ICCV 2023 Papers — Page 13
IEEE/CVF International Conference on Computer Vision · 2156 papers
MixSynthFormer: A Transformer Encoder-like Structure with Mixed Synthetic Self-attention for Efficient Human Pose Estimation
Yuran Sun (University of Hong Kong), Chuan Wu (University of Hong Kong)
Pose EstimationTransformerVideoMesh
🎯 What it does: This paper proposes MixSynthFormer, a lightweight transformer-encoder structure that achieves human pose estimation, body mesh recovery, and motion prediction in videos through a mixed spatial and temporal synthesized attention mechanism, utilizing a recover-refine pipeline on sampled frames.
MMST-ViT: Climate Change-aware Crop Yield Prediction via Multi-Modal Spatial-Temporal Vision Transformer
Fudong Lin (University of Delaware), Nian-Feng Tzeng (University of Louisiana at Lafayette)
TransformerContrastive LearningMultimodalityTime SeriesAgriculture Related
🎯 What it does: Designed and implemented a multi-modal spatial-temporal visual Transformer (MMST-ViT) model for crop yield prediction at the county level in the United States.
MMVP: Motion-Matrix-Based Video Prediction
Yiqi Zhong (University of Southern California), Ulrich Neumann (University of Southern California)
GenerationData SynthesisOptimizationComputational EfficiencyConvolutional Neural NetworkOptical FlowVideo
🎯 What it does: A two-stream video prediction framework named MMVP is proposed, which first decouples motion information and appearance information through a motion matrix, and then combines the predicted motion matrix with multi-scale appearance features using matrix multiplication to generate future frames.
MODA: Mapping-Once Audio-driven Portrait Animation with Dual Attentions
Yunfei Liu (International Digital Economy Academy), Yu Li (International Digital Economy Academy)
GenerationData SynthesisTransformerGenerative Adversarial NetworkOptical FlowVideoMultimodalityAudio
🎯 What it does: A three-stage audio-driven portrait animation system is designed, which includes a one-shot mapping network (MODA), a facial synthesis network, and a renderer with temporal position encoding, achieving multi-modal high-fidelity voice-driven portrait video synthesis.
Modality Unifying Network for Visible-Infrared Person Re-Identification
Hao Yu (Nanjing University of Information Science and Technology), Guoying Zhao (University of Oulu)
RecognitionRetrievalConvolutional Neural NetworkContrastive LearningImageMultimodality
🎯 What it does: This paper proposes the Modality Unifying Network (MUN), which achieves visible-infrared person re-identification by generating auxiliary modalities and jointly learning distinguishable features.
Model Calibration in Dense Classification with Adaptive Label Perturbation
Jiawei Liu (Australian National University), Nick Barnes (Australian National University)
ClassificationObject DetectionSegmentationConvolutional Neural NetworkImage
🎯 What it does: Proposes Adaptive Stochastic Label Perturbation (ASLP) and Self-Correcting Binary Cross-Entropy (SC-BCE) loss to enhance the probability calibration of dense binary classification models.
ModelGiF: Gradient Fields for Model Functional Distance
Jie Song (Zhejiang University), Mingli Song (Zhejiang University)
OptimizationSafty and PrivacyConvolutional Neural NetworkImage
🎯 What it does: Proposes the Model Gradient Field (ModelGiF) method, using the gradient field as a unified representation to measure the functional similarity of pre-trained models, and applies it to task similarity assessment, intellectual property protection, and model forgetting verification.
Modeling the Relative Visual Tempo for Self-supervised Skeleton-based Action Recognition
Yisheng Zhu (Nanjing University of Posts and Telecommunications), Guangcan Liu (Southeast University)
RecognitionRepresentation LearningGraph Neural NetworkContrastive LearningVideo
🎯 What it does: A skeleton action recognition framework based on self-supervised contrastive learning, RVTCLR, and its improved version RVTCLR+ are proposed. The framework jointly trains skeleton features through relative visual tempo learning and appearance consistency tasks, and further enhances high-level semantic representation by adding a Distribution-Consistency branch.
MolGrapher: Graph-based Visual Recognition of Chemical Structures
Lucas Morin (IBM Research), Fisher Yu (ETH Zurich)
ClassificationRecognitionGraph Neural NetworkImageBenchmark
🎯 What it does: Designed and implemented MolGrapher, achieving complete recognition and reconstruction of chemical structure images through keypoint detection, hypergraph construction, and graph neural network classification.
Moment Detection in Long Tutorial Videos
Ioana Croitoru (Adobe Research), Trung Bui (Adobe Research)
RecognitionObject DetectionTransformerLarge Language ModelVideoText
🎯 What it does: A method for moment detection based on long tutorial videos is proposed, and two new datasets (Behance Moment Detection, YouTube Chapters) are constructed.
Monocular 3D Object Detection with Bounding Box Denoising in 3D by Perceiver
Xianpeng Liu (North Carolina State University), Tianfu Wu (North Carolina State University)
Object DetectionAutonomous DrivingTransformerPoint Cloud
🎯 What it does: A two-stage monocular 3D object detection framework called MonoXiver is proposed, which first generates 3D box candidates using a low-level detector and then performs 3D→2D denoising verification on these candidates to achieve refined 3D center localization.
MonoDETR: Depth-guided Transformer for Monocular 3D Object Detection
Renrui Zhang (Chinese University of Hong Kong), Peng Gao (Shanghai Artificial Intelligence Laboratory)
Object DetectionAutonomous DrivingTransformerImage
🎯 What it does: A monocular 3D object detection framework called MonoDETR based on Transformer is proposed, which utilizes the predicted foreground depth map to guide the detection process and achieve global feature aggregation.
