arXivSub Start free trial

ICCV 2025 Papers — Page 9

IEEE/CVF International Conference on Computer Vision · 2701 papers

Exploring Weather-aware Aggregation and Adaptation for Semantic Segmentation under Adverse Conditions

Yuwen Pan (University of Science and Technology of China), Tianzhu Zhang (University of Science and Technology of China)

SegmentationDomain AdaptationKnowledge DistillationPrompt EngineeringImage

🎯 What it does: A weather-aware aggregation and adaptation network (WA2Net) is designed for unsupervised domain adaptation of semantic segmentation under adverse weather conditions.

Expressive Talking Human from Single-Image with Imperfect Priors

Jun Xiang (University of Science and Technology of China), Juyong Zhang (Hong Kong Polytechnic University)

GenerationPose EstimationDiffusion modelGaussian SplattingImageVideoMesh

🎯 What it does: Building an animated full-body talking avatar from a single image

Extending Foundational Monocular Depth Estimators to Fisheye Cameras with Calibration Tokens

Suchisrit Gangopadhyay (Yale University), Alex Wong (Yale University)

Depth EstimationDomain AdaptationTransformerImage

🎯 What it does: This paper proposes a Calibration Tokens mechanism that extends existing foundation monocular depth estimators (FMDE) trained on perspective images to fisheye camera images without retraining or fine-tuning the model.

External Knowledge Injection for CLIP-Based Class-Incremental Learning

Da-Wei Zhou (Nanjing University), De-Chuan Zhan (Nanjing University)

ClassificationRecognitionTransformerLarge Language ModelContrastive LearningImageText

🎯 What it does: The ENGINE method is proposed, which utilizes GPT-4 to extract fine-grained category descriptions and integrates external knowledge into CLIP through a dual-branch injection unit to achieve sample-free memory incremental learning.

Extrapolated Urban View Synthesis Benchmark

Xiangyu Han (New York University), Yiming Li (NVIDIA)

Data SynthesisAutonomous DrivingNeural Radiance FieldGaussian SplattingImageVideoBenchmark

🎯 What it does: The first public EUVS benchmark has been established, evaluating the performance of NVS in extrapolated views using real urban data with multiple traversals, multiple agents, and multiple cameras, and quantitatively comparing it with existing 3D Gaussian scattering and NeRF methods.

EYE3:Turn Anything into Naked-eye 3D

Yingde Song (Beijing University of Posts and Telecommunications), Xunbo Yu (Beijing University of Posts and Telecommunications)

GenerationDepth EstimationDiffusion modelImageVideoTextPoint Cloud

🎯 What it does: This paper proposes the EYE 3 framework, which can automatically convert a single 2D image, video, or text into high-quality naked-eye 3D content required for light field display.

F-Bench: Rethinking Human Preference Evaluation Metrics for Benchmarking Face Generation, Customization, and Restoration

Lu Liu (Shanghai Jiao Tong University), Guangtao Zhai (Bilibili Inc.)

RestorationGenerationLarge Language ModelMixture of ExpertsImageMultimodalityBenchmark

🎯 What it does: An AI-generated facial image quality assessment database, FaceQ, has been constructed for three types of tasks: generation, customization, and restoration. Based on this, a comprehensive human preference benchmark, F-Bench, and a multidimensional evaluation model, F-Eval, have been proposed.

FA: Forced Prompt Learning of Vision-Language Models for Out-of-Distribution Detection

Xinhua Lu (Sun Yat-sen University), Ruixuan Wang (Sun Yat-sen University)

Anomaly DetectionPrompt EngineeringVision Language ModelContrastive LearningImage

🎯 What it does: This paper proposes a Forced Prompt Learning framework based on CLIP to enhance out-of-distribution (OOD) detection performance by learning richer class descriptions under few-shot conditions.

Face Retouching with Diffusion Data Generation and Spectral Restorement

Zhidan Xu (Wenzhou University), Shijian Lu (Nanyang Technological University)

RestorationGenerationData SynthesisTransformerDiffusion modelImageBenchmark

🎯 What it does: This paper proposes a high-resolution facial beautification method and constructs a high-quality facial beautification benchmark dataset of 25,000 pairs;

FaceCraft4D: Animated 3D Facial Avatar Generation from a Single Image

Fei Yin (University of Cambridge), Varun Jampani (Stability AI)

GenerationData SynthesisDiffusion modelGenerative Adversarial NetworkGaussian SplattingImageVideo

🎯 What it does: This paper proposes an end-to-end framework based on multiple priors (geometric, image, video) that can generate high-quality, animatable 4D avatars from a single facial photo.

