ICCV 2023 Papers — Page 8
IEEE/CVF International Conference on Computer Vision · 2156 papers
Feature Proliferation -- the "Cancer" in StyleGAN and its Treatments
Shuang Song (Cardiff University), Yipeng Qin (Cardiff University)
RestorationGenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: A post-processing method is proposed to eliminate artifacts in StyleGAN synthesized images by identifying and scaling 'cancer' features, enhancing image quality without sacrificing diversity.
FeatureNeRF: Learning Generalizable NeRFs by Distilling Foundation Models
Jianglong Ye (University of California San Diego), Xiaolong Wang
Object DetectionSegmentationKnowledge DistillationDiffusion modelNeural Radiance FieldImage
🎯 What it does: A FeatureNeRF framework has been constructed that can learn general 3D semantic features from a single image, and knowledge distillation is used to transfer 2D visual foundation models (such as DINO and Latent Diffusion) into 3D space.
Federated Learning Over Images: Vertical Decompositions and Pre-Trained Backbones Are Difficult to Beat
Erdong Hu (Rice University), Chris Jermaine (Rice University)
ClassificationFederated LearningConvolutional Neural NetworkImageBenchmark
🎯 What it does: This paper systematically evaluates the performance of various federated learning algorithms in image classification tasks, constructing a benchmark that includes six fine-grained datasets, with a focus on pre-trained backbone networks, communication/computation costs, and the no-tuning 'plug-and-play' effect.
FedPD: Federated Open Set Recognition with Parameter Disentanglement
Chen Yang (City University of Hong Kong), Yixuan Yuan (Chinese University of Hong Kong)
RecognitionFederated LearningImage
🎯 What it does: Proposes the FedOSR (Federated Open-Set Recognition) problem and introduces the FedPD algorithm to achieve open set recognition in a federated environment.
FedPerfix: Towards Partial Model Personalization of Vision Transformers in Federated Learning
Guangyu Sun (University of Central Florida), Chen Chen (University of Pittsburgh)
ClassificationFederated LearningTransformerImage
🎯 What it does: This paper studies partial model personalization of Vision Transformers (ViT) in federated learning. It first identifies that the self-attention layers and classification heads are the most sensitive layers through empirical research, and then proposes the FedPerfix method, which inserts a Prefix plugin into the global self-attention layer and uses parallel adapters for stable initialization, thereby achieving a mix of local and global attention learning.
FemtoDet: An Object Detection Baseline for Energy Versus Performance Tradeoffs
Peng Tu (MicroBT Inc), Yefeng Zheng (Tencent)
Object DetectionConvolutional Neural NetworkImage
🎯 What it does: Designed and implemented a lightweight object detector FemtoDet, systematically evaluating and optimizing activation functions, convolution operations, and neck structures to achieve a trade-off between energy consumption and performance.
FerKD: Surgical Label Adaptation for Efficient Distillation
Zhiqiang Shen (Mohamed bin Zayed University of Artificial Intelligence)
Object DetectionKnowledge DistillationConvolutional Neural NetworkTransformerSupervised Fine-TuningImage
🎯 What it does: Proposes FerKD, a fast knowledge distillation framework that calibrates hard/soft labels through image regions obtained from random cropping;
Few Shot Font Generation Via Transferring Similarity Guided Global Style and Quantization Local Style
Wei Pan (Wuhan University of Technology), Shilin Li (Wuhan University of Technology)
GenerationData SynthesisTransformerAuto EncoderContrastive LearningImage
🎯 What it does: This paper proposes a few-shot font generation method that combines global and local style aggregation;
Few-Shot Common Action Localization via Cross-Attentional Fusion of Context and Temporal Dynamics
Juntae Lee (Qualcomm AI Research), Sungrack Yun (Qualcomm AI Research)
RecognitionObject DetectionConvolutional Neural NetworkVideo
🎯 What it does: Proposes a three-stage cross attention (CDC-CA) and a relation classifier to locate common actions in long untrimmed query videos using only a small amount of unlabeled supporting videos.
Few-shot Continual Infomax Learning
Ziqi Gu (Nanjing University of Science and Technology), Zhen Cui (Nanjing University of Science and Technology)
ClassificationRepresentation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A Few-shot Continual Infomax Learning (FCIL) framework is proposed, which utilizes information maximization to continuously learn new categories from a small number of samples and effectively alleviates catastrophic forgetting.
Few-Shot Dataset Distillation via Translative Pre-Training
Songhua Liu (National University of Singapore), Xinchao Wang (National University of Singapore)
ClassificationKnowledge DistillationAuto EncoderImage
🎯 What it does: In this paper, the authors propose a framework for Few-Shot Dataset Distillation aimed at a small number of networks (even just one network);
Few-Shot Physically-Aware Articulated Mesh Generation via Hierarchical Deformation
Xueyi Liu (Tsinghua University), Li Yi (Beijing Institute for General Artificial Intelligence)
GenerationOptimizationMesh
🎯 What it does: This study investigates the generation of deformable meshes under few-sample physical constraints, proposing a framework for hierarchical mesh deformation and physical correction.