MonoNeRD: NeRF-like Representations for Monocular 3D Object Detection
Junkai Xu (Zhejiang University), Deng Cai (Zhejiang University)
Object DetectionAutonomous DrivingNeural Radiance FieldPoint CloudBenchmark
🎯 What it does: This paper proposes MonoNeRD, a monocular 3D object detection framework that utilizes NeRF-like implicit 3D representations (SDF + volume rendering).
MonoNeRF: Learning a Generalizable Dynamic Radiance Field from Monocular Videos
Fengrui Tian (Xi'an Jiaotong University), Yueqi Duan (Tsinghua University)
GenerationData SynthesisConvolutional Neural NetworkNeural Radiance FieldOptical FlowVideoOrdinary Differential Equation
🎯 What it does: This paper proposes MonoNeRF, which can learn transferable dynamic radiance fields from multiple segments of monocular video, enabling novel view synthesis, frame interpolation, and scene editing.
Monte Carlo Linear Clustering with Single-Point Supervision is Enough for Infrared Small Target Detection
Boyang Li (Aviation University of Air Force), Yulan Guo (National University of Defense Technology)
Object DetectionImage
🎯 What it does: A method for infrared small target detection is proposed, which uses only single-point annotations. It generates pseudo pixel-level masks through Monte Carlo linear clustering, thereby transforming a fully supervised network into a weakly supervised network.
MoreauGrad: Sparse and Robust Interpretation of Neural Networks via Moreau Envelope
Jingwei Zhang (Chinese University of Hong Kong), Farzan Farnia (Chinese University of Hong Kong)
Explainability and InterpretabilityConvolutional Neural NetworkImage
🎯 What it does: A Moreau envelope-based explanation framework called MoreauGrad is proposed, which can generate smooth and sparse neural network explanations.
MosaiQ: Quantum Generative Adversarial Networks for Image Generation on NISQ Computers
Daniel Silver (Northeastern University), Devesh Tiwari (Northeastern University)
GenerationGenerative Adversarial NetworkImage
🎯 What it does: A quantum generative adversarial network called MosaiQ, suitable for NISQ computers, has been designed and implemented for high-quality image generation.
MOSE: A New Dataset for Video Object Segmentation in Complex Scenes
Henghui Ding (Nanyang Technological University), Song Bai
Object DetectionSegmentationVideoBenchmark
🎯 What it does: A MOSE dataset has been constructed and released, focusing on the performance of video object segmentation in complex scenes.
Most Important Person-Guided Dual-Branch Cross-Patch Attention for Group Affect Recognition
Hongxia Xie (National Yang Ming Chiao Tung University), Wen-Huang Cheng (National Taiwan University)
RecognitionTransformerImage
🎯 What it does: This paper proposes a Dual-Branch Cross-Patch Attention Transformer (DCAT) for group emotion recognition, utilizing the Most Important Person (MIP) and global images.
MOST: Multiple Object Localization with Self-Supervised Transformers for Object Discovery
Sai Saketh Rambhatla (Meta), Abhinav Shrivastava (University of Maryland)
Object DetectionTransformerContrastive LearningImage
🎯 What it does: A multi-object localization method called MOST based on self-supervised Transformer is proposed, which can locate multiple foreground objects in images without labels and is used for subsequent tasks such as detection and discovery.
MoTIF: Learning Motion Trajectories with Local Implicit Neural Functions for Continuous Space-Time Video Super-Resolution
Yi-Hsin Chen (National Yang Ming Chiao Tung University), Wen-Hsiao Peng (National Yang Ming Chiao Tung University)
RestorationSuper ResolutionOptical FlowVideo
🎯 What it does: This paper proposes a continuous spatiotemporal video super-resolution method called MoTIF, based on spatiotemporal local implicit neural functions, which reconstructs high-resolution frames using forward motion trajectories and reliability-aware splatting.
Motion-Guided Masking for Spatiotemporal Representation Learning
David Fan (Amazon Prime Video), Xinyu Li (Amazon Prime Video)
Representation LearningTransformerAuto EncoderVideo
🎯 What it does: A three-dimensional occlusion algorithm MGM based on motion vectors is designed to enhance the effectiveness of video self-supervised learning.
MotionBERT: A Unified Perspective on Learning Human Motion Representations
Wentao Zhu (Peking University), Yizhou Wang (Peking University)
Pose EstimationTransformerSupervised Fine-TuningVideoPoint CloudMesh
🎯 What it does: Learned motion representation through a pre-trained 2D-to-3D recovery task, and then fine-tuned to complete three types of tasks: 3D pose estimation, action recognition, and mesh recovery, achieving multiple SOTA results.
MotionDeltaCNN: Sparse CNN Inference of Frame Differences in Moving Camera Videos with Spherical Buffers and Padded Convolutions
Mathias Parger (Graz University of Technology), Markus Steinberger (Graz University of Technology)
SegmentationPose EstimationComputational EfficiencyConvolutional Neural NetworkVideo
🎯 What it does: Proposes the MotionDeltaCNN framework, which utilizes sparse frame differences and spherical buffering to achieve efficient CNN inference for videos from moving cameras.
