ECCV 2024 Papers — Page 9
European Conference on Computer Vision · 2387 papers
FinePseudo: Improving Pseudo-Labelling through Temporal-Alignablity for Semi-Supervised Fine-Grained Action Recognition
Ishan Rajendrakumar Dave (University of Central Florida), Mubarak Shah (University of Central Florida)
ClassificationRecognitionTransformerContrastive LearningVideo
🎯 What it does: This paper proposes a semi-supervised fine-grained action recognition framework called FinePseudo, which improves model performance by generating pseudo labels through time-aligned metric learning.
FipTR: A Simple yet Effective Transformer Framework for Future Instance Prediction in Autonomous Driving
Xingtai Gui, Chi Zhang
Autonomous DrivingTransformerImage
🎯 What it does: Proposed a fully end-to-end Transformer framework called FipTR, which directly predicts BEV instance segmentation and motion states in future perspectives using instance queries, eliminating auxiliary outputs and post-processing;
Fisher Calibration for Backdoor-Robust Heterogeneous Federated Learning
Wenke Huang (Wuhan University), Dacheng Tao (Nanyang Technological University)
Federated LearningAdversarial AttackImage
🎯 What it does: Propose the Self-Driven Fisher Calibration (SDFC) method, which in heterogeneous federated learning measures the difference in parameter importance between local and validation distributions through the Fisher information matrix, then introduces FDReg regularization during local training and assigns weighted aggregation during model aggregation based on the differences, thereby defending against backdoor attacks.
FisherRF: Active View Selection and Mapping with Radiance Fields using Fisher Information
Wen Jiang (University of Pennsylvania), Kostas Daniilidis (University of Pennsylvania)
OptimizationNeural Radiance FieldGaussian SplattingSimultaneous Localization and MappingImageMesh
🎯 What it does: Quantify the observed information of radiance field model parameters using Fisher information, and propose an active view selection and mapping algorithm based on Expected Information Gain (EIG).
Flash Cache: Reducing Bias in Radiance Cache Based Inverse Rendering
Benjamin Attal (Carnegie Mellon University), Pratul Srinivasan
RestorationOptimizationNeural Radiance FieldImageBenchmarkPhysics Related
🎯 What it does: Propose an unbiased illumination cache inverse rendering method, leveraging volume rendering combined with physical rendering to recover geometry, material, and lighting;
Flash-Splat: 3D Reflection Removal with Flash Cues and Gaussian Splats
Mingyang Xie (University of Maryland), Christopher A. Metzler (Massachusetts Institute of Technology)
RestorationGaussian SplattingImage
🎯 What it does: Propose the Flash-Splat method, which utilizes unpaired flash/non-flash multi-view images to achieve light-aware reflection separation and 3D reconstruction through 3D Gaussian Splatting.
FlashSplat: 2D to 3D Gaussian Splatting Segmentation Solved Optimally
Qiuhong Shen (National University of Singapore), Xinchao Wang (National University of Singapore)
SegmentationOptimizationGaussian SplattingImage
🎯 What it does: This paper proposes a method to efficiently map 2D segmentation masks to 3D Gaussian splatting rendering, completing 3D segmentation with a single global optimal solution.
FlashTex: Fast Relightable Mesh Texturing with LightControlNet
Kangle Deng (Carnegie Mellon University), Maneesh Agrawala (Carnegie Mellon University)
GenerationData SynthesisComputational EfficiencyKnowledge DistillationDiffusion modelScore-based ModelImageTextMeshBenchmark
🎯 What it does: Automatically generate textures for input 3D meshes that can be correctly relit under any lighting conditions based on user-provided text prompts.
FLAT: Flux-aware Imperceptible Adversarial Attacks on 3D Point Clouds
Keke Tang (Guangzhou University), Zhihong Tian (Guangzhou University)
Adversarial AttackPoint Cloud
🎯 What it does: Studied a flux-aware imperceptible adversarial attack framework named FLAT, aiming to generate more stealthy attack point clouds by controlling the flux of point cloud perturbation vector fields to maintain local uniformity.
Flatness-aware Sequential Learning Generates Resilient Backdoors
Hoang Pham (VinUniversity), Khoa D Doan
OptimizationAdversarial AttackImage
🎯 What it does: This paper proposes a Sequential Backdoor Learning (SBL) framework that utilizes continual learning techniques to actively resist fine-tuning defenses during training, maintaining high attack success rates for backdoor models.
FlexAttention for Efficient High-Resolution Vision-Language Models
Junyan Li (University of Massachusetts Amherst), Chuang Gan (University of Massachusetts Amherst)
Computational EfficiencyTransformerLarge Language ModelVision Language ModelMultimodality
🎯 What it does: Propose the FlexAttention mechanism for high-resolution vision-language models, which can dynamically select important high-resolution image features and fuse them with low-resolution features and text features, significantly reducing the computational cost of self-attention.
Flexible Distribution Alignment: Towards Long-tailed Semi-supervised Learning with Proper Calibration
Emanuel Sanchez Aimar (Linköping University), Michael Felsberg (Linköping University)
ClassificationDomain AdaptationConvolutional Neural NetworkImage
🎯 What it does: Propose a framework named ADELLO for long-tailed semi-supervised learning, with the core being Flexible Distribution Alignment (FlexDA) and complementary consistency regularization, dynamically aligning the distribution of unlabeled data and achieving model calibration.
FlexiEdit: Frequency-Aware Latent Refinement for Enhanced Non-Rigid Editing
Gwanhyeong Koo (Korea Advanced Institute of Science and Technology), Chang D. Yoo (Korea Advanced Institute of Science and Technology)
GenerationDiffusion modelImage
🎯 What it does: Propose the FlexiEdit method, achieving non-rigid image editing through frequency-aware latent refinement and a re-inversion three-branch network.
