CVPR 2026 Papers with Code — Page 4

FailureAtlas: Mapping the Failure Landscape of T2I Models via Active Exploration

Muxi Chen (Chinese University of Hong Kong), Qiang Xu (Chinese University of Hong Kong)

🎯 What it does: Proposed the FAILUREATLAS framework for actively exploring and automating the discovery of failure slices in text-to-image models, and constructed a large-scale entity-attribute corpus to structure the search space.

FairLLaVA: Fairness-Aware Parameter-Efficient Fine-Tuning for Large Vision-Language Assistants

Mahesh Bhosale (University at Buffalo), Xuan Gong (Harvard Medical School)

🎯 What it does: Proposes FairLLaVA, a parameter-efficient fine-tuning method for medical multimodal large language models, leveraging mutual information minimization to eliminate potential sensitive attribute shortcuts in images;

Faithful Contouring: Near-Lossless 3D Voxel Representation Free from Iso-surface

Yihao Luo (Imperial College London), ChoonHwai Yap (Imperial College London)

🎯 What it does: Propose a sparse voxelization representation called Faithful Contouring, which can directly convert any triangle mesh into a near-lossless voxel representation and is applicable to high-resolution (2048+) reconstruction and compression.

FAPE-IR: Frequency-Aware Planning and Execution Framework for All-in-One Image Restoration

Jingren Liu (Tianjin University), Zhong Ji (Tianjin University)

🎯 What it does: Proposes a frequency-aware planning and execution framework, FAPE-IR, capable of uniformly handling multiple image restoration tasks including denoising, dehazing, deraining, desnowing, deblurring, low-light enhancement, and super-resolution.

Fast Markov Random Field Optimisation for Topologically Noisy 3D Shape Matching

Paul Roetzer (University of Bonn), Florian Bernard (University of Bonn)

🎯 What it does: Propose a 3D shape matching framework based on Markov Random Fields (MRF), specifically addressing shape correspondence problems under topological noise (e.g., changes in genus).

FAST: Topology-Aware Frequency-Domain Distribution Matching for Coreset Selection

Jin Cui (Xi'an Jiaotong University), Pengju Ren (Xi'an Jiaotong University)

🎯 What it does: This paper proposes a DNN-free core subset selection method called FAST, which achieves complete distribution alignment with the original dataset through frequency domain distribution matching.

Faster-GS: Analyzing and Improving Gaussian Splatting Optimization

Florian Hahlbohm (TU Braunschweig), Marcus Magnor (University of New Mexico)

🎯 What it does: Propose Faster-GS, an efficient training framework based on 3D Gaussian splatting, significantly accelerating optimization and reducing VRAM usage.

FastLightGen: Fast and Light Video Generation with Fewer Steps and Parameters

Shitong Shao (Hong Kong University of Science and Technology), Zeke Xie (Hong Kong University of Science and Technology)

🎯 What it does: Propose FastLightGen, a three-stage joint distillation framework, achieving video generation with fewer steps and parameter compression through identifying irrelevant layers, dynamic pruning, and distribution matching.

FastRef: Fast Prototype Refinement for Few-shot Industrial Anomaly Detection

Yufei Li (Xidian University), Xiyang Liu (Xidian University)

🎯 What it does: Proposes a fast prototype refinement module called FastRef, which enhances the representativeness of prototypes in few-shot industrial defect detection by iteratively updating transformation matrices and transport probabilities.

FB-CLIP: Fine-Grained Zero-Shot Anomaly Detection with Foreground-Background Disentanglement

Ming Hu (Xi'an Institute of Optics and Precision Mechanics Chinese Academy of Sciences), Quan Wang (Xi'an Institute of Optics and Precision Mechanics Chinese Academy of Sciences)

🎯 What it does: In the unannotated anomaly detection task, the FB-CLIP framework is proposed, significantly enhancing CLIP's zero-shot performance in fine-grained anomaly localization through multi-strategy text feature fusion, foreground-background separation, and background suppression.