MotionLM: Multi-Agent Motion Forecasting as Language Modeling
Ari Seff (Waymo), Benjamin Sapp (Waymo)
Autonomous DrivingTransformerMultimodality
🎯 What it does: This paper transforms the multi-agent motion prediction task into a language modeling problem, using discrete action tokens and generating joint trajectories through autoregressive decoding.
Movement Enhancement toward Multi-Scale Video Feature Representation for Temporal Action Detection
Zixuan Zhao (Shanghai Jiao Tong University), Xu Zhao (Shanghai Jiao Tong University)
RecognitionObject DetectionConvolutional Neural NetworkRecurrent Neural NetworkVideo
🎯 What it does: This paper proposes MENet (Movement Enhance Network) for temporal action detection, addressing the issues of action features being overwhelmed by the background and the difficulty of multi-scale action localization.
MPCViT: Searching for Accurate and Efficient MPC-Friendly Vision Transformer with Heterogeneous Attention
Wenxuan Zeng (Peking University), Ru Huang (Peking University)
Computational EfficiencyKnowledge DistillationNeural Architecture SearchTransformerImage
🎯 What it does: A visual Transformer model for secure multi-party computation, MPCViT, is proposed, achieving low-latency and high-accuracy inference through heterogeneous attention and NAS, and further extended to MPCViT+.
MPI-Flow: Learning Realistic Optical Flow with Multiplane Images
Yingping Liang (Beijing Institute of Technology), Ying Fu (Beijing Institute of Technology)
GenerationData SynthesisOptical FlowImage
🎯 What it does: Constructing Multi-Plane Images (MPI) from single-view images, generating realistic perspective images through volume rendering, and calculating the corresponding optical flow to build an optical flow dataset from real images.
MRM: Masked Relation Modeling for Medical Image Pre-Training with Genetics
Qiushi Yang (City University of Hong Kong), Yixuan Yuan (Chinese University of Hong Kong)
ClassificationSegmentationRepresentation LearningSpiking Neural NetworkTransformerContrastive LearningMultimodalityBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper proposes a self-supervised pre-training framework for medical images and genetic data—Masked Relation Modeling (MRM), which enhances feature representation while maintaining disease-related semantics through a dual mechanism of relation masking and relation matching.
MRN: Multiplexed Routing Network for Incremental Multilingual Text Recognition
Tianlun Zheng (Fudan University), Yu-Gang Jiang (Fudan University)
RecognitionRecurrent Neural NetworkTransformerText
🎯 What it does: Proposes the Incremental Multilingual Text Recognition (IMLTR) task and designs a Multi-Route Network (MRN) to address the replay imbalance issue.
MSI: Maximize Support-Set Information for Few-Shot Segmentation
Seonghyeon Moon (Rutgers University), Mubbasir Kapadia (Rutgers University)
SegmentationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a novel few-shot segmentation method called MSI (Maximize Support-Set Information), which maximizes support set information by simultaneously utilizing occluded support images (STF) and complete support images (SIF), thereby addressing the performance bottlenecks of traditional methods in scenarios involving target occlusion, small sizes, incomplete boundaries, and background interference.
MST-compression: Compressing and Accelerating Binary Neural Networks with Minimum Spanning Tree
Quang Hieu Vo (Kyung Hee University), Choong Seon Hong (Kyung Hee University)
CompressionComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a compression and acceleration scheme for binary neural networks (BNN) based on the Minimum Spanning Tree (MST), utilizing MST to rearrange the computation order of convolution output channels, reducing XNOR operations and further minimizing MST distance during the learning phase, thereby decreasing the number of parameters and computational load, while also implementing a corresponding hardware accelerator.
MULLER: Multilayer Laplacian Resizer for Vision
Zhengzhong Tu (Google Research), Hossein Talebi (Google Research)
Image TranslationObject DetectionSegmentationConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes a minimalist learnable image scaling module called MULLER, which achieves size transformation while preserving details through multi-layer Laplacian decomposition. It is trained jointly as a front layer with existing visual models (such as MaxViT, ResNet, EfficientNet, MobileNet, etc.) to enhance the performance of various visual tasks.
Multi-body Depth and Camera Pose Estimation from Multiple Views
Andrea Porfiri Dal Cin (Politecnico di Milano), Luca Magri (Politecnico di Milano)
Pose EstimationDepth EstimationAutonomous DrivingConvolutional Neural NetworkSimultaneous Localization and MappingOptical FlowImagePoint Cloud
🎯 What it does: A framework for depth and camera pose estimation in multi-body dynamic scenes is proposed, capable of simultaneously handling the motion of multiple rigid bodies and outputting a unified scale dense depth map and accurate poses.
Multi-Directional Subspace Editing in Style-Space
Chen Naveh (Reichman University)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: Proposes the MDSE framework, which constructs mutually orthogonal multi-dimensional subspaces in the latent space of StyleGAN for multi-directional editing of a single facial attribute;
Multi-Event Video-Text Retrieval
Gengyuan Zhang (Ludwig Maximilian University of Munich), Volker Tresp (Ludwig Maximilian University of Munich)
RetrievalTransformerVision Language ModelContrastive LearningVideoText
🎯 What it does: This paper studies the multi-event video-text retrieval (MeVTR) task and proposes the CLIP-based MeRetriever model.
Multi-Frequency Representation Enhancement with Privilege Information for Video Super-Resolution
Fei Li (China Agricultural University), Zhenbo Li (China Agricultural University)
RestorationSuper ResolutionConvolutional Neural NetworkVideo
🎯 What it does: Proposes a multi-frequency representation enhancement module MFE based on the frequency domain and privileged training PT to improve video super-resolution models.