Flow-Assisted Motion Learning Network for Weakly-Supervised Group Activity Recognition
Muhammad Adi Nugroho (Korea Advanced Institute of Science and Technology), Changick Kim (Korea Advanced Institute of Science and Technology)
RecognitionConvolutional Neural NetworkTransformerContrastive LearningOptical FlowVideo
🎯 What it does: Designed and trained Flaming-Net, a weakly supervised group activity recognition network that uses optical flow during training to guide the motion perception encoder and employs a dual-path relationship module to aggregate individual long-term dynamics and short-term spatiotemporal relationships.
FlowCon: Out-of-Distribution Detection using Flow-based Contrastive Learning
Saandeep Aathreya (University of South Florida), Shaun Canavan (University of South Florida)
Anomaly DetectionFlow-based ModelContrastive LearningImage
🎯 What it does: Proposed a flow-based contrastive learning method called FlowCon for out-of-distribution (OOD) detection without modifying the pre-trained classifier.
Flowed Time of Flight Radiance Fields
Mikhail Okunev (Brown University), James Tompkin (Brown University)
RestorationDepth EstimationNeural Radiance FieldOptical FlowVideoPhysics Related
🎯 What it does: Propose a method to correct motion artifacts in continuous-wave time-of-flight (C-ToF) imaging through 4D volume field reconstruction, utilizing a physically differentiable C-ToF simulator and weak optical flow supervision to estimate scene geometry and motion;
Flying with Photons: Rendering Novel Views of Propagating Light
Anagh Malik (University of Toronto), David B. Lindell
GenerationRepresentation LearningNeural Radiance FieldImageVideoTime SeriesPhysics Related
🎯 What it does: Propose a neural volume rendering method based on transient fields, utilizing transient video data captured by multi-view ultrafast cameras to generate light propagation videos from arbitrary moving viewpoints.
FMBoost: Boosting Latent Diffusion with Flow Matching
Johannes S Fischer, Björn Ommer (LMU Munich)
GenerationData SynthesisSuper ResolutionConvolutional Neural NetworkDiffusion modelFlow-based ModelImageTextMultimodalityOrdinary Differential Equation
🎯 What it does: This paper proposes FMBoost, a high-resolution image synthesis framework that combines Latent Diffusion Models (LDM) with Flow Matching (FM). It first generates semantic information in a low-resolution latent space using a small-scale diffusion model, then maps low-resolution latent vectors to a high-resolution latent space via a flow matching model, and finally reconstructs pixel space through a pre-trained convolutional decoder, achieving fast, scalable high-resolution generation while maintaining diversity.
FocusDiffuser: Perceiving Local Disparities for Camouflaged Object Detection
Jianwei Zhao (University of Electronic Science and Technology of China), Hong Cheng (University of Electronic Science and Technology of China)
Object DetectionDiffusion modelImage
🎯 What it does: Proposed a covert target detection framework called FocusDiffuser based on diffusion models.
Follow the Rules: Reasoning for Video Anomaly Detection with Large Language Models
Yuchen Yang (Johns Hopkins University), Shao-Yuan Lo (Honda Research Institute USA)
Anomaly DetectionTransformerLarge Language ModelVision Language ModelVideoText
🎯 What it does: Developed a rule-based reasoning framework for anomaly video detection called AnomalyRuler, which induces LLMs to generate detection rules using a small number of normal samples and identifies anomalies based on rules during inference.
FontStudio: Shape-Adaptive Diffusion Model for Coherent and Consistent Font Effect Generation
Xinzhi Mu (Microsoft), Yuhui Yuan (Microsoft)
GenerationTransformerLarge Language ModelDiffusion modelMultimodality
🎯 What it does: Developed a system named FontStudio that utilizes shape-adaptive diffusion models to generate multilingual font effects on arbitrary shape font canvases, ensuring consistent multi-glyph effects through a training-free effect transfer method;
Forbes: Face Obfuscation Rendering via Backpropagation Refinement Scheme
Jintae Kim (Korea University), Chang-Su Kim (Korea University)
Safty and PrivacyConvolutional Neural NetworkImageMultimodality
🎯 What it does: This paper proposes a multi-modal image fusion method that unifies eight image preprocessing and fusion operations, including mosaic, mean filtering, color scaling, distortion, frequency domain sine fusion, and random noise suppression, into a single energy minimization framework.
Forecasting Future Videos from Novel Views via Disentangled 3D Scene Representation
Sudhir Yarram (State University of New York at Buffalo), Junsong Yuan (State University of New York at Buffalo)
GenerationDepth EstimationAutonomous DrivingVideoPoint Cloud
🎯 What it does: This paper proposes a spatiotemporal video extrapolation method based on discretized 3D point clouds, capable of predicting future frames from given historical frames and rendering them from new viewpoints.
Forest2Seq: Revitalizing Order Prior for Sequential Indoor Scene Synthesis
Qi Sun (USTC), Houqiang Li (USTC)
GenerationData SynthesisRepresentation LearningTransformerImageMesh
🎯 What it does: Reorganize unordered collections of indoor scene objects into a hierarchical tree/forest structure, and use a Transformer-based autoregressive model to generate realistic 3D indoor scenes based on this order.
