CVPR 2024 Papers — Page 17
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2716 papers
Multi-Scale 3D Gaussian Splatting for Anti-Aliased Rendering
Zhiwen Yan (National University of Singapore), Gim Hee Lee (National University of Singapore)
RestorationComputational EfficiencyGaussian SplattingPoint Cloud
🎯 What it does: A multi-scale 3D Gaussian splitting algorithm is proposed to address the issues of anti-aliasing and rendering speed degradation of the original 3D Gaussian at low resolutions or distant perspectives.
Multi-scale Dynamic and Hierarchical Relationship Modeling for Facial Action Units Recognition
Zihan Wang (Shenzhen University), Linlin Shen (Shenzhen University)
RecognitionConvolutional Neural NetworkTransformerVideo
🎯 What it does: A multi-scale dynamic and hierarchical relationship modeling (MDHR) framework is proposed for the precise recognition of facial action units (AUs) in videos.
Multi-Scale Video Anomaly Detection by Multi-Grained Spatio-Temporal Representation Learning
Menghao Zhang (Beijing University of Posts and Telecommunications), Jianxin Liao (Beijing University of Posts and Telecommunications)
Anomaly DetectionRepresentation LearningConvolutional Neural NetworkContrastive LearningVideo
🎯 What it does: A multi-scale video anomaly detection method is proposed, which learns spatiotemporal features at three granularities (coarse and fine) using a self-supervised task of video continuity, and performs missing frame contrast estimation in the feature space to train a robust anomaly detection model.
Multi-Session SLAM with Differentiable Wide-Baseline Pose Optimization
Lahav Lipson (Princeton University), Jia Deng (Princeton University)
Pose EstimationOptimizationRobotic IntelligenceRecurrent Neural NetworkSimultaneous Localization and MappingOptical FlowImageVideo
🎯 What it does: This paper proposes an end-to-end system for multi-session monocular RGB SLAM, capable of simultaneously achieving visual odometry, two-view pose estimation, and global optimization across multiple segments of discrete video.
Multi-Space Alignments Towards Universal LiDAR Segmentation
Youquan Liu (ShanghaiTech University), Yuexin Ma (ShanghaiTech University)
SegmentationAutonomous DrivingVision Language ModelImageTextMultimodalityPoint Cloud
🎯 What it does: A general LiDAR segmentation framework named M3Net is proposed, capable of performing multi-task, multi-dataset, and multi-modal LiDAR semantic/panoptic segmentation under a single set of parameters.
Multi-Task Dense Prediction via Mixture of Low-Rank Experts
Yuqi Yang (Nankai University), Bo Li (vivo Mobile Communication Co., Ltd)
Object DetectionSegmentationConvolutional Neural NetworkMixture of ExpertsImage
🎯 What it does: A decoder framework for multi-task dense prediction, MLoRE, is proposed, utilizing a Mixture-of-Low-Rank-Experts to achieve dynamic combination and global association of task features;
Multi-view Aggregation Network for Dichotomous Image Segmentation
Qian Yu (Dalian University of Technology), Huchuan Lu (Dalian University of Technology)
SegmentationTransformerImage
🎯 What it does: Proposes a Multi-View Aggregation Network (MVANet) that achieves high-precision binary image segmentation through a single-stream, single-stage structure.
Multi-View Attentive Contextualization for Multi-View 3D Object Detection
Xianpeng Liu (North Carolina State University), Tianfu Wu (Ant Group)
Object DetectionAutonomous DrivingTransformerPoint Cloud
🎯 What it does: A multi-view attention contextual module MvACon is proposed to improve the 2D→3D feature enhancement process in query-based multi-view 3D object detection.
Multiagent Multitraversal Multimodal Self-Driving: Open MARS Dataset
Yiming Li (New York University), Chen Feng (New York University)
RetrievalAutonomous DrivingNeural Radiance FieldSimultaneous Localization and MappingVideoMultimodalityPoint CloudBenchmark
🎯 What it does: The MARS dataset has been proposed and released, which includes multi-vehicle collaboration, repeated traversals, and multimodal (LiDAR + panoramic RGB) data for studying multi-agent systems, memory retrieval, and neural reconstruction in autonomous driving.
MultiDiff: Consistent Novel View Synthesis from a Single Image
Norman Müller (Meta Reality Labs), Peter Kontschieder (Technical University of Munich)
GenerationData SynthesisDepth EstimationDiffusion modelImage
🎯 What it does: Synthesize a multi-view image sequence consistent with camera trajectory from a single RGB image.
MULTIFLOW: Shifting Towards Task-Agnostic Vision-Language Pruning
Matteo Farina (University of Trento), Elisa Ricci (University of Trento)
GenerationRetrievalTransformerVision Language ModelMultimodality
🎯 What it does: A new task-agnostic visual-language model pruning method, called MULTIFLOW, is proposed, aiming to obtain a sparse model that can be transferred to multiple unknown downstream tasks through a single pruning.
Multimodal Industrial Anomaly Detection by Crossmodal Feature Mapping
Alex Costanzino (University of Bologna), Luigi Di Stefano (University of Bologna)
Anomaly DetectionTransformerContrastive LearningImageMultimodalityPoint Cloud
🎯 What it does: A cross-modal feature mapping framework is proposed to detect defects in industrial anomaly detection by utilizing the mutual mapping of RGB images and point cloud features.
