CVPR 2026 Papers — Page 15

Generate, Analyze, and Refine: Training-Free Sound Source Localization via MLLM Meta-Reasoning

Subin Park (Kyung Hee University), Jung Uk Kim (Kyung Hee University)

🎯 What it does: Proposed a training-free generation-analysis-refinement (GAR) sound source localization framework that utilizes a multimodal large language model for cross-modal reasoning

Generating Humanless Environment Walkthroughs from Egocentric Walking Tour Videos

Yujin Ham (Rice University), Guha Balakrishnan (Rice University)

🎯 What it does: Propose CrowdEraser, an algorithm based on video diffusion models, which can realistically remove people and their shadows in first-person walking tour videos, generating crowd-free roaming scenes.

Generative Adversarial Perturbations with Cross-paradigm Transferability on Localized Crowd Counting

Alabi Mehzabin Anisha (University of South Florida), Sriram Chellappan (University of South Florida)

🎯 What it does: Propose a U-Net based generative adversarial perturbation framework that can achieve cross-paradigm attacks on both density map and point regression human crowd localization models.

Generative Diffusion Priors for 3D Mapping of the Dark Universe

Brandon Zhao (California Institute of Technology), Katherine L. Bouman (California Institute of Technology)

🎯 What it does: Recover the three-dimensional distribution of cosmic dark matter using weak lensing observations and diffusion model priors based on N-body simulations.

Generative Modeling of Weights: Generalization or Memorization?

Boya Zeng (Princeton University), Zhuang Liu (Princeton University)

🎯 What it does: This paper conducts experiments on four mainstream weight generation methods (Hyper-Representations, G.pt, HyperDiffusion, P-diff), verifying that they primarily replicate or perform linear interpolation on training checkpoints when generating new model weights, rather than achieving true generalization;

Generative Neural Video Compression via Video Diffusion Prior

Qi Mao (Communication University of China), Siwei Ma (Peking University)

🎯 What it does: Propose GNVC-VD, a generative neural video compression framework based on video diffusion Transformers, unifying spatiotemporal latent compression with sequence-level generation refinement;

Generative Point Tracking and Forecasting

Xuanchen Lu (Cornell University), Andrew Owens (University of Michigan)

🎯 What it does: Propose a unified generative model that simultaneously performs point tracking and trajectory prediction, enabling task switching through video conditions.

Generative Video Compression with One-Dimensional Latent Representation

Zihan Zheng (University of Science and Technology of China), Yan Lu (Microsoft Research Asia)

🎯 What it does: This paper proposes a generative video compression method based on one-dimensional (1D) latent representations (GVC1D), which compresses video frames into a small number of freely aggregable 1D latent tokens and employs a 1D memory module for long-term context modeling;

Generative Video Motion Editing with 3D Point Tracks

Yao-Chih Lee (Adobe Research), Zhengqi Li (Adobe Research)

🎯 What it does: Proposes the Edit-by-Track framework, achieving joint editing of camera and object motion in videos through 3D point trajectories.

GeneVAR: Causal MeanFlow for Autoregressive Gene-to-WSI Tile Synthesis

Jianwei Zhao (UESTC), Hong Cheng (Brown)

🎯 What it does: Propose GeneVAR, an RNA-Seq-based whole slide image (WSI) autoregressive generation framework that continuously enhances gene expression guidance during multi-scale iterative processes to generate high-quality H&E histopathological images.

GenHOI: Towards Object-Consistent Hand-Object Interaction with Temporally Balanced and Spatially Selective Object Injection

Xuan Huang (Baidu Inc.), Jingdong Wang (Baidu Inc.)

🎯 What it does: Propose the GenHOI framework, which adds a lightweight module to a pre-trained video generation model to achieve more consistent and natural hand-object interaction (HOI) video recreation.

GenieDrive: Towards Physics-Aware Driving World Model with 4D Occupancy Guided Video Generation

Zhenya Yang (The University Of Hong Kong), Hengshuang Zhao (Huawei Noah's Ark Lab)

🎯 What it does: Proposed GenieDrive, a two-stage driving world model that first predicts a physically consistent 4D occupancy grid using a 3-plane VAE, then projects it and guides multi-view video generation through Normalized Multi-View Attention, achieving controllable, physically consistent long-duration driving videos;

GeniNav: Generative Model Driven Image-Goal Navigation via Imagination-Guided Consistency Flow Matching

Yuqi Chen (Nanyang Technological University), Mir Feroskhan (Nanyang Technological University)

🎯 What it does: This paper proposes GeniNav, a generative framework for image goal navigation, combining potential subgoals inferred by VLM, continuous action generation based on multi-segment consistency flow matching, and trajectory selection through hybrid semantic and geometric evaluation, while constructing the GeniBench evaluation benchmark.

