CVPR 2023 Papers — Page 15
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2353 papers
NeRF in the Palm of Your Hand: Corrective Augmentation for Robotics via Novel-View Synthesis
Allan Zhou (Stanford University), Chelsea Finn (Stanford University)
Data SynthesisRobotic IntelligenceConvolutional Neural NetworkReinforcement LearningNeural Radiance FieldImage
🎯 What it does: A NeRF-based offline data augmentation method called SPARTN has been developed to generate corrective noise in eye-hand camera visual tricks and enhance the robustness of behavior cloning.
NeRF-DS: Neural Radiance Fields for Dynamic Specular Objects
Zhiwen Yan (National University of Singapore), Gim Hee Lee (National University of Singapore)
GenerationData SynthesisNeural Radiance FieldVideo
🎯 What it does: This paper proposes NeRF-DS, a neural radiance field model for dynamic reflective objects, capable of reconstructing and rendering high-quality novel view images from monocular RGB videos.
NeRF-RPN: A General Framework for Object Detection in NeRFs
Benran Hu (Hong Kong University of Science and Technology), Chi-Keung Tang (Hong Kong University of Science and Technology)
Object DetectionConvolutional Neural NetworkTransformerNeural Radiance FieldPoint CloudBenchmark
🎯 What it does: Proposes the NeRF-RPN framework, which directly regresses 3D bounding boxes on the NeRF voxel representation, achieving 3D object detection without the need for rendering.
NeRF-Supervised Deep Stereo
Fabio Tosi (University of Bologna), Matteo Poggi (University of Bologna)
Data SynthesisDepth EstimationNeural Radiance FieldImage
🎯 What it does: A framework for training depth stereo matching networks without deep annotations and without a stereo camera is proposed—NeRF-Supervised Stereo.
NeRFInvertor: High Fidelity NeRF-GAN Inversion for Single-Shot Real Image Animation
Yu Yin (Northeastern University), Yun Fu (Northeastern University)
GenerationNeural Radiance FieldGenerative Adversarial NetworkImage
🎯 What it does: A general NeRF-GAN inversion method is proposed, capable of generating high-fidelity, three-dimensional consistent animations that maintain identity from a single real human face image.
Nerflets: Local Radiance Fields for Efficient Structure-Aware 3D Scene Representation From 2D Supervision
Xiaoshuai Zhang, Kyle Genova
SegmentationGenerationAutonomous DrivingNeural Radiance FieldImage
🎯 What it does: This paper proposes a 3D scene representation based on local neural radiance fields (nerflets), which can achieve novel view synthesis, semantic and instance segmentation, and scene editing using only 2D image supervision.
NeRFLight: Fast and Light Neural Radiance Fields Using a Shared Feature Grid
Fernando Rivas-Manzaneque (Universidad Polit' ecnica de Madrid), Angela Ribeiro (Centre for Automation and Robotics CSIC-UPM)
GenerationComputational EfficiencyNeural Radiance FieldImage
🎯 What it does: A lightweight real-time NeRF model called NeRFLight is proposed, achieving rendering speeds of over 150 FPS through shared feature grids and multi-density decoders.
NeRFLix: High-Quality Neural View Synthesis by Learning a Degradation-Driven Inter-Viewpoint MiXer
Kun Zhou (Chinese University of Hong Kong Shenzhen), Jiangbo Lu (SmartMore Corporation)
RestorationData SynthesisConvolutional Neural NetworkNeural Radiance FieldImageVideo
🎯 What it does: We propose NeRFLiX, a general NeRF visual restoration framework that significantly enhances the quality of rendered images without modifying the original NeRF model.
NeRFVS: Neural Radiance Fields for Free View Synthesis via Geometry Scaffolds
Chen Yang (Shanghai Jiao Tong University), Wei Shen (Shanghai Jiao Tong University)
GenerationData SynthesisNeural Radiance FieldPoint Cloud
🎯 What it does: A method for free-viewpoint synthesis of indoor scenes based on NeRF, called NeRFVS, is proposed.
NerVE: Neural Volumetric Edges for Parametric Curve Extraction From Point Cloud
Xiangyu Zhu (Chinese University of Hong Kong Shenzhen), Xiaoguang Han (Chinese University of Hong Kong Shenzhen)
SegmentationRepresentation LearningConvolutional Neural NetworkPoint Cloud
🎯 What it does: The NerVE network is proposed, which directly predicts structured voxelized edge representations, thereby extracting parametric curves from point clouds.
Network Expansion for Practical Training Acceleration
Ning Ding (Peking University), Yunhe Wang (Huawei)
SegmentationOptimizationComputational EfficiencyConvolutional Neural NetworkTransformerImage
🎯 What it does: By first extracting a sparse subnetwork from a dense network and gradually expanding the width or depth during the training process, dynamic network expansion is achieved, significantly accelerating training.
Network-Free, Unsupervised Semantic Segmentation With Synthetic Images
Qianli Feng, Aleix Martinez
SegmentationGenerative Adversarial NetworkImage
🎯 What it does: Using the style mixing attributes of StyleGAN2, semantic segmentation is achieved through pixel color correlation clustering under unsupervised conditions, without the need for additional networks, manual annotations, or supervised training.