Few-Shot Video Classification via Representation Fusion and Promotion Learning
Haifeng Xia (Tulane University), Zhengming Ding (Tulane University)
ClassificationMeta LearningReinforcement LearningVideo
🎯 What it does: This paper proposes the RFPL framework, which enhances the performance of few-shot video classification through two submodules: meta-action learning and reinforced image representation.
Fg-T2M: Fine-Grained Text-Driven Human Motion Generation via Diffusion Model
Yin Wang (Beihang University), Xiaohui Liang (University of Durham)
GenerationData SynthesisPose EstimationGraph Neural NetworkDiffusion modelVideoText
🎯 What it does: This paper proposes a fine-grained text-driven human action generation framework called Fg-T2M, which can generate 3D motion sequences that are highly consistent with natural language descriptions.
Fine-grained Unsupervised Domain Adaptation for Gait Recognition
Kang Ma (Beijing Institute of Technology), Yongzhen Huang (Beihang University)
RecognitionDomain AdaptationVideo
🎯 What it does: A fine-grained unsupervised domain adaptation (Fine-Grained UDA) framework is proposed for gait recognition. The framework includes an offline clustering phase to generate pseudo-labels and an online training phase that learns gait features across viewpoints and clothing variations through dynamic mixed memory.
Fine-grained Visible Watermark Removal
Li Niu (Shanghai Jiao Tong University), Liqing Zhang (Shanghai Jiao Tong University)
RestorationTransformerImage
🎯 What it does: A query-based multi-task network is designed to subdivide the watermark area into various visual sub-components and restore them separately, achieving fine-grained visible watermark removal.
FineDance: A Fine-grained Choreography Dataset for 3D Full Body Dance Generation
Ronghui Li (Tsinghua University), Xiu Li (Tsinghua University)
GenerationRetrievalGraph Neural NetworkTransformerDiffusion modelVideoMultimodality
🎯 What it does: A large-scale 3D full-body dance dataset called FineDance has been released, and the FineNet network has been proposed to achieve multi-style and expressive dance generation.
FineRecon: Depth-aware Feed-forward Network for Detailed 3D Reconstruction
Noah Stier (Apple), Baptiste Angles (Apple)
RestorationSegmentationDepth EstimationConvolutional Neural NetworkPoint Cloud
🎯 What it does: An end-to-end deep learning framework called FineRecon is proposed, which directly predicts the scene TSDF from multi-view images with known poses, achieving high-fidelity 3D reconstruction.
Fingerprinting Deep Image Restoration Models
Yuhui Quan (South China University of Technology), Hui Ji (National University of Singapore)
RestorationImage
🎯 What it does: A non-intrusive fingerprinting method is proposed for image recovery models to verify model ownership.
First Session Adaptation: A Strong Replay-Free Baseline for Class-Incremental Learning
Aristeidis Panos (University of Cambridge), Richard E. Turner (University of Cambridge)
ClassificationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: Designed the First Session Adaptation (FSA) baseline, which adapts the pre-trained model only during the first learning phase and then remains unchanged, implementing a replay-free class-incremental learning with an LDA head.
FishNet: A Large-scale Dataset and Benchmark for Fish Recognition, Detection, and Functional Trait Prediction
Faizan Farooq Khan (King Abdullah University of Science and Technology), Mohamed Elhoseiny (King Abdullah University of Science and Technology)
ClassificationObject DetectionConvolutional Neural NetworkTransformerImageBenchmark
🎯 What it does: A FishNet dataset was constructed, collecting 94,532 fish images covering 17,357 species, providing category hierarchical annotations, bounding boxes, and 22 functional ecological features. Based on this, three benchmark tasks were proposed: fish classification, detection, and functional feature prediction.
Flatness-Aware Minimization for Domain Generalization
Xingxuan Zhang (Tsinghua University), Peng Cui (Tsinghua University)
Domain AdaptationOptimizationConvolutional Neural NetworkImage
🎯 What it does: A unified flatness-aware optimization framework FAD is proposed, which can efficiently optimize both zeroth-order and first-order flatness while enhancing model robustness in domain generalization tasks; a systematic evaluation of the performance of common optimizers (such as Adam, SGD, etc.) in DG tasks is also conducted.
FLatten Transformer: Vision Transformer using Focused Linear Attention
Dongchen Han (Tsinghua University), Gao Huang (Tsinghua University)
RecognitionObject DetectionSegmentationComputational EfficiencyTransformerImage
🎯 What it does: A new Focused Linear Attention module is proposed to improve the efficiency and expressiveness of visual Transformer models.
Flexible Visual Recognition by Evidential Modeling of Confusion and Ignorance
Lei Fan (Northwestern University), Gang Hua (Wormpex AI Research)
ClassificationRecognitionConvolutional Neural NetworkImage
🎯 What it does: This paper distinguishes between confusion and ignorance as two types of uncertainty in visual recognition, and achieves unknown sample rejection and multi-class prediction through plastic visual recognition.
FLIP: Cross-domain Face Anti-spoofing with Language Guidance
Koushik Srivatsan (Mohamed Bin Zayed University of Artificial Intelligence), Karthik Nandakumar (Mohamed Bin Zayed University of Artificial Intelligence)
RecognitionDomain AdaptationTransformerVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: This paper studies a cross-domain face spoofing detection method based on a vision-language pre-training model.