GenMask: Adapting DiT for Segmentation via Direct Mask Generation

Yuhuan Yang (Shanghai Jiao Tong University), Yanfeng Wang (Shanghai Jiao Tong University)

🎯 What it does: Train a Diffusion Transformer (DiT) to directly generate binary segmentation masks in the RGB space while maintaining a unified training objective with image generation.

GenMatter: Perceiving Physical Objects with Generative Matter Models

Eric Li (MIT), Joshua B. Tenenbaum (MIT)

🎯 What it does: Proposed and implemented a hierarchical generative model called GenMatter for motion-based object perception and segmentation in various scenarios such as random points, concealed textures, and natural RGB videos.

GenSplat: Bridging the Generalization Gap in 3DGS Language Comprehension

Fang Liu (City University of Hong Kong), Rynson W. H. Lau (City University of Hong Kong)

GenTract: Generative Global Tractography

Alec Sargood (University College London), Daniel C. Alexander (University College London)

🎯 What it does: Designed and implemented GenTract, a global fiber tracking model based on conditional generation, which can generate coordinates for all brain fibers at once, avoiding error accumulation in traditional step-by-step tracking.

Geo2: Geometry-Guided Cross-view Geo-Localization and Image Synthesis

Yancheng Zhang (University of Central Florida), Chen Chen (University of Central Florida)

🎯 What it does: Propose a unified framework Geo2 that jointly accomplishes cross-view geolocation (CVGL) and bidirectional cross-view image synthesis (CVIS), and extracts 3D geometric priors through geometric foundation models (GFMs).

GeoAgent: Learning to Geolocate Everywhere with Reinforced Geographic Characteristics

Modi Jin (Nankai University), Qibin Hou (Nankai University)

🎯 What it does: Propose the GeoAgent model, leveraging reinforcement learning's chain-of-thought process similar to human reasoning to achieve fine-grained geolocation.

GeoBridge: A Semantic-Anchored Multi-View Foundation Model Bridging Images and Text for Geo-Localization

Zixuan Song (Jilin University), Bo Du (Wuhan University)

🎯 What it does: This paper proposes GeoBridge, a cross-view and cross-modal localization framework that connects multi-view images (drone, street view, satellite) with text through semantic anchors, and constructs the GeoLoc dataset containing over 50,000 triple-view images along with corresponding textual descriptions.

GeoCoT: Towards Reliable Remote Sensing Reasoning with Manifold Perspective

Daixun Li (Xidian University), Leyuan Fang (Xidian University)

🎯 What it does: Built GeoCoT, a remote sensing multi-modal large language model that integrates Manifold-driven Mixture-of-Experts (Mani-MoE) and Chain-of-Thought (CoT) reasoning, supporting unified inference for multi-task scenarios (classification, counting, detection, image description, etc.).

GeodesicNVS: Probability Density Geodesic Flow Matching for Novel View Synthesis

Xuqin Wang (Technical University of Munich), Daniel Cremers (Huawei Hilbert Research Center)

🎯 What it does: Proposes a conditional flow matching framework based on probability density geometric flow matching (PDG-FM) for single-view novel view synthesis, which can learn smooth interpolation paths along high-density regions in the latent space, significantly improving view consistency and structural fidelity.

GeoDexGrasp: Geometry-aware Generation for Data-efficient and Physics-plausible Dexterous Grasping

Bing Han (Xi'an Jiaotong University), Zhi Zhai (Xi'an Jiaotong University)

🎯 What it does: Proposes the GeoDexGrasp framework based on object geometry representation, achieving geometry-aware grasp generation.

GeoDiff4D: Geometry-Aware Diffusion for 4D Head Avatar Reconstruction

Chao Xu (Tsinghua University), Yebin Liu (Tsinghua University)

🎯 What it does: Developed a framework for reconstructing an animatable 4D head avatar from a single portrait image.

GeoDiT: A Diffusion-based Vision-Language Model for Geospatial Understanding

Jiaqi Liu (Jilin University), Bo Yang (Jilin University)

🎯 What it does: Propose GeoDiT, a diffusion-based vision-language model for structured description, object localization, and question answering on remote sensing images.

GeoFlow: Real-Time Fine-Grained Cross-View Geolocalization via Iterative Flow Prediction

Ayesh Abu Lehyeh (University of Vermont), Safwan Wshah (University of Vermont)

🎯 What it does: Proposed the GeoFlow framework, which leverages the flow matching idea to achieve fine-grained cross-view localization between ground images and satellite images, and employs Iterative Refinement Sampling (IRS) to continuously optimize multiple initial hypotheses, ultimately obtaining high-precision 2D position information.