FaceLift: Learning Generalizable Single Image 3D Face Reconstruction from Synthetic Heads

Weijie Lyu (University of California), Zhixin Shu (Adobe Research)

RestorationGenerationTransformerDiffusion modelImage

🎯 What it does: Proposes the FaceLift framework, which uses a multi-view diffusion model to generate sparse views and then reconstructs a complete 3D Gaussian head representation through a Transformer.

FaceShield: Defending Facial Image against Deepfake Threats

Jaehwan Jeong (Korea University), Sangpil Kim (Korea University)

Adversarial AttackDiffusion modelGenerative Adversarial NetworkImage

🎯 What it does: This paper proposes FaceShield, an adversarial invisible facial image protection method that prevents facial images from being exploited in deepfakes.

FaceXFormer: A Unified Transformer for Facial Analysis

Kartik Narayan (Johns Hopkins University), Vishal M. Patel (Johns Hopkins University)

ClassificationRecognitionSegmentationPose EstimationTransformerImage

🎯 What it does: This paper presents FaceXFormer, a unified Transformer model capable of simultaneously performing ten facial analysis tasks (face segmentation, keypoint detection, pose estimation, attribute recognition, age/gender/race prediction, emotion recognition, face recognition, and visibility prediction) within the same framework.

Factorized Learning for Temporally Grounded Video-Language Models

Wenzheng Zeng (National University of Singapore), Hwee Tou Ng (National University of Singapore)

TransformerVision Language ModelVideoTextBenchmark

🎯 What it does: The D2VLM framework is proposed, which separates the temporal localization of video-language models and text responses into a two-step learning process: first localization, then response.

Failure Cases Are Better Learned But Boundary Says Sorry: Facilitating Smooth Perception Change for Accuracy-Robustness Trade-Off in Adversarial Training

Yanyun Wang (Hong Kong University of Science and Technology), Li Liu (Hong Kong University of Science and Technology)

ClassificationAdversarial AttackConvolutional Neural NetworkContrastive LearningImage

🎯 What it does: A new Robust Perception objective is proposed in adversarial training (AT), and based on this objective, the RPAT method is designed to alleviate the accuracy-robustness trade-off problem in AT.

Fair Generation without Unfair Distortions: Debiasing Text-to-Image Generation with Entanglement-Free Attention

Jeonghoon Park (Korea Advanced Institute of Science and Technology), Jindong Gu (University of Oxford)

GenerationTransformerDiffusion modelImageText

🎯 What it does: To address the bias issue in text-to-image diffusion models, the Entanglement-Free Attention (EFA) method is proposed, achieving a fair distribution of target attributes (such as race and gender) without modifying the original model parameters, while maintaining the integrity of non-target attributes (such as background and scene).

FairGen: Enhancing Fairness in Text-to-Image Diffusion Models via Self-Discovering Latent Directions

Yilei Jiang (Chinese University of Hong Kong), Xiangyu Yue (Chinese University of Hong Kong)

GenerationData SynthesisDiffusion modelImage

🎯 What it does: A lightweight, pluggable framework called 'FairGen' is proposed, which learns attribute latent directions in diffusion models through self-supervised methods, achieving debiasing and distribution control for gender, race, and their intersectional attributes.

FairHuman: Boosting Hand and Face Quality in Human Image Generation with Minimum Potential Delay Fairness in Diffusion Models

Yuxuan Wang (Beijing University of Posts and Telecommunications), Zhanyu Ma (Beijing University of Posts and Telecommunications)

GenerationOptimizationDiffusion modelImage

🎯 What it does: To address the issue of distortion in facial and hand details during portrait image generation, a multi-objective fine-tuning framework called FairHuman is proposed. It constructs global and local (facial, hand) losses and achieves dynamic gradient weight allocation through the Minimum Potential Delay (MPD) fairness principle, enhancing the balance between local detail and overall quality.

FakeRadar: Probing Forgery Outliers to Detect Unknown Deepfake Videos

Zhaolun Li (Guilin University of Electronic Technology), Rushi Lan (Guangxi Key Laboratory of Image and Graphic Intelligent Processing)

Anomaly DetectionTransformerContrastive LearningVideo

🎯 What it does: The FakeRadar framework is proposed for deepfake video detection, emphasizing cross-domain generalization and actively probing potential fakes using large-scale pre-trained models.

FALCON: Resolving Visual Redundancy and Fragmentation in High-resolution Multimodal Large Language Models via Visual Registers

Renshan Zhang (Harbin Institute of Technology), Liqiang Nie (Huawei Noah's Ark Lab)

CompressionRepresentation LearningTransformerLarge Language ModelVision Language ModelImageTextMultimodality

🎯 What it does: To address the issues of visual redundancy and fragmentation in high-resolution multimodal large language models (MLLM), the FALCON model is proposed, utilizing visual register technology to achieve compression and continuity in visual encoding.