NeuDA: Neural Deformable Anchor for High-Fidelity Implicit Surface Reconstruction
Bowen Cai (Alibaba Group), Huan Fu (Alibaba Group)
RestorationGenerationNeural Radiance FieldPoint CloudMesh
🎯 What it does: This paper proposes a neural implicit surface reconstruction method called Neural Deformable Anchor (NeuDA), which utilizes learnable 3D anchors in a multi-level voxel grid to adaptively capture surface details, and combines hierarchical frequency encoding to achieve high-frequency detail recovery.
NeUDF: Leaning Neural Unsigned Distance Fields With Volume Rendering
Yu-Tao Liu, Lin Gao
GenerationDepth EstimationConvolutional Neural NetworkPoint CloudMesh
🎯 What it does: This paper proposes a watertight surface reconstruction method based on multi-view images, utilizing depth estimation, voxel convolution, and Poisson reconstruction techniques to generate high-quality 3D meshes.
NeuFace: Realistic 3D Neural Face Rendering From Multi-View Images
Mingwu Zheng (Beihang University), Di Huang (Beihang University)
GenerationData SynthesisNeural Radiance FieldImage
🎯 What it does: This paper presents NeuFace, an end-to-end 3D neural facial rendering framework that simultaneously recovers facial geometry, diffuse reflection, specular reflection, and lighting from multi-view images.
Neumann Network With Recursive Kernels for Single Image Defocus Deblurring
Yuhui Quan (South China University of Technology), Hui Ji (National University of Singapore)
RestorationConvolutional Neural NetworkImage
🎯 What it does: Proposed NRKNet, an end-to-end single image defocus deblurring method.
NeuMap: Neural Coordinate Mapping by Auto-Transdecoder for Camera Localization
Shitao Tang (Simon Fraser University), Yasutaka Furukawa (Simon Fraser University)
Pose EstimationCompressionTransformerSupervised Fine-TuningSimultaneous Localization and MappingImage
🎯 What it does: An end-to-end neural coordinate mapping (NeuMap) method is designed, utilizing learnable spatial codes and a Transformer self-decoder to perform cross-attention on image features, thereby regressing 3D scene coordinates and achieving camera localization within a minimal storage space.
Neural Congealing: Aligning Images to a Joint Semantic Atlas
Dolev Ofri-Amar (Weizmann Institute of Science), Tali Dekel (Weizmann Institute of Science)
Image TranslationDomain AdaptationOptimizationTransformerContrastive LearningImage
🎯 What it does: This paper proposes Neural Congealing, a method that achieves automatic alignment of semantically common content across multiple images through self-supervised optimization learning of joint 2D maps and dense mappings during testing.
Neural Dependencies Emerging From Learning Massive Categories
Ruili Feng (University of Science and Technology of China), Zheng-Jun Zha (University of Science and Technology of China)
ClassificationConvolutional Neural NetworkTransformerSupervised Fine-TuningImage
🎯 What it does: This study investigates the phenomenon of neural dependence that occurs in large-scale image classification models, where the logits of certain categories can be obtained by a linear combination of a few other categories.
Neural Fields Meet Explicit Geometric Representations for Inverse Rendering of Urban Scenes
Zian Wang (NVIDIA), Sanja Fidler (NVIDIA)
Autonomous DrivingOptimizationNeural Radiance FieldImagePoint CloudMesh
🎯 What it does: We propose FEGR, a hybrid rendering framework that integrates neural fields and explicit meshes to reverse reconstruct the geometry, spatially transformed materials, and HDR lighting of large-scale urban scenes from images captured by pose-aware cameras, supporting lighting variations and virtual object insertion.
Neural Fourier Filter Bank
Zhijie Wu (University of British Columbia), Kwang Moo Yi (University of British Columbia)
RestorationGenerationData SynthesisNeural Radiance FieldImagePoint Cloud
🎯 What it does: A neural Fourier filter bank has been designed and implemented, capable of performing spatial partitioning and frequency encoding simultaneously on a multi-scale grid, thereby achieving efficient and high-quality 2D image fitting, 3D shape reconstruction, and NeRF view synthesis.
Neural Intrinsic Embedding for Non-Rigid Point Cloud Matching
Puhua Jiang (Tsinghua University), Ruqi Huang (Tsinghua University)
Representation LearningGraph Neural NetworkPoint Cloud
🎯 What it does: A Neural Intrinsic Embedding (NIE) model is proposed, which learns high-dimensional embedding representations directly from point clouds, and based on this embedding, a weakly supervised NIM matching network is constructed to achieve registration of non-rigid point clouds.
Neural Kaleidoscopic Space Sculpting
Byeongjoo Ahn (Carnegie Mellon University), Aswin C. Sankaranarayanan (Carnegie Mellon University)
GenerationOptimizationNeural Radiance FieldImage
🎯 What it does: This paper proposes a single-mirror panoramic 3D reconstruction method based on neural implicit surfaces (SDF), which achieves complete 3D reconstruction around an object using only the contours, background, foreground, and texture information of a single mirrored image, without explicitly estimating mirror labels.
Neural Kernel Surface Reconstruction
Jiahui Huang (NVIDIA), Francis Williams (NVIDIA)
RestorationGenerationAutonomous DrivingConvolutional Neural NetworkPoint Cloud
🎯 What it does: Reconstruct high-quality 3D implicit surfaces from sparse and noisy point clouds, achieving multi-domain generalization on a single model.