GeoFree-CoSeg: Unsupervised Point Cloud-Image Cross-Modal Co-Segmentation Without Geometric Alignment

Xin Duan (Beijing Institute Of Technology), Liyuan Pan (Beijing Institute Of Technology)

🎯 What it does: This paper proposes an unsupervised point cloud-image cross-modal co-segmentation framework named GeoFree-CoSeg, which can identify and segment shared semantic objects in multiple point clouds and images without requiring geometric alignment or segmentation annotations.

GeoGuide: Hierarchical Geometric Guidance for Open-Vocabulary 3D Semantic Segmentation

Xujing Tao (University of Science and Technology of China), Tianzhu Zhang (University of Science and Technology of China)

🎯 What it does: This paper proposes a GeoGuide framework for open-vocabulary 3D semantic segmentation, aiming to enhance segmentation performance on unknown categories by guiding 2D-3D knowledge distillation through geometric priors.

Geoint-R1: Formalizing Multimodal Geometric Reasoning with Dynamic Auxiliary Constructions

Jingxuan Wei (Key Laboratory of Computing Power Network and Information Security Ministry of Education Shandong Computer Science Center National Supercomputer Center in Jinan Qilu University of Technology Shandong Academy of Sciences), Cheng Tan (Shanghai Artificial Intelligence Laboratory)

🎯 What it does: Proposed a multi-modal framework called Geoint-R1 for formal geometric reasoning, which can automatically identify, construct, and visualize auxiliary geometric elements, generating complete reasoning processes verifiable on Lean4.

Geometric Neural Distance Fields for Learning Human Motion Priors

Zhengdi Yu (Imperial College London), Tolga Birdal (Imperial College London)

🎯 What it does: This paper proposes Neural Riemannian Motion Fields (NRMF), an unconditional human motion prior that models pose, velocity, and acceleration as third-order zero-level sets using neural distance fields, and provides adaptive projection and geometric integration algorithms, enabling its application to tasks such as denoising, interpolation, test-time optimization, and motion generation.

Geometric-Aware Hypergraph Reasoning for Novel Class Discovery in Point Cloud Segmentation

Zihao Zhang (Tianjin University), Jialie Shen (Tianjin University)

🎯 What it does: Proposed a geometry-aware hypergraph reasoning framework to achieve novel class discovery in point cloud segmentation.

Geometric-Photometric Event-based 3D Gaussian Ray Tracing

Kai Kohyama (Keio University), Shintaro Shiba (Keio University)

🎯 What it does: To address the sparse event streams from event cameras, the GPERT framework is proposed, which splits 3D Gaussian splatting rendering into event-level geometry rendering and snapshot-level radiance rendering.

Geometrically-Constrained Agent for Spatial Reasoning

Zeren Chen (School of Software Beihang University), Lu Sheng (School of Software Beihang University)

🎯 What it does: Designed and implemented a geometry-constrained untrained agent (GCA) to address spatial reasoning tasks in vision-language models (VLMs).

Geometry-Aligned and Anomaly-Aware Reconstruction for 3D Anomaly Detection

Linchun Wu (Wuhan University), Zhongyuan Wang (Wuhan University)

🎯 What it does: Proposed the AARD framework, achieving high-fidelity 3D point cloud anomaly detection and reconstruction through geometrically aligned noise sampling and anomaly-aware Transformer;

Geometry-as-context: Modulating Explicit 3D in Scene-consistent Video Generation to Geometry Context

JiaKui Hu (Peking University), Yanye Lu (Peking University)

🎯 What it does: Propose the Geometry-as-Context (GaC) framework, integrating geometry estimation, rendering, and image restoration into a differentiable autoregressive generative model to achieve single-end-to-end scene video generation.

Geometry-Aware Cross-Modal Graph Alignment for Referring Segmentation in 3D Gaussian Splatting

Yuwen Tao (Tsinghua University), Liyuan Wang (Tsinghua University)

🎯 What it does: Propose a geometry-aware cross-modal graph alignment framework (GeoCGA), which constructs semantic-space graphs and geometric graphs from natural language descriptions and 3D Gaussian fields, respectively, and performs fine-grained graph matching between them to achieve precise localization and segmentation of target objects in 3D scenes.

Geometry-driven OOD Detectors Are Class-Incremental Learners

Wangwang Jia (National University of Defense Technology), Kele Xu (National University of Defense Technology)

🎯 What it does: A geometry-driven OOD detector named GOD is proposed for class-incremental learning, enabling each task head to not only recognize its own task classes but also reject out-of-distribution (OOD) samples;

Geometry-Guided 3D Visual Token Pruning for Video-Language Models

Han Li (Beihang University), Si Liu (Beihang University)

🎯 What it does: Proposed a geometry-guided 3D visual token pruning framework called Geo3DPruner, which can significantly reduce the number of visual tokens in video-language models.