Fast Globally Optimal and Geometrically Consistent 3D Shape Matching

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

OptimizationMesh

🎯 What it does: A global optimal geometric consistent 3D shape matching method based on surface loop sets and super product graphs is proposed, solved using integer linear programming, and integer optimal solutions are obtained in experiments.

Fast Image Super-Resolution via Consistency Rectified Flow

Jiaqi Xu (Chinese University of Hong Kong), Pheng-Ann Heng (Chinese University of Hong Kong)

RestorationSuper ResolutionDiffusion modelRectified FlowGenerative Adversarial NetworkImage

🎯 What it does: Proposes FlowSR, a single-step generation method that transforms image super-resolution into rectified flow.

Faster and Better 3D Splatting via Group Training

Chengbo Wang (Hunan University), Yizhen Lao (Hunan University)

OptimizationComputational EfficiencyGaussian SplattingPoint CloudBenchmark

🎯 What it does: Proposes the Group Training method, which accelerates the training of 3D Gaussian Splatting through periodic grouping and caching of Gaussian atoms, while improving rendering quality.

FastJSMA: Accelerating Jacobian-based Saliency Map Attacks through Gradient Decoupling

Zhenghao Gao (Huazhong University of Science and Technology), Changxin Gao (Huazhong University of Science and Technology)

Computational EfficiencyAdversarial AttackImage

🎯 What it does: Proposes the FastJSMA method, reconstructing the Jacobian computation of JSMA as gradient decoupling to achieve faster adversarial attacks.

FastPoint: Accelerating 3D Point Cloud Model Inference via Sample Point Distance Prediction

Donghyun Lee (Seoul National University), Hongil Yoon (Google)

OptimizationComputational EfficiencyPoint Cloud

🎯 What it does: Proposes the FastPoint software acceleration technology, which accelerates FPS and neighborhood search by predicting the distance curve between sampling points, significantly improving the inference speed of 3D point cloud models.

FastVAR: Linear Visual Autoregressive Modeling via Cached Token Pruning

Hang Guo (Tsinghua University), Luca Benini (ETH Zurich)

GenerationComputational EfficiencyTransformerImage

🎯 What it does: This paper proposes FastVAR, a post-training acceleration method that enhances the efficiency of visual autoregressive (VAR) models in high-resolution image generation by caching token pruning.

FB-Diff: Fourier Basis-guided Diffusion for Temporal Interpolation of 4D Medical Imaging

Xin You (Shanghai Jiao Tong University), Nassir Navab (Technical University of Munich)

RestorationGenerationData SynthesisDiffusion modelAuto EncoderImageBiomedical DataMagnetic Resonance ImagingComputed Tomography

🎯 What it does: This paper proposes a diffusion model FB-Diff based on Fourier frequency priors for temporal interpolation of 4D medical images, which can better simulate the nonlinear and quasi-periodic motion during the breathing process.

FDPT: Federated Discrete Prompt Tuning for Black-Box Visual-Language Models

Jiaqi Wu (Tsinghua University), Zijian Tian (China University of Mining and Technology)

Federated LearningExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageMultimodality

🎯 What it does: This paper proposes a Federated Black-box Discrete Prompt Tuning framework (FDPT), which enhances the task performance of visual language models without sharing user data and model parameters.

FE-CLIP: Frequency Enhanced CLIP Model for Zero-Shot Anomaly Detection and Segmentation

Tao Gong (University of Science and Technology of China), Nenghai Yu (University of Science and Technology of China)

SegmentationAnomaly DetectionTransformerVision Language ModelContrastive LearningImageBiomedical DataMagnetic Resonance Imaging

🎯 What it does: This paper proposes a method called FE-CLIP, which injects frequency domain information into the CLIP visual encoder for zero-shot anomaly detection and segmentation.

Feather the Throttle: Revisiting Visual Token Pruning for Vision-Language Model Acceleration

Mark Endo (Stanford University), Serena Yeung-Levy (Stanford University)

OptimizationComputational EfficiencyTransformerLarge Language ModelVision Language ModelImageMultimodality

🎯 What it does: This paper studies the acceleration of inference in visual language models by pre-pruning visual tokens and analyzes the limitations of existing methods such as FastV, proposing the FEATHER scheme.

Feature Coding in the Era of Large Models: Dataset, Test Conditions, and Benchmark

Changsheng Gao (Nanyang Technological University), Weisi Lin (Nanyang Technological University)

ClassificationSegmentationDepth EstimationCompressionTransformerAuto EncoderImageTextMultimodalityBenchmark

🎯 What it does: A feature dataset was constructed that includes three major types of large models (DINOv2, Llama3, SD3) across five tasks (image classification, semantic segmentation, depth estimation, common sense reasoning, text-to-image synthesis), and a unified testing condition, bit rate metric (BPFP), and benchmark evaluation process were proposed; two baseline encoders (VTM and Hyperprior) were also provided, and benchmark experiments were completed.