Multimodal Pathway: Improve Transformers with Irrelevant Data from Other Modalities
Yiyuan Zhang (Chinese University of Hong Kong), Xiangyu Yue (Chinese University of Hong Kong)
ClassificationObject DetectionSegmentationRepresentation LearningTransformerImageVideoMultimodalityPoint CloudAudio
🎯 What it does: This paper proposes a Multimodal Pathway Transformer (M2PT) that enhances the performance of unimodal Transformers on the target modality by using cross-modal unrelated data during the training phase and leveraging Cross-Modal Re-parameterization.
Multimodal Prompt Perceiver: Empower Adaptiveness Generalizability and Fidelity for All-in-One Image Restoration
Yuang Ai, Ran He
RestorationPrompt EngineeringDiffusion modelImageMultimodalityBenchmark
🎯 What it does: A multi-modal prompt learning framework MPerceiver is proposed, utilizing the generative prior of Stable Diffusion to achieve one-click image restoration.
Multimodal Representation Learning by Alternating Unimodal Adaptation
Xiaohui Zhang (University of North Carolina at Chapel Hill), Huaxiu Yao (University of North Carolina at Chapel Hill)
Representation LearningMultimodality
🎯 What it does: A multi-modal learning framework MLA is proposed to avoid the modality laziness problem through alternating unimodal adaptation.
Multimodal Sense-Informed Forecasting of 3D Human Motions
Zhenyu Lou (Zhejiang University), Hong Zhou (Zhejiang University)
Pose EstimationAutonomous DrivingGraph Neural NetworkTransformerMultimodalityPoint Cloud
🎯 What it does: A multi-modal perception-based 3D human motion prediction framework SIF3D is proposed, which can predict future motion trajectories and postures using external 3D scene point clouds and internal gaze information.
MultiPhys: Multi-Person Physics-aware 3D Motion Estimation
Nicolas Ugrinovic (Institut de Robotica i Informatica Industrial), Leonidas Guibas (Stanford University)
Pose EstimationRobotic IntelligenceReinforcement LearningVideoPhysics Related
🎯 What it does: This paper proposes the MultiPhys framework, which corrects multi-person 3D motion estimation through a physics simulator, eliminating physically unreasonable phenomena such as body penetration and foot sliding, and enhancing the coherence of spatial placement.
Multiplane Prior Guided Few-Shot Aerial Scene Rendering
Zihan Gao (Xidian University), Yuwei Guo (Xidian University)
GenerationData SynthesisTransformerNeural Radiance FieldImage
🎯 What it does: This paper proposes a multi-plane prior-guided neural radiance field (MPNeRF) for few-shot aerial scene view synthesis.
Multiple View Geometry Transformers for 3D Human Pose Estimation
Ziwei Liao (University of Toronto), Steven L. Waslander (University of Toronto)
Pose EstimationTransformerImage
🎯 What it does: In multi-view 3D human pose estimation, this paper proposes MVGFormer, a Transformer framework that combines a learned appearance module with a non-learned geometric module to achieve iterative refinement of multi-camera images.
MultiPLY: A Multisensory Object-Centric Embodied Large Language Model in 3D World
Yining Hong (University of California, Los Angeles), Chuang Gan (University of Massachusetts Amherst)
RetrievalRobotic IntelligenceTransformerLarge Language ModelSupervised Fine-TuningVision-Language-Action ModelMultimodalityAudio
🎯 What it does: This study presents MultiPLY, a multi-sensory embodied large language model that can interact with actions and a three-dimensional environment, encoding visual, audio, tactile, and thermal information into the LLM.
MultiPly: Reconstruction of Multiple People from Monocular Video in the Wild
Zeren Jiang (ETH Zurich), Jie Song (Microsoft)
Object DetectionSegmentationPose EstimationNeural Radiance FieldVideo
🎯 What it does: Achieve complete and detailed 3D human geometry and appearance reconstruction of multiple individuals through monocular video, providing temporally and spatially consistent 3D avatars.
Multiscale Vision Transformers Meet Bipartite Matching for Efficient Single-stage Action Localization
Ioanna Ntinou (Queen Mary University London), Georgios Tzimiropoulos (Queen Mary University London)
RecognitionObject DetectionComputational EfficiencyTransformerVideo
🎯 What it does: This paper proposes a single-stage action localization method called BMViT, which directly utilizes the spatiotemporal tokens output by Vision Transformers (such as MViTv2-S, ViT-B) to predict bounding boxes, actor confidence, and action categories through a simple MLP head, and trains the entire network using a bidirectional matching loss without the need for additional decoders or learnable queries.
Multiview Aerial Visual RECognition (MAVREC): Can Multi-view Improve Aerial Visual Perception?
Aritra Dutta (University of Central Florida), Mubarak Shah (University of Central Florida)
RecognitionObject DetectionVideo
🎯 What it does: This paper constructs a large-scale multi-view drone visual recognition dataset MAVREC and explores geographic information-based aerial visual detection methods.
Multiway Point Cloud Mosaicking with Diffusion and Global Optimization
Shengze Jin (ETH Zurich), Daniel Barath (ETH Zurich)
Pose EstimationOptimizationTransformerDiffusion modelPoint CloudOrdinary Differential Equation
🎯 What it does: An end-to-end multi-view point cloud stitching framework is proposed. First, high-precision dual point cloud registration is achieved through ODIN, followed by global pose rotation averaging, robust translation re-estimation, translation averaging, and diffusion graph optimization, ultimately unifying multiple partially overlapping point clouds into the same coordinate system.