Forget More to Learn More: Domain-specific Feature Unlearning for Semi-supervised and Unsupervised Domain Adaptation
Hritam Basak (Stony Brook University), Zhaozheng Yin (Stony Brook University)
ClassificationDomain AdaptationConvolutional Neural NetworkTransformerAuto EncoderContrastive LearningImage
🎯 What it does: Designed a three-stage 'learn-forget-then-learn more' framework, first learning domain-specific features for source and target domains separately via a dedicated Encoder-Classifier, then actively forgetting these domain-specific information using a reconstruction network and uncertainty loss, and finally learning domain-agnostic task features through Gaussian-guided Latent Alignment (GLA) combined with supervised classification loss.
Formula-Supervised Visual-Geometric Pre-training
Ryosuke Yamada (AIST), Yutaka Satoh (AIST)
ClassificationObject DetectionSegmentationData SynthesisTransformerImagePoint Cloud
🎯 What it does: This paper proposes a formula-based supervised visual-geometric pre-training method called FSVGP, enabling images and point clouds to be pre-trained on a unified Transformer model.
Foster Adaptivity and Balance in Learning with Noisy Labels
Mengmeng Sheng (Nanjing University of Science and Technology), Yazhou Yao (Nanjing University of Science and Technology)
ClassificationImage
🎯 What it does: Propose a noise label learning framework SED that combines adaptive and class-balanced sample selection, mean teacher label correction, and dynamic truncated normal distribution weighting.
FoundPose: Unseen Object Pose Estimation with Foundation Features
Evin Pınar Örnek (Technical University of Munich), Tomas Hodan (Meta Reality Labs)
Pose EstimationRetrievalOptimizationTransformerContrastive LearningImageRetrieval-Augmented Generation
🎯 What it does: Proposes a model called FoundPose for 6D pose estimation, which utilizes frozen DINOv2 intermediate layer features to achieve fast template retrieval, correspondence establishment, and untrained pose refinement, enabling high-precision localization of unseen objects from a single RGB image.
Four Ways to Improve Verbo-visual Fusion for Dense 3D Visual Grounding
Ozan Unal (ETH Zurich), Luc Van Gool (ETH Zurich)
Object DetectionSegmentationConvolutional Neural NetworkTransformerVision Language ModelContrastive LearningTextPoint Cloud
🎯 What it does: Propose a Dense 3D visual localization network called ConcreteNet based on point clouds, which directly outputs instance-level masks instead of only bounding boxes, and enhances the handling of repeated instances, viewpoint-dependent descriptions, and multi-view uncertainties through four innovative modules.
FouriScale: A Frequency Perspective on Training-Free High-Resolution Image Synthesis
Linjiang Huang (Chinese University of Hong Kong MMLab), Hongsheng Li (SenseTime Research)
GenerationConvolutional Neural NetworkDiffusion modelImageText
🎯 What it does: This paper proposes a training-free method called FouriScale, which utilizes dilated convolutions and low-pass filtering in the frequency domain to achieve structural and scale consistency with pre-trained diffusion models, thereby generating high-resolution images of arbitrary sizes without retraining.
FRDiff : Feature Reuse for Universal Training-free Acceleration of Diffusion Models
Junhyuk So (Pusan National University), Eunhyeok Park (Pusan National University)
Super ResolutionConvolutional Neural NetworkDiffusion modelOptical FlowVideo
🎯 What it does: This paper studies the time sampling strategy in video super-resolution and proposes a training framework based on unified sampling and dynamic α weight;
Free Lunch for Gait Recognition: A Novel Relation Descriptor
Jilong Wang (University of Science and Technology of China), Liang Wang (Institute of Automation, Chinese Academy of Sciences)
ClassificationRecognitionContrastive LearningVideo
🎯 What it does: This paper proposes a gait relationship descriptor based on classifier weights to enhance gait recognition performance.
Free-ATM: Harnessing Free Attention Masks for Representation Learning on Diffusion-Generated Images
David Junhao Zhang (National Univeristy of Singapore), Mike Zheng Shou (ByteDance)
Representation LearningVision Language ModelDiffusion modelContrastive LearningImageMultimodality
🎯 What it does: Leverage synthetic images generated by diffusion models and their cross-attention masks to enhance three types of representation learning methods: contrastive learning, masked image modeling, and vision-language pre-training, proposing the Free-ATM scheme;
Free-Editor: Zero-shot Text-driven 3D Scene Editing
Nazmul Karim (University of Central Florida), Jing Hua (Wayne State University)
GenerationTransformerVision Language ModelNeural Radiance FieldImageText
🎯 What it does: Propose a zero-shot text-driven 3D scene editing method called Free-Editor, which achieves style transfer across the entire scene by editing a single view without requiring retraining.
Free-Viewpoint Video of Outdoor Sports Using a Drone
Zhengdong Hong (Zhejiang University)
Image TranslationGenerationData SynthesisPose EstimationTransformerNeural Radiance FieldImageVideo
🎯 What it does: Proposed a system based on a single-camera drone that automatically orbits around athletes to capture dynamic athletes and large-scale outdoor scenes, achieving 4D reconstruction and generating 360° free-viewpoint videos at arbitrary times.
Free-VSC: Free Semantics from Visual Foundation Models for Unsupervised Video Semantic Compression
Yuan Tian (Shanghai Jiao Tong University), Guangtao Zhai (Shanghai Jiao Tong University)
CompressionTransformerPrompt EngineeringGenerative Adversarial NetworkVideo
🎯 What it does: Propose the Free-VSC framework, leveraging the rich semantics provided by visual foundation models (VFM) to achieve unsupervised video semantic compression, and design a prompt alignment layer (Prom-SAL) and dynamic trajectory entropy model to achieve the goal of maintaining semantic integrity during compression.