GeoMMBench and GeoMMAgent: Toward Expert-Level Multimodal Intelligence in Geoscience and Remote Sensing

Aoran Xiao (RIKEN AIP), Naoto Yokoya (RIKEN AIP)

🎯 What it does: This work proposes the GeoMMBench benchmark and the GeoMMAgent multi-agent framework to evaluate and enhance the specialized capabilities of multimodal large language models in the fields of geoscience and remote sensing.

GeoMotion: Rethinking Motion Segmentation via Latent 4D Geometry

Xiankang He (Zhejiang University of Technology), Xiaoqin Zhang (Zhejiang University of Technology)

🎯 What it does: This paper proposes a full end-to-end GeoMotion framework that fuses 4D prior geometry with optical flow to directly infer dynamic object masks from videos, avoiding explicit correspondence and iterative optimization;

GeoPredict: Leveraging Predictive Kinematics and 3D Gaussian Geometry for Precise VLA Manipulation

Jingjing Qian (Chinese University of Hong Kong Shenzhen), Li Jiang (Chinese University of Hong Kong Shenzhen)

🎯 What it does: Propose the GeoPredict framework, integrating trajectory prediction and 3D Gaussian structures into a VLA model to achieve precise prediction and planning of robot actions.

GeoRelight: Learning Joint Geometrical Relighting and Reconstruction with Flexible Multi-Modal Diffusion Transformers

Yuxuan Xue (Codec Avatars Lab, Meta), Javier Romero (Codec Avatars Lab, Meta)

🎯 What it does: Generate realistic relighting images, surface albedo, normals, and high-quality 3D geometry point clouds from a single photograph;

GeoRK2: Geometry-Guided Runge-Kutta Integration for Diffusion Transformer Acceleration

Chaoqun Sun (Bilibili Inc), Chenyu Wang (Bilibili Inc)

🎯 What it does: Proposed a training-agnostic GeoRK2 framework that significantly accelerates diffusion Transformer sampling by implementing second-order Runge-Kutta integration on Riemannian feature flows.

GeoSAM2: Unleashing the Power of SAM2 for 3D Part Segmentation

Ken Deng (VAST), Yan-Pei Cao (VAST)

🎯 What it does: Propose GeoSAM2, which transforms the 3D part segmentation task into multi-view 2D mask prediction, achieving fine-grained, controllable segmentation for texture-missing 3D models through simple 2D click or bounding box prompts.

GeoSANE: Learning Geospatial Representations from Models, Not Data

Joëlle Hanna (University of St.Gallen), Damian Borth (University of St.Gallen)

🎯 What it does: This study proposes GeoSANE, a model generation framework based on the weight space, which utilizes a collection of weights from publicly available remote sensing foundational models (RSFM). It trains a sequence-to-sequence autoencoder to map models with different architectures, tasks, and sensors into a unified latent space, and generates weights on demand that conform to the target architecture and task, directly applicable to downstream classification, segmentation, and detection tasks.

GeoSemba: Reconstructing State Space Model for Cross Paradigm Representation in Medical Image Segmentation

Xutao Sun (Liaoning Normal University), Yonggong Ren (Liaoning Normal University)

🎯 What it does: Proposed the GeoSemba framework, integrating the Mamba linear state space model, utilizing SSR and CAR modules to achieve cross-layer geometric-semantic interaction and cross-dimensional spatial-channel dependencies, significantly enhancing medical image segmentation accuracy and efficiency.

GeoSURGE: Geo-localization using Semantic Fusion with Hierarchy of Geographic Embeddings

Angel Daruna (SRI International), Rakesh Kumar (SRI International)

🎯 What it does: This paper proposes the GeoSURGE method, which achieves global image localization by leveraging semantic fusion and geographic embedding hierarchies.

GeoTikzBridge: Advancing Multimodal Code Generation for Geometric Perception and Reasoning

Jiayin Sun (China Mobile), Junlan Feng

🎯 What it does: Investigate the shortcomings of multimodal large language models in geometric perception and reasoning, proposing the GeoTikzBridge framework to enhance geometric reasoning capabilities through TikZ code generation.

GeoViS: Geospatially Rewarded Visual Search for Remote Sensing Visual Grounding

Peirong Zhang (Aerospace Information Research Institute Chinese Academy of Sciences), Lei Wang (Aerospace Information Research Institute Chinese Academy of Sciences)

🎯 What it does: Perform visual grounding in remote sensing images by proposing the GeoViS framework, decomposing the task into geographically reward-driven visual search and conditional localization.

GeoWorld: Geometric World Models

Zeyu Zhang (Australian National University), Richard Hartley (Australian National University)

🎯 What it does: Designed and implemented GeoWorld, a hyperbolic space-based energy-predicting world model for multi-step visual planning;

GFRRN: Explore the Gaps in Single Image Reflection Removal

Yu Chen (Zhejiang University), Zhe-Ming Lu (Zhejiang University)

🎯 What it does: Propose GFRRN (Gap-Free Reflection Removal Network), a new network architecture for single-image reflection removal.