Neural Koopman Pooling: Control-Inspired Temporal Dynamics Encoding for Skeleton-Based Action Recognition
Xinghan Wang (Peking University), Yadong Mu (Peking University)
ClassificationRecognitionGraph Neural NetworkVideo
🎯 What it does: A high-order pooling module based on Koopman theory (Koopman pooling) is proposed to replace traditional average/max pooling, directly modeling the linear dynamics of skeletal action sequences for classification.
Neural Lens Modeling
Wenqi Xian (Cornell University), Christoph Lassner (Meta Reality Labs Research)
Data SynthesisOptimizationNeural Radiance FieldImage
🎯 What it does: A neural lens model based on reversible residual networks (i-ResNet) is proposed, which can simultaneously support projection and ray casting, and can be freely differentiated within an end-to-end optimization framework.
Neural Map Prior for Autonomous Driving
Xuan Xiong (Shanghai Qi Zhi Institute), Hang Zhao (Tsinghua University)
Autonomous DrivingRecurrent Neural NetworkSimultaneous Localization and MappingPoint Cloud
🎯 What it does: A Neural Map Prior (NMP) framework based on neural networks is proposed for the online learning and updating of HD semantic maps in autonomous driving, enabling bidirectional fusion of local map inference and global map prior during each frame of driving.
Neural Part Priors: Learning To Optimize Part-Based Object Completion in RGB-D Scans
Aleksei Bokhovkin, Angela Dai
SegmentationOptimizationPoint Cloud
🎯 What it does: Learning an optimizable neural part prior space in RGB-D scanning to achieve complete geometric recovery and segmentation of object parts.
Neural Pixel Composition for 3D-4D View Synthesis From Multi-Views
Aayush Bansal (Reality Labs Research), Michael Zollhöfer
GenerationData SynthesisDepth EstimationNeural Radiance FieldImageVideo
🎯 What it does: This paper studies a neural pixel synthesis method (NPC) for generating continuous 3D-4D views from sparse multi-view images.
Neural Preset for Color Style Transfer
Zhanghan Ke (City University of Hong Kong), Rynson W.H. Lau (City University of Hong Kong)
Image TranslationConvolutional Neural NetworkImage
🎯 What it does: A Neural Preset technique is proposed, achieving artifact-free, high-resolution color style transfer.
Neural Rate Estimator and Unsupervised Learning for Efficient Distributed Image Analytics in Split-DNN Models
Nilesh Ahuja (Intel Labs), Omesh Tickoo (Intel Labs)
SegmentationCompressionComputational EfficiencyKnowledge DistillationConvolutional Neural NetworkAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: A neural rate estimator based on variational autoencoders is proposed, utilizing a low-complexity bottleneck layer for unsupervised training in the Split-DNN model, achieving efficient image feature compression and distributed inference.
Neural Residual Radiance Fields for Streamably Free-Viewpoint Videos
Liao Wang, Minye Wu (KULeuven)
CompressionNeural Radiance FieldVideo
🎯 What it does: This paper designs a Residual Radiance Field (ReRF) framework for efficient compression and real-time streaming rendering of long sequence free-viewpoint videos.
Neural Scene Chronology
Haotong Lin (Zhejiang University), Noah Snavely (Cornell University)
GenerationData SynthesisNeural Radiance FieldSimultaneous Localization and MappingImage
🎯 What it does: Construct a time-varying 3D scene model using internet photos, achieving realistic rendering with independent control over perspective, time, and lighting.
Neural Texture Synthesis With Guided Correspondence
Yang Zhou (Shenzhen University), Hui Huang (Shenzhen University)
GenerationData SynthesisConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: In example texture synthesis, the CNN-MRF model is improved by proposing the Guided Correspondence distance and loss, achieving higher quality texture generation.
Neural Transformation Fields for Arbitrary-Styled Font Generation
Bin Fu (Shenzhen Institute of Advanced Technology), Yu Qiao (Shenzhen Institute of Advanced Technology)
GenerationGenerative Adversarial NetworkImage
🎯 What it does: A few-shot font generation model based on Neural Transformation Fields (NTF) is proposed, viewing font generation as a continuous transformation process of pixel creation and dissipation.
Neural Vector Fields: Implicit Representation by Explicit Learning
Xianghui Yang (University of Sydney), Luping Zhou (University of Sydney)
Representation LearningTransformerPoint CloudMesh
🎯 What it does: This paper proposes a Neural Vector Fields (NVF) method that directly predicts the displacement from query points to the surface by learning vector fields, achieving both implicit representation and explicit mesh deformation, capable of handling arbitrary topology and high-resolution surface reconstruction.
Neural Video Compression With Diverse Contexts
Jiahao Li (Microsoft Research), Yan Lu (Microsoft Research)
CompressionOptical FlowVideo
🎯 What it does: A new neural video encoder (DCVC-DC) has been designed within the deep contextual video compression framework to enhance coding efficiency through hierarchical quality structure, cross-group offset diversification, and quadtree partition entropy model.
Neural Volumetric Memory for Visual Locomotion Control
Ruihan Yang (University of California San Diego), Xiaolong Wang (Massachusetts Institute of Technology)
Robotic IntelligenceConvolutional Neural NetworkTransformerReinforcement LearningContrastive LearningPoint Cloud
🎯 What it does: Designed and implemented Neural Volumetric Memory (NVM), which constructs three-dimensional feature voxels through continuous observation using a forward-looking depth camera, providing input to the legged robot's walking controller, thereby enabling adaptive walking of the robot on complex terrains.