FlipNeRF: Flipped Reflection Rays for Few-shot Novel View Synthesis
Seunghyeon Seo (Seoul National University), Nojun Kwak (Seoul National University)
GenerationData SynthesisNeural Radiance FieldImage
🎯 What it does: This paper proposes FlipNeRF, a regularization method that utilizes flipped reflection rays and uncertainty-aware emptiness loss along with bottleneck feature consistency loss, significantly improving the rendering quality of neural radiance fields from a limited number of viewpoints.
Focal Network for Image Restoration
Yuning Cui (Technical University of Munich), Alois Knoll (Technical University of Munich)
RestorationConvolutional Neural NetworkImage
🎯 What it does: An efficient image restoration network named FocalNet is proposed, primarily targeting tasks such as dehazing, snow removal, and deblurring.
FocalFormer3D: Focusing on Hard Instance for 3D Object Detection
Yilun Chen (Chinese University of Hong Kong), Jose M. Alvarez (NVIDIA)
Object DetectionObject TrackingAutonomous DrivingTransformerGaussian SplattingMultimodalityPoint Cloud
🎯 What it does: This paper proposes Hard Instance Probing (HIP) and FocalFormer3D, specifically targeting the issue of missed detections in 3D object detection, enhancing recall through multi-stage heatmap encoding and box-level deformable decoding.
Focus on Your Target: A Dual Teacher-Student Framework for Domain-Adaptive Semantic Segmentation
Xinyue Huo (University of Science and Technology of China), Qi Tian (Huawei Inc.)
SegmentationDomain AdaptationAutonomous DrivingImage
🎯 What it does: Proposes a dual teacher-student framework and introduces a bidirectional learning strategy, separating the two capabilities of learning and adaptation, enhancing the performance of unsupervised domain adaptive semantic segmentation.
Focus the Discrepancy: Intra- and Inter-Correlation Learning for Image Anomaly Detection
Xincheng Yao (Shanghai Jiao Tong University), Chongyang Zhang (Shanghai Jiao Tong University)
Anomaly DetectionTransformerImage
🎯 What it does: A three-view anomaly detection framework FOD based on Transformer self-attention is proposed, which can simultaneously capture patch-level representation differences, intra-image correlations, and inter-image correlations, thereby achieving detection and localization of anomalies in industrial images.
Forecast-MAE: Self-supervised Pre-training for Motion Forecasting with Masked Autoencoders
Jie Cheng (Hong Kong University of Science and Technology), Ming Liu (Hong Kong University of Science and Technology)
Autonomous DrivingTransformerAuto EncoderPoint CloudBenchmark
🎯 What it does: A self-supervised pre-training framework called Forecast-MAE based on Masked Autoencoder is proposed for traffic motion prediction tasks.
Foreground and Text-lines Aware Document Image Rectification
Heng Li (PengCheng Laboratory), Qianjin Xiang (PengCheng Laboratory)
RecognitionRestorationConvolutional Neural NetworkTransformerImageBenchmark
🎯 What it does: This paper proposes a cross-attention fusion framework based on foreground and text lines to achieve geometric distortion removal of document images and improve readability.
Foreground Object Search by Distilling Composite Image Feature
Bo Zhang (Shanghai Jiao Tong University), Li Niu (Xian Jiao Tong University)
RetrievalKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: A foreground object retrieval method called DiscoFOS is proposed, which significantly improves retrieval performance by transferring features from a composite image discriminator to foreground and background encoders through knowledge distillation.
Foreground-Background Distribution Modeling Transformer for Visual Object Tracking
Dawei Yang (University of Science and Technology of China), Tianzhu Zhang (University of Science and Technology of China)
Object TrackingTransformerVideo
🎯 What it does: A foreground-background distribution modeling Transformer (F-BDMTrack) is proposed, which achieves object tracking through foreground-background proxy learning and distribution-aware attention modules.
Foreground-Background Separation through Concept Distillation from Generative Image Foundation Models
Mischa Dombrowski (Friedrich-Alexander-Universität Erlangen-Nürnberg), Bernhard Kainz (Imperial College London)
SegmentationGenerationDiffusion modelImageBiomedical Data
🎯 What it does: Automatically generate foreground-background segmentation masks through a pre-trained latent diffusion model and text prompts, achieving unsupervised segmentation model training.
Forward Flow for Novel View Synthesis of Dynamic Scenes
Xiang Guo (Northwestern Polytechnical University), Jingdong Wang (Baidu Inc.)
GenerationData SynthesisNeural Radiance FieldOptical FlowImageVideo
🎯 What it does: A new perspective synthesis method for dynamic scenes based on forward flow is proposed, which generates dynamic views by performing forward warping on the complete radiance field in canonical space.
FPR: False Positive Rectification for Weakly Supervised Semantic Segmentation
Liyi Chen (Hong Kong Polytechnic University), Lei Zhang (Hong Kong Polytechnic University)
SegmentationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A False Positive Rectification (FPR) method is proposed to correct the misactivation problem caused by class co-occurrence in weakly supervised semantic segmentation, utilizing the false positive information from CAM to learn foreground/background prototypes and introducing contrastive and pixel-level rectification losses.