MuRF: Multi-Baseline Radiance Fields
Haofei Xu (ETH Zurich), Fisher Yu
Data SynthesisConvolutional Neural NetworkTransformerNeural Radiance FieldImage
🎯 What it does: A multi-baseline radiance field (MuRF) framework is proposed for sparse view synthesis, capable of handling both small and large baseline inputs simultaneously.
MuseChat: A Conversational Music Recommendation System for Videos
Zhikang Dong (Stony Brook University), Peng Zhang (Bytedance)
Recommendation SystemTransformerLarge Language ModelContrastive LearningVideoTextMultimodality
🎯 What it does: The first multi-turn music recommendation dialogue system for videos, MuseChat, has been constructed, and explanations are generated through tri-modal fusion and LLM.
MV-Adapter: Multimodal Video Transfer Learning for Video Text Retrieval
Xiaojie Jin (Bytedance Inc), Jiashi Feng (Bytedance Inc)
RetrievalTransformerVision Language ModelVideoTextMultimodality
🎯 What it does: A parameter-efficient video-text retrieval framework MV-Adapter is proposed, specifically designed for lightweight adaptation of the video/text branches of CLIP;
MVBench: A Comprehensive Multi-modal Video Understanding Benchmark
Kunchang Li (Shenzhen Institute of Advanced Technology Chinese Academy of Sciences), Yu Qiao (Shenzhen Institute of Advanced Technology Chinese Academy of Sciences)
TransformerLarge Language ModelVision Language ModelVideoMultimodalityBenchmark
🎯 What it does: Designed the MVBench video understanding benchmark and proposed VideoChat2, a video multimodal LLM;
MVCPS-NeuS: Multi-view Constrained Photometric Stereo for Neural Surface Reconstruction
Hiroaki Santo (Osaka University), Yasuyuki Matsushita (Osaka University)
Depth EstimationOptimizationNeural Radiance FieldPoint CloudMesh
🎯 What it does: This paper proposes a multi-view constrained illumination photometric stereo reconstruction method called MVCPS-NeuS, which utilizes the constraint of synchronized movement between the light source and the camera, combined with neural surface reconstruction to achieve high-precision 3D reconstruction under sparse views and lighting conditions.
MVD-Fusion: Single-view 3D via Depth-consistent Multi-view Generation
Hanzhe Hu (Carnegie Mellon University), Shubham Tulsiani (Carnegie Mellon University)
GenerationDepth EstimationTransformerDiffusion modelImage
🎯 What it does: MVD-Fusion generates multi-view RGB-D images from a single RGB input, enabling single-view 3D predictions.
MVHumanNet: A Large-scale Dataset of Multi-view Daily Dressing Human Captures
Zhangyang Xiong (Chinese University of Hong Kong, Shenzhen), Xiaoguang Han (Chinese University of Hong Kong, Shenzhen)
RecognitionSegmentationGenerationPose EstimationNeural Radiance FieldImageVideoTextMultimodality
🎯 What it does: This paper constructs MVHumanNet, the largest multi-view human capture dataset, covering 4,500 identities, 9,000 sets of everyday clothing, 645 million frames, and complete annotations;
MVIP-NeRF: Multi-view 3D Inpainting on NeRF Scenes via Diffusion Prior
Honghua Chen (Nanyang Technological University), Xingang Pan (Nanyang Technological University)
RestorationGenerationDiffusion modelNeural Radiance FieldImage
🎯 What it does: This paper proposes a NeRF method for multi-view consistent 3D completion using diffusion models, performing global 3D reconstruction of missing areas in the scene.
Named Entity Driven Zero-Shot Image Manipulation
Zhida Feng (Wuhan University of Science and Technology), Shikun Feng (Baidu Inc.)
GenerationData SynthesisGenerative Adversarial NetworkContrastive LearningImageText
🎯 What it does: This paper proposes StyleEntity, a model that utilizes named entities as training proxies to achieve zero-shot image editing, and further enhances stability during inference through Prompt Ensemble Latent Averaging (PELA).
NAPGuard: Towards Detecting Naturalistic Adversarial Patches
Siyang Wu (Beihang University), Xianglong Liu (Zhongguancun Laboratory)
Object DetectionAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: Designed and implemented the NAPGuard framework for detecting Natural Adversarial Patches (NAP) by learning aggressive feature alignment at high frequency (AFAL) during the training phase and using a feature shield module to suppress natural features (NFSI) during the inference phase, significantly improving detection accuracy and generalization ability.
Narrative Action Evaluation with Prompt-Guided Multimodal Interaction
Shiyi Zhang (Tsinghua University), Yansong Tang (Tsinghua University)
RecognitionGenerationTransformerPrompt EngineeringVision Language ModelVideoTextMultimodalityBenchmark
🎯 What it does: Proposes the Narrative Action Evaluation (NAE) task, generating specialized natural language comments that include scores and action details.
NARUTO: Neural Active Reconstruction from Uncertain Target Observations
Ziyue Feng (OPPO US Research Center), Yi Xu (OPPO US Research Center)
OptimizationRobotic IntelligenceConvolutional Neural NetworkNeural Radiance FieldSimultaneous Localization and MappingPoint Cloud
🎯 What it does: Based on known positioning, the active reconstruction system NARUTO, guided by learned uncertainty, achieves high-precision 3D reconstruction of indoor scenes in 6DoF free space;
Naturally Supervised 3D Visual Grounding with Language-Regularized Concept Learners
Chun Feng (Stanford University), Jiajun Wu (Stanford University)
Object DetectionRepresentation LearningLarge Language ModelPoint Cloud
🎯 What it does: This paper proposes a Language Regularization Concept Learner (LARC), which achieves the localization and relational reasoning of 3D objects by training a neural symbolic 3D visual orientation model under natural supervision (with only scenes and question-answer pairs).