Multi-grained Temporal Prototype Learning for Few-shot Video Object Segmentation
Nian Liu (Mohamed bin Zayed University of Artificial Intelligence), Fahad Shahbaz Khan (Linkoping University)
SegmentationTransformerVideo
🎯 What it does: A few-shot video object segmentation method based on multi-granularity temporal prototypes, VIPMT, is proposed.
Multi-granularity Interaction Simulation for Unsupervised Interactive Segmentation
Kehan Li (Peking University), Jie Chen (Peking University)
SegmentationTransformerContrastive LearningImage
🎯 What it does: A multi-granularity interactive simulation (MIS) framework is proposed, which automatically generates semantically consistent regions from unlabeled images to train interactive segmentation models.
Multi-interactive Feature Learning and a Full-time Multi-modality Benchmark for Image Fusion and Segmentation
Jinyuan Liu (Dalian University of Technology), Xin Fan (Dalian University of Technology)
SegmentationConvolutional Neural NetworkTransformerImageMultimodalityBenchmark
🎯 What it does: This study investigates the joint task of multimodal image fusion and semantic segmentation, proposing the SegMiF architecture.
Multi-label Affordance Mapping from Egocentric Vision
Lorenzo Mur-Labadia (Universidad de Zaragoza), Ruben Martinez-Cantin (Universidad de Zaragoza)
Object DetectionSegmentationConvolutional Neural NetworkSupervised Fine-TuningSimultaneous Localization and MappingVideo
🎯 What it does: In first-person perspective videos, multi-label pixel-level affordance segmentation is achieved through automated interactive 3D geometry playback, and the largest EPIC-Aff dataset is constructed based on this.
Multi-Label Knowledge Distillation
Penghui Yang (Nanjing University of Aeronautics and Astronautics), Sheng-Jun Huang (Nanjing University of Aeronautics and Astronautics)
Knowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: This paper studies multi-label knowledge distillation methods and proposes the L2D framework, which combines multi-label logits distillation and label-level embedding distillation to enhance the performance of the student model.
Multi-Label Self-Supervised Learning with Scene Images
Ke Zhu (Nanjing University), Jianxin Wu (Nanjing University)
ClassificationObject DetectionSegmentationRepresentation LearningContrastive LearningImage
🎯 What it does: A method called Multi-Label Self-Supervised (MLS) is proposed, treating self-supervised learning of scene images as a multi-label classification task. It generates pseudo-labels using two dictionaries and trains with binary cross-entropy loss to learn high-quality image representations.
Multi-Metrics Adaptively Identifies Backdoors in Federated Learning
Siquan Huang (South China University of Technology), Ying Gao (South China University of Technology)
Federated LearningTextTabularFinance Related
🎯 What it does: A multi-metric adaptive defense framework is proposed to detect and isolate malicious gradients from backdoor attacks in a federated learning environment.
Multi-Modal Continual Test-Time Adaptation for 3D Semantic Segmentation
Haozhi Cao (Nanyang Technological University), Lihua Xie (Nanyang Technological University)
SegmentationDomain AdaptationAutonomous DrivingContrastive LearningMultimodalityPoint Cloud
🎯 What it does: A multi-modal continuous test-time adaptation (MM-CTTA) framework is proposed for 3D semantic segmentation.
Multi-Modal Gated Mixture of Local-to-Global Experts for Dynamic Image Fusion
Bing Cao (Tianjin University), Qinghua Hu (Tianjin University)
Image TranslationObject DetectionConvolutional Neural NetworkMixture of ExpertsImageMultimodality
🎯 What it does: A dynamic image fusion framework based on multimodal gated local-to-global expert mixture (MoE-Fusion) is proposed to achieve adaptive fusion of infrared and visible light images.
Multi-Modal Neural Radiance Field for Monocular Dense SLAM with a Light-Weight ToF Sensor
Xinyang Liu (Zhejiang University), Zhaopeng Cui (Google)
Depth EstimationNeural Radiance FieldSimultaneous Localization and MappingImageMultimodality
🎯 What it does: A dense SLAM system is proposed that uses only a monocular camera and a lightweight ToF sensor.
Multi-Object Discovery by Low-Dimensional Object Motion
Sadra Safadoust (Koc University), Fatma Güney (Koc University)
Object DetectionSegmentationDepth EstimationConvolutional Neural NetworkTransformerOptical FlowImageVideo
🎯 What it does: This paper proposes an unsupervised multi-object segmentation method based on a single RGB image, utilizing pixel-level depth prediction and low-dimensional optical flow subspace constraints to reconstruct optical flow, thereby learning object segmentation.
Multi-Object Navigation with Dynamically Learned Neural Implicit Representations
Pierre Marza (INSA Lyon), Christian Wolf (Naver Labs Europe)
Robotic IntelligenceTransformerReinforcement LearningPoint Cloud
🎯 What it does: This paper proposes and online learns two types of neural implicit representations (semantic locator and occupancy/exploration implicit representation) in multi-object navigation tasks, and combines them with reinforcement learning agents to achieve instant map construction and target localization in unknown environments.
Multi-Scale Bidirectional Recurrent Network with Hybrid Correlation for Point Cloud Based Scene Flow Estimation
Wencan Cheng (Sungkyunkwan University), Jong Hwan Ko (Sungkyunkwan University)
Depth EstimationAutonomous DrivingRecurrent Neural NetworkOptical FlowPoint Cloud
🎯 What it does: A multi-scale bidirectional recurrent network based on point clouds, MSBRN, is proposed for scene flow estimation.