Feature Decomposition-Recomposition in Large Vision-Language Model for Few-Shot Class-Incremental Learning

Zongyao Xue (Institute of Computing Technology, Chinese Academy of Sciences), Xilin Chen (Institute of Computing Technology, Chinese Academy of Sciences)

ClassificationRecognitionTransformerVision Language ModelContrastive LearningImage

🎯 What it does: A CLIP-based Semantic Feature Decomposition-Recombination (FDR) method is proposed for few-shot class incremental learning, addressing the issues of new class prototype bias and catastrophic forgetting.

Feature Extraction and Representation of Pre-training Point Cloud Based on Diffusion Models

Chang Qiu (Southeast University), Zilei Zhang (Southeast University)

ClassificationSegmentationRepresentation LearningTransformerDiffusion modelPoint Cloud

🎯 What it does: Proposes the PreDifPoint pre-training framework, which utilizes diffusion models for self-supervised pre-training of point clouds, generating high-quality point clouds and significantly improving the performance of downstream tasks such as classification and segmentation.

Feature Purification Matters: Suppressing Outlier Propagation for Training-Free Open-Vocabulary Semantic Segmentation

Shuo Jin (Xi'an Jiaotong-Liverpool University), Jimin Xiao (Xi'an Jiaotong-Liverpool University)

SegmentationTransformerVision Language ModelImage

🎯 What it does: This paper proposes a training-free open vocabulary semantic segmentation framework SFP, aimed at suppressing outlier propagation in CLIP attention and enhancing semantic feature representation.

FED-PsyAU: Privacy-Preserving Micro-Expression Recognition via Psychological AU Coordination and Dynamic Facial Motion Modeling

Jingting Li (Chinese Academy of Sciences), Su-Jing Wang (Chinese Academy of Sciences)

RecognitionFederated LearningSafty and PrivacyGraph Neural NetworkTransformerOptical FlowImageVideo

🎯 What it does: A privacy-preserving micro-expression recognition framework (FED-PsyAU) that integrates psychological priors, dynamic AU relationships, and federated learning has been designed and implemented.

FedAGC: Federated Continual Learning with Asymmetric Gradient Correction

Chengchao Zhang (Tianjin University), Wei Feng (Tianjin University)

Federated LearningImage

🎯 What it does: This paper proposes FedAGC, a framework that integrates Federated Learning and Continual Learning, addressing the space-time catastrophic forgetting problem in federated continual learning through anti-gradient correction, centroid-based representative sample selection, and group-level personalized aggregation.

FedDifRC: Unlocking the Potential of Text-to-Image Diffusion Models in Heterogeneous Federated Learning

Huan Wang (University of Wollongong), Jun Shen (University of Wollongong)

GenerationFederated LearningDiffusion modelContrastive LearningImage

🎯 What it does: By combining diffusion models with federated learning, the FedDifRC framework is proposed, utilizing Text-Driven Diffusion Contrastive Learning (TDCL) and Noise-Driven Diffusion Consistency Regularization (NDCR) to address the issue of data heterogeneity.

Federated Continual Instruction Tuning

Haiyang Guo (University of Chinese Academy of Sciences), Cheng-Lin Liu (University of Chinese Academy of Sciences)

Federated LearningLarge Language ModelTextMultimodalityBenchmark

🎯 What it does: Proposed the FCIT benchmark and designed the DISCO framework, integrating federated learning with continuous instruction tuning to address data heterogeneity and catastrophic forgetting in LMM during distributed continuous learning.

Federated Continuous Category Discovery and Learning

Lixu Wang (Nanyang Technological University), Wei Dong (Nanyang Technological University)

Federated LearningConvolutional Neural NetworkContrastive LearningImage

🎯 What it does: This paper proposes the Federated Continuous Category Discovery and Learning (FC DL 2) framework, which can continuously discover and learn newly emerging categories in a federated learning environment while retaining the performance of learned categories. The core method is Global Prototype Alignment (GPA), which includes generating potential prototypes through local K-means, estimating the number of new categories and obtaining global prototypes on the server side using PPM (Prototype Merge), using Semantic Weighted Loss (SWL) during local training, and jointly optimizing with Contrastive Weighted Loss (CWL) and data/model mixup.

Federated Domain Generalization with Domain-specific Soft Prompts Generation

Jianhan Wu (Ping An Technology), Jianzong Wang (Ping An Technology)

Domain AdaptationFederated LearningTransformerPrompt EngineeringGenerative Adversarial NetworkImage

🎯 What it does: In a federated learning environment, addressing the distribution differences between the source domain and the target domain, domain-specific soft prompts are learned on the CLIP model, and generative models are utilized to generate prompts for unknown domains to enhance the model's generalization performance in the target domain.