FRAug: Tackling Federated Learning with Non-IID Features via Representation Augmentation
Haokun Chen (Ludwig Maximilian University of Munich), Volker Tresp (Ludwig Maximilian University of Munich)
Federated LearningGenerative Adversarial NetworkImageBiomedical Data
🎯 What it does: Proposes Federated Representation Augmentation (FRAug), which achieves client feature space augmentation through a shared generator and local RTNet to address the non-IID feature distribution problem in FL.
FreeCOS: Self-Supervised Learning from Fractals and Unlabeled Images for Curvilinear Object Segmentation
Tianyi Shi (Huazhong University of Science and Technology), Xin Yang (Huazhong University of Science and Technology)
SegmentationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes an unsupervised curve structure segmentation method called FreeCOS, which extracts robust features from fractals and unlabeled images through self-supervised learning to achieve segmentation of curved objects such as blood vessels and cracks.
FreeDoM: Training-Free Energy-Guided Conditional Diffusion Model
Jiwen Yu (Peking University), Jian Zhang (Peking University)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: Proposes a training-free energy-guided conditional diffusion model FreeDoM, which can utilize pre-trained networks to control generation under various conditions.
Frequency Guidance Matters in Few-Shot Learning
Hao Cheng (Nanyang Technological University), Bihan Wen (Nanyang Technological University)
ClassificationMeta LearningContrastive LearningImage
🎯 What it does: This paper proposes a few-shot learning framework guided by frequency information (FGFL), which enhances the model's discriminability and generalization performance through task-specific frequency masks and multi-level metrics.
Frequency-aware GAN for Adversarial Manipulation Generation
Peifei Zhu (LINE Corporation), Tsubasa Takahashi (LINE Corporation)
Object DetectionGenerationAdversarial AttackGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a task for adversarial attacks on image operation detectors, called AMG, and designs a covert perturbation generation method based on frequency-aware GAN.
From Chaos Comes Order: Ordering Event Representations for Object Recognition and Detection
Nikola Zubić (University of Zurich), Davide Scaramuzza (University of Zurich)
RecognitionObject DetectionOptimizationRepresentation LearningHyperparameter SearchImage
🎯 What it does: This paper studies a method for rapid evaluation of dense representations of event cameras, using Gromov-Wasserstein discrepancy (GWD) as a substitute for traditional network training and validation;
From Knowledge Distillation to Self-Knowledge Distillation: A Unified Approach with Normalized Loss and Customized Soft Labels
Zhendong Yang (Tsinghua University), Yu Li (International Digital Economy Academy)
Object DetectionKnowledge DistillationConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes two unified knowledge distillation methods: Normalized KD (NKD) and Universal Self-Knowledge Distillation (USKD), which are used for teacher-assisted and teacher-free training scenarios, respectively.
From Sky to the Ground: A Large-scale Benchmark and Simple Baseline Towards Real Rain Removal
Yun Guo (Huazhong University of Science and Technology), Luxin Yan (Huazhong University of Science and Technology)
RestorationObject DetectionSegmentationTransformerImageVideoBenchmark
🎯 What it does: A large-scale high-quality real rain dataset, LHP-Rain, has been constructed, and the RLRTR video de-raining and SCD-Former single image de-raining models have been proposed.
FrozenRecon: Pose-free 3D Scene Reconstruction with Frozen Depth Models
Guangkai Xu (University of Science and Technology of China), Feng Zhao
Depth EstimationOptimizationSimultaneous Localization and MappingVideo
🎯 What it does: By freezing the pre-trained homogeneous invariant deep model and jointly optimizing a small number of parameters (about dozens), dense 3D reconstruction of monocular videos is achieved without offline camera pose.
FS-DETR: Few-Shot DEtection TRansformer with Prompting and without Re-Training
Adrian Bulat (Samsung AI Cambridge), Georgios Tzimiropoulos (Samsung AI Cambridge)
Object DetectionTransformerPrompt EngineeringImage
🎯 What it does: This paper proposes a few-shot object detection method (FS-DETR) that does not require retraining and supports an arbitrary number of new categories and samples, capable of detecting multiple new objects in a single forward inference step.
FSAR: Federated Skeleton-based Action Recognition with Adaptive Topology Structure and Knowledge Distillation
Jingwen Guo (Peking University), Chenyang Si (Nanyang Technological University)
RecognitionPose EstimationFederated LearningSafty and PrivacyKnowledge DistillationGraph Neural NetworkVideo
🎯 What it does: A federated learning framework called FSAR is proposed for skeleton action recognition while protecting privacy.
FSI: Frequency and Spatial Interactive Learning for Image Restoration in Under-Display Cameras
Chengxu Liu (Xi'an Jiaotong University), Xueming Qian (Xi'an Jiaotong University)
RestorationConvolutional Neural NetworkImage
🎯 What it does: A UDC image restoration network FSI that combines frequency domain and spatial domain learning is designed to address diffraction and color shift issues.
Full-Body Articulated Human-Object Interaction
Nan Jiang (Peking University), Siyuan Huang (Beijing Institute of General Artificial Intelligence)
Pose EstimationConvolutional Neural NetworkReinforcement LearningImageVideo
🎯 What it does: A dataset called CHAIRS was constructed, and a method for object pose reconstruction based on full-body interaction was proposed.