Neural Voting Field for Camera-Space 3D Hand Pose Estimation
Lin Huang (University at Buffalo), Zicheng Liu (Microsoft)
Pose EstimationImagePoint Cloud
🎯 What it does: A model named Neural Voting Field (NVF) is proposed, which utilizes a single RGB image to achieve 3D hand pose estimation in camera space through a 3D implicit function and dense voting.
Neuralangelo: High-Fidelity Neural Surface Reconstruction
Zhaoshuo Li (NVIDIA Research), Chen-Hsuan Lin (NVIDIA Research)
RestorationGenerationNeural Radiance FieldImage
🎯 What it does: The Neuralangelo framework is proposed, utilizing multi-resolution hash encoding and neural SDF for high-fidelity RGB image surface reconstruction.
NeuralDome: A Neural Modeling Pipeline on Multi-View Human-Object Interactions
Juze Zhang (ShanghaiTech University), Jingya Wang (ShanghaiTech University)
Object TrackingSegmentationGenerationPose EstimationNeural Radiance FieldVideoMultimodality
🎯 What it does: This paper constructs a large-scale multi-view human-object interaction dataset called HODome and proposes the NeuralDome pipeline, which enables multi-camera multi-person pose tracking, hierarchical neural reconstruction of dynamic humans and static objects, and free-viewpoint rendering.
NeuralEditor: Editing Neural Radiance Fields via Manipulating Point Clouds
Jun-Kun Chen (University of Illinois), Yu-Xiong Wang (University of Illinois)
GenerationData SynthesisNeural Radiance FieldPoint CloudBenchmark
🎯 What it does: Proposes NeuralEditor, which allows shape editing of NeRF through manipulation of point clouds.
NeuralField-LDM: Scene Generation With Hierarchical Latent Diffusion Models
Seung Wook Kim (NVIDIA), Sanja Fidler (Vector Institute)
GenerationData SynthesisAutonomous DrivingDiffusion modelScore-based ModelAuto EncoderImage
🎯 What it does: A 3D scene generation framework based on a hierarchical latent diffusion model is designed, capable of automatically generating high-quality open-world 3D scenes from multi-view images and depth data, and supports various post-processing such as conditional generation, scene editing, and style transfer.
Neuralizer: General Neuroimage Analysis Without Re-Training
Steffen Czolbe (University of Copenhagen), Adrian V. Dalca (Massachusetts Institute of Technology)
RestorationSegmentationConvolutional Neural NetworkMultimodalityBiomedical DataMagnetic Resonance Imaging
🎯 What it does: We propose Neuralizer, a single model that can perform various neuroimaging tasks without the need for retraining or fine-tuning by providing a set of task contexts;
NeuralLift-360: Lifting an In-the-Wild 2D Photo to a 3D Object With 360deg Views
Dejia Xu (University of Texas at Austin), Zhangyang Wang (University of Texas at Austin)
GenerationData SynthesisDepth EstimationDiffusion modelNeural Radiance FieldImage
🎯 What it does: We propose NeuralLift-360, which can elevate a single real scene photo into a complete 3D object and generate 360° new perspective renderings.
NeuralPCI: Spatio-Temporal Neural Field for 3D Point Cloud Multi-Frame Non-Linear Interpolation
Zehan Zheng (Tongji University), Changjun Jiang (Tongji University)
Data SynthesisAutonomous DrivingOptimizationNeural Radiance FieldPoint Cloud
🎯 What it does: A multi-frame point cloud interpolation method based on a 4D spatiotemporal neural field, NeuralPCI, is proposed to address the problem of 3D point cloud interpolation under nonlinear large motion.
NeuralUDF: Learning Unsigned Distance Fields for Multi-View Reconstruction of Surfaces With Arbitrary Topologies
Xiaoxiao Long (University of Hong Kong), Wenping Wang (Texas A&M University)
GenerationData SynthesisNeural Radiance FieldImagePoint CloudMesh
🎯 What it does: This study proposes the NeuralUDF method, which achieves 3D surface reconstruction of multi-view images by learning unsigned distance fields (UDF), supporting arbitrary topological structures.
Neuro-Modulated Hebbian Learning for Fully Test-Time Adaptation
Yushun Tang (Southern University of Science and Technology), Zhihai He (Southern University of Science and Technology)
Domain AdaptationImage
🎯 What it does: A Hebbian learning framework based on neural modulation is proposed for unsupervised full test-time adaptation;
NeurOCS: Neural NOCS Supervision for Monocular 3D Object Localization
Zhixiang Min (Stevens Institute of Technology), Manmohan Chandraker (NEC Laboratories America)
Object DetectionAutonomous DrivingNeural Radiance FieldPoint Cloud
🎯 What it does: This paper proposes a monocular 3D object localization framework called NeurOCS, which is based on Neural NOCS supervision. By utilizing instance masks and 3D bounding box information, it combines NeRF to achieve category-level shape learning and generates dense NOCS supervision through differentiable rendering, directly training single-pixel 3D coordinate regression for monocular 3D localization.
Neuron Structure Modeling for Generalizable Remote Physiological Measurement
Hao Lu (Hong Kong University of Science and Technology), Ying-Cong Chen
Domain AdaptationConvolutional Neural NetworkContrastive LearningVideo
🎯 What it does: The NEST method is proposed to solve the domain transfer problem in rPPG measurement through neural structure modeling.