Federated Prompt-Tuning with Heterogeneous and Incomplete Multimodal Client Data

Thu Hang Phung (Hanoi University of Science and Technology), Phi Le Nguyen (Hanoi University of Science and Technology)

Federated LearningTransformerPrompt EngineeringImageTextMultimodality

🎯 What it does: A federated prompt-tuning framework (FED-PRIME) is proposed for multimodal heterogeneous data with missing features, capable of simultaneously learning and aggregating two types of prompts on different clients: inter-client prompts for missing patterns and intra-client prompts for missing modalities.

Federated Representation Angle Learning

Liping Yi (Nankai University), Xiaoxiao Li (University of British Columbia)

Federated LearningComputational EfficiencyRepresentation LearningImage

🎯 What it does: This study proposes a model heterogeneous federated learning framework called FedRAL, which enhances model generalization and communication/computation efficiency through representation angle learning, adaptive diagonal sparsification, and element-weighted aggregation mechanisms.

FedMeNF: Privacy-Preserving Federated Meta-Learning for Neural Fields

Junhyeog Yun (Seoul National University), Gunhee Kim (Seoul National University)

Federated LearningSafty and PrivacyMeta LearningNeural Radiance FieldImageVideoMultimodality

🎯 What it does: Proposes FedMeNF, a privacy-preserving neural field training framework based on federated meta-learning.

FedMVP: Federated Multimodal Visual Prompt Tuning for Vision-Language Models

Mainak Singha, Elisa Ricci

Federated LearningTransformerPrompt EngineeringVision Language ModelImageMultimodality

🎯 What it does: This paper proposes FedMVP, a federated multimodal visual prompt tuning framework that utilizes PromptFormer to generate dynamic multimodal prompts through visual features and textual attributes, allowing federated learning to be completed by training only lightweight modules.

FedPall: Prototype-based Adversarial and Collaborative Learning for Federated Learning with Feature Drift

Yong Zhang (Shenzhen MSU-BIT University), Xiping Hu (Shenzhen MSU-BIT University)

Federated LearningAdversarial AttackConvolutional Neural NetworkContrastive LearningImage

🎯 What it does: This paper proposes the FedPall framework, which utilizes prototype adversarial learning and collaborative learning to unify the feature space across clients, addressing the issue of feature drift in federated learning and constructing a global classifier on the server side.

FedVLA: Federated Vision-Language-Action Learning with Dual Gating Mixture-of-Experts for Robotic Manipulation

Cui Miao (National University of Defense Technology), Xiaodong Wang (National University of Defense Technology)

Federated LearningRobotic IntelligenceTransformerMixture of ExpertsVision-Language-Action ModelImageMultimodality

🎯 What it does: This paper proposes FedVLA, a vision-language-action (VLA) model based on distributed federated learning, which can understand and execute robotic operation instructions without sharing raw data.

FedWSQ: Efficient Federated Learning with Weight Standardization and Distribution-Aware Non-Uniform Quantization

Seung-Wook Kim (Pukyong National University), Se-Ho Lee (Jeonbuk National University)

Federated LearningComputational EfficiencyImage

🎯 What it does: A new federated learning framework, FedWSQ, has been developed, combining weight normalization and distribution-aware non-uniform quantization to enhance model convergence and communication efficiency.

FedXDS: Leveraging Model Attribution Methods to counteract Data Heterogeneity in Federated Learning

Maximilian Andreas Hoefler (Fraunhofer Heinrich Hertz Institute), Wojciech Samek (Fraunhofer Heinrich Hertz Institute)

Federated LearningSafty and PrivacyExplainability and InterpretabilityImage

🎯 What it does: FedXDS proposes a federated learning framework that utilizes XAI feature attribution for interpretable feature selection and sharing to alleviate data heterogeneity.

Feed-Forward SceneDINO for Unsupervised Semantic Scene Completion

Aleksandar Jevtić, Daniel Cremers

SegmentationDomain AdaptationAutonomous DrivingTransformerContrastive LearningImage

🎯 What it does: This paper proposes SceneDINO, which utilizes a single image to simultaneously predict the complete 3D scene geometry and high-dimensional feature fields during forward inference, achieving unsupervised semantic scene completion through 3D feature distillation.

FEVER-OOD: Free Energy Vulnerability Elimination for Robust Out-of-Distribution Detection

Brian K.S. Isaac-Medina (Durham University), Toby P. Breckon (Durham University)

Object DetectionAnomaly DetectionGenerative Adversarial NetworkImage

🎯 What it does: This paper proposes FEVER-OOD, a complete set of improvement strategies designed to address vulnerabilities in free energy-based OOD detection methods.