FreeAugment: Data Augmentation Search Across All Degrees of Freedom
Tom Bekor (Technion - Israel Institute of Technology), Lihi Zelnik-Manor (Technion - Israel Institute of Technology)
Hyperparameter SearchData-Centric LearningImage
🎯 What it does: Proposed a fully differentiable data augmentation search framework called FreeAugment, which can simultaneously learn four degrees of freedom of the augmentation policy: depth, type, order, and intensity.
FreeCompose: Generic Zero-Shot Image Composition with Diffusion Prior
Zhekai Chen (Zhejiang University), Chunhua Shen (Zhejiang University)
GenerationDiffusion modelImage
🎯 What it does: Proposes FreeCompose, which achieves general zero-shot image synthesis by leveraging the generative prior of pre-trained diffusion models.
FreeDiff: Progressive Frequency Truncation for Image Editing with Diffusion Models
Wei WU, Antoni Chan (City University of Hong Kong)
Image TranslationPrompt EngineeringDiffusion modelImageText
🎯 What it does: Propose a text-driven image editing method called FreeDiff, which does not require fine-tuning and is based on frequency domain truncation. It uses advanced frequency truncation to progressively refine the guidance of diffusion models, enabling unified processing for various editing tasks.
FreeInit: Bridging Initialization Gap in Video Diffusion Models
Tianxing Wu (Nanyang Technological University), Ziwei Liu (Nanyang Technological University)
GenerationDiffusion modelVideo
🎯 What it does: This paper analyzes the problem of poor temporal consistency caused by mismatched initial noise distributions in video diffusion models during training and inference, and proposes a novel, no-training, parameter-free sampling strategy called FreeInit to address this gap.
FreeMotion: A Unified Framework for Number-free Text-to-Motion Synthesis
Ke Fan (Shanghai Jiao Tong University), Lizhuang Ma (Shanghai Jiao Tong University)
GenerationData SynthesisTransformerLarge Language ModelVision Language ModelDiffusion modelTextMultimodality
🎯 What it does: This paper proposes a unified 'FreeMotion' framework that recursively utilizes conditional motion distribution to generate human motion with arbitrary numbers of people, supporting spatial control for multiple individuals.
FreeMotion: MoCap-Free Human Motion Synthesis with Multimodal Large Language Models
Zhikai Zhang (Tsinghua University), Li Yi (Tsinghua University)
GenerationData SynthesisTransformerLarge Language ModelAuto EncoderWorld ModelTextMultimodality
🎯 What it does: Proposed FreeMotion, an open-source framework for human motion synthesis that completely does not rely on motion capture (mocap) data, generating keyframes based on natural language instructions using multimodal large language models (MLLMs), and achieving continuous motion through interpolation with physical motion tracking;
FreestyleRet: Retrieving Images from Style-Diversified Queries
Hao Li (Peking University), Li Yuan
RetrievalTransformerPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: Proposed a image retrieval task based on diverse-style queries, constructed the Diverse-Style Retrieval dataset, and designed a lightweight FreestyleRet framework
Freeview Sketching: View-Aware Fine-Grained Sketch-Based Image Retrieval
Aneeshan Sain (University of Surrey), Yi-Zhe Song (University of Surrey)
RetrievalConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: Designed a perspective-aware fine-grained sketch image retrieval system capable of flexible retrieval in both perspective-agnostic and perspective-specific modes.
FreeZe: Training-free zero-shot 6D pose estimation with geometric and vision foundation models
Andrea Caraffa (Fondazione Bruno Kessler), Fabio Poiesi (Fondazione Bruno Kessler)
Pose EstimationTransformerImagePoint CloudBenchmark
🎯 What it does: Propose a training-free zero-shot 6D object pose estimation method called FreeZe, which extracts point-level features using pre-trained geometric and visual foundation models, and achieves target pose prediction through RANSAC-based 3D-3D matching and symmetry-aware fine-tuning.
FrePolad: Frequency-Rectified Point Latent Diffusion for Point Cloud Generation
Chenliang Zhou (University of Cambridge), A. Cengiz Oztireli
GenerationData SynthesisConvolutional Neural NetworkDiffusion modelFlow-based ModelAuto EncoderPoint Cloud
🎯 What it does: Propose FrePolad, a point cloud generation framework that combines VAE with latent DDPM and introduces spherical harmonic frequency correction during VAE training to enhance the preservation of high-frequency details, supporting the generation of point clouds with arbitrary numbers of points.
Frequency-Spatial Entanglement Learning for Camouflaged Object Detection
Yanguang Sun (Nanjing University of Science and Technology), Lei Luo (Nanjing University of Science and Technology)
Object DetectionConvolutional Neural NetworkTransformerImage
🎯 What it does: Proposed a frequency-spectrum and spatial entanglement learning framework (FSEL) that combines frequency domain and spatial domain features for concealed object detection.
FREST: Feature RESToration for Semantic Segmentation under Multiple Adverse Conditions
Sohyun Lee (POSTECH), Suha Kwak (POSTECH)
SegmentationDomain AdaptationTransformerContrastive LearningImage
🎯 What it does: Proposed a source-free domain adaptation feature recovery framework (FREST), which achieves robust improvement for semantic segmentation under various adverse conditions by alternately learning conditional embedding space and adversarial feature recovery.
FRI-Net: Floorplan Reconstruction via Room-wise Implicit Representation
Honghao Xu (Shenzhen University), Ruizhen Hu (Shenzhen University)
GenerationTransformerPoint Cloud
🎯 What it does: For reconstructing 2D floor plans from 3D point clouds, FRI-Net is proposed, which generates more regular room polygons by leveraging implicit representations and structured regularization for each room.