NeuWigs: A Neural Dynamic Model for Volumetric Hair Capture and Animation
Ziyan Wang (Carnegie Mellon University), Christoph Lassner (Meta Reality Labs)
GenerationData SynthesisAuto EncoderVideoPoint Cloud
🎯 What it does: A two-stage end-to-end data-driven pipeline is proposed, which first compresses and tracks the hair state in multi-view videos using a volumetric autoencoder, and then trains a dynamic transfer network based on head motion and gravity to generate realistic hair animations without hair observations.
NewsNet: A Novel Dataset for Hierarchical Temporal Segmentation
Haoqian Wu (Tencent), Bernard Ghanem (National Tsing Hua University)
SegmentationTransformerVideoMultimodality
🎯 What it does: The researchers constructed a large-scale news video dataset called NewsNet and conducted a hierarchical temporal segmentation study on this dataset.
Next3D: Generative Neural Texture Rasterization for 3D-Aware Head Avatars
Jingxiang Sun (Tsinghua University), Yebin Liu (Tsinghua University)
GenerationData SynthesisPose EstimationGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a 3D GAN framework that can unsupervisedly learn high-quality, 3D-consistent, and animatable facial avatars from unstructured 2D images, supporting fine-grained control over full head rotation, expressions, eye blinking, gaze, and achieving high-fidelity rendering.
NICO++: Towards Better Benchmarking for Domain Generalization
Xingxuan Zhang (Tsinghua University), Peng Cui (Tsinghua University)
Domain AdaptationConvolutional Neural NetworkImageBenchmark
🎯 What it does: This paper presents a large-scale, domain-wide NICO++ dataset and provides a more reasonable evaluation protocol for the domain generalization problem.
NIFF: Alleviating Forgetting in Generalized Few-Shot Object Detection via Neural Instance Feature Forging
Karim Guirguis (Robert Bosch GmbH), Jürgen Beyerer (Robert Bosch GmbH)
Object DetectionKnowledge DistillationImage
🎯 What it does: A data-free knowledge distillation framework called NIFF is proposed for generalized few-shot object detection (G-FSOD) under privacy-preserving and low-memory conditions, alleviating the forgetting problem of base classes by synthesizing instance-level features instead of images.
Nighttime Smartphone Reflective Flare Removal Using Optical Center Symmetry Prior
Yuekun Dai (Nanyang Technological University), Chen Change Loy (Nanyang Technological University)
RestorationConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes a method for removing reflection spots from nighttime mobile phone camera images based on optical center symmetry priors, and constructs a new BracketFlare dataset.
NIKI: Neural Inverse Kinematics With Invertible Neural Networks for 3D Human Pose and Shape Estimation
Jiefeng Li (Shanghai Jiao Tong University), Cewu Lu (Shanghai Jiao Tong University)
Pose EstimationFlow-based ModelImage
🎯 What it does: A method for inverse kinematics based on reversible neural networks, NIKI, is proposed for robust and pixel-aligned 3D human pose and shape estimation from monocular images.
NIPQ: Noise Proxy-Based Integrated Pseudo-Quantization
Juncheol Shin (POSTECH), Eunhyeok Park (POSTECH)
Object DetectionSuper ResolutionOptimizationConvolutional Neural NetworkTransformerImage
🎯 What it does: A noise proxy-based integrated pseudo-quantization (NIPQ) method is proposed, which achieves quantization-aware training of low-precision networks by replacing STE with pseudo-quantization noise during the training process, and supports unified quantization of weights and activations as well as mixed precision scheduling.
NIRVANA: Neural Implicit Representations of Videos With Adaptive Networks and Autoregressive Patch-Wise Modeling
Shishira R. Maiya (University of Maryland), Abhinav Shrivastava (University of Maryland)
CompressionVideo
🎯 What it does: A self-regressive Patch-wise video implicit representation (INR) framework called NIRVANA is designed for efficient video compression.
NLOST: Non-Line-of-Sight Imaging With Transformer
Yue Li (University of Science and Technology of China), Zhiwei Xiong (University of Science and Technology of China)
RestorationTransformerImage
🎯 What it does: This paper proposes NLOST, a Transformer-based non-line-of-sight (NLOS) imaging reconstruction network that can recover three-dimensional hidden scenes from time-resolved indirect reflection signals.
No One Left Behind: Improving the Worst Categories in Long-Tailed Learning
Yingxiao Du (Nanjing University), Jianxin Wu (Nanjing University)
ClassificationRecognitionOptimizationSupervised Fine-TuningImage
🎯 What it does: A post-training fine-tuning plugin is proposed that can be directly applied to existing long-tail recognition models, enhancing the recall rate for extreme categories through retraining the classifier.
Noisy Correspondence Learning With Meta Similarity Correction
Haochen Han (Xi'an Jiaotong University), Minnan Luo (Xi'an Jiaotong University)
RetrievalMeta LearningImageText
🎯 What it does: This study investigates similarity correction and model robustness enhancement in image-text retrieval under the presence of noise through a meta-learning framework.
NoisyQuant: Noisy Bias-Enhanced Post-Training Activation Quantization for Vision Transformers
Yijiang Liu (Nanjing University), Shanghang Zhang (Peking University)
ClassificationObject DetectionTransformerImage
🎯 What it does: Proposes NoisyQuant: In post-training quantization (PTQ), a fixed uniform noise bias is added to the activation of each layer, and then the original output is restored through denoising bias after multiplication, significantly reducing quantization error.