Fed-ADE: Adaptive Learning Rate for Federated Post-adaptation under Distribution Shift

Heewon Park (Sungkyunkwan University), Minhae Kwon (Sungkyunkwan University)

🎯 What it does: Propose an unsupervised federated post-deployment adaptation framework Fed-ADE, which dynamically adjusts the learning rate in a multi-client distribution drift environment to improve model performance on unlabeled data streams;

FedAdamom: Adaptive Momentum for Improved Generalization in Federated Optimization

Wenjie Hou (Beihang University), Wei Yang Bryan Lim (Nanyang Technological University)

🎯 What it does: This paper proposes a new federated learning adaptive optimizer, FedAdamom, which enhances the generalization performance of federated models by adaptively adjusting the global momentum at the parameter level to escape saddle points quickly while favoring flat minima.

FedAFD: Multimodal Federated Learning via Adversarial Fusion and Distillation

Min Tan (Hangzhou Dianzi University), Zhou Yu (Hangzhou Dianzi University)

🎯 What it does: Proposes the FedAFD framework, achieving cross-modal and cross-task collaborative training in multi-modal federated learning while balancing client personalization and server global performance.

FedBPrompt: Federated Domain Generalization Person Re-Identification via Body Distribution Aware Visual Prompts

Xin Xu (Wuhan University of Science and Technology), Kui Jiang (Harbin Institute of Technology Zhengzhou Research Institute)

🎯 What it does: This paper proposes the FedBPrompt framework, which inserts learnable Body Distribution Aware Visual Prompts (BAPM) into ViT to achieve background suppression and view alignment in federated domain generalization for person Re-ID, and designs a Prompt-based Fine-Tuning Strategy (PFTS) to significantly reduce communication costs.

Federated Active Learning Under Extreme Non-IID and Global Class Imbalance

Chen-Chen Zong (Nanjing University of Aeronautics and Astronautics), Sheng-Jun Huang (Nanjing University of Aeronautics and Astronautics)

🎯 What it does: Propose a federated active learning framework named FairFAL to efficiently reduce annotation costs under extreme non-independent and identically distributed (non-IID) and globally imbalanced class distribution scenarios.

FedMOP: Achieving Enhanced Privacy and Performance in Federated Learning via Momentum Orthogonal Projection

Yunlong Zhao (Central South University), Xiu Su (Central South University)

🎯 What it does: Propose the FedMOP method, achieving privacy protection and performance improvement in federated learning by using momentum-based orthogonal projection initialization offset before client training.

FedRG: Unleashing the Representation Geometry for Federated Learning with Noisy Clients

Tian Wen (Chinese Academy of Sciences), Bo Han (Hong Kong Baptist University)

🎯 What it does: Propose the FedRG method, which identifies noisy labels on unlabeled spherical representations by leveraging the representation geometry priority principle, and achieves robust optimization through a personalized noise absorption matrix.

FedSDR: Federated Graph Learning with Structural Noise Detection and Reconstruction

Jiaqi Liu (Wuhan University), Mang Ye (Wuhan University)

🎯 What it does: This paper proposes FedSDR, a federated graph learning framework designed for high structural noise scenarios. It can detect structural noise in client graphs and perform structural reconstruction, thereby enhancing the collaborative effectiveness between the global model and damaged clients.

FilterGS: Traversal-Free Parallel Filtering and Adaptive Shrinking for Large-Scale LoD 3D Gaussian Splatting

Yixian Wang (Beijing Institute of Technology), Yi Yang (Beijing Institute of Technology)

🎯 What it does: Propose the FilterGS method, leveraging non-traversal parallel filtering and adaptive Gaussian contraction to achieve efficient rendering of large-scale 3D Gaussian splatting scenes.

Fine-Grained Post-Training Quantization for Large Vision Language Models with Quantization-Aware Integrated Gradients

Ziwei Xiang (CASIA), Xu-Yao Zhang (CASIA)

🎯 What it does: Proposes a fine-grained post-training quantization method for large vision-language models, which uses quantization-aware integrated gradients (QIG) to measure token-level quantization errors, and subsequently performs channel-balanced quantization of weights and activations.