From Fake to Real: Pretraining on Balanced Synthetic Images to Prevent Spurious Correlations in Image Recognition
Maan Qraitem (Boston University), Bryan A. Plummer (Boston University)
ClassificationRecognitionData SynthesisConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: Propose a two-stage training pipeline FFR, pre-training on balanced synthetic images first, then fine-tuning on real images to reduce spurious correlation in visual recognition models;
From Pixels to Objects: A Hierarchical Approach for Part and Object Segmentation Using Local and Global Aggregation
Yunfei Xie (Huazhong University of Science and Technology), Jieru Mei (Johns Hopkins University)
SegmentationTransformerImage
🎯 What it does: Proposed LGFormer, a hierarchical Transformer model that can simultaneously perform object segmentation and part segmentation tasks.
Frontier-enhanced Topological Memory with Improved Exploration Awareness for Embodied Visual Navigation
Xinru Cui (Shanghai Jiao Tong University), Hesheng Wang (Shanghai Jiao Tong University)
Robotic IntelligenceRecurrent Neural NetworkGraph Neural NetworkNeural Radiance FieldSimultaneous Localization and MappingImage
🎯 What it does: Proposed Frontier-Enhanced Topological Memory (FTM), which integrates ghost nodes with perceptual features into topological maps, predicts visual information for them using an online-trained NeRF, and achieves end-to-end image goal navigation with a multi-stage memory extraction module.
FroSSL: Frobenius Norm Minimization for Efficient Multiview Self-Supervised Learning
Oscar Skean (University of Kentucky), Luis G Sanchez Giraldo (University of Kentucky)
Computational EfficiencyRepresentation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: Proposed a new self-supervised learning objective, FroSSL, combining Frobenius norm regularization with mean squared error to maintain view invariance, aiming to achieve faster convergence through multi-view learning and explicit control of the eigenvalues of the covariance matrix.
Frugal 3D Point Cloud Model Training via Progressive Near Point Filtering and Fused Aggregation
Donghyun Lee (Seoul National University), Hongil Yoon (Google)
Computational EfficiencyPoint Cloud
🎯 What it does: Accelerate the most time-consuming FPS and aggregation operations in 3D point cloud model training by proposing two optimization techniques: L-FPS and fused aggregation.
FSD-BEV: Foreground Self-Distillation for Multi-view 3D Object Detection
Zheng Jiang (Beihang University), Yunhong Wang (Beihang University)
Object DetectionDepth EstimationAutonomous DrivingKnowledge DistillationImagePoint Cloud
🎯 What it does: Propose a Foreground Self-Distillation (FSD) framework to enhance the performance of Bird’s-Eye-View (BEV) 3D object detection based on multi-view cameras.
FSGS: Real-Time Few-shot View Synthesis using Gaussian Splatting
Zehao Zhu, Zhangyang Wang
GenerationData SynthesisNeural Radiance FieldGaussian SplattingImagePoint Cloud
🎯 What it does: This paper proposes a few-view synthesis framework called FSGS based on 3D Gaussian grating, achieving real-time and high-quality novel view rendering.
FTBC: Forward Temporal Bias Correction for Optimizing ANN-SNN Conversion
Xiaofeng Wu (City University of Macau), Kai Zou (ProtagoLabs Inc)
ClassificationConvolutional Neural NetworkSpiking Neural NetworkImage
🎯 What it does: This paper proposes a forward temporal bias correction (FTBC) method to improve the conversion from artificial neural networks (ANN) to spiking neural networks (SNN), significantly enhancing the classification accuracy of SNN within limited time steps.
Fully Authentic Visual Question Answering Dataset from Online Communities
Chongyan Chen (University of Texas at Austin), Danna Gurari (University of Texas at Austin)
Large Language ModelVision Language ModelMultimodalityBenchmark
🎯 What it does: This paper constructs a visual question answering (VQAonline) dataset entirely sourced from a real online community and evaluates the performance of six state-of-the-art vision-language models on this dataset;
Fully Sparse 3D Occupancy Prediction
Haisong Liu (Nanjing University), Limin Wang (Nanjing University)
Autonomous DrivingTransformerImagePoint Cloud
🎯 What it does: Designed a fully sparse 3D occupancy prediction network called SparseOcc, which first recovers a sparse 3D structure from multi-view images using a sparse voxel decoder, and then predicts semantic/instance occupancy information within the sparse space through a mask transformer.
Functional Transform-Based Low-Rank Tensor Factorization for Multi-Dimensional Data Recovery
Jian-Li Wang, Xi-Le Zhao
RestorationImageVideo
🎯 What it does: Propose a functional transform (Functional Transform)-driven low-rank tensor decomposition framework (FLRTF) for multi-dimensional data recovery, including tasks such as video frame interpolation/extrapolation, MSI band interpolation, and spectral super-resolution.
Fundamental Matrix Estimation Using Relative Depths
Yaqing Ding (Czech Technical University in Prague), Zuzana Kukelova (Czech Technical University in Prague)
Pose EstimationDepth EstimationImage
🎯 What it does: Proposed a minimal solver for estimating the fundamental matrix of dual cameras using relative depth information, and presented three solving methods: 4p4d, 4p3d, and 3p3d;
FunQA: Towards Surprising Video Comprehension
Binzhu Xie (Beijing University of Posts and Telecommunications), Ziwei Liu (Nanyang Technological University)
TransformerLarge Language ModelPrompt EngineeringVision Language ModelVideoTextMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: Propose the FunQA video question-answering dataset, focusing on humorous, creative, and magic surprise videos, and design multi-task evaluation.
