recursive Motion Extraction (RME) module and a Bidirectional Feature Alignment Fusion (FAF) module.
Motion Keyframe Interpolation for Any Human Skeleton using Point Cloud-based Human Motion Data Homogenisation
Clinton A Mo (University of Sydney), Zhiyong Wang (Meta Reality Labs)
Pose EstimationRepresentation LearningGraph Neural NetworkTransformerPoint Cloud
🎯 What it does: Proposes an unsupervised motion representation learning method based on point clouds, PC-MRL, enabling keyframe interpolation for any skeleton.
Motion Mamba: Efficient and Long Sequence Motion Generation
Zeyu Zhang (Monash University), Hao Tang (Peking University)
GenerationDiffusion modelAuto EncoderTextSequential
🎯 What it does: Propose Motion Mamba—a text-to-motion generation framework based on the selective state space model (Mamba), combining hierarchical temporal Mamba (HTM) and bidirectional spatial Mamba (BSM) modules to achieve efficient long-sequence motion generation;
Motion-Guided Latent Diffusion for Temporally Consistent Real-world Video Super-resolution
Xi Yang (Hong Kong Polytechnic University), Lei Zhang (Hong Kong Polytechnic University)
RestorationSuper ResolutionConvolutional Neural NetworkDiffusion modelGenerative Adversarial NetworkOptical FlowVideo
🎯 What it does: This paper proposes a method for real-scene video super-resolution using a motion-guided latent diffusion model (MGLD).