NoisyTwins: Class-Consistent and Diverse Image Generation Through StyleGANs
Harsh Rangwani (Indian Institute of Science Bangalore), R. Venkatesh Babu (Indian Institute of Science Bangalore)
GenerationData SynthesisGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: The NoisyTwins method is proposed to improve the mode collapse and class confusion issues in long-tail conditional generation with StyleGAN.
Non-Contrastive Learning Meets Language-Image Pre-Training
Jinghao Zhou (Microsoft), Furu Wei (Microsoft)
RetrievalRepresentation LearningTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: This paper explores the introduction of non-contrastive learning methods into language-image pre-training and proposes the xCLIP multi-task framework, which integrates the advantages of CLIP and nCLIP to achieve stronger zero-shot transfer and representation learning.
Non-Contrastive Unsupervised Learning of Physiological Signals From Video
Jeremy Speth (University of Notre Dame), Adam Czajka (University of Notre Dame)
Representation LearningConvolutional Neural NetworkVideoBiomedical Data
🎯 What it does: An unsupervised, non-contrastive learning framework SiNC is proposed to directly regress blood volume pulse signals from facial videos.
Non-Line-of-Sight Imaging With Signal Superresolution Network
Jianyu Wang (Tsinghua University), Xing Fu (Tsinghua University)
RestorationSuper ResolutionConvolutional Neural NetworkSupervised Fine-TuningImagePoint Cloud
🎯 What it does: A two-step learning-based NLOS imaging pipeline is proposed, where a neural network first recovers high-resolution signals from sparsely sampled low-resolution instantaneous signals, and then existing high-resolution imaging algorithms are used to reconstruct hidden objects.
NoPe-NeRF: Optimising Neural Radiance Field With No Pose Prior
Wenjing Bian (Active Vision Lab, University of Oxford), Victor Adrian Prisacariu (Active Vision Lab, University of Oxford)
Pose EstimationDepth EstimationOptimizationNeural Radiance FieldImage
🎯 What it does: This paper proposes NoPe-NeRF, an end-to-end method for jointly estimating camera pose and NeRF, capable of achieving high-quality novel view synthesis without pose priors.
Normal-Guided Garment UV Prediction for Human Re-Texturing
Yasamin Jafarian (University of Minnesota), Hyun Soo Park (Adobe Research)
Image TranslationGenerationOptical FlowImageVideo
🎯 What it does: A self-supervised method based on geometric isometric constraints is proposed, which can predict continuous UV maps of clothing from a single image or video, enabling physically feasible texture editing without 3D reconstruction.
Normalizing Flow Based Feature Synthesis for Outlier-Aware Object Detection
Nishant Kumar (TU Dresden), Stefan Gumhold (Carl Zeiss Meditec AG)
Object DetectionAnomaly DetectionFlow-based ModelImageVideo
🎯 What it does: This paper proposes a feature synthesis framework (FFS) based on normalized flows, which generates low-likelihood synthetic anomalous features by learning the joint distribution of features in the inlier class, thereby achieving accurate identification of outlier samples in object detection.
Not All Image Regions Matter: Masked Vector Quantization for Autoregressive Image Generation
Mengqi Huang (University of Science and Technology of China), Yongdong Zhang (Beijing University of Posts and Telecommunications)
GenerationData SynthesisTransformerAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a two-stage autoregressive image generation framework. In the first stage, the Mask-Vector-Quantization VAE (MQ-VAE) quantizes only the important region features through an adaptive mask module. In the second stage, Stackformer predicts the quantized codes and their positions in the 2D feature map, achieving more efficient and higher quality image generation.
Novel Class Discovery for 3D Point Cloud Semantic Segmentation
Luigi Riz (Fondazione Bruno Kessler), Fabio Poiesi (Fondazione Bruno Kessler)
Object DetectionSegmentationConvolutional Neural NetworkPoint Cloud
🎯 What it does: This paper proposes a new category discovery framework NOPS for 3D point cloud semantic segmentation based on online clustering and uncertainty quantification, which can automatically identify and segment unlabelled new category points under the premise of having only base category labels.
Novel-View Acoustic Synthesis
Changan Chen (University of Texas at Austin), Andrea Vedaldi (FAIR Meta AI)
GenerationData SynthesisConvolutional Neural NetworkVideoMultimodalityAudio
🎯 What it does: A new task called Novel-View Acoustic Synthesis (NVAS) is proposed, and a Visually-Guided Acoustic Synthesis (ViGAS) network is designed to synthesize scene audio from unobserved viewpoints using visual information.
NS3D: Neuro-Symbolic Grounding of 3D Objects and Relations
Joy Hsu (Stanford University), Jiajun Wu (Stanford University)
Object DetectionRepresentation LearningTransformerLarge Language ModelPoint Cloud
🎯 What it does: This paper presents NS3D, a framework that parses natural language into neural symbolic programs and performs object and relationship localization in 3D scenes.
NULL-Text Inversion for Editing Real Images Using Guided Diffusion Models
Ron Mokady (Google Research), Daniel Cohen-Or (Google Research)
RestorationGenerationData SynthesisPrompt EngineeringDiffusion modelImageStochastic Differential Equation
🎯 What it does: This paper proposes a method to map real images into the text-guided diffusion model space through Null-text Inversion and utilizes Prompt-to-Prompt for text editing.