Navigate Beyond Shortcuts: Debiased Learning Through the Lens of Neural Collapse
Yining Wang (Fudan University), Min Yang (Fudan University)
ClassificationData-Centric LearningConvolutional Neural NetworkImage
🎯 What it does: This study investigates how neural networks form shortcut learning in the early stages of training on biased datasets with attribute imbalance, leading to the absence of Neural Collapse. It proposes the 'ETF-Prime' mechanism to guide the model to skip shortcut learning using simple shortcut features before training, until it learns the intrinsic relationships, thus achieving debiasing without additional costs.
Navigating Beyond Dropout: An Intriguing Solution towards Generalizable Image Super Resolution
Hongjun Wang (University of Tokyo), Tieyong Zeng (Chinese University of Hong Kong)
RestorationSuper ResolutionConvolutional Neural NetworkImage
🎯 What it does: A regularization strategy is proposed to enhance the generalization ability of blind image super-resolution models by aligning the first and second statistical moments of features from different degraded images of the same content.
NAYER: Noisy Layer Data Generation for Efficient and Effective Data-free Knowledge Distillation
Minh-Tuan Tran (Monash University), Dinh Phung (Monash University)
ClassificationKnowledge DistillationConvolutional Neural NetworkLarge Language ModelImage
🎯 What it does: This paper proposes a data-independent knowledge distillation method called NAYER, which quickly synthesizes high-quality pseudo-samples for teacher-student distillation by transferring the noise source from the input to the noise layer and using label text embeddings generated by a pre-trained language model as input to the generator.
NB-GTR: Narrow-Band Guided Turbulence Removal
Yifei Xia (Peking University), Boxin Shi (Peking University)
RestorationTransformerImage
🎯 What it does: Proposes the NB-GTR method, which utilizes pairs of simultaneously captured RGB images and narrowband images to remove atmospheric turbulence through a dual encoder, a re-disturbance module, and a multi-scale Transformer.
NC-SDF: Enhancing Indoor Scene Reconstruction Using Neural SDFs with View-Dependent Normal Compensation
Ziyi Chen (Zhejiang University), Yu Zhang (Zhejiang University)
SegmentationGenerationNeural Radiance FieldPoint Cloud
🎯 What it does: The NC-SDF framework is proposed, utilizing view-dependent normal compensation, information sampling, and a mixed geometric model to achieve neural SDF reconstruction of indoor scenes.
NC-TTT: A Noise Constrastive Approach for Test-Time Training
David Osowiechi (École nationale supérieure de technologie de Montréal), Christian Desrosiers (École nationale supérieure de technologie de Montréal)
Domain AdaptationContrastive LearningImage
🎯 What it does: A method for unsupervised test-time training (NC-TTT) based on noise contrastive estimation is proposed, which achieves adaptation to the target domain by learning to distinguish between the inner and outer domains of noisy feature maps.
Nearest is Not Dearest: Towards Practical Defense against Quantization-conditioned Backdoor Attacks
Boheng Li (Zhejiang University), Yiming Li (Zhejiang University)
OptimizationAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a new defense framework called EFRAP, designed to eliminate quantization condition backdoor attacks that are activated after model quantization.
NEAT: Distilling 3D Wireframes from Neural Attraction Fields
Nan Xue (Ant Group), Yujun Shen (Zhejiang University)
Knowledge DistillationRepresentation LearningNeural Radiance FieldPoint Cloud
🎯 What it does: A method called NEAT for unmatched 3D wireframe reconstruction is proposed, which directly recovers 3D line segments and nodes from 2D wireframe detection results using neural field rendering and a global node perceiver.
NECA: Neural Customizable Human Avatar
Junjin Xiao (Sun Yat-sen University), Wei-Shi Zheng (Sun Yat-sen University)
GenerationData SynthesisPose EstimationNeural Radiance FieldVideo
🎯 What it does: This paper proposes a method for learning fully customizable neural human avatars from monocular or sparse multi-view videos, capable of achieving realistic rendering under arbitrary poses, viewpoints, lighting, shapes, textures, and shadows.
Neighbor Relations Matter in Video Scene Detection
Jiawei Tan (Chongqing University), Zhangbin Qian (Chongqing University)
RecognitionSegmentationGraph Neural NetworkSupervised Fine-TuningVideo
🎯 What it does: We propose NeighborNet, which re-evaluates shot-to-shot similarity by introducing semantic neighbor and temporal neighbor relationships, and recursively transmits contextual information on feature maps and temporal graphs to achieve more accurate video scene detection.
NeISF: Neural Incident Stokes Field for Geometry and Material Estimation
Chenhao Li (Sony Semiconductor Solutions Corporation), Yusuke Moriuchi (Sony Semiconductor Solutions Corporation)
GenerationData SynthesisNeural Radiance FieldImage
🎯 What it does: This paper studies a multi-view inverse rendering method based on polarization information, NeISF, which can simultaneously recover the geometric shape of objects, surface reflection properties (diffuse albedo and roughness), and the polarization state of multiple scattered light.