Fine-Tuning Impairs the Balancedness of Foundation Models in Long-tailed Personalized Federated Learning

Shihao Hou (Xiamen University), Yang Lu (Xiamen University)

🎯 What it does: This paper proposes the FedPuReL method, which maintains the balance of pre-trained models in long-tailed personalized federated learning through gradient purification, and achieves unbiased personalization via residual learning.

Flow4DGS-SLAM: Optical Flow-Guided 4D Gaussian Splatting SLAM

Yunsong Wang (National University of Singapore), Gim Hee Lee (National University of Singapore)

🎯 What it does: Propose Flow4DGS-SLAM, a flow-guided 4D Gaussian Splatting SLAM system that simultaneously tracks camera pose and reconstructs dynamic scenes in real-time.

FlowDIS: Language-Guided Dichotomous Image Segmentation with Flow Matching

Andranik Sargsyan (Picsart AI Research), Shant Navasardyan (Picsart AI Research)

🎯 What it does: Proposes a binary image segmentation framework based on flow matching called FlowDIS, and introduces a position-aware instance pairing (PAIP) training strategy to achieve controllable language-guided segmentation.

FluidGaussian: Propagating Simulation-Based Uncertainty Toward Functionally-Intelligent 3D Reconstruction

Yuqiu Liu (Simon Fraser University), Michael W. Mahoney (University Of California Berkeley)

🎯 What it does: Proposed FluidGaussian, a plugin that integrates fluid-structure interaction information into 3D Gaussian Splatting reconstruction, achieving both visual and physical consistency.

FluxMem: Adaptive Hierarchical Memory for Streaming Video Understanding

Yiweng Xie (Fudan University), Zuxuan Wu (Fudan University)

🎯 What it does: Propose FluxMem, a training-free, hierarchical memory framework adapted for multi-modal large language models, designed for streaming video understanding.

FMPose3D: monocular 3D pose estimation via flow matching

Ti Wang (École Polytechnique Fédérale de Lausanne), Mackenzie Weygandt Mathis (École Polytechnique Fédérale de Lausanne)

🎯 What it does: Proposed a monocular 3D pose estimation framework FMPose3D based on flow matching, which can directly convert Gaussian noise into a 3D joint distribution under a conditional ODE, achieving multi-hypothesis generation and efficient inference.

Focus on Background: Exploring SAM's Potential in Few-shot Medical Image Segmentation with Background-centric Prompting

Yuntian Bo (Nanjing University of Science and Technology), Haofeng Zhang (Nanjing University of Science and Technology)

🎯 What it does: Propose FoB, a background-centric prompt generator that provides precise background point prompts for the Segment Anything Model (SAM), significantly enhancing the performance of few-shot medical image segmentation (FSMIS).

Focus-to-Perceive Representation Learning: A Cognition-Inspired Hierarchical Framework for Endoscopic Video Analysis

Yuan Zhang (Xiangtan University), Xieping Gao (Hunan Normal University)

🎯 What it does: Proposed a cognitive-inspired hierarchical framework FPRL, which enhances endoscopic video representation learning by first focusing on the static semantics of lesions and then perceiving their temporal evolution.

Focus, Don't Prune: Identifying Instruction-Relevant Regions for Information-Rich Image Understanding

Mincheol Kwon (Korea University), Jinkyu Kim (Korea University)

🎯 What it does: This paper proposes PinPoint, a two-stage framework that first locates instruction-related image regions and then refines visual feature extraction, aiming to improve the reasoning efficiency and accuracy of large vision-language models on information-rich images.

ForeAct: Steering Your VLA with Efficient Visual Foresight Planning

Zhuoyang Zhang (MIT), Song Han (MIT)

🎯 What it does: Propose the Visual Foresight Planning (ForeAct) framework, which guides the Vision-Language-Action (VLA) model to achieve closed-loop control by generating future observation images and subtask descriptions.