FULLER: Unified Multi-modality Multi-task 3D Perception via Multi-level Gradient Calibration
Zhijian Huang (Sun Yat-sen University), Xiaodan Liang (Sun Yat-sen University)
Object DetectionSegmentationAutonomous DrivingTransformerMultimodalityPoint Cloud
🎯 What it does: A unified multimodal multi-task 3D perception framework called Fuller is proposed, capable of simultaneously performing 3D detection and map segmentation on a single BEV representation.
Fully Attentional Networks with Self-emerging Token Labeling
Bingyin Zhao (Clemson University), Jose M. Alvarez (NVIDIA)
ClassificationObject DetectionSegmentationTransformerSupervised Fine-TuningImage
🎯 What it does: A self-generating Token Labeling (STL) framework is proposed, which enhances pre-training performance through two-stage training on the Fully Attentional Network (FAN).
FunnyBirds: A Synthetic Vision Dataset for a Part-Based Analysis of Explainable AI Methods
Robin Hesse (Technische Universität Darmstadt), Stefan Roth (Technische Universität Darmstadt)
Data SynthesisExplainability and InterpretabilityConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper presents a synthetic avian visual dataset named FunnyBirds and constructs a multidimensional XAI evaluation framework to automate and systematize the assessment of various explanation methods in terms of completeness, correctness, and distinctiveness.
G2L: Semantically Aligned and Uniform Video Grounding via Geodesic and Game Theory
Hongxiang Li (Peking University), Yuexian Zou (Peking University)
RetrievalContrastive LearningVideoText
🎯 What it does: A method for achieving video-text semantic alignment through geographic distance and game theory is proposed, addressing the issues of semantic overlap and sparse annotations in video grounding.
GACE: Geometry Aware Confidence Enhancement for Black-Box 3D Object Detectors on LiDAR-Data
David Schinagl (Graz University of Technology), Horst Bischof (Graz University of Technology)
Object DetectionAutonomous DrivingPoint Cloud
🎯 What it does: Proposes the GACE method, which performs geometry information-driven post-processing on the confidence estimation of black-box LiDAR 3D detectors to enhance detection performance.
GAFlow: Incorporating Gaussian Attention into Optical Flow
Ao Luo (Megvii Technology), Shuaicheng Liu (University of Electronic Science and Technology of China)
TransformerOptical FlowImageVideo
🎯 What it does: This paper proposes the GAFlow framework, which introduces a learnable Gaussian attention module in optical flow estimation, used to enhance the local discriminability of feature representations (Gaussian-Constrained Layer, GCL) and the motion affinity during the matching process (Gaussian-Guided Attention Module, GGAM).
GAIT: Generating Aesthetic Indoor Tours with Deep Reinforcement Learning
Desai Xie (Stony Brook University), Arie E. Kaufman (Stony Brook University)
GenerationReinforcement LearningVideo
🎯 What it does: GAIT is proposed, an agent based on deep reinforcement learning that can automatically generate aesthetic viewpoint sequences of 3D indoor scenes from any initial camera pose.
GameFormer: Game-theoretic Modeling and Learning of Transformer-based Interactive Prediction and Planning for Autonomous Driving
Zhiyu Huang (Nanyang Technological University), Chen Lv (Nanyang Technological University)
Autonomous DrivingTransformerMultimodality
🎯 What it does: A Transformer model called GameFormer is constructed through hierarchical game theory to achieve interactive prediction and planning.
GaPro: Box-Supervised 3D Point Cloud Instance Segmentation Using Gaussian Processes as Pseudo Labelers
Tuan Duc Ngo (VinAI Research), Khoi Nguyen (VinAI Research)
Object DetectionSegmentationPoint Cloud
🎯 What it does: This paper proposes the GaPro method, which utilizes axis-aligned 3D bounding box supervision to generate pseudo-instance masks through Gaussian processes, and uses these to train a 3D point cloud instance segmentation network.
GasMono: Geometry-Aided Self-Supervised Monocular Depth Estimation for Indoor Scenes
Chaoqiang Zhao (East China University of Science and Technology), Stefano Mattoccia (University of Bologna)
Depth EstimationKnowledge DistillationTransformerSimultaneous Localization and MappingImage
🎯 What it does: The GasMono framework is proposed to address the issues of large rotations and low texture in indoor scenes through geometric-assisted self-supervised monocular depth estimation.
GECCO: Geometrically-Conditioned Point Diffusion Models
Michał J Tyszkiewicz, Eduard Trulls (Google Research)
GenerationData SynthesisTransformerDiffusion modelPoint Cloud
🎯 What it does: This paper proposes a point cloud generation framework based on diffusion models, capable of generating 3D point clouds under both unconditional and image-conditioned settings.
GEDepth: Ground Embedding for Monocular Depth Estimation
Xiaodong Yang (QCraft), Zhe Ren
Depth EstimationAutonomous DrivingConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes a ground-embedded module GEDepth, which decouples camera parameters and image features to enhance the generalization ability of monocular depth estimation.
Gender Artifacts in Visual Datasets
Nicole Meister (Stanford University), Olga Russakovsky (Princeton University)
ClassificationObject DetectionConvolutional Neural NetworkImage
🎯 What it does: This paper systematically identifies and quantifies various gender-related 'artifacts' (such as color, shape, background, posture, and contextual objects) in visual datasets by training classifiers and applying various masking and downsampling techniques, exploring the impact of these artifacts on gender judgment.