Few-Shot Image Quality Assessment via Adaptation of Vision-Language Models

Xudong Li (Xiamen University), Rongrong Ji (Xiamen University)

Meta LearningTransformerPrompt EngineeringVision Language ModelImage

🎯 What it does: A GRMP-IQA framework based on CLIP for meta-learning soft prompt initialization and gradient regularization is proposed for few-shot blind image quality assessment.

Few-Shot Pattern Detection via Template Matching and Regression

Eunchan Jo (Pohang University of Science and Technology), Minsu Cho (Pohang University of Science and Technology)

RecognitionObject DetectionTransformerImage

🎯 What it does: A few-shot pattern detection method based on template matching and regression (TMR) is proposed, and a new dataset RPINE containing various non-object repetitive patterns is constructed.

Fewer Denoising Steps or Cheaper Per-Step Inference: Towards Compute-Optimal Diffusion Model Deployment

Zhenbang Du (Georgia Institute of Technology), Yingyan Celine Lin

CompressionOptimizationComputational EfficiencyDiffusion modelImage

🎯 What it does: This paper studies the effectiveness of reducing the number of denoising steps and the inference cost per step in a post-training deployment environment without fine-tuning, and proposes the PostDiff framework that achieves training-free compression through mixed-resolution denoising and mixed module caching.

FG-OrIU: Towards Better Forgetting via Feature-Gradient Orthogonality for Incremental Unlearning

Qian Feng (Zhejiang University), Hui Qian (Zhejiang University)

ClassificationRecognitionImage

🎯 What it does: This paper proposes an incremental forgetting framework called FG-OrIU, based on orthogonality of features and gradients, to achieve irreversible deep forgetting on pre-trained models.

FICGen: Frequency-Inspired Contextual Disentanglement for Layout-driven Degraded Image Generation

Wenzhuang Wang (Beihang University), Jia Li (Beihang University)

RestorationGenerationDiffusion modelImage

🎯 What it does: This paper proposes a layout-driven image generation method for degradation scenarios called FICGen, which utilizes frequency information to achieve separation and reconstruction of instances and backgrounds.

FiffDepth: Feed-forward Transformation of Diffusion-Based Generators for Detailed Depth Estimation

Yunpeng Bai (University of Texas at Austin), Qixing Huang (University of Texas at Austin)

Depth EstimationDiffusion modelImage

🎯 What it does: By transforming the diffusion model into a feedforward network and combining DINOv2 pseudo-labels, FiffDepth is proposed to achieve high-fidelity, low-cost monocular depth estimation.

Find a Scapegoat: Poisoning Membership Inference Attack and Defense to Federated Learning

Wenjin Mo (Yale University), Mingwei Fang (Guangdong Polytechnic Normal University)

Federated LearningSafty and PrivacyAdversarial AttackImage

🎯 What it does: This paper proposes a poisoning-based membership inference attack (FedPoisonMIA) in federated learning achieved through gradient manipulation, as well as an angle-trimmed robust aggregation method (ATM) to counter this attack.

Find Any Part in 3D

Ziqi Ma (California Institute of Technology), Georgia Gkioxari (California Institute of Technology)

RecognitionSegmentationTransformerContrastive LearningPoint Cloud

🎯 What it does: Developed the FIND3D model to achieve open-source 3D component segmentation under any text query;

FIND: Few-Shot Anomaly Inspection with Normal-Only Multi-Modal Data

Yiting Li (Institute for Infocomm Research), Xulei Yang (Institute for Infocomm Research)

Anomaly DetectionKnowledge DistillationTransformerMultimodalityPoint Cloud

🎯 What it does: A dual-student framework called FIND is proposed for few-shot multimodal anomaly detection, achieving fine-grained localization by combining reverse distillation and cross-modal feature mapping.

Fine-Grained 3D Gaussian Head Avatars Modeling from Static Captures via Joint Reconstruction and Registration

Yuan Sun (Xi'an Jiaotong University), Fei Wang (Xi'an Jiaotong University)

GenerationData SynthesisGaussian SplattingImage

🎯 What it does: Using a small number of static images, combined with joint reconstruction and registration of prior and non-prior models, we generate high-resolution, fine-grained 3D Gaussian head avatars and achieve animation through binding information.

Fine-grained Abnormality Prompt Learning for Zero-shot Anomaly Detection

Jiawen Zhu (Singapore Management University), Guansong Pang (Singapore Management University)

Anomaly DetectionTransformerPrompt EngineeringVision Language ModelImageBiomedical DataMagnetic Resonance ImagingComputed Tomography

🎯 What it does: A zero-shot anomaly detection framework based on fine-grained anomaly prompt learning, FAPrompt, is proposed to enhance the detection and localization capabilities for diverse anomalies.