Forecast the Principal, Stabilize the Residual: Subspace-Aware Feature Caching for Diffusion Transformers

Guantao Chen (Shanghai Jiao Tong University), Linfeng Zhang (Shanghai Jiao Tong University)

🎯 What it does: Propose a training-free feature caching method called SVD-Cache, which first performs singular value decomposition (SVD) on the hidden features of diffusion Transformers, dividing them into a principal subspace and a residual subspace. The principal subspace is predicted using exponential moving average (EMA), while the residual subspace is directly reused, significantly accelerating inference while maintaining image/video quality.

Forecasting 3D Scanpaths in Egocentric Video

Fiona Ryan (Georgia Institute of Technology), Calvin Murdock (Meta Reality Labs Research)

🎯 What it does: Propose and implement a task of predicting future 3D gaze paths (scanpaths) in egocentric videos, inferring future gaze positions and durations by leveraging historical video frames, head pose, and past 3D gaze points.

ForeHOI: Feed-forward 3D Object Reconstruction from Daily Hand-Object Interaction Videos

Yuantao Chen (Chinese University of Hong Kong), Xiaoguang Han (Chinese University of Hong Kong)

🎯 What it does: Developed an end-to-end feed-forward 3D object reconstruction model called ForeHOI, which can directly generate complete object geometry from monocular hand-object interaction videos.

Forensic-Friendly Image Manipulation via Controllable Latent Diffusion

Hanyu Chen (University of Electronic Science and Technology of China), Jiantao Zhou (University of Macau)

🎯 What it does: Proposes a plug-and-play control denoising framework called Forensic-Friendly Image Manipulation (FFIM), which actively controls noise during the editing process of diffusion models. This enables the generated images to meet user editing requirements while significantly improving the detection and localization accuracy of third-party forensic algorithms.

Foundation Encoders Are All You Need for Preference-Aware Personalization

Hyungjin Kim (Inha University), Young-Duk Seo (Inha University)

🎯 What it does: Achieve preference-aware personalized image generation using a base encoder without requiring additional structures or fine-tuning.

Foundation Model Priors Enhance Object Focus in Feature Space for Source-Free Object Detection

Sairam VCR (IIT Hyderabad), Muhammad Haris Khan (MBZUAI)

🎯 What it does: In source-agnostic object detection, class-agnostic segmentation masks from large vision foundation models are used to spatially regularize the feature space of the detector, combined with imbalance-robust pseudo-label learning to achieve self-supervised adaptation to the target domain.

FoundIR-v2: Optimizing Pre-Training Data Mixtures for Image Restoration Foundation Model

Xiang Chen (Nanjing University of Science and Technology), Jinhui Tang (Nanjing Forestry University)

🎯 What it does: Developed FoundIR-v2, a high-capacity diffusion image restoration foundation model that dynamically optimizes data mixing ratios and employs Mixture-of-Experts (MoE) scheduling, capable of handling over 50 subtasks in a single pass.

Fourier Angle Alignment for Oriented Object Detection in Remote Sensing

Changyu Gu (Beijing Institute of Technology), Ying Fu (Beijing Institute of Technology)

🎯 What it does: In remote sensing images, frequency domain analysis is utilized to estimate and align the main orientation of targets, thereby enhancing directional consistency and angle regression performance;

FoV-Net: Rotation-Invariant CAD B-rep Learning via Field-of-View Ray Casting

Matteo Ballegeer (Ghent University), Dries F. Benoit (Ghent University)

🎯 What it does: Propose the FoV-Net framework, combining local reference frame UV grids and field-of-view ray grids to enable rotation-invariant learning for B-rep.

FOZO: Forward-Only Zeroth-Order Prompt Optimization for Test-Time Adaptation

Xingyu Wang (Sichuan University), Tao Wang (Sichuan University)

🎯 What it does: Propose a Forward-only Zeroth-Order Gradient Prompt Optimization (FOZO) framework for test-time adaptation without gradient backpropagation, inserting learnable prompts into ViT inputs and adapting to the target domain through zeroth-order gradient estimation and dynamic perturbation.