NUWA-LIP: Language-Guided Image Inpainting With Defect-Free VQGAN
Minheng Ni (Harbin Institute of Technology), Wangmeng Zuo (Peng Cheng Laboratory)
RestorationGenerationTransformerGenerative Adversarial NetworkImageText
🎯 What it does: A language-guided image inpainting method called N WA-LIP is proposed, which can fill in missing areas based on text descriptions while keeping the undamaged regions unchanged.
NVTC: Nonlinear Vector Transform Coding
Runsen Feng (University of Science and Technology of China), Zhibo Chen (University of Science and Technology of China)
CompressionImage
🎯 What it does: This paper proposes a novel neural network image compression framework—Nonlinear Vector Transformation Coding (NVTC), which achieves efficient compression through multi-stage product vector quantization, nonlinear vector transformation, and entropy-constrained vector quantization.
Objaverse: A Universe of Annotated 3D Objects
Matt Deitke (Allen Institute for AI), Ali Farhadi (University of Washington)
Object DetectionSegmentationGenerationRobotic IntelligenceVision Language ModelContrastive LearningPoint CloudMesh
🎯 What it does: This paper presents Objaverse 1.0, a large dataset containing 818K high-quality 3D models and their natural language descriptions, and validates its potential value in 3D generation, instance segmentation, embodied AI, and robustness evaluation of visual models through four experiments.
Object Detection With Self-Supervised Scene Adaptation
Zekun Zhang (Stony Brook University), Minh Hoai (VinAI Research)
Object DetectionDomain AdaptationConvolutional Neural NetworkContrastive LearningVideo
🎯 What it does: A self-supervised scene-adaptive object detection framework is proposed, utilizing a pre-trained detector to generate pseudo-labels combined with tracking, location-aware artifact-free Mixup, and dynamic background extraction, significantly improving detection accuracy in fixed camera scenarios.
Object Discovery From Motion-Guided Tokens
Zhipeng Bao (Carnegie Mellon University), Martial Hebert (Carnegie Mellon University)
Object DetectionSegmentationAutonomous DrivingConvolutional Neural NetworkRecurrent Neural NetworkContrastive LearningVideo
🎯 What it does: An unsupervised video object discovery framework based on motion-guided discrete tokens (MoTok) is proposed.
Object Pop-Up: Can We Infer 3D Objects and Their Poses From Human Interactions Alone?
Ilya A. Petrov (University of Tuebingen), Gerard Pons-Moll (University of Tuebingen)
Object DetectionPose EstimationSupervised Fine-TuningPoint Cloud
🎯 What it does: Predict the position and posture of 3D objects from only human body point clouds and target object category information.
Object Pose Estimation With Statistical Guarantees: Conformal Keypoint Detection and Geometric Uncertainty Propagation
Heng Yang (NVIDIA Research), Marco Pavone (NVIDIA Research)
Object DetectionPose EstimationImageStochastic Differential Equation
🎯 What it does: Transform heatmap keypoint detection into a statistically reliable set of circle/ellipse predictions, and propagate these uncertainties to 6D poses, providing a computable worst-case error upper bound.
Object-Aware Distillation Pyramid for Open-Vocabulary Object Detection
Luting Wang (Beihang University), Si Liu (Beihang University)
Object DetectionKnowledge DistillationConvolutional Neural NetworkTransformerVision Language ModelImage
🎯 What it does: Proposes the Object-Aware Distillation Pyramid (OADP) framework to extract complete and pure object knowledge from the pre-trained Vision-and-Language model (CLIP) for open vocabulary object detection, and transfers this knowledge to the Faster R-CNN detector through a three-level distillation (global, block-level, object-level).
Object-Goal Visual Navigation via Effective Exploration of Relations Among Historical Navigation States
Heming Du (Australian National University), Xin Yu (Netease Fuxi AI Lab)
Robotic IntelligenceTransformerReinforcement LearningMultimodality
🎯 What it does: A historical heuristic navigation learning framework HiNL is proposed for object goal visual navigation.
ObjectMatch: Robust Registration Using Canonical Object Correspondences
Can Gümeli (Technical University of Munich), Matthias Nießner (Technical University of Munich)
Object DetectionPose EstimationConvolutional Neural NetworkSimultaneous Localization and MappingImagePoint Cloud
🎯 What it does: This paper presents ObjectMatch, a method that utilizes the Normalized Object Coordinate System (NOCS) to construct indirect correspondences in low-overlap RGB-D image pairs, thereby achieving more robust camera pose estimation and SLAM.
ObjectStitch: Object Compositing With Diffusion Model
Yizhi Song (Purdue University), Daniel Aliaga (Adobe Research)
Image HarmonizationGenerationData SynthesisTransformerDiffusion modelImage
🎯 What it does: A self-supervised object synthesis framework based on diffusion models (ObjectStitch) is proposed, capable of achieving geometric correction, color harmonization, shadow generation, and perspective synthesis without the need for manual annotations, enabling one-stop realistic synthesis.
Observation-Centric SORT: Rethinking SORT for Robust Multi-Object Tracking
Jinkun Cao (Carnegie Mellon University), Kris Kitani (Carnegie Mellon University)
Object TrackingAutonomous DrivingVideo
🎯 What it does: This paper proposes an observation-centric multi-object tracking framework (OC-SORT) that improves robustness against occlusion and nonlinear motion through observation-driven re-update and momentum consistency based on Kalman filtering.