Multi-Scale Residual Low-Pass Filter Network for Image Deblurring
Jiangxin Dong (Nanjing University of Science and Technology), Jinhui Tang (Nanjing University of Science and Technology)
RestorationConvolutional Neural NetworkTransformerImage
🎯 What it does: A multi-scale residual low-pass filtering network (MRLPFNet) is proposed to recover clear images from motion-blurred images, capable of simultaneously modeling low-frequency structures and high-frequency details.
Multi-Task Learning with Knowledge Distillation for Dense Prediction
Yangyang Xu (Wuhan University), Lefei Zhang (Wuhan University)
SegmentationDepth EstimationKnowledge DistillationTransformerImage
🎯 What it does: A new knowledge distillation framework KDAM is proposed for multi-task dense prediction (semantic segmentation, depth estimation, normal estimation, boundary detection, etc.), which uses a single powerful multi-task teacher model to guide a smaller student model during the training process.
Multi-task View Synthesis with Neural Radiance Fields
Shuhong Zheng (University of Illinois Urbana Champaign), Yu-Xiong Wang (University of Illinois Urbana Champaign)
GenerationData SynthesisTransformerNeural Radiance FieldImage
🎯 What it does: Proposes the Multi-Task View Synthesis (MTVS) task and designs the MuvieNeRF framework to achieve joint synthesis of various scene attributes (RGB, normals, semantics, edges, etc.) on NeRF.
Multi-View Active Fine-Grained Visual Recognition
Ruoyi Du (Beijing University of Posts and Telecommunications), Zhanyu Ma (Beijing University of Posts and Telecommunications)
ClassificationRecognitionConvolutional Neural NetworkRecurrent Neural NetworkReinforcement LearningImage
🎯 What it does: This paper proposes the problem of Multi-view Active Fine-grained Visual Recognition (MAFR) and conducts research by collecting a multi-view fine-grained vehicle dataset (MvCars), designing experiments to validate the necessity and research value of MAFR.
Multi-view Self-supervised Disentanglement for General Image Denoising
Hao Chen (University of Birmingham), Jianbo Jiao (University of Birmingham)
RestorationTransformerContrastive LearningImage
🎯 What it does: This paper proposes a multi-view self-supervised separation framework (MeD) for image denoising, which utilizes multiple images of the same scene containing only noise to separate scene features from noise features in the latent space, achieving denoising without the need for clean images.
Multi-view Spectral Polarization Propagation for Video Glass Segmentation
Yu Qiao, Xin Yang
SegmentationConvolutional Neural NetworkVideo
🎯 What it does: This paper proposes a PGVS-Net network that utilizes multi-view optical polarization information for coherent segmentation of glass regions in RGB-P video sequences.
Multi-weather Image Restoration via Domain Translation
Prashant W. Patil (Deakin University), Subrahmanyam Murala (Trinity College Dublin)
RestorationObject DetectionDepth EstimationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A unified multi-weather image restoration framework based on domain translation is proposed, which utilizes multi-weather variants to learn weather-invariant features, thereby restoring clear images under various weather conditions such as rain, fog, and snow.
Multi3DRefer: Grounding Text Description to Multiple 3D Objects
Yiming Zhang (Simon Fraser University), Angel X. Chang (Simon Fraser University)
Object DetectionRobotic IntelligenceTransformerContrastive LearningTextPoint Cloud
🎯 What it does: This paper proposes a multi-object 3D visual localization task called Multi3DRefer, along with a corresponding dataset and evaluation metrics.
Multimodal Distillation for Egocentric Action Recognition
Gorjan Radevski (KU Leuven University), Tinne Tuytelaars (KU Leuven University)
RecognitionComputational EfficiencyKnowledge DistillationConvolutional Neural NetworkOptical FlowVideoMultimodality
🎯 What it does: This paper proposes a framework based on multi-modal knowledge distillation, training an RGB-only student model for first-person action recognition, where the student uses only RGB video during inference.
Multimodal Garment Designer: Human-Centric Latent Diffusion Models for Fashion Image Editing
Alberto Baldrati (University of Florence), Rita Cucchiara (University of Modena and Reggio Emilia)
Image TranslationGenerationDiffusion modelImageTextMultimodality
🎯 What it does: A multi-modal clothing designer based on a latent diffusion model is proposed, enabling clothing image editing under three conditions: text, human pose, and clothing sketches.
Multimodal High-order Relation Transformer for Scene Boundary Detection
Xi Wei (University of Science and Technology of China), Lei Xiao (Huawei Cloud)
Object DetectionSegmentationTransformerVideoMultimodality
🎯 What it does: A multi-modal high-order relation transformer is proposed for video scene boundary detection.
Multimodal Motion Conditioned Diffusion Model for Skeleton-based Video Anomaly Detection
Alessandro Flaborea (Sapienza University of Rome), Fabio Galasso (Sapienza University of Rome)
Anomaly DetectionGraph Neural NetworkDiffusion modelAuto EncoderVideoMultimodality
🎯 What it does: This paper proposes a skeleton video anomaly detection method called MoCoDAD based on a diffusion probability model. It judges anomalies by conditioning on past actions to generate multimodal future poses and comparing them with real future poses.