Fine-Grained Evaluation of Large Vision-Language Models in Autonomous Driving

Yue Li (University of Science and Technology of China), Xinhai Zhao (Huawei Noah's Ark Lab)

Autonomous DrivingTransformerVision Language ModelMultimodalityBenchmark

🎯 What it does: A fine-grained evaluation benchmark for autonomous driving scenarios, VLADBench, is proposed to conduct detailed assessments of large visual-language models (VLMs) across five key areas (traffic knowledge understanding, general element recognition, traffic map generation, target attribute understanding, and owner decision planning), and to enhance model performance in each area through training on domain-specific data.

Fine-grained Spatiotemporal Grounding on Egocentric Videos

Shuo Liang (Chinese University of Hong Kong), Liwei Wang (Chinese University of Hong Kong)

Object DetectionSegmentationLarge Language ModelSupervised Fine-TuningVideoTextBenchmark

🎯 What it does: The first pixel-level head perspective video localization benchmark EgoMask and its large-scale training set EgoMask-Train have been constructed, and a systematic evaluation of existing spatiotemporal localization models has been conducted.

Fine-structure Preserved Real-world Image Super-resolution via Transfer VAE Training

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

RestorationSuper ResolutionDiffusion modelAuto EncoderImage

🎯 What it does: Proposes the Transfer VAE Training (TVT) method, which migrates 8×VAE to 4×VAE and aligns with the Stable Diffusion pre-trained UNet to enhance detail preservation in Real-ISR;

Fine-Tuning Visual Autogressive Models for Subject-Driven Generation

Jiwoo Chung (Sungkyunkwan University), Jae-Pil Heo (Sungkyunkwan University)

GenerationData SynthesisComputational EfficiencyKnowledge DistillationTransformerSupervised Fine-TuningDiffusion modelImage

🎯 What it does: A topic-driven generation method based on a large-scale Visual Autoregressive (VAR) model is proposed, along with three techniques: selective layer fine-tuning, scale-weighted fine-tuning, and prior distillation.

FineMotion: A Dataset and Benchmark with both Spatial and Temporal Annotation for Fine-grained Motion Generation and Editing

Bizhu Wu (University of Nottingham), Linlin Shen (Shenzhen University)

GenerationRetrievalTransformerLarge Language ModelSupervised Fine-TuningDiffusion modelVideoTextMultimodalityBenchmark

🎯 What it does: The FineMotion dataset is proposed, along with an automated text annotation pipeline that enhances text-driven action generation and editing through fine-grained descriptions of body part movements.

FinMMR: Make Financial Numerical Reasoning More Multimodal, Comprehensive, and Challenging

Zichen Tang (Beijing University of Posts and Telecommunications), Shiyao Peng (Beijing University of Posts and Telecommunications)

Recommendation SystemAnomaly DetectionOptimizationData-Centric LearningTransformerLarge Language ModelPrompt EngineeringImageTextMultimodalityBenchmarkFinance RelatedChain-of-Thought

🎯 What it does: Proposed the FinMMR financial multimodal numerical reasoning benchmark, which includes three major challenges: images, text, and numerical computation;

Fish2Mesh Transformer: 3D Human Mesh Recovery from Egocentric Vision

Tianma Shen (Santa Clara University), David C. Jeong (Santa Clara University)

Pose EstimationTransformerImageMesh

🎯 What it does: This paper proposes a Transformer-based Fish2Mesh model for recovering 3D human meshes from the first-person perspective of a head-mounted fisheye camera.

FiVE-Bench: A Fine-grained Video Editing Benchmark for Evaluating Emerging Diffusion and Rectified Flow Models

Minghan Li (Harvard University), Mengyu Wang (Harvard University)

GenerationData SynthesisVision Language ModelDiffusion modelRectified FlowVideoBenchmark

🎯 What it does: A fine-grained video editing benchmark FiVE-Bench and evaluation metric FiVE-Acc are proposed, along with the development of two editing methods, Pyramid-Edit and Wan-Edit, based on Rectified Flow, which are non-reversible and non-trained.

Fix-CLIP: Dual-Branch Hierarchical Contrastive Learning via Synthetic Captions for Better Understanding of Long Text

Bingchao Wang (Shanghai Jiao Tong University), Wei Liu (Shanghai Jiao Tong University)

GenerationRetrievalTransformerLarge Language ModelContrastive LearningImageTextMultimodality

🎯 What it does: Through dual-branch training, short texts are aligned with masked images, and long texts are aligned with original images. Learnable region prompts and unidirectional masks are designed to extract local information, and a hierarchical feature alignment module is added to enhance the understanding of long texts while retaining the performance of short texts based on CLIP.

FixTalk: Taming Identity Leakage for High-Quality Talking Head Generation in Extreme Cases

Shuai Tan (Shanghai Jiao Tong University), Ye Pan (Shanghai Jiao Tong University)

GenerationGenerative Adversarial NetworkVideo

🎯 What it does: Proposes the FixTalk framework to address identity leakage and rendering distortion issues in GAN-based speaker avatar generation, achieving high-quality, real-time, and separable control of speaker animations.