Frame2Freq: Spectral Adapters for Fine-Grained Video Understanding

Thinesh Thiyakesan Ponbagavathi (University of Stuttgart), Alina Roitberg

🎯 What it does: Developed a frequency domain adaptive module (Frame2Freq) to transfer frozen visual foundation models (e.g., CLIP, DINOv2) to fine-grained video understanding tasks;

Free-Lunch Long Video Generation via Layer-Adaptive O.O.D Correction

Jiahao Tian, Chi Zhang

🎯 What it does: Achieve long video generation on pre-trained DiT video diffusion models using the training-agnostic FreeLOC framework.

Fresco: Frequency-Spatial Consistent Optimization for Fine-Grained Head Avatar Modeling

Shikun Zhang (Monash University), Cunjian Chen (Monash University)

🎯 What it does: Proposed the Fresco optimization framework, combining frequency domain curriculum learning with UV space consistency constraints, significantly enhancing the detail quality and cross-view consistency of head animation models.

From Exploration to Exploitation: A Two-Stage Entropy RLVR Approach for Noise-Tolerant MLLM Training

Donglai Xu (Independent Researcher), Lai-Man Po (City University of Hong Kong)

🎯 What it does: Propose a two-stage RLVR training method based on token entropy to enhance the robustness of multimodal large language models under noisy annotations.

From Few-way to Many-way: Rethinking Few-shot Fine-grained Image Classification

Li-Jun Zhao (Shandong University), Xin-Shun Xu (Shandong University)

🎯 What it does: This paper proposes an SCEG method for few-shot fine-grained image classification, expanding the problem for the first time from traditional few-way scenarios to more challenging many-way scenarios.

From Indoor to Open World: Revealing the Spatial Reasoning Gap in MLLMs

Mingrui Wu (University of Chinese Academy of Sciences), Tong Zhang (University of Chinese Academy of Sciences)

🎯 What it does: Proposes OSi-Bench, an outdoor spatial reasoning benchmark based on multi-sensor pedestrian perspective videos, evaluating relational, metric, and motion reasoning capabilities of multi-modal large language models (MLLM).

From Infusion to Assimilation Distillation for Medical Image Segmentation

Jiankang Hong (Tongji University), Junsong Yuan (State University of New York at Buffalo)

🎯 What it does: Designed a two-phase knowledge distillation framework named IAD for compressing and enhancing medical image segmentation models.

From Pairs to Sequences: Track-Aware Policy Gradients for Keypoint Detection

Yepeng Liu (Wuhan University), Yongchao Xu (Xiaomi EV)

🎯 What it does: Propose a trajectory-aware keypoint detection framework TraqPoint based on reinforcement learning, directly optimizing the long-term trackability of keypoints on image sequences;

From Sketch to Fresco: Efficient Diffusion Transformer with Progressive Resolution

Shikang Zheng (Shanghai Jiao Tong University), Linfeng Zhang (Shanghai Jiao Tong University)

🎯 What it does: Proposes a training-free progressive resolution sampling framework called Fresco, which achieves high-resolution generation by utilizing a unified noise field and variance-guided progressive upsampling.

From Static to Dynamic: Exploring Self-supervised Image-to-Video Representation Transfer Learning

Yang Liu (University of Chinese Academy of Sciences), Qingming Huang (University of Chinese Academy of Sciences)

🎯 What it does: This paper proposes the Co-Settle framework, which performs self-supervised image-to-video representation transfer learning using a lightweight projection layer on a frozen image pre-trained encoder;

FUSER: Feed-Forward Multiview 3D Registration Transformer and SE(3)$^N$ Diffusion Refinement

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

🎯 What it does: The study proposes FUSER, a fully feed-forward multi-view 3D registration Transformer that directly predicts global pose in a unified latent space, avoiding traditional pairwise matching and synchronization steps.

FusionRegister: Every Infrared and Visible Image Fusion Deserves Registration

Congcong Bian (Jiangnan University), Xiao-jun Wu (Jiangnan University)

🎯 What it does: Proposes a post-registration framework called FusionRegister based on visual priors to correct residual spatial mismatches during infrared-visible image fusion, thereby enhancing the structural accuracy and detail preservation of the fused results.