Multimodal Optimal Transport-based Co-Attention Transformer with Global Structure Consistency for Survival Prediction
Yingxue Xu (Hong Kong University of Science and Technology), Hao Chen (Hong Kong University of Science and Technology)
ClassificationSegmentationOptimizationTransformerMultimodalityBiomedical Data
🎯 What it does: This paper proposes a multi-modal optimal transport (OT) co-attention transformer (MOTCat), which matches whole slide image (WSI) patches with gene expression vectors through OT to filter out key information related to the tumor microenvironment, and aggregates multi-modal features using a transformer to ultimately achieve cancer survival prediction.
Multimodal Variational Auto-encoder based Audio-Visual Segmentation
Yuxin Mao (Northwestern Polytechnical University), Yuchao Dai (Northwestern Polytechnical University)
SegmentationAuto EncoderVideoMultimodalityAudio
🎯 What it does: An Explicit Conditional Multimodal Variational Auto-Encoder (ECMVAE) is proposed for audio-visual segmentation, which fully utilizes audio instructions by explicitly decomposing shared and exclusive representations in the latent space and incorporating orthogonal and mutual information constraints.
Multiple Instance Learning Framework with Masked Hard Instance Mining for Whole Slide Image Classification
Wenhao Tang (Chongqing University), Bo Liu (Walmart Global Tech)
ClassificationContrastive LearningImageBiomedical Data
🎯 What it does: A multi-instance learning framework based on masked hard sample mining (MHIM-MIL) is proposed for whole slide image classification.
Multiple Planar Object Tracking
Zhicheng Zhang (Nankai University), Jufeng Yang (Nankai University)
Object TrackingConvolutional Neural NetworkVideoBenchmark
🎯 What it does: This paper proposes a multi-plane object tracking framework called PRTrack and constructs the first large-scale benchmark dataset for multi-plane object tracking, MPOT-3K.
Multiscale Representation for Real-Time Anti-Aliasing Neural Rendering
Dongting Hu (University of Melbourne), Mingming Gong (University of Melbourne)
GenerationComputational EfficiencyNeural Radiance FieldImage
🎯 What it does: Proposes Mip-VoG, a multi-scale voxel grid, which achieves real-time anti-aliasing rendering by combining a delayed NeRF architecture;
Multiscale Structure Guided Diffusion for Image Deblurring
Mengwei Ren (New York University), Peyman Milanfar (Google Research)
RestorationConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: This work proposes a multi-scale structure-guided image conditional diffusion model for single image deblurring, which enhances the deblurring effect by injecting multi-scale grayscale structural information into the latent space using a regression network, showing more robustness especially on unseen domain data.
Muscles in Action
Mia Chiquier (Columbia University), Carl Vondrick (Columbia University)
Pose EstimationRecommendation SystemConvolutional Neural NetworkTransformerVideoMultimodality
🎯 What it does: This paper presents a new multimodal dataset MIA and constructs a bidirectional learning framework that maps motion in videos to electromyographic (EMG) activation, enabling muscle activation prediction, motion reconstruction, and muscle-targeted motion recommendation.
MUter: Machine Unlearning on Adversarially Trained Models
Junxu Liu (Renmin University of China), Zhan Qin (Zhejiang University)
OptimizationAdversarial AttackConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: A new method called MUter is proposed for machine unlearning of adversarial training models (ATM), addressing the issue that traditional unlearning methods cannot fully eliminate data influence under a two-layer optimization structure.
MUVA: A New Large-Scale Benchmark for Multi-View Amodal Instance Segmentation in the Shopping Scenario
Zhixuan Li (Peking University), Tiejun Huang (Peking University)
Object DetectionSegmentationTransformerImageBenchmark
🎯 What it does: This paper proposes the Multi-Angle Invisible Instance Segmentation (MAIS) task, constructs a large-scale synthetic dataset for shopping scenes called MUVA, and introduces the MASFormer model based on Transformer for multi-angle fusion.
MV-DeepSDF: Implicit Modeling with Multi-Sweep Point Clouds for 3D Vehicle Reconstruction in Autonomous Driving
Yibo Liu (Huawei Noah's Ark Lab), Jinjun Shan (York University)
GenerationAutonomous DrivingPoint Cloud
🎯 What it does: Using multi-view sparse point clouds, a 3D vehicle reconstruction framework based on implicit modeling (MV-DeepSDF) is proposed, which effectively integrates multi-scan point cloud information by transforming shape estimation into a feature extraction problem of 'elements to sets'.
MV-Map: Offboard HD-Map Generation with Multi-view Consistency
Ziyang Xie (University of Illinois Urbana-Champaign), Yu-Xiong Wang (Fudan University)
SegmentationGenerationAutonomous DrivingNeural Radiance FieldSimultaneous Localization and MappingVideo
🎯 What it does: This paper proposes an offline high-definition map (HD map) generation framework called MV-Map based on multi-view consistency, which aggregates information from multiple camera frames and outputs a unified BEV semantic map.
MVPSNet: Fast Generalizable Multi-view Photometric Stereo
Dongxu Zhao (University of North Carolina), Soumyadip Sengupta (University of North Carolina)
RestorationGenerationDepth EstimationComputational EfficiencyConvolutional Neural NetworkMesh
🎯 What it does: An end-to-end fast general multi-view photometric stereo reconstruction network MVPSNet is proposed, capable of generating 3D meshes with excellent global shape and texture details in just a few seconds of inference time.