GGBench: A Geometric Generative Reasoning Benchmark for Unified Multimodal Models

Jingxuan Wei (Qilu University of Technology), Cheng Tan (Shanghai Artificial Intelligence Laboratory)

🎯 What it does: Propose GGBench, a benchmark for evaluating the geometric construction reasoning capabilities of unified multimodal models, containing text, executable GeoGebra code, and rendered images with three-modal alignment.

GGPT: Geometry-Grounded Point Transformer

Yutong Chen (ETH Zurich), Siyu Tang (Delft University of Technology)

🎯 What it does: High-precision sparse geometry point clouds are obtained using dense feature matching and lightweight Bundle Adjustment. These sparse geometries are then fused with the original dense predictions in 3D space via a 3D Point Transformer, ultimately achieving globally geometrically consistent and refined dense 3D reconstruction results.

GH-NAF: Grid-Adaptive Hash-Level-Attended Neural Attenuation Fields for Discrepancy-Aware CBCT

Seong Je Oh (Seoul National University), Kyungsu Kim (Seoul National University)

🎯 What it does: Propose a new sparse-view CBCT reconstruction framework, GH-NAF, which utilizes multi-resolution hash encoding, grid-adaptive hash-level attention, error-aware rendering, and uncertainty-weighted supervision to significantly improve reconstruction quality under projection inconsistency environments.

Ghost-FWL: A Large-Scale Full-Waveform LiDAR Dataset for Ghost Detection and Removal

Kazuma Ikeda (Keio University), Kentaro Yoshioka (Sony Semiconductor Solutions)

🎯 What it does: Propose the Ghost-FWL dataset and develop a ghost detection and removal framework based on full-waveform LiDAR.

Ghosts in the Point Clouds: De-glaring LiDAR in the Transient Domain

Avery Gump (University of Wisconsin-Madison), Mohit Gupta (University of Wisconsin-Madison)

🎯 What it does: Proposes a training-agnostic algorithm that suppresses internal multipath glare in solid-state LiDAR before low-level peak detection;

GHPT: Real-Time Relightable Gaussian Splatting using Hybrid Path Tracing

Jinyang Bo (Northwest University), Guohua Geng (Northwest University)

🎯 What it does: Propose a three-stage inverse rendering framework (PGSR→differentiable rendering→FIPT) that can generate 3D Gaussian splatting (GS) models capable of real-time relighting.

GIFSplat: Generative Prior-Guided Iterative Feed-Forward 3D Gaussian Splatting from Sparse Views

Tianyu Chen (La Trobe University), Ramana Rao Kompella (Cisco Research)

🎯 What it does: Proposes an iterative forward 3D Gaussian splatting framework called GIFSplat, which achieves high-quality 3D scene reconstruction under sparse views by leveraging observation residuals and prior hints from frozen diffusion models.

GIFT: Global Irreplaceability Frame Targeting for Efficient Video Understanding

Junpeng Ma (Fudan University), Jian Pu (Zhejiang University)

🎯 What it does: Propose a training-free keyframe selection framework called GIFT, which evaluates and selects video frames based on global replaceability to reduce computational costs in video large language models.

GLINT: Modeling Scene-Scale Transparency via Gaussian Radiance Transport

Youngju Na (KAIST), Suyong Yeon (NAVER LABS)

🎯 What it does: The GLINT framework achieves physically consistent inverse rendering and geometric reconstruction for transparent scenes by explicitly decomposing the scene's Gaussian primitives into interface, transmission, and reflection components, combined with a hybrid rendering approach that integrates rasterization and ray tracing.

Global Information Thresholding for Sufficient and Necessary Circuits

Jegyeong Cho (KAIST)

🎯 What it does: This paper proposes a circuit extraction method based on a global threshold, utilizing integrated gradients to obtain edge importance scores, then adaptively selecting thresholds through a performance retention criterion, ultimately achieving a subgraph that is both sufficient and necessary.

Global Prior Meets Local Consistency: Dual-Memory Augmented Vision-Language-Action Model for Efficient Robotic Manipulation

Zaijing Li (Harbin Institute of Technology), Liqiang Nie (Harbin Institute of Technology)

🎯 What it does: Proposed a dual-memory Vision-Language-Action (VLA) framework OptimusVLA, combining Global Prior Memory (GPM) and Local Consistency Memory (LCM), to enhance the action generation efficiency and robustness of robotic manipulation.

Global Structure-from-Motion Meets Feedforward Reconstruction

Linfei Pan (ETH Zurich), Marc Pollefeys (ETH Zurich)

🎯 What it does: This paper proposes a full-process reconstruction system called GLUEMAP, which integrates classical structured light reconstruction (SfM) with an end-to-end feedforward learning method, maintaining high precision and scalability under challenging scenarios such as sparsity, low texture, and low viewpoint overlap.