General Image-to-Image Translation with One-Shot Image Guidance
Bin Cheng (NetEase Games AI Lab), Yue Lin (NetEase Games AI Lab)
Image TranslationGenerationDiffusion modelImage
🎯 What it does: This paper proposes a framework called Visual Concept Translator (VCT) for general image-to-image translation tasks given only a reference image, capable of transferring visual concepts from the reference image while preserving the content of the source image.
General Planar Motion from a Pair of 3D Correspondences
Juan Carlos Dibene (Stevens Institute of Technology), Enrique Dunn (Stevens Institute of Technology)
Pose EstimationAutonomous DrivingSimultaneous Localization and MappingPoint Cloud
🎯 What it does: A geometric closed-form solver based on two 3D-3D corresponding points is proposed to estimate the relative pose (5 degrees of freedom) of the camera and the motion plane under unknown motion planes.
Generalizable Decision Boundaries: Dualistic Meta-Learning for Open Set Domain Generalization
Xiran Wang (Nanjing University), Yinghuan Shi (Nanjing University)
Domain AdaptationMeta LearningConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a dual meta-learning framework MEDIC for open set domain generalization, which jointly learns model parameters through domain-level and class-level gradient matching, and constructs decision boundaries for each known class using multiple binary classifiers.
Generalizable Neural Fields as Partially Observed Neural Processes
Jeffrey Gu (Stanford University), Serena Yeung (Stanford University)
RestorationGenerationMeta LearningConvolutional Neural NetworkTransformerImageComputed Tomography
🎯 What it does: By designing a partially observable neural process (PONP) framework, the goal of generalized learning of various neural fields under sparse observations from sensors is achieved.
Generalized Differentiable RANSAC
Tong Wei (Czech Technical University in Prague), Daniel Barath (ETH Zurich)
OptimizationReinforcement LearningPoint Cloud
🎯 What it does: A differentiable RANSAC framework, ∇-RANSAC, is proposed and implemented, capable of learning matching confidence, sampling distribution, and minimum solvers, achieving end-to-end training from feature matching to model estimation.
Generalized Few-Shot Point Cloud Segmentation via Geometric Words
Yating Xu (National University of Singapore), Gim Hee Lee (National University of Singapore)
SegmentationPoint Cloud
🎯 What it does: A general few-shot point cloud semantic segmentation framework is proposed, which can maintain segmentation capabilities for base classes while effectively recognizing new classes with only a small number of new category samples.
Generalized Lightness Adaptation with Channel Selective Normalization
Mingde Yao (University of Science and Technology of China), Zhiwei Xiong (University of Science and Technology of China)
Image TranslationImage HarmonizationRestorationImage
🎯 What it does: A general photometric adaptive Channel Selective Normalization (CSNorm) module is proposed, which can effectively enhance, color-correct, and beautify images under unknown photometric conditions after being trained with a single photometric condition.
Generalized Sum Pooling for Metric Learning
Yeti Z. Gürbüz (RSiM, TU Berlin), A. Aydin Alatan (Intel Labs)
RetrievalMeta LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A learnable Generalized Average Pooling (GSP) module is proposed for deep metric learning, replacing the traditional Global Average Pooling (GAP), which automatically selects and weights feature subsets through an optimal transport framework.
Generalizing Event-Based Motion Deblurring in Real-World Scenarios
Xiang Zhang (Wuhan University), Gui-Song Xia (Wuhan University)
RestorationConvolutional Neural NetworkImageVideo
🎯 What it does: An adaptive scale event-driven motion deblurring network (SAN) is proposed, achieving generalization to different spatial resolutions and temporal blurs in real scenes through a two-stage self-supervised learning approach.
Generalizing Neural Human Fitting to Unseen Poses With Articulated SE(3) Equivariance
Haiwen Feng (Max Planck Institute for Intelligent Systems), Victoria Fernandez Abrevaya
Pose EstimationPoint Cloud
🎯 What it does: A new neural network framework called ArtEq is proposed for estimating the shape and pose of the SMPL human model from point cloud data, particularly focusing on generalization to unseen poses.
Generating Dynamic Kernels via Transformers for Lane Detection
Ziye Chen (University of Melbourne), Kate Smith-Miles (University of Melbourne)
Object DetectionAutonomous DrivingTransformerImage
🎯 What it does: A dynamic convolution kernel generation network based on Transformer is proposed for lane line detection.
Generating Instance-level Prompts for Rehearsal-free Continual Learning
Dahuin Jung (Seoul National University), Hwanjun Song (Amazon Web Services)
Domain AdaptationTransformerPrompt EngineeringImage
🎯 What it does: A no-pool domain adaptive prompt (DAP) framework is proposed, capable of generating prompts instantaneously at the instance level to adjust the frozen ViT backbone, enabling replay-free continual learning.
Generating Realistic Images from In-the-wild Sounds
Taegyeong Lee (Ulsan National Institute of Science and Technology), Taehwan Kim (Ulsan National Institute of Science and Technology)
GenerationData SynthesisTransformerDiffusion modelImageTextAudio
🎯 What it does: By combining audio-to-text (audio subtitles) with diffusion models, wild sounds are converted into high-quality images that match the sounds.