FuseTeacher: Modality-fused Encoders are Strong Vision Supervisors
Chen-Wei Xie (Alibaba Group), Yun Zheng (University Of Science And Technology Of China)
ClassificationRetrievalKnowledge DistillationRepresentation LearningTransformerVision Language ModelContrastive LearningMultimodality
🎯 What it does: By introducing a lightweight fusion encoder to perform multi-modal fusion of images and text, and using the fused representation for classification distillation and retrieval distillation in the visual encoder, the FuseTeacher method achieves stronger visual representation learning.
FutureDepth: Learning to Predict the Future Improves Video Depth Estimation
Rajeev Yasarla (Qualcomm AI Research), Fatih Porikli (Qualcomm AI Research)
Depth EstimationConvolutional Neural NetworkTransformerAuto EncoderVideo
🎯 What it does: Proposes a video depth estimation method called FutureDepth, which learns to predict future frame features (F-Net) and reconstructs multi-frame features using an adaptive mask (R-Net) during training, thereby injecting motion and correspondence information into the decoder and enhancing depth details through a refinement network.
FYI: Flip Your Images for Dataset Distillation
Byunggwan Son (Yonsei University), Bumsub Ham (Yonsei University)
ClassificationKnowledge DistillationImage
🎯 What it does: Proposed the FYI method, which introduces horizontal flipping during dataset distillation and concatenates the flipped images with original synthetic images, addressing the symmetry issues caused by bilateral equivalence;
G2fR: Frequency Regularization in Grid-based Feature Encoding Neural Radiance Fields
Shuxiang Xie (University of Tokyo), Takeshi Oishi (National Institute of Advanced Industrial Science and Technology)
GenerationNeural Radiance FieldImage
🎯 What it does: The paper explains the expressive power of NeRF based on Grid-Based Feature Encoding (GFE) through Fourier analysis theory, and proposes a general grid frequency regularization method called G2fR to address the issues of local minima and generalization in camera pose optimization and sparse sample reconstruction.
G3R: Gradient Guided Generalizable Reconstruction
Yun Chen (Waabi), Raquel Urtasun (Waabi)
OptimizationComputational EfficiencyRepresentation LearningConvolutional Neural NetworkNeural Radiance FieldGaussian SplattingImage
🎯 What it does: Propose a generalizable large-scale scene reconstruction method called G3R, which utilizes a gradient-guided learning optimizer to rapidly reconstruct high-quality 3D Gaussian representations from multi-view images and initial geometric structures.
GalLop: Learning global and local prompts for vision-language models
Marc Lafon (Conservatoire national des arts et métiers), Nicolas Thome (Sorbonne Université)
ClassificationAnomaly DetectionConvolutional Neural NetworkTransformerPrompt EngineeringVision Language ModelImageText
🎯 What it does: Propose the GalLoP method, which improves the accuracy and robustness of Vision-Language Models (VLM) in few-shot classification, domain generalization, and OOD detection by learning diverse global and local prompts.
GAMMA-FACE: GAussian Mixture Models Amend Diffusion Models for Bias Mitigation in Face Images
Basudha Pal (Johns Hopkins University), Rama Chellappa (University of Texas at Dallas)
ClassificationGenerationDiffusion modelImage
🎯 What it does: Propose a method that utilizes a Gaussian Mixture Model (GMM) to segment the latent space of diffusion models, generating debiased facial images without retraining the model, and using these images for data augmentation in downstream classification tasks.
GAReT: Cross-view Video Geolocalization with Adapters and Auto-Regressive Transformers
Manu S Pillai (University of Central Florida), Mubarak Shah (University of Central Florida)
RetrievalDomain AdaptationTransformerContrastive LearningImageVideo
🎯 What it does: This paper proposes the GAReT method, which migrates a Vision Transformer-based image geolocation model to video input through a GeoAdapter, and generates temporally consistent GPS trajectories using TransRetriever.
GarmentAligner: Text-to-Garment Generation via Retrieval-augmented Multi-level Corrections
Shiyue Zhang, Xiaodan Liang (Lenovo Research)
GenerationRetrievalDiffusion modelContrastive LearningImageTextRetrieval-Augmented Generation
🎯 What it does: Propose a text-to-fabric image generation model named GarmentAligner, which enhances the fine-grained alignment of generated results using retrieval-enhanced multi-level correction techniques.
GarmentCodeData: A Dataset of 3D Made-to-Measure Garments With Sewing Patterns
Maria Korosteleva (ETH Zurich), Olga Sorkine-Hornung (ETH Zurich)
GenerationData SynthesisMesh
🎯 What it does: This paper proposes and releases the first large-scale synthetic dataset, GarmentCodeData, containing 115,000 3D garment models and their corresponding sewing patterns, along with a complete generation pipeline;
Gated Temporal Diffusion for Stochastic Long-term Dense Anticipation
Olga Zatsarynna (University of Bonn), Jürgen Gall
GenerationData SynthesisConvolutional Neural NetworkDiffusion modelVideo
🎯 What it does: Proposes the Gated Temporal Diffusion (GTD) model, which jointly models the uncertainty of observations and future actions using diffusion models to generate diverse long-term action predictions;
GAURA: Generalizable Approach for Unified Restoration and Rendering of Arbitrary Views
Vinayak Gupta (Indian Institute of Technology Madras), Kaushik Mitra (Massachusetts Institute of Technology)
RestorationTransformerNeural Radiance FieldImage
🎯 What it does: Propose GAURA, an end-to-end learning framework capable of performing 3D scene reconstruction and novel view rendering under any type of degradation.