Global Underwater Geolocation from Time-Lapse Polarization Imagery

Sara Aghajanzadeh (University of Illinois Urbana Champaign), Viktor Gruev (University of Illinois Urbana Champaign)

🎯 What it does: This paper proposes a global underwater localization method that recovers the solar elevation curve and predicts geographic coordinates using underwater polarization time series images.

Global-Aware Edge Prioritization for Pose Graph Initialization

Tong Wei (Czech Technical University in Prague), Daniel Barath (ETH Zurich)

🎯 What it does: Propose a camera pose graph initialization method based on global edge priority prediction and multi-minimum spanning tree construction using graph neural networks (GNNs), which can build a sparse and connected pose graph without requiring complete geometric validation beforehand.

Global-Graph Guided and Local-Graph Weighted Contrastive Learning for Unified Clustering on Incomplete and Noise Multi-View Data

Hongqing He (Guangxi Normal University), Xiaofeng Zhu (Hainan University)

🎯 What it does: Propose a global-local graph-guided contrastive learning framework named GLGC for unified handling of incomplete and noisy multi-view clustering.

Globally Optimal Pose from Orthographic Silhouettes

Agniva Sengupta (Freie Universitaet Berlin), Stefan Zachow (Zuse Institute Berlin)

🎯 What it does: The study proposes a method for achieving globally optimal pose estimation by leveraging the orthographic/perspective projection silhouettes of objects.

Globscope: Toward a Global View of the Loss Landscape

Mashiat Mustaq (Purdue University), Xavier M. Tricoche (Purdue University)

🎯 What it does: Propose the Globscope framework, which uses an autoencoder to learn a low-dimensional nonlinear manifold of the neural network parameter space, enabling the visualization and reconstruction of the global loss landscape.

Gloria: Consistent Character Video Generation via Content Anchors

Yuhang Yang, Zheng-Jun Zha (University Of Science And Technology Of China)

🎯 What it does: Propose Gloria, a long-term consistency human video generation framework based on content anchors, which can maintain multi-view appearance and expression consistency in videos longer than 10 minutes.

Glove2Hand: Synthesizing Natural Hand-Object Interaction from Multi-Modal Sensing Gloves

Xinyu Zhang (Meta Reality Labs), Li Guan

🎯 What it does: Propose the Glove2Hand framework, which converts sensor-equipped glove videos into realistic bare-hand videos while synchronously preserving tactile and IMU signals.

GlyphPrinter: Region-Grouped Direct Preference Optimization for Glyph-Accurate Visual Text Rendering

Xincheng Shuai (Fudan University), Dacheng Tao (Nanyang Technological University)

🎯 What it does: Proposed a visual text rendering method called GlyphPrinter based on preference optimization, aiming to significantly improve the glyph accuracy of rendered text.

GM-R^2: Generative Matching Learning for Unsupervised Geometric Representation and Registration

Haobo Jiang (Nanyang Technological University), Jianmin Zheng (Nanyang Technological University)

🎯 What it does: Propose GM-R²: an unsupervised geometric descriptor framework based on generative alignment learning, which indirectly supervises feature learning through cross-view image generation conditioned on geometry, thereby achieving high-quality point cloud registration.

GMT: Effective Global Framework for Multi-Camera Multi-Target Tracking

Yihao Zhen, Huijie Fan (Chinese Academy Of Sciences)

🎯 What it does: Proposed a global multi-camera multi-object tracking framework GMT that constructs global trajectories across all perspectives and performs trajectory-target matching directly.

Goal Force: Teaching Video Models To Accomplish Physics-Conditioned Goals

Nate Gillman (Brown University), Chen Sun (Brown University)

🎯 What it does: This paper proposes and implements a framework called Goal Force, enabling video generation models to plan prior actions to achieve physical goals by inputting a target force vector;

Goal-Driven Reward by Video Diffusion Models for Reinforcement Learning

Qi Wang (Shanghai Jiao Tong University), Wenjun Zeng (Ningbo Key Laboratory of Spatial Intelligence and Digital Derivative, Ningbo Institute of Digital Twin, Eastern Institute of Technology)

🎯 What it does: This paper proposes the GenReward framework, which uses a pre-trained video diffusion model to provide goal-based video and frame-level rewards for reinforcement learning.

Goldilocks Test Sets for Face Verification

Haiyu Wu (University of Notre Dame), Kevin Bowyer

🎯 What it does: This paper creates three entirely new facial verification test sets: Hadrian (focused on facial hair differences), Eclipse (focused on exposure differences), and ND-Twins (focused on monozygotic twins), and evaluates them under the LFW 10-fold cross-validation framework.