NeLF-Pro: Neural Light Field Probes for Multi-Scale Novel View Synthesis
Zinuo You (ETH Zurich), Anpei Chen (University of Tübingen)
GenerationData SynthesisOptimizationNeural Radiance FieldImageVideo
🎯 What it does: We propose NeLF-Pro, a sparse scalable light field representation based on local light field probes, capable of achieving high-quality novel view synthesis in scenes with varying scales and spatial granularity;
NeRF Analogies: Example-Based Visual Attribute Transfer for NeRFs
Michael Fischer (University College London), Tobias Ritschel (Meta Reality Labs Research)
Image TranslationGenerationData SynthesisTransformerNeural Radiance FieldContrastive LearningPoint Cloud
🎯 What it does: Proposes the NeRF Analogies method, which transfers the appearance of the source NeRF to the target geometry, generating a new 3D view-consistent NeRF.
NeRF Director: Revisiting View Selection in Neural Volume Rendering
Wenhui Xiao (Queensland University of Technology), Leo Lebrat (CSIRO Data61)
GenerationData SynthesisOptimizationNeural Radiance FieldImage
🎯 What it does: This study investigates the perspective selection problem in NeRF training and evaluation, proposing two perspective sampling methods: FVS and IGS.
NeRF On-the-go: Exploiting Uncertainty for Distractor-free NeRFs in the Wild
Weining Ren (ETH Zurich), Songyou Peng (ETH Zurich)
GenerationOptimizationNeural Radiance FieldImageVideo
🎯 What it does: A NeRF training method based on uncertainty prediction has been developed, which can automatically remove dynamic elements from 'wild' image sequences containing dynamic interfering objects and train high-quality static scene NeRF.
NeRF-HuGS: Improved Neural Radiance Fields in Non-static Scenes Using Heuristics-Guided Segmentation
Jiahao Chen (Sun Yat-sen University), Guanbin Li (Sun Yat-sen University)
RestorationSegmentationNeural Radiance FieldSimultaneous Localization and MappingImage
🎯 What it does: To address the issue of transient interfering objects (such as moving objects and shadows) when training NeRF in non-static scenes, this paper proposes a 'Heuristic Guided Segmentation' (HuGS) paradigm that combines handcrafted heuristics with segmentation models. It first uses two types of heuristics, Structure-from-Motion (SfM) and color residuals, to roughly separate static and transient regions, and then employs the Segment Anything Model (SAM) for fine segmentation, ultimately generating high-quality static image weights for NeRF training.
NeRFCodec: Neural Feature Compression Meets Neural Radiance Fields for Memory-Efficient Scene Representation
Sicheng Li (Zhejiang University), Lu Yu (Zhejiang University)
CompressionNeural Radiance FieldImage
🎯 What it does: Proposed NeRFCodec, an end-to-end compression framework for planar hybrid NeRF;
NeRFDeformer: NeRF Transformation from a Single View via 3D Scene Flows
Zhenggang Tang (University of Illinois Urbana-Champaign), Alexander Schwing (University of Illinois Urbana-Champaign)
GenerationData SynthesisNeural Radiance FieldOptical FlowImageMesh
🎯 What it does: Proposes the NeRFDeformer method, which uses a single RGBD image to non-rigidly deform an existing NeRF scene into a new scene, allowing for rendering from any viewpoint and generating the deformed mesh.
NeRFiller: Completing Scenes via Generative 3D Inpainting
Ethan Weber (Google Research), Angjoo Kanazawa (University of California Berkeley)
RestorationGenerationDiffusion modelNeural Radiance FieldImagePoint Cloud
🎯 What it does: Proposes NeRFiller, a generative 3D inpainting method that utilizes a 2D generative diffusion model to complete missing areas in 3D scenes.
NeRSP: Neural 3D Reconstruction for Reflective Objects with Sparse Polarized Images
Yufei Han (Beijing University of Posts and Telecommunications), Yunpeng Jia (Beijing University of Posts and Telecommunications)
GenerationDepth EstimationNeural Radiance FieldImagePoint Cloud
🎯 What it does: A NeRSP method utilizing sparse polarized images for 3D reconstruction of reflective surfaces is proposed.
NetTrack: Tracking Highly Dynamic Objects with a Net
Guangze Zheng (University of Hong Kong), Jia Pan (University of Hong Kong)
Object TrackingVideoBenchmark
🎯 What it does: A multi-object tracking framework named NetTrack is proposed to address the challenges of tracking high-dynamic objects in an open world.
NeuRAD: Neural Rendering for Autonomous Driving
Adam Tonderski (Zenseact), Christoffer Petersson (Zenseact)
Data SynthesisAutonomous DrivingNeural Radiance FieldPoint Cloud
🎯 What it does: We propose NeuRAD, an editable neural rendering method for dynamic autonomous driving scenarios that can synthesize realistic sensor data from cameras and LiDAR, supporting pose and subject editing.
Neural 3D Strokes: Creating Stylized 3D Scenes with Vectorized 3D Strokes
Hao-Bin Duan (Beihang University), Yong-Liang Yang (University of Bath)
GenerationData SynthesisOptimizationNeural Radiance FieldImage
🎯 What it does: Utilizing differentiable 3D stroke representations (basic geometries and B-Bézier curves) learned from multi-view 2D images, the stroke parameters are optimized directly through gradient descent to generate 3D scene renderings that maintain consistent geometry and style from any viewpoint.
Neural Clustering based Visual Representation Learning
Guikun Chen (Zhejiang University), Wenguan Wang (ETH Zurich)
Object DetectionSegmentationRepresentation LearningImage
🎯 What it does: A visual feature extraction framework FEC based on neural clustering is proposed, treating the feature extraction process as an adaptive selection and updating of pixel representatives, replacing traditional grid-based partitioning.