Name Your Colour For the Task: Artificially Discover Colour Naming via Colour Quantisation Transformer
Shenghan Su (Shanghai Jiao Tong University), Tatsuya Harada (University of Tokyo)
ClassificationObject DetectionTransformerImage
🎯 What it does: This paper proposes a Transformer-based color quantization model called CQFormer, which can maintain a low-bit color space while balancing visual perceptual structure and machine recognition accuracy.
NAPA-VQ: Neighborhood-Aware Prototype Augmentation with Vector Quantization for Continual Learning
Tamasha Malepathirana (University of Melbourne), Saman Halgamuge (University of Melbourne)
ClassificationKnowledge DistillationImage
🎯 What it does: Proposes the NAPA-VQ method, which maintains old knowledge in class-incremental learning without examples through neighborhood-aware vector quantization and prototype augmentation.
Narrator: Towards Natural Control of Human-Scene Interaction Generation via Relationship Reasoning
Haibiao Xuan (Tianjin University), Kun Li (Tianjin University)
GenerationTransformerAuto EncoderPoint Cloud
🎯 What it does: A generative model named Narrator based on relational reasoning is proposed, which can naturally and controllably generate diverse human-scene interactions based on natural language descriptions and 3D scenes.
Navigating to Objects Specified by Images
Jacob Krantz (Oregon State University), Devendra Singh Chaplot
Object DetectionRobotic IntelligenceGraph Neural NetworkReinforcement LearningImage
🎯 What it does: A fully modular instance image target navigation system is proposed, capable of recognizing and locating target instances in unknown environments through foreground keypoint matching and projection, and completing navigation tasks.
NaviNeRF: NeRF-based 3D Representation Disentanglement by Latent Semantic Navigation
Baao Xie (Tianjin University), Wenjun Zeng (Eastern Institute of Technology)
GenerationData SynthesisNeural Radiance FieldGenerative Adversarial NetworkImage
🎯 What it does: Proposes the NaviNeRF model, which utilizes NeRF and self-supervised navigation to achieve unsupervised 3D fine-grained decoupling and attribute editing.
NCHO: Unsupervised Learning for Neural 3D Composition of Humans and Objects
Taeksoo Kim (Seoul National University), Hanbyul Joo (Seoul National University)
GenerationData SynthesisPoint CloudMesh
🎯 What it does: This study investigates how to unsupervisedly separate and synthesize the human body and objects (such as backpacks, coats, and scarves) from real 3D scans to generate controllable 3D virtual dolls.
NDC-Scene: Boost Monocular 3D Semantic Scene Completion in Normalized Device Coordinates Space
Jiawei Yao (University of Washington), Hongsheng Li (Chinese University of Hong Kong)
SegmentationDepth EstimationAutonomous DrivingConvolutional Neural NetworkPoint Cloud
🎯 What it does: The NDC-Scene framework is proposed, utilizing Normalized Device Coordinates space and a depth-adaptive dual decoder to achieve monocular 3D semantic scene completion.
NDDepth: Normal-Distance Assisted Monocular Depth Estimation
Shuwei Shao, Zhengguo Li
Depth EstimationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a new method to address a specific computer vision task, with specific details not provided.
Nearest Neighbor Guidance for Out-of-Distribution Detection
Jaewoo Park (Yonsei University), Andrew Beng Jin Teoh (Yonsei University)
Anomaly DetectionImage
🎯 What it does: We propose the Nearest Neighbor Guidance (NNGuide) method, which improves the OOD detection score by incorporating nearest neighbor feature similarity into classifier confidence.
Neglected Free Lunch - Learning Image Classifiers Using Annotation Byproducts
Dongyoon Han, Seong Joon Oh (Google)
ClassificationConvolutional Neural NetworkTransformerSupervised Fine-TuningImage
🎯 What it does: This paper proposes to utilize the no-cost annotation byproducts (such as mouse trajectories and click positions) during the image annotation process as intangible information to improve the learning of image classification models.
NeILF++: Inter-Reflectable Light Fields for Geometry and Material Estimation
Jingyang Zhang (Hong Kong University of Science and Technology), Long Quan (Hong Kong University of Science and Technology)
GenerationOptimizationNeural Radiance FieldImage
🎯 What it does: Jointly estimate scene geometry, material, and lighting from multi-view images, and propose a reciprocal light field framework.
NeMF: Inverse Volume Rendering with Neural Microflake Field
Youjia Zhang (Huazhong University of Science and Technology), Wei Yang (Tencent)
GenerationData SynthesisNeural Radiance FieldImage
🎯 What it does: The Neural Microflake Field (NeMF) model is proposed, achieving inverse volume rendering to recover the distribution, density, chromaticity, and roughness of volumetric microcrystals from multi-view images, and supports relighting, material editing, and volumetric scattering effects.
NEMTO: Neural Environment Matting for Novel View and Relighting Synthesis of Transparent Objects
Dongqing Wang (École Polytechnique Fédérale de Lausanne), Sabine Süsstrunk
RestorationGenerationData SynthesisNeural Radiance FieldImage
🎯 What it does: We propose NEMTO, an end-to-end neural rendering pipeline that can recover the geometry and lighting of transparent objects from multi-view natural lighting images under unknown refractive indices, and synthesize new views and re-lit results.