FVBench: Benchmarking Deepfake Video Detection Capability of Large Multimodal Models

Jiarui Wang (Shanghai Jiao Tong University), Xiongkuo Min (Shanghai Jiao Tong University)

🎯 What it does: This paper constructs FVBench, a deepfake detection benchmark comprising 120k videos, and evaluates the performance of traditional detection models and large multilingual models (LMM) on real, AI-edited, and AI-generated videos.

G-MIXER: Geodesic Mixup-based Implicit Semantic Expansion and Explicit Semantic Re-ranking for Zero-Shot Composed Image Retrieval

Jiyoung Lim (Sungkyunkwan University), Jee-Hyong Lee (Sungkyunkwan University)

🎯 What it does: Proposed a training-agnostic zero-shot synthetic image retrieval method called G-MIXER, which enhances implicit semantics through geodesic spherical mixing between image and text features and improves retrieval diversity and accuracy by leveraging explicit semantics for re-ranking.

Gamba: Mamba-based graph convolutional network with dynamic graph topology learning for action recognition

Rouyi Zhou (Shenzhen University), Linlin Shen (Shenzhen University)

🎯 What it does: This paper proposes the Gamba network, which combines Mamba and GCN to achieve dynamic graph topology learning and long-term temporal modeling for human action recognition.

GaussianDWM: 3D Gaussian Driving World Model for Unified Scene Understanding and Multi-Modal Generation

Tianchen Deng (Shanghai Jiao Tong University), Hesheng Wang (Shanghai Jiao Tong University)

🎯 What it does: Developed a unified driving world model based on 3D Gaussian fields (GaussianDWM), achieving 3D perception and multimodal generation in driving scenarios.

GaussianMatch: Semi-Supervised Regression with Pseudo-Label Filtering via Multi-View Gaussian Consistency

Yin Wang (Zhejiang University), Shuiguang Deng (Zhejiang University)

🎯 What it does: Proposed GaussianMatch, a framework for semi-supervised regression that utilizes multi-view prediction consistency to filter high-quality pseudo labels;

GDPO-SR: Group Direct Preference Optimization for One-Step Generative Image Super-Resolution

Qiaosi Yi (Hong Kong Polytechnic University), Lei Zhang (Hong Kong Polytechnic University)

🎯 What it does: Proposed a single-step generative image super-resolution framework GDPO (Group Direct Preference Optimization) based on reinforcement learning

GeCo: Geometry-Consistent Regularization for Domain Generalized Semantic Segmentation

Qi Zang (Hefei University of Technology), Meng Wang (Hefei University of Technology)

🎯 What it does: This paper addresses the domain generalization semantic segmentation task by proposing Geometry-Consistent Regularization (GeCo), which significantly enhances the model's robustness and unknown category recognition capability on unseen domains through geometry-consistent regularization of lightweight adapters (e.g., LoRA) for Vision Foundation Models (VFMs).

GEM: Generating LiDAR World Model via Deformable Mamba

Yang Wu (NJUST), Jin Xie (Nanjing University)

🎯 What it does: Proposes GEM—a generative LiDAR world model based on Mamba, which discretizes LiDAR scans using a customized LiDAR scene tokenizer, employs an unsupervised dynamic-static separator, and adaptively scans, aggregates, and fuses dynamic, static, and general features through a three-path deformable Mamba, achieving high-precision prediction and controllable generation of future LiDAR sequences;

Generalizable Knowledge Distillation from Vision Foundation Models for Semantic Segmentation

Chonghua Lv (Xidian University), Zhun Zhong (Hefei University of Technology)

🎯 What it does: This paper studies how to distill the general knowledge of Vision Foundation Models (VFMs) into lightweight semantic segmentation models to enhance cross-domain generalization capabilities.

Generalizable Video Quality Assessment via Weak-to-Strong Learning

Linhan Cao (Shanghai Jiao Tong University), Xiongkuo Min (Shanghai Jiao Tong University)

🎯 What it does: Explored an unsupervised weak-to-strong learning framework that trains stronger student models using pseudo-labels generated by weak VQA teachers, achieving video quality assessment without requiring human annotations.