Neural Directional Encoding for Efficient and Accurate View-Dependent Appearance Modeling
Liwen Wu (University of California San Diego), Ravi Ramamoorthi (University of California San Diego)
Data SynthesisComputational EfficiencyNeural Radiance FieldImage
🎯 What it does: This paper proposes a Neural Direction Encoding (NDE) that extends the spatial encoding of feature grids into the directional domain to better model high-frequency view-dependent reflections, such as those from smooth metals.
Neural Exposure Fusion for High-Dynamic Range Object Detection
Emmanuel Onzon (Torc Robotics), Felix Heide (Princeton University)
Object DetectionAutonomous DrivingImage
🎯 What it does: This paper addresses object detection in outdoor high dynamic range scenes by proposing a method for fusing multi-exposure images in the feature domain.
Neural Fields as Distributions: Signal Processing Beyond Euclidean Space
Daniel Rebain (University of British Columbia), Andrea Tagliasacchi (Google DeepMind)
Image TranslationRestorationNeural Radiance FieldImage
🎯 What it does: By treating the filtering operation as a convolution of probability distributions, filtering is integrated into the training process of neural fields, allowing the model to achieve filtering effects during inference without additional convolution operations.
Neural Implicit Morphing of Face Images
Guilherme Schardong (Institute of Systems and Robotics, University of Coimbra), Nuno Gonçalves (Institute of Systems and Robotics, University of Coimbra)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: Utilizing a coordinate-based neural network to achieve implicit deformation and blending of facial images, completing continuous time facial modeling.
Neural Implicit Representation for Building Digital Twins of Unknown Articulated Objects
Yijia Weng (NVIDIA), Stan Birchfield (NVIDIA)
Object DetectionPose EstimationRepresentation LearningPoint CloudMesh
🎯 What it does: Solving the digital twin reconstruction of unknown multi-joint objects from RGB-D scans of two different joint states.
Neural Lineage
Runpeng Yu (National University of Singapore), Xinchao Wang (National University of Singapore)
Object DetectionSegmentationTransformerImage
🎯 What it does: This paper proposes a task for detecting generational relationships in neural networks and provides two methods: a no-learning method based on NTK linear approximation and a learning method based on Transformer.
Neural Markov Random Field for Stereo Matching
Tongfan Guan (Chinese University of Hong Kong), Yun-Hui Liu (Chinese University of Hong Kong)
Depth EstimationOptimizationConvolutional Neural NetworkGraph Neural NetworkTransformerImage
🎯 What it does: A completely data-driven Neural Markov Random Field (Neural MRF) framework is proposed for stereo matching, introducing a Disparity Proposal Network (DPN) to prune candidate disparity space at a coarse level, followed by refinement at a fine level.
Neural Modes: Self-supervised Learning of Nonlinear Modal Subspaces
Jiahong Wang (ETH Zurich), Bernhard Thomaszewski (ETH Zurich)
OptimizationComputational EfficiencyMeshPhysics Related
🎯 What it does: A self-supervised learning method for nonlinear modal subspaces (Neural Modes) is proposed for real-time physical simulation, directly minimizing mechanical energy during the training process, eliminating the reliance on manually simulated data.
Neural Parametric Gaussians for Monocular Non-Rigid Object Reconstruction
Devikalyan Das (Max Planck Institute for Informatics), Jan Eric Lenssen (Max Planck Institute for Informatics)
GenerationData SynthesisGaussian SplattingOptical FlowVideo
🎯 What it does: A two-stage neural parametric Gaussian model (NPGs) is proposed for reconstructing non-rigid objects from monocular videos and generating high-quality new views.
Neural Point Cloud Diffusion for Disentangled 3D Shape and Appearance Generation
Philipp Schröppel (University of Freiburg), Thomas Brox (University of Freiburg)
GenerationData SynthesisTransformerDiffusion modelNeural Radiance FieldAuto EncoderPoint Cloud
🎯 What it does: A neural point cloud diffusion model is proposed, capable of generating the shape and appearance of objects in three-dimensional space, and supports independent control of shape and appearance.
Neural Redshift: Random Networks are not Random Functions
Damien Teney (Idiap Research Institute), Ehsan Abbasnejad (University of Adelaide)
TransformerSequential
🎯 What it does: This study investigates the function bias (simplicity bias) of random weight neural networks and verifies that the architecture itself can determine the network's generalization ability before and after training.
Neural Refinement for Absolute Pose Regression with Feature Synthesis
Shuai Chen (University of Oxford), Victor Adrian Prisacariu (University of Oxford)
Pose EstimationNeural Radiance FieldImage
🎯 What it does: A method is proposed to refine the Absolute Pose Regression (APR) model through neural feature synthesis during testing.
Neural Sign Actors: A Diffusion Model for 3D Sign Language Production from Text
Vasileios Baltatzis (Imperial College London), Stefanos Zafeiriou (Queen's University Belfast)
GenerationPose EstimationRecurrent Neural NetworkGraph Neural NetworkDiffusion modelVideoText
🎯 What it does: A 3D sign language generation method based on diffusion models (Neural Sign Actors) is proposed, which can directly map English text to realistic and semantically accurate 3D sign language action sequences, generating animatable sign language avatars.
Neural Spline Fields for Burst Image Fusion and Layer Separation
Ilya Chugunov (Princeton University), Felix Heide (Princeton University)
Optical FlowImageVideo
🎯 What it does: A dual-layer alpha-composite image representation based on neural spline fields is proposed, which jointly performs high-resolution image fusion and separation of layers such as occlusion/reflection/shadow in mobile long bursts.