Generating Visual Scenes from Touch
Fengyu Yang (University of Michigan), Andrew Owens (University of Michigan)
GenerationData SynthesisDiffusion modelContrastive LearningImageMultimodalityStochastic Differential Equation
🎯 What it does: A visual-tactile cross-modal synthesis framework based on latent diffusion models is proposed, capable of generating natural scene images from tactile signals without visual input, while also supporting tasks such as tactile-driven image style transfer, hand-free image generation, and tactile-based shadow estimation.
Generative Action Description Prompts for Skeleton-based Action Recognition
Wangmeng Xiang (Hong Kong Polytechnic University), Lei Zhang (Hong Kong Polytechnic University)
RecognitionGraph Neural NetworkLarge Language ModelContrastive LearningVideoMultimodality
🎯 What it does: The Generative Action-description Prompts (GAP) framework is proposed, which automatically generates action descriptions using large language models and enhances skeleton action recognition performance through multimodal training and multi-part contrastive loss, with no additional cost during inference.
Generative Gradient Inversion via Over-Parameterized Networks in Federated Learning
Chi Zhang (Institute of High Performance Computing), Yong Liu (Institute of High Performance Computing)
RestorationGenerationFederated LearningConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This study investigates the recovery of high-resolution, large-batch, and complex image samples through gradient inversion in federated learning, proposing a CI-Net generator based on over-parameterized convolutional networks.
Generative Multiplane Neural Radiance for 3D-Aware Image Generation
Amandeep Kumar (Mohamed bin Zayed University of AI), Fahad Shahbaz Khan (Linkoping University)
GenerationData SynthesisNeural Radiance FieldGenerative Adversarial NetworkImage
🎯 What it does: A high-resolution 3D-perception image generation framework GMNR is proposed, capable of generating perspective-consistent realistic images from multiple camera poses.
Generative Novel View Synthesis with 3D-Aware Diffusion Models
Eric R. Chan (Stanford University), Gordon Wetzstein (NVIDIA)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: A new 3D-aware view synthesis framework based on diffusion models is proposed, capable of generating diverse and geometrically consistent novel views from a single or a few input images, and supports autoregressive long sequence generation.
Generative Prompt Model for Weakly Supervised Object Localization
Yuzhong Zhao (University of Chinese Academy of Sciences), Fang Wan (University of Chinese Academy of Sciences)
Object DetectionTransformerVision Language ModelDiffusion modelImage
🎯 What it does: A generative prompt model called GenPromp is proposed, redefining the weakly supervised object localization problem as a conditional image denoising task. By learning category representative prompt embeddings and fusing them with CLIP discriminative embeddings, high-quality localization maps are generated using multi-scale cross-attention.
Geometric Viewpoint Learning with Hyper-Rays and Harmonics Encoding
Zhixiang Min (Stevens Institute of Technology), Enrique Dunn (Stevens Institute of Technology)
SegmentationRepresentation LearningPoint Cloud
🎯 What it does: An end-to-end deep learning framework is proposed, which learns viewpoint preferences through perspective ray encoding and captures geometric and semantic context using a Harmonic Ray Encoder (HRE) learned on point clouds.
Geometrized Transformer for Self-Supervised Homography Estimation
Jiazhen Liu (Renmin University of China), Xirong Li (Renmin University of China)
TransformerImage
🎯 What it does: Proposes GeoFormer, a detector-free homography estimation method based on Transformer.
Geometry-guided Feature Learning and Fusion for Indoor Scene Reconstruction
Ruihong Yin (University of Amsterdam), Theo Gevers (University of Amsterdam)
Depth EstimationRepresentation LearningTransformerPoint Cloud
🎯 What it does: By integrating geometric information at three levels: feature learning, feature fusion, and network supervision, 3D reconstruction of indoor scenes has been achieved.
GeoMIM: Towards Better 3D Knowledge Transfer via Masked Image Modeling for Multi-view 3D Understanding
Jihao Liu (Chinese University of Hong Kong), Hongsheng Li (Chinese University of Hong Kong)
Object DetectionSegmentationDomain AdaptationAutonomous DrivingTransformerSupervised Fine-TuningPoint Cloud
🎯 What it does: Through a pre-training-fine-tuning framework, the Geometry Enhanced Masked Image Modeling (GeoMIM) method is proposed, which utilizes the BEV features of a pre-trained LiDAR detection model to transfer knowledge to a multi-view camera 3D detection model.
GeoUDF: Surface Reconstruction from 3D Point Clouds via Geometry-guided Distance Representation
Siyu Ren (City University of Hong Kong), Wenping Wang (Texas A&M University)
GenerationOptimizationGraph Neural NetworkPoint CloudMesh
🎯 What it does: A learning-based surface reconstruction framework called GeoUDF is proposed, which can directly reconstruct closed or non-closed 3D surfaces from sparse point clouds and generate high-quality triangular meshes.
GePSAn: Generative Procedure Step Anticipation in Cooking Videos
Mohamed A. Abdelsalam (Samsung AI Centre), Afsaneh Fazly (Samsung AI Centre)
GenerationDomain AdaptationTransformerAuto EncoderVideoTextMultimodality
🎯 What it does: This paper proposes a generative model named GEPSAN, which predicts multiple possible next steps from procedural videos (using cooking as an example) and achieves zero-shot transfer under video input.