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

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.

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.

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.

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.

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.

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;

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.

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.

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.

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.

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.

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.

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.

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.

GS-CLIP: Zero-shot 3D Anomaly Detection by Geometry-Aware Prompt and Synergistic View Representation Learning

Zehao Deng (Soochow University), Yan Wang (Tsinghua University)

🎯 What it does: Designed the GS-CLIP framework, enhancing CLIP's zero-shot 3D anomaly detection capability through geometric-aware text prompts and synergistic perspective representation learning.

GS^2: Graph-based Spatial Distribution Optimization for Compact 3D Gaussian Splatting

Xianben Yang (Beijing Jiaotong University), Haibin Ling (Westlake University)

🎯 What it does: Propose a graph-based spatial distribution optimization framework GSˆ2, which can significantly compress the number of 3D Gaussian points while improving the quality of novel view synthesis.

GuardTrace-VL: Detecting Unsafe Multimodel Reasoning via Iterative Safety Supervision

Yuxiao Xiang (University of Science and Technology of China), Nenghai Yu (University of Science and Technology of China)

🎯 What it does: Propose GuardTrace-VL, a security auditor specifically designed to monitor unsafe content throughout the entire process (question-thinking-answer) of multimodal reasoning models (MLRM).

GuideFlow: Constraint-Guided Flow Matching for Planning in End-to-End Autonomous Driving

Lin Liu (Beijing Jiaotong University), Yadan Luo (Beijing Academy of Artificial Intelligence)

🎯 What it does: Proposes an end-to-end planning framework called GuideFlow based on flow matching, which can directly incorporate safety and physical constraints during the generation process, addressing the issues of multi-modal mode collapse and insufficient generation safety in traditional imitation learning.

Guiding a Diffusion Model by Swapping Its Tokens

Weijia Zhang (Shanghai Jiao Tong University), Chao Ma (Shanghai Jiao Tong University)

🎯 What it does: Proposed an unconditional guidance method called Self-Swap Guidance (SSG) during diffusion model inference, generating perturbations by swapping the least semantically similar tokens in the model to guide sampling and enhance image quality.

Guiding Diffusion Models with Semantically Degraded Conditions

Shilong Han (National University of Defense Technology), Hongxia Wang (National University of Defense Technology)

🎯 What it does: Propose Condition-Degradation Guidance (CDG), which improves CFG by replacing empty prompts with negative conditions based on semantic degradation.

Guiding Diffusion-based Reconstruction with Contrastive Signals for Balanced Visual Representation

Boyu Han (Institute of Computing Technology, Chinese Academy of Sciences), Qingming Huang (University of Chinese Academy of Sciences)

🎯 What it does: Propose the Diffusion Contrastive Reconstruction (DCR) framework, which integrates contrastive learning with diffusion reconstruction to enhance the discriminative ability (D-Ability) and detail perception ability (P-Ability) of the CLIP visual encoder.

HAD: Heterogeneity-Aware Distillation for Lifelong Heterogeneous Learning

Xuerui Zhang (Southern University of Science and Technology), Yu Zhang (Technical University of Munich)

🎯 What it does: This paper proposes a new scenario called Lifelong Heterogeneous Learning (LHL) and designs a method named Lifelong Heterogeneous Distillation (HAD) for dense prediction tasks, addressing the problem of catastrophic forgetting when continuously learning tasks with different output structures.

Harmonious Parameter Adaptation in Continual Visual Instruction Tuning for Safety-Aligned MLLMs

Ziqi Wang (Hefei University of Technology), Meng Wang (Tsinghua University)

🎯 What it does: Propose a post-training framework HPA for continuous visual instruction fine-tuning (post-SA CVIT) on safe-aligned multi-modal large language models (MLLMs), minimizing task forgetting while ensuring safety.

Harnessing Chain-of-Thought Reasoning in Multimodal Large Language Models for Face Anti-Spoofing

Honglu Zhang (Didi Chuxing), Zhaofeng He (Beijing University of Posts and Telecommunications)

🎯 What it does: Constructed the FaceCoT large-scale chain-of-thought (CoT) visual question answering (VQA) dataset and proposed a CoT-Enhanced Progressive Learning (CEPL) training strategy to achieve high accuracy and interpretability in facial disguise detection.