Neural Super-Resolution for Real-time Rendering with Radiance Demodulation
Jia Li (Shandong University), Lei Zhang (Hong Kong Polytechnic University)
RestorationSuper ResolutionConvolutional Neural NetworkRecurrent Neural NetworkVideo
🎯 What it does: This paper proposes a complete method for 4×4 super-resolution in real-time rendering, which first splits the rendered radiance into illumination and material components, performs super-resolution on the smooth illumination, and then overlaps it with high-resolution materials.
Neural Underwater Scene Representation
Yunkai Tang (Peking University), Boxin Shi (Peking University)
RestorationGenerationData SynthesisNeural Radiance FieldVideo
🎯 What it does: This paper proposes a neural radiance field (NeRF) model for underwater scenes that can simultaneously handle water attenuation, lighting variations, and dynamic objects, achieving high-quality underwater novel view synthesis.
Neural Video Compression with Feature Modulation
Jiahao Li (Microsoft Research), Yan Lu (Microsoft Research)
CompressionOptical FlowVideo
🎯 What it does: A neural video compression model DCVC-FM based on conditional coding is proposed, utilizing feature modulation to achieve a single model with a wide quality range, long prediction chain processing, support for RGB/YUV color spaces, and low-precision inference.
Neural Visibility Field for Uncertainty-Driven Active Mapping
Shangjie Xue (Georgia Institute of Technology), Danfei Xu (Georgia Institute of Technology)
OptimizationNeural Radiance FieldGaussian SplattingImage
🎯 What it does: Proposes the Neural Visibility Field (NVF), providing a probabilistic reasoning framework based on visibility and uncertainty for active 3D reconstruction and next best view selection.
NICE: Neurogenesis Inspired Contextual Encoding for Replay-free Class Incremental Learning
Mustafa Burak Gurbuz (Georgia Institute of Technology), Constantine Dovrolis (Cyprus Institute)
ClassificationConvolutional Neural NetworkImage
🎯 What it does: NICE is proposed, a replay-free class-incremental learning architecture inspired by adult neurogenesis, capable of achieving zero forgetting and automatically locating the required subnetworks during testing.
NIFTY: Neural Object Interaction Fields for Guided Human Motion Synthesis
Nilesh Kulkarni (University of Michigan), Leonidas Guibas (Stanford University)
GenerationData SynthesisPose EstimationTransformerDiffusion modelPoint Cloud
🎯 What it does: Proposes the NIFTY framework, which uses neural interaction fields and conditional diffusion models to generate interactions between humans and a single object; supplements training data through an automated data synthesis pipeline.
NightCC: Nighttime Color Constancy via Adaptive Channel Masking
Shuwei Li (National University of Singapore), Robby T. Tan (ASUS Intelligent Cloud Services National University of Singapore)
Image TranslationDomain AdaptationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: To address the color constancy (white balance) problem in nighttime scenes, an algorithm is proposed that utilizes daytime labeled data for unsupervised domain adaptation.
NIVeL: Neural Implicit Vector Layers for Text-to-Vector Generation
Vikas Thamizharasan (University of Massachusetts Amherst), Michal Lukac (Adobe Research)
GenerationData SynthesisDiffusion modelScore-based ModelImageText
🎯 What it does: A text-to-vector graphic generation method based on Neural Implicit Vector Layer (NIVeL) is proposed, capable of producing hierarchical, editable, and resolution-independent 2D shapes and colors.
No More Ambiguity in 360deg Room Layout via Bi-Layout Estimation
Yu-Ju Tsai (University of California Merced), Ming-Hsuan Yang (University of California Merced)
SegmentationDepth EstimationConvolutional Neural NetworkImage
🎯 What it does: Proposes the Bi-Layout model, which can simultaneously predict two types of layouts (enclosed and extended), addressing the ambiguity issue in 360° image indoor layout annotation.
No Time to Train: Empowering Non-Parametric Networks for Few-shot 3D Scene Segmentation
Xiangyang Zhu (City University of Hong Kong), Peng Gao (Tencent Youtu Lab)
SegmentationPoint Cloud
🎯 What it does: This paper proposes an untrained non-parametric network Seg-NN and a lightweight parametric version Seg-PN for few-shot 3D scene segmentation, completely omitting the pre-training and meta-learning phases.
NoiseCLR: A Contrastive Learning Approach for Unsupervised Discovery of Interpretable Directions in Diffusion Models
Yusuf Dalva (Virginia Tech), Pinar Yanardag (Virginia Tech)
GenerationData SynthesisExplainability and InterpretabilityDiffusion modelContrastive LearningImage
🎯 What it does: An unsupervised contrastive learning framework called NoiseCLR is proposed to automatically discover interpretable editing directions in pre-trained text-to-image diffusion models (such as Stable Diffusion) and achieve decoupled multi-directional editing both within and across domains.
NoiseCollage: A Layout-Aware Text-to-Image Diffusion Model Based on Noise Cropping and Merging
Takahiro Shirakawa (Kyushu University), Seiichi Uchida (Kyushu University)
GenerationDiffusion modelImageText
🎯 What it does: This paper presents NoiseCollage, a layout-aware diffusion model for multi-object image generation through noise cropping and merging under text-layout conditions.
Noisy One-point Homographies are Surprisingly Good
Yaqing Ding (Czech Technical University in Prague), Viktor Larsson (Lund University)
OptimizationImageBenchmark
🎯 What it does: A minimum solver is proposed that can estimate the global homography matrix using only a single feature point with scale and orientation information, and it is integrated with existing multi-point solvers within the RANSAC framework.