Occlusion-Free Scene Recovery via Neural Radiance Fields
Chengxuan Zhu (Peking University), Boxin Shi (Peking University)
RestorationData SynthesisNeural Radiance FieldImage
🎯 What it does: Proposes an unsupervised occlusion removal method based on NeRF, capable of synthesizing occlusion-free scene images from multiple viewpoints.
OCELOT: Overlapped Cell on Tissue Dataset for Histopathology
Jeongun Ryu (Lunit Inc), Sérgio Pereira (Lunit Inc)
Object DetectionSegmentationConvolutional Neural NetworkImageBiomedical Data
🎯 What it does: The OCELOT dataset was constructed, and a multi-task learning method based on cell detection and tissue segmentation was developed to improve the accuracy of cell detection in pathological images.
OCTET: Object-Aware Counterfactual Explanations
Mehdi Zemni (Valeo AI), Matthieu Cord (Sorbonne Université)
Autonomous DrivingExplainability and InterpretabilityAdversarial AttackGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes an object-oriented adversarial explanation framework called OCTET, designed to generate interpretable adversarial examples for visual models.
OcTr: Octree-Based Transformer for 3D Object Detection
Chao Zhou (Beihang University), Di Huang (Beihang University)
Object DetectionAutonomous DrivingTransformerPoint Cloud
🎯 What it does: This paper proposes a Transformer based on octrees (OcTr), which significantly enhances global context modeling in sparse point clouds through the OctAttn mechanism that adaptively sparsifies attention in a multi-scale feature pyramid.
Octree Guided Unoriented Surface Reconstruction
Chamin Hewa Koneputugodage (Australian National University), Stephen Gould (Australian National University)
GenerationOptimizationPoint CloudMeshBenchmark
🎯 What it does: Construct a discrete octree from an undirected point cloud, label internal/external leaves, and use this as a prior to guide the training of implicit neural representations (INR), ultimately obtaining a closed and high-precision 3D surface.
Omni Aggregation Networks for Lightweight Image Super-Resolution
Hang Wang (Shanghai Jiao Tong University), Jinfan Liu (Shanghai Jiao Tong University)
RestorationSuper ResolutionTransformerImage
🎯 What it does: A lightweight visual Transformer super-resolution framework called Omni-SR is proposed, aiming to enhance the performance of small models in high-quality image reconstruction.
Omni3D: A Large Benchmark and Model for 3D Object Detection in the Wild
Garrick Brazil (Meta AI), Georgia Gkioxari (Caltech)
Object DetectionImageBenchmark
🎯 What it does: A large-scale multi-scene 3D object detection benchmark, OMNI3D, is proposed, and a general Cube R‑CNN model is trained on it;
OmniAL: A Unified CNN Framework for Unsupervised Anomaly Localization
Ying Zhao (Ricoh Software Research Center)
Anomaly DetectionConvolutional Neural NetworkImage
🎯 What it does: This paper presents OmniAL, a unified unsupervised anomaly localization framework for multi-class industrial images, capable of performing detection and localization tasks with just one model.
OmniAvatar: Geometry-Guided Controllable 3D Head Synthesis
Hongyi Xu (ByteDance Inc), Linjie Luo (ByteDance Inc)
GenerationData SynthesisNeural Radiance FieldGenerative Adversarial NetworkMesh
🎯 What it does: Trained a controllable 3D head generation model that supports independent adjustment of parameters such as camera pose, facial expression, head shape, neck, and chin joints.
OmniCity: Omnipotent City Understanding With Multi-Level and Multi-View Images
Weijia Li (Sun Yat-Sen University), Dahua Lin (Wuhan University)
Object DetectionSegmentationDepth EstimationConvolutional Neural NetworkImageBenchmark
🎯 What it does: The OmniCity dataset is proposed, integrating satellite images with street view panoramas/single views, containing 25K geographic locations, 100K pixel-level fine-grained building instances, and multi-task annotations such as height;
OmniMAE: Single Model Masked Pretraining on Images and Videos
Rohit Girdhar (Meta AI), Ishan Misra (Meta AI)
Computational EfficiencyRepresentation LearningTransformerAuto EncoderImageVideoMultimodality
🎯 What it does: This paper proposes a single Vision Transformer (ViT) model called OmniMAE, which jointly pre-trains on both image and video visual modalities through unsupervised masked autoencoding (MAE) and directly transfers to downstream tasks.
Omnimatte3D: Associating Objects and Their Effects in Unconstrained Monocular Video
Mohammed Suhail, Forrester Cole
RestorationObject DetectionSegmentationConvolutional Neural NetworkSupervised Fine-TuningVideo
🎯 What it does: This paper proposes a hierarchical decomposition method for unconstrained monocular videos, dividing the video into a background layer and multiple foreground layers, and separating objects in the foreground layers along with their shadows, reflections, and other related effects.
OmniObject3D: Large-Vocabulary 3D Object Dataset for Realistic Perception, Reconstruction and Generation
Tong Wu (SenseTime Research), Ziwei Liu (Shanghai Artificial Intelligence Laboratory)
Object DetectionSegmentationGenerationNeural Radiance FieldVideoPoint CloudMeshBenchmark
🎯 What it does: This paper presents OmniObject3D—a dataset consisting of 6,000 real scanned 3D objects across 190 categories, equipped with rich annotations such as textured meshes, point clouds, multi-view renderings, videos + masks + camera poses, and establishes benchmarks and evaluation tracks for four tasks (3D perception, view synthesis, neural surface reconstruction, 3D generation).