HATS: Hardness-Aware Trajectory Synthesis for GUI Agents

Rui Shao (Harbin Institute of Technology), Gongwei Chen (Harbin Institute of Technology)

🎯 What it does: Propose a closed-loop trajectory synthesis framework called HATS, designed to generate high-quality, semantically aligned GUI trajectories, aiding in training more robust GUI agents.

Head-wise Adaptive Rotary Positional Encoding for Fine-Grained Image Generation

Jiaye Li (Fudan University), Siyu Zhu (Fudan University)

🎯 What it does: Propose a Head-Adaptive Rotating Positional Encoding (HARoPE), introducing a learnable linear transformation into multi-dimensional Transformer positional encoding to enhance fine-grained image generation and understanding.

HERO: Hierarchical Embedding-Refinement for Open-Vocabulary Temporal Sentence Grounding in Videos

Tingting Han (Hangzhou Dianzi University), Zhou Yu (Hangzhou Dianzi University)

🎯 What it does: This paper proposes the open-vocabulary video sentence grounding (OV-TSGV) task and designs the HERO framework to address this problem; meanwhile, it constructs the first open-vocabulary benchmarks Charades-OV and ActivityNet-OV.

HeSS: Head Sensitivity Score for Sparsity Redistribution in VGGT

Yongsung Kim (Seoul National University), Sungroh Yoon (Seoul National University)

🎯 What it does: Sparse the global attention layers of VGGT by introducing a two-phase sparsification process: first, use Head Sensitivity Score (HeSS) to evaluate the sparsity sensitivity of each attention head, then redistribute the attention budget based on HeSS during inference.

Heuristic-inspired Reasoning Priors Facilitate Data-Efficient Referring Object Detection

Xu Zhang (University of Sydney), Dacheng Tao (Nanyang Technological University)

🎯 What it does: Propose HeROD, which improves referring object detection under data-scarce conditions by injecting spatial and semantic reasoning priors.

Hg-I2P: Bridging Modalities for Generalizable Image-to-Point-Cloud Registration via Heterogeneous Graphs

Pei An (Huazhong University of Science and Technology), Liangliang Nan (Huazhong University of Science and Technology)

🎯 What it does: Propose a heterogeneous graph-based image-to-point cloud registration framework named Hg-I2P.

HG-Lane: High-Fidelity Generation of Lane Scenes under Adverse Weather and Lighting Conditions without Re-annotation

Daichao Zhao (Shanghai Jiao Tong University), Qiankun Li (Shanghai Jiao Tong University)

🎯 What it does: This paper proposes HG-Lane, which synthesizes high-fidelity lane images under adverse weather and lighting conditions using a two-stage ControlNet diffusion framework without re-annotation, and constructs a lane dataset containing 30,000 images across six weather/lighting categories.

Hi-Lo Prune: Look at What You'll Lose before Pruning with Hierarchical Token Selection

Zixun Sun, Yi Yang (State Key Lab Of Brain Machine Intelligence Zhejiang University)

🎯 What it does: Propose a training-agnostic visual token pruning method called Hi-Lo Prune, which enables high-ratio pruning in early layers without fine-tuning the model

HiconAgent: History Context-aware Policy Optimization for GUI Agents

Xurui Zhou (Harbin Institute of Technology), Rui Shao (Shenzhen Loop Area Institute)

🎯 What it does: Propose HiconAgent, which improves the utilization of historical information in GUI agents by using HCPO with dynamic context sampling and Anchor-guided historical compression.

HiDRA: Hierarchical Degradation Representation and Adaptation with Generative Priors for Enhancing Infrared Vision

Zihang Chen (Dalian University of Technology), Risheng Liu (Dalian University of Technology)

🎯 What it does: Propose the HiDRA framework, which divides thermal image enhancement into degradation representation learning and adaptive fine-tuning based on pre-trained diffusion models;