Noisy-Correspondence Learning for Text-to-Image Person Re-identification
Yang Qin (Sichuan University), Peng Hu (Sichuan University)
RecognitionRetrievalContrastive LearningImageText
🎯 What it does: A robust dual embedding method RDE is proposed to address the noise correspondence (NC) problem in text-image person re-identification, utilizing CCD to filter clean samples and employing TAL for stable triplet learning.
Non-autoregressive Sequence-to-Sequence Vision-Language Models
Kunyu Shi (Amazon Web Services AI Labs), Stefano Soatto (Amazon Web Services AI Labs)
GenerationComputational EfficiencyKnowledge DistillationTransformerVision Language ModelImageTextMultimodality
🎯 What it does: A non-autoregressive visual-language sequence-to-sequence model, NARVL, has been designed, which can decode the entire output sequence in parallel at once, significantly reducing inference latency.
Non-Rigid Structure-from-Motion: Temporally-Smooth Procrustean Alignment and Spatially-Variant Deformation Modeling
Jiawei Shi (Northwestern Polytechnical University), Yuchao Dai (Northwestern Polytechnical University)
Pose EstimationOptimizationVideo
🎯 What it does: A non-rigid structure reconstruction framework based on spatiotemporal modeling is proposed to address camera motion/rotation ambiguity and extreme spatial deformation issues.
NOPE: Novel Object Pose Estimation from a Single Image
Van Nguyen Nguyen (Ecole des Ponts), Vincent Lepetit (EPFL)
Pose EstimationConvolutional Neural NetworkAuto EncoderImage
🎯 What it does: A method is proposed that can predict the relative 3D pose of a target object in a new image using only a single reference image of the object, without the need for a known 3D model or training for new objects.
Normalizing Flows on the Product Space of SO(3) Manifolds for Probabilistic Human Pose Modeling
Olaf Dünkel, Florian Pfaff (University of Stuttgart)
Pose EstimationFlow-based ModelMultimodality
🎯 What it does: This paper proposes and implements a normalized flow model HuProSO3 on the SO(3) product space for learning the probability distribution of human poses, and applies it to tasks such as unconditional pose priors, inverse kinematics, partially observed inverse kinematics, and 2D to 3D lifting.
Not All Classes Stand on Same Embeddings: Calibrating a Semantic Distance with Metric Tensor
Jae Hyeon Park (Dongguk University), Sung In Cho (Dongguk University)
ClassificationRepresentation LearningImage
🎯 What it does: A consistency regularization method based on metric tensors is proposed, utilizing global and local metric tensors to calibrate the semantic distances of different categories in the embedding space, thereby enhancing the classification performance of semi-supervised learning.
Not All Prompts Are Secure: A Switchable Backdoor Attack Against Pre-trained Vision Transfomers
Sheng Yang, Shu-Tao Xia
Adversarial AttackTransformerImage
🎯 What it does: The paper explores a specific problem in the field of computer vision and proposes a new solution.
Not All Voxels Are Equal: Hardness-Aware Semantic Scene Completion with Self-Distillation
Song Wang (Zhejiang University), Jianke Zhu (Udeer.ai)
SegmentationAutonomous DrivingKnowledge DistillationTransformerPoint Cloud
🎯 What it does: This paper proposes a hardness-aware semantic scene completion framework (HASSC) that utilizes hardness voxel mining and self-distillation techniques to enhance the performance of semantic scene completion under single/multiple camera visual inputs.
Novel Class Discovery for Ultra-Fine-Grained Visual Categorization
Yu Liu (Dalian University of Technology), Nan Pu (University of Trento)
ClassificationRecognitionConvolutional Neural NetworkContrastive LearningImageAgriculture Related
🎯 What it does: This paper proposes the Ultra-Fine-Grained Novel Class Discovery (UFG-NCD) task and introduces the Region-Aligned Proxy Learning (RAPL) framework for discovering unlabeled ultra-fine-grained categories under partially labeled data.
Novel View Synthesis with View-Dependent Effects from a Single Image
Juan Luis Gonzalez Bello (Korea Advanced Institute of Science and Technology), Munchurl Kim (Korea Advanced Institute of Science and Technology)
Data SynthesisDepth EstimationConvolutional Neural NetworkTransformerImageVideo
🎯 What it does: This paper proposes a Novel View Synthesis (NVS) method based on a single image, while simultaneously learning View-Dependent Effects (VDE).
NRDF: Neural Riemannian Distance Fields for Learning Articulated Pose Priors
Yannan He (University of Tübingen), Gerard Pons-Moll (Max Planck Institute for Informatics)
GenerationPose EstimationPoint Cloud
🎯 What it does: Proposes Neural Riemannian Distance Fields (NRDF), which utilizes implicit neural distance fields to learn pose priors in high-dimensional quaternion product space, and achieves projection through adaptive step size Riemannian gradient descent.
NTO3D: Neural Target Object 3D Reconstruction with Segment Anything
Xiaobao Wei (Peking University), Shanghang Zhang (Peking University)
SegmentationNeural Radiance FieldPoint Cloud
🎯 What it does: This study proposes a neural target object 3D reconstruction method called NTO3D based on the Segment Anything Model (SAM). It utilizes a 3D occupancy field to elevate multi-view 2D SAM segmentation masks into a unified 3D space, and further maps the features of the SAM encoder to a 3D feature field, allowing users to obtain high-quality 3D reconstruction results of target objects with prompts given from a single viewpoint.