Good Can Sometimes be Bad: A Unified Attack against 3D Point Cloud Classifier by a Flexible Isotropic Resampling

Linkun Fan (Henan University), Daojun Han (Henan University)

🎯 What it does: Proposed a unified 3D point cloud attack framework called UAtt3D, which can simultaneously perform adversarial attacks during the inference phase and backdoor attacks during the training phase;

GOR-IS: 3D Gaussian Object Removal In the Intrinsic Space

Yonghao Zhao (Nankai University), Beibei Wang (Nanjing University)

🎯 What it does: Propose a 3D object removal framework GOR-IS based on 3D Gaussian projection, achieving physically consistent and visually coherent 3D object removal in the Intrinsic space.

GP-4DGS: Probabilistic 4D Gaussian Splatting from Monocular Video via Variational Gaussian Processes

Mijeong Kim (Seoul National University), Bohyung Han (Seoul National University)

🎯 What it does: This paper combines Gaussian processes with 4D Gaussian splatting to propose GP-4DGS for probabilistic dynamic scene reconstruction from monocular videos.

GPFlow: Gaussian Prototype Probability Flow for Unsupervised Multi-Modal Anomaly Detection

Yiting Li (Institute for Infocomm Research, A*STAR), Fayao Liu (Institute for Infocomm Research, A*STAR)

🎯 What it does: Propose GPFlow, a multimodal unsupervised few-shot anomaly detection framework based on learnable Gaussian prototypes and posterior mean path (PMP) routing.

gQIR: Generative Quanta Image Reconstruction

Aryan Garg (University of Wisconsin-Madison), Mohit Gupta (University of Wisconsin-Madison)

🎯 What it does: Proposed a three-stage framework gQIR for reconstructing high-quality color images and videos from extremely low-light quantum frames (single-photon counters), integrating VAE denoising/de-mosaicking, LoRA fine-tuned diffusion models, and spatiotemporal Transformer fusion;

GR-Gauge: Cost-efficient Training Configuration By Gauging the Gradient Redundancy

Guanjie Wang (Shanghai Jiao Tong University), Chen Chen (Shanghai Jiao Tong University)

🎯 What it does: Propose a method called GR-Gauge, which achieves efficient automated hyperparameter tuning for learning rate and batch size by monitoring gradient redundancy (GR_T and GR_S) during training, supporting early stopping, search direction adjustment, and intermediate state reuse.

Gradient Knows Best: Mixed-Precision Quantization via Gradient-Guided Bit Allocation for Super-Resolution

Jun Young Kim (Dongguk University), Sung In Cho (Sogang University)

🎯 What it does: Proposed a mixed-precision quantization framework based on post-training quantization (PTQ), specifically designed for image super-resolution (SR) models.

Granulon: Awakening Pixel-Level Visual Encoders with Adaptive Multi-Granularity Semantics for MLLM

Junyuan Mao (National University of Singapore), Yueming Jin (National University of Singapore)

🎯 What it does: Proposes Granulon, a multi-scale controllable visual encoder based on DINOv3, combining a text-driven granularity controller and adaptive token aggregation to achieve multi-granularity reasoning from pixel-level to coarse-grained in a single forward pass.

Graph Attention Prototypical Network for Robust Few-Shot Classification

Tingyun Liu (Hunan University), C. L. Philip Chen (South China University Of Technology)

🎯 What it does: Proposed a Graph Attention Prototype Network (GAPNet), which extracts global features through group attention generalized learning, models intra- and inter-class relationships via pseudo label guided graph construction combined with edge attention, and suppresses prototype drift caused by label noise using an adaptive noise-robust prototype generator.

Graph-to-Frame RAG: Visual-Space Knowledge Fusion for Training-Free and Auditable Video Reasoning

Songyuan Yang (National University of Defense Technology), Nong Xiao (Sun Yat-sen University)

🎯 What it does: Propose Graph-to-Frame Retrieval-Augmented Generation (G2F-RAG), which attaches retrieved knowledge as visual frames to videos within a training-agnostic multi-agent framework, achieving knowledge fusion in the visual space;

Graph2Eval: Automatic Multimodal Task Generation for Agents via Knowledge Graphs

Yurun Chen (Zhejiang University), Shengyu Zhang (Zhejiang University)

🎯 What it does: Proposed the GRAPH2EVAL framework, which generates multi-modal tasks using knowledge graphs and constructed a benchmark dataset named GRAPH2EVAL-BENCH with 1,319 questions to evaluate the capabilities of RAG and Web Agents.

GraPHFormer: A Multimodal Graph Persistent Homology Transformer for the Analysis of Neuroscience Morphologies

Uzair Shah (Hamad Bin Khalifa University), Mowafa Househ (Hamad Bin Khalifa University)

🎯 What it does: Proposes GraPHFormer, a multimodal neuromorphic analysis framework that simultaneously leverages tree graph structures and multi-channel persistence images through CLIP-style contrastive learning.