GaussCtrl: Multi-View Consistent Text-Driven 3D Gaussian Splatting Editing
Jing Wu (University of Oxford), Victor Adrian Prisacariu (University of Oxford)
GenerationVision Language ModelDiffusion modelGaussian SplattingTextPoint Cloud
🎯 What it does: Edit 3D Gaussian Splatting scenes using text instructions, proposing the GaussCtrl method, which achieves multi-view consistent 3D editing through depth-conditioned ControlNet and attention alignment.
Gaussian Frosting: Editable Complex Radiance Fields with Real-Time Rendering
Antoine Guédon (Ecole des Ponts), Vincent Lepetit (Ecole des Ponts)
GenerationComputational EfficiencyGaussian SplattingImagePoint CloudMesh
🎯 What it does: We propose Gaussian Frosting, a hybrid representation that constructs variable-thickness Gaussian layers on a mesh, enabling real-time rendering and supporting editing and animation with traditional tools;
Gaussian Grouping: Segment and Edit Anything in 3D Scenes
Mingqiao Ye (ETH Zurich), Lei Ke (ETH Zurich)
SegmentationGenerationGaussian SplattingImage
🎯 What it does: Proposes Gaussian Grouping, combining 3D Gaussian Splatting with 2D masks generated by SAM to achieve unsupervised instance and stuff-level segmentation for complete 3D scenes, supporting various local edits (deletion, filling, reorganization, coloring, etc.).
Gaussian in the wild: 3D Gaussian Splatting for Unconstrained Image Collections
Dongbin Zhang (Tsinghua University), Haoqian Wang (Tsinghua University)
GenerationData SynthesisConvolutional Neural NetworkGaussian SplattingImage
🎯 What it does: This paper proposes a method called 'Gaussian in the Wild (GS‑W)' based on 3D Gaussian Splatting, for synthesizing high-quality images with varying viewpoints from unconstrained real-world image collections.
Gaussian Splatting on the Move: Blur and Rolling Shutter Compensation for Natural Camera Motion
Otto Seiskari (Spectacular AI), Arno Solin (Spectacular AI)
RestorationGenerationPose EstimationNeural Radiance FieldGaussian SplattingSimultaneous Localization and MappingOptical FlowImageVideo
🎯 What it does: This paper proposes a method for high-quality scene reconstruction and novel view synthesis under handheld camera motion using Gaussian Splatting, with compensation for motion blur and rolling shutter distortion.
GaussianFormer: Scene as Gaussians for Vision-Based 3D Semantic Occupancy Prediction
Yuanhui Huang (Tsinghua University), Jiwen Lu (Tsinghua University)
Autonomous DrivingComputational EfficiencyTransformerGaussian SplattingImage
🎯 What it does: Proposes GaussianFormer, which converts multi-view images into 3D semantic occupancy predictions using sparse 3D Gaussian representations, and generates dense occupancy maps through local splatting from Gaussians to voxels.
GaussianImage: 1000 FPS Image Representation and Compression by 2D Gaussian Splatting
Xinjie Zhang (Hong Kong University of Science and Technology), Jun Zhang (Hong Kong University of Science and Technology)
CompressionGaussian SplattingImage
🎯 What it does: Proposed a GaussianImage method based on 2D Gaussian distribution for image representation and compression, achieving an image encoder with low memory usage, fast training, and extremely high rendering speed.
GaussReg: Fast 3D Registration with Gaussian Splatting
Jiahao Chang (Chinese University of Hong Kong Shenzhen), Xiaoguang Han (Chinese University of Hong Kong Shenzhen)
Pose EstimationConvolutional Neural NetworkTransformerGaussian SplattingPoint Cloud
🎯 What it does: Proposes the GaussReg framework for fast and accurate registration of 3D scenes based on Gaussian Splatting (GS);
Gaze Target Detection Based on Head-Local-Global Coordination
Yaokun Yang (Beihang University), Feng Lu (Beihang University)
Pose EstimationDepth EstimationConvolutional Neural NetworkContrastive LearningVideoBenchmark
🎯 What it does: Proposes an end-to-end framework that leverages head, local (FOV-based) views, and global views for collaborative prediction of gaze targets, introducing position and representation consistency mechanisms among the three views.
GazeXplain: Learning to Predict Natural Language Explanations of Visual Scanpaths
Xianyu Chen (University of Minnesota), Qi Zhao (University of Minnesota)
Explainability and InterpretabilityTransformerLarge Language ModelImageTextMultimodality
🎯 What it does: This paper proposes the GazeXplain model, which can simultaneously predict visual scan paths and generate corresponding natural language explanations, and constructs and annotates an eye-tracking dataset containing explanations.
General and Task-Oriented Video Segmentation
Mu Chen (University of Technology Sydney), Yi Yang (Zhejiang University)
SegmentationTransformerContrastive LearningVideo
🎯 What it does: Designed a unified task-oriented video segmentation framework called GvSeg, capable of simultaneously handling four types of tasks: example-guided, instance, semantic, and panoptic video segmentation.
General Geometry-aware Weakly Supervised 3D Object Detection
Guowen Zhang (The Hong Kong Polytechnic University), Lei Zhang (The Hong Kong Polytechnic University)
Object DetectionAutonomous DrivingConvolutional Neural NetworkLarge Language ModelImagePoint Cloud
🎯 What it does: Proposes a generic geometry-aware weakly supervised 3D object detection framework GGA, which learns 3D bounding boxes by leveraging 2D boxes and shape ratio priors generated by LLMs.
GeneralAD: Anomaly Detection Across Domains by Attending to Distorted Features
Luc P.J. Sträter (University of Amsterdam), Yuki M. Asano (University of Amsterdam)
Anomaly DetectionTransformerImage
🎯 What it does: Designed a general-purpose anomaly detection framework called GeneralAD, which leverages pre-trained Vision Transformer features to generate pseudo-anomalous features and achieves image-level and pixel-level anomaly detection and localization through a cross-patch attention discriminator.