Get the Best of Both Worlds: Improving Accuracy and Transferability by Grassmann Class Representation
Haoqi Wang (EPFL), Wayne Zhang (SenseTime Research)
ClassificationRepresentation LearningConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: Generalize the category vector to a linear subspace and use subspace projection in the final fully connected layer to compute logits, constructing the Grassmann Class Representation (GCR);
GeT: Generative Target Structure Debiasing for Domain Adaptation
Can Zhang (National University of Singapore), Gim Hee Lee (National University of Singapore)
Domain AdaptationImage
🎯 What it does: The GeT framework is proposed, which combines an online target domain classifier with structural similarity regularization to automatically generate pseudo-labels that are free from source domain bias and have balanced class distribution, thereby enhancing multi-scene domain adaptation performance.
GET: Group Event Transformer for Event-Based Vision
Yansong Peng (University of Science and Technology of China), Feng Wu (University of Science and Technology of China)
ClassificationObject DetectionTransformerImage
🎯 What it does: A visual backbone network called Group Event Transformer (GET) is proposed, specifically designed to extract features from asynchronous event streams generated by event cameras, and key modules such as Group Token representation, Event Dual Self-Attention (EDSA), and Group Token Aggregation (GTA) are designed.
Get3DHuman: Lifting StyleGAN-Human into a 3D Generative Model Using Pixel-Aligned Reconstruction Priors
Zhangyang Xiong (Chinese University of Hong Kong Shenzhen), Xiaoguang Han (Chinese University of Hong Kong Shenzhen)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: Proposes the Get3DHuman 3D human generation framework, which utilizes the 2D prior of StyleGAN-Human and the 3D reconstruction prior of PIFu to train high-quality, multi-pose 3D textured human models from single-view images.
GETAvatar: Generative Textured Meshes for Animatable Human Avatars
Xuanmeng Zhang, Jiashi Feng
GenerationData SynthesisPose EstimationGenerative Adversarial NetworkMesh
🎯 What it does: Designed and trained an explicit texture 3D mesh generation model GETAvatar that can generate controllable full-body human animated avatars.
GIFD: A Generative Gradient Inversion Method with Feature Domain Optimization
Hao Fang (Harbin Institute of Technology), Shu-Tao Xia (Tsinghua University)
GenerationFederated LearningAdversarial AttackGenerative Adversarial NetworkImage
🎯 What it does: A gradient inversion attack method based on Generative Adversarial Networks (GAN) is proposed (GIFD), which recovers the private data corresponding to the uploaded gradients in federated learning by searching the feature domain of the generator layer by layer.
GLA-GCN: Global-local Adaptive Graph Convolutional Network for 3D Human Pose Estimation from Monocular Video
Bruce X.B. Yu (Hong Kong Polytechnic University), Chang Wen Chen (Hong Kong Polytechnic University)
Pose EstimationGraph Neural NetworkVideo
🎯 What it does: A global-local adaptive graph convolutional network (GLA-GCN) is proposed for 3D human pose estimation in monocular videos, capturing spatiotemporal structures in the global layer and performing fine regression for each joint in the local layer.
Global Adaptation Meets Local Generalization: Unsupervised Domain Adaptation for 3D Human Pose Estimation
Wenhao Chai (University of Washington), Gaoang Wang (Zhejiang University)
Pose EstimationDomain AdaptationGenerative Adversarial NetworkImageVideo
🎯 What it does: A framework named PoseDA is proposed for unsupervised domain adaptation to transfer 2D-3D human pose enhancement models to unseen target datasets.
Global Balanced Experts for Federated Long-Tailed Learning
Yaopei Zeng (Chinese University of Hong Kong), Baoyuan Wu (Chinese University of Hong Kong)
Federated LearningMixture of ExpertsImage
🎯 What it does: This paper proposes a Global Balanced Multi-Expert framework (GBME) that utilizes Global Proxy Information (GPI) to achieve balanced training on long-tail data in federated learning, and further enhances performance through a multi-expert architecture and client grouping.
Global Features are All You Need for Image Retrieval and Reranking
Shihao Shao (Peking University), Bingyi Cao (Google Research)
RetrievalConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: Proposes the SuperGlobal system, which utilizes improved global features to complete image retrieval and re-ranking, completely independent of local features;
Global Knowledge Calibration for Fast Open-Vocabulary Segmentation
Kunyang Han (Beijing Jiaotong University), Yunchao Wei (Beijing Jiaotong University)
SegmentationComputational EfficiencyKnowledge DistillationTransformerImageVideo
🎯 What it does: A fast open-vocabulary semantic segmentation model is proposed—Global Knowledge Calibration, which utilizes text diversification and text-guided knowledge distillation to avoid overfitting on base class names during training, and no longer uses an additional CLIP visual encoder during the inference phase;
Global Perception Based Autoregressive Neural Processes
Jinyang Tai (Shanghai University)
GenerationData SynthesisMeta LearningRecurrent Neural NetworkTransformerAuto EncoderImageTime Series
🎯 What it does: A self-autoregressive neural process framework AENPs and CAENPs is proposed, improving the latent distribution and deterministic paths of NPs, allowing the model to better capture the global and local relationships of contextual sample points.