GraphVLM: Benchmarking Vision Language Models for Multimodal Graph Learning

Jiajin Liu (NYU Shanghai), Qiaoyu Tan (Rice University)

🎯 What it does: Propose the GraphVLM benchmark, systematically evaluating the three roles of vision-language models (VLM) in multimodal graph learning (VLM-as-Encoder, VLM-as-Aligner, VLM-as-Predictor), and conducting unified experiments on six multimodal graph datasets.

GraspALL: Adaptive Structural Compensation from Illumination Variation for Robotic Garment Grasping in Any Low-Light Conditions

Haifeng Zhong (Jilin University), Yixing Gao (University of Birmingham)

🎯 What it does: Propose the GraspALL model to address the robustness of clothing grasping by service robots in low-light environments; estimate scene illumination through a learnable parametric luminance curve, which guides adaptive feature compensation for RGB and non-RGB (depth) modalities, ultimately achieving high-precision semantic segmentation and grasp point prediction.

GraspGen-X: Cross-Embodiment 6-DOF Diffusion-based Grasping

Beining Han (NVIDIA), Adithyavairavan Murali (NVIDIA)

🎯 What it does: Proposed a cross-robotic-arm 6-DOF grasping model GraspGen-X, capable of achieving zero-shot grasping on unknown objects and unknown grippers

GraspLDP: Towards Generalizable Grasping Policy via Latent Diffusion

Enda Xiang (Beihang University), Di Huang (Beihang University)

🎯 What it does: Proposes GraspLDP, a framework that generates general grasping strategies by leveraging latent diffusion models combined with priors on grasp poses and grasp strength maps.

Gravitation-Driven Semantic Alignment for Text Video Retrieval

Yi Yang (Tongji University), Heng Tao Shen (Tongji University)

🎯 What it does: Proposed the GraviAlign framework, which achieves cross-modal alignment by probabilistically embedding videos and texts through two orthogonal factors inspired by physical gravity: semantic attraction and geometric overlap.

Grid Distillation: Compositional Image Distillation via Structured Generative Grids

Biplab Ch Das (Samsung R&D Institute), Viswanath Gopalakrishnan (IIIT)

🎯 What it does: Propose the Grid Distillation framework, which constructs structured grids using spectral submodular optimization and recovers fine details through single-step diffusion, achieving compression and distillation on large-scale datasets.

GrOCE : Graph-Guided Online Concept Erasure for Text-to-Image Diffusion Models

Ning Han (Xiangtan University), Jingjing Chen (Fudan University)

🎯 What it does: Proposed a training-free, online concept elimination framework called GrOCE, which utilizes dynamic semantic graphs to cluster and identify target concepts in text prompts and precisely sever their influence in the text embedding space.

Ground Reaction Inertial Poser: Physics-based Human Motion Capture from Sparse IMUs and Insole Pressure Sensors

Ryosuke Hori (Carnegie Mellon University), Kris Kitani (Carnegie Mellon University)

🎯 What it does: Propose the Ground Reaction Inertial Poser (GRIP) method, which utilizes four wearable IMUs and insole pressure sensors to achieve full-body motion capture through deep networks and physics simulation;

Grounded 3D-Aware Spatial Vision-Language Modeling

An-Chieh Cheng (UCSD), Sifei Liu (NVIDIA)

🎯 What it does: Propose the GR3D framework, integrating explicit 2D localization, implicit 2D localization (dynamic region insertion), and monocular 3D localization to achieve unified reasoning for spatial inference and 3D detection.

Grounded Chain-of-Thought for Multimodal Large Language Models

Qiong Wu (Key Laboratory of Multimedia Trusted Perception and Efficient Computing, Ministry of Education of China, Xiamen University), Rongrong Ji (Key Laboratory of Multimedia Trusted Perception and Efficient Computing, Ministry of Education of China, Xiamen University)

🎯 What it does: Proposed the 'Grounded Chain-of-Thought (GCoT)' task and the MM-GCoT benchmark to evaluate the performance of multi-modal large language models (MLLMs) in visual spatial reasoning and visual hallucination;

Grounded Latents for Entity-Centric 4D Scene Generation

Jinhyung Park (Carnegie Mellon University), Kris Kitani (Carnegie Mellon University)

🎯 What it does: Propose LatentWorld, which utilizes sparse, instantiated 3D latent vectors for controllable 3D and 4D scene generation, and achieves fine geometric reconstruction through Gaussian splitting.

Grounding Everything in Tokens for Multimodal Large Language Models

Xiangxuan Ren (Shanghai Jiao Tong University), Chao Ma (Shanghai Jiao Tong University)

🎯 What it does: Propose the GETok framework, introducing learnable grids and offset tokens into multimodal large language models to achieve unified localization and segmentation in 2D space;