Generalizable Facial Expression Recognition
Yuhang Zhang (Beijing University of Posts and Telecommunications), Weihong Deng (Beijing University of Posts and Telecommunications)
RecognitionDomain AdaptationVision Language ModelContrastive LearningImage
🎯 What it does: This paper proposes a general expression recognition method that utilizes fixed CLIP facial features and a learnable Sigmoid mask, enabling training only on the training set and zero-shot inference on multi-domain test sets.
Generalizable Human Gaussians for Sparse View Synthesis
YoungJoong Kwon, Fernando de la Torre
GenerationPose EstimationConvolutional Neural NetworkDiffusion modelNeural Radiance FieldGaussian SplattingImageMesh
🎯 What it does: Propose a Generalizable Human Gaussians (GHG) model based on the human UV space, which can quickly generate high-quality, realistic new views of people in a reasoning manner with only three sparse input perspectives.
Generalizable Symbolic Optimizer Learning
Xiaotian Song (Sichuan University), Andy Song (RMIT University)
OptimizationHyperparameter SearchImageTextGraph
🎯 What it does: The paper proposes Symbolic Optimizer Learner (SOL), a method that directly learns and outputs interpretable symbolic optimizers on target tasks.
Generalized Coverage for More Robust Low-Budget Active Learning
Wonho Bae (University of British Columbia), Danica J. Sutherland (University of British Columbia)
ClassificationComputational EfficiencyImage
🎯 What it does: This paper proposes a low-budget active learning method called MaxHerding based on general coverage, which selects labeled samples by greedily maximizing the coverage of the data distribution.
Generalizing to Unseen Domains via Text-guided Augmentation
Daiqing Qi (University of Virginia), Sheng Li (University of Virginia)
Data SynthesisDomain AdaptationRepresentation LearningVision Language ModelContrastive LearningImageText
🎯 What it does: This paper proposes TEAM, a training-free, text-guided feature-level data augmentation method that aligns modal directions in CLIP's multimodal embedding space, enabling training-free augmentation of source domain images to any unseen domains;
GenerateCT: Text-Conditional Generation of 3D Chest CT Volumes
Ibrahim Ethem Hamamci (University Of Zurich), Bjoern Menze (Istanbul Medipol University)
GenerationData SynthesisTransformerSupervised Fine-TuningPrompt EngineeringVision Language ModelDiffusion modelTextBiomedical DataComputed Tomography
🎯 What it does: This study proposes the GenerateCT framework, which can generate high-resolution, three-dimensional chest CT volumes based on free-text prompts;
Generating 3D House Wireframes with Semantics
Xueqi Ma (Shenzhen University), Hui Huang (Shenzhen University)
GenerationGraph Neural NetworkTransformerMesh
🎯 What it does: Proposed a self-attention Transformer model that generates 3D house wireframes through a unified line (wire) sequence, incorporating semantic information during generation to enable direct segmentation of the wireframe into semantic blocks such as walls, roofs, and rooms.
Generating Human Interaction Motions in Scenes with Text Control
Hongwei Yi (NVIDIA), Davis Rempe (NVIDIA)
GenerationLarge Language ModelVision-Language-Action ModelDiffusion modelImageTextMultimodalityPoint CloudMesh
🎯 What it does: Generate diverse and physics-conforming human-robot interaction motions in 3D scenes with text control.
Generating Physically Realistic and Directable Human Motions from Multi-Modal Inputs
Aayam Shrestha (Oregon State University), Alan Fern (Oregon State University)
GenerationReinforcement LearningGenerative Adversarial NetworkMultimodality
🎯 What it does: Proposed a Masked Humanoid Controller (MHC) capable of generating physically realistic human motions based on multimodal, sparse, or underspecified instructions.
Generative Camera Dolly: Extreme Monocular Dynamic Novel View Synthesis
Basile Van Hoorick (Columbia University), Carl Vondrick (Toyota Research Institute)
GenerationData SynthesisDiffusion modelVideo
🎯 What it does: Fine-tuned on a large diffusion model (Stable Video Diffusion), the Generative Camera Dolly (GCD) system is proposed to generate synchronized dynamic videos from monocular videos at arbitrary extreme perspectives.
Generative End-to-End Autonomous Driving
Wenzhao Zheng (Tsinghua University), Long Chen (Chinese Academy of Sciences)
GenerationAutonomous DrivingRecurrent Neural NetworkAuto EncoderMultimodality
🎯 What it does: This paper proposes the GenAD framework, treating autonomous driving as a generative problem, and simultaneously performs motion prediction and planning in the structural latent space using an instance-centric scene tokenizer and a VAE+GRU structure.
GENIXER: Empowering Multimodal Large Language Models as a Powerful Data Generator
Henry Hengyuan Zhao, Mike Zheng Shou (National University of Singapore)
Data SynthesisTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringVision Language ModelContrastive LearningImageTextBenchmark
🎯 What it does: Propose the Genixer data generation pipeline, train LLaVA1.5 and Shikra to generate vision instruction tuning data, construct two synthetic datasets (Genixer-915K and Genixer-350K), and verify their effectiveness in improving the performance of multimodal large language models.
GenQ: Quantization in Low Data Regimes with Generative Synthetic Data
Yuhang Li (Yale University), Priyadarshini Panda (Yale University)
Data SynthesisComputational EfficiencyDiffusion modelImage
🎯 What it does: Propose GenQ, which uses Stable Diffusion to generate high-quality synthetic images and employs energy, BN/Patch filtering for quantization training with low data volume;