NeO 360: Neural Fields for Sparse View Synthesis of Outdoor Scenes
Muhammad Zubair Irshad (Georgia Institute of Technology), Rares Ambrus (Toyota Research Institute)
GenerationData SynthesisNeural Radiance FieldImage
🎯 What it does: We propose NeO 360, a generalized 3D sparse view synthesis framework that can generate complete 360° unrestricted outdoor scene views using only 1 to 5 RGB images with known poses;
NeRF-Det: Learning Geometry-Aware Volumetric Representation for Multi-View 3D Object Detection
Chenfeng Xu (University of California), Masayoshi Tomizuka (University of California)
Object DetectionNeural Radiance FieldPoint Cloud
🎯 What it does: This work proposes NeRF-Det, which combines NeRF with a 3D detection network, using pose RGB images to learn a geometry-aware voxel representation through a shared MLP, achieving indoor 3D object detection.
NeRF-LOAM: Neural Implicit Representation for Large-Scale Incremental LiDAR Odometry and Mapping
Junyuan Deng (Shanghai Jiao Tong University), Ling Pei (Shanghai Jiao Tong University)
Pose EstimationAutonomous DrivingOptimizationNeural Radiance FieldSimultaneous Localization and MappingPoint Cloud
🎯 What it does: This paper proposes NeRF-LOAM, a laser radar incremental odometry and mapping method based on neural implicit representation, achieving simultaneous pose estimation and dense 3D mesh generation.
NeRF-MS: Neural Radiance Fields with Multi-Sequence
Peihao Li (Shenzhen International Graduate School Tsinghua University), Haoqian Wang (Shenzhen International Graduate School Tsinghua University)
RestorationGenerationNeural Radiance FieldImageVideo
🎯 What it does: This paper proposes NeRF-MS, which can train neural radiance fields on collections of images captured by multiple sequences, multiple time phases, and multiple sensors, achieving more complete and consistent 3D reconstruction and view synthesis.
NerfAcc: Efficient Sampling Accelerates NeRFs
Ruilong Li, Angjoo Kanazawa
OptimizationComputational EfficiencyNeural Radiance FieldImage
🎯 What it does: This paper presents the NerfAcc toolbox, which achieves efficient sampling of the NeRF training process (such as occupancy grids and proposal networks) through a unified transmittance estimator, significantly improving training speed while maintaining or even enhancing visual quality.
Nerfbusters: Removing Ghostly Artifacts from Casually Captured NeRFs
Frederik Warburg (University of California), Angjoo Kanazawa (University of California)
RestorationDiffusion modelNeural Radiance FieldPoint Cloud
🎯 What it does: The Nerfbusters method is proposed, utilizing local 3D diffusion priors to eliminate floating objects and erroneous geometry in NeRF.
NeRFrac: Neural Radiance Fields through Refractive Surface
Yifan Zhan (University of Tokyo), Yinqiang Zheng (University of Tokyo)
RestorationGenerationData SynthesisNeural Radiance FieldImageVideo
🎯 What it does: A model called NeRFrac is proposed, which incorporates an MLP refraction field into the NeRF framework to achieve novel view synthesis and reconstruction of refractive surfaces (such as water surfaces).
NeSS-ST: Detecting Good and Stable Keypoints with a Neural Stability Score and the Shi-Tomasi detector
Konstantin Pakulev (Skolkovo Institute of Science and Technology), Gonzalo Ferrer (Skolkovo Institute of Science and Technology)
Pose EstimationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper proposes NeSS-ST, a keypoint detection method that combines the Shi-Tomasi hand-crafted detector with neural network learning, which can be trained using only raw images without the need for corresponding labels.
NeTO:Neural Reconstruction of Transparent Objects with Self-Occlusion Aware Refraction-Tracing
Zongcheng Li (Wuhan University), Chunxia Xiao (Wuhan University)
GenerationOptimizationNeural Radiance FieldPoint CloudMesh
🎯 What it does: Using implicit Signed Distance Function (SDF) and volume rendering techniques, the 3D geometry of solid transparent objects is reconstructed based on the known correspondence between camera-projected light rays and background points.
Neural Characteristic Function Learning for Conditional Image Generation
Shengxi Li (Beihang University), Li Li (Beihang University)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a conditionally generated adversarial network based on characteristic functions (CCF-GAN), which learns the differences in joint distributions through neural characteristic functions (NCF) to achieve more stable conditional generation.
Neural Collage Transfer: Artistic Reconstruction via Material Manipulation
Ganghun Lee (Seoul National University), Byoung-Tak Zhang (Seoul National University)
Image TranslationGenerationData SynthesisReinforcement LearningGenerative Adversarial NetworkImage
🎯 What it does: Achieved collage generation based on material shear and paste through reinforcement learning, proposing a collage MDP and a model-based Soft Actor-Critic training framework.
Neural Deformable Models for 3D Bi-Ventricular Heart Shape Reconstruction and Modeling from 2D Sparse Cardiac Magnetic Resonance Imaging
Meng Ye (Rutgers University), Dimitris Metaxas (Rutgers University)
SegmentationGenerationImagePoint CloudMeshMagnetic Resonance ImagingOrdinary Differential Equation
🎯 What it does: A neural deformable model (NDM) is proposed, utilizing superquadric parameterization and neural ODE to reconstruct from sparse 2D CMR images, generate high-quality 3D biventricular meshes, and achieve dense correspondence registration.
Neural Fields for Structured Lighting
Aarrushi Shandilya (Carnegie Mellon University), Matthew O'toole
Depth EstimationNeural Radiance FieldImage
🎯 What it does: This study proposes a volume rendering framework based on NeRF, combining a physical model of structured light imaging. It utilizes the raw structured light and ambient light images from Intel RealSense to achieve high-precision depth, normal, direct/indirect light decomposition, and complete scene reconstruction under sparse viewpoints.