Flash-VStream: Efficient Real-Time Understanding for Long Video Streams

Haoji Zhang (Tsinghua University), Xiaojie Jin (Beijing Jiaotong University)

RecognitionComputational EfficiencyTransformerLarge Language ModelVision Language ModelVideoMultimodality

🎯 What it does: This paper proposes Flash-VStream, an efficient real-time long video understanding model that utilizes a two-process asynchronous framework and Flash Memory for online processing and instant responses to long video streams.

FlashDepth: Real-time Streaming Video Depth Estimation at 2K Resolution

Gene Chou (Netflix Eyeline Studios), Paul Debevec (Cornell University)

Depth EstimationRecurrent Neural NetworkTransformerVideo

🎯 What it does: Achieved real-time streaming video depth estimation at 2K resolution and 24 FPS.

FlexGen: Flexible Multi-View Generation from Text and Image Inputs

Xinli Xu (Hong Kong University of Science and Technology), Ying-Cong Chen (Hong Kong University of Science and Technology)

GenerationData SynthesisTransformerLarge Language ModelDiffusion modelImageTextMultimodality

🎯 What it does: This paper presents FlexGen, a framework that generates controllably consistent multi-view images based on single-view images, text, or a combination of both inputs.

Flexi-FSCIL: Adaptive Knowledge Retention for Breaking the Stability-Plasticity Dilemma in Few-Shot Class-Incremental Learning

Wufei Xie (Central South University), Xue Yang (Shanghai Jiao Tong University)

Knowledge DistillationConvolutional Neural NetworkContrastive LearningImage

🎯 What it does: Proposes the Flexi-FSCIL framework to address the stability-plasticity dilemma in few-shot incremental learning;

FLOAT: Generative Motion Latent Flow Matching for Audio-driven Talking Portrait

Taekyung Ki (Korea Advanced Institute of Science and Technology), Gyeongsu Chae (DeepBrain AI Inc)

GenerationData SynthesisTransformerFlow-based ModelVideoAudio

🎯 What it does: FLOAT is proposed, which utilizes flow matching to generate audio-driven speaker videos, capable of producing natural and expressive continuous videos from a single image and audio.

FLOSS: Free Lunch in Open-vocabulary Semantic Segmentation

Yasser Benigmim (Inria), Raoul de Charette (Inria)

SegmentationVision Language ModelImage

🎯 What it does: A text template selection method called FLOSS is proposed, which does not require training or labeling, to improve the Open Vocabulary Semantic Segmentation (OVSS) model.

Flow Stochastic Segmentation Networks

Fabio De Sousa Ribeiro (Imperial College London), Ben Glocker (Imperial College London)

SegmentationGenerationDiffusion modelFlow-based ModelImageBiomedical DataMagnetic Resonance Imaging

🎯 What it does: A new generative segmentation model called Flow-SSN is proposed, which can learn high-order pixel correlations and sample efficiently;

Flow to the Mode: Mode-Seeking Diffusion Autoencoders for State-of-the-Art Image Tokenization

Kyle Sargent (Stanford University), Jiajun Wu (Stanford University)

GenerationCompressionTransformerDiffusion modelRectified FlowAuto EncoderImage

🎯 What it does: FlowMo is proposed, a diffusion autoencoder based on Transformer for image discretization.

Flow-MIL: Constructing Highly-expressive Latent Feature Space For Whole Slide Image Classification Using Normalizing Flow

Yingfan Ma (Fudan University), Manning Wang (Fudan University)

ClassificationFlow-based ModelImage

🎯 What it does: This paper proposes a multi-instance learning framework called Flow-MIL based on regularized flows, which utilizes flow networks to map the features of pathological image patches into a highly expressive latent semantic space, and generates instance-level pseudo-labels through Gaussian mixture prototypes, achieving dual classification at both the sliding window level and instance level for whole slide images (WSI).

Flow4Agent: Long-form Video Understanding via Motion Prior from Optical Flow

Ruyang Liu (Peking University), Ge Li (Peking University)

RecognitionRetrievalOptimizationTransformerLarge Language ModelOptical FlowVideoMultimodality

🎯 What it does: This paper proposes Flow4Agent, a framework that utilizes optical flow priors to extract key content from long videos and assist large language models in understanding long videos.

FlowChef: Steering of Rectified Flow Models for Controlled Generations

Maitreya Patel (Arizona State University), Yezhou Yang (Arizona State University)

GenerationFlow-based ModelRectified FlowImageBenchmarkOrdinary Differential Equation

🎯 What it does: We propose FlowChef, a scheme for controlling Rectified Flow Models (RFMs) during inference that is gradient-independent and does not require backpropagation, enabling efficient completion of tasks such as linear inverse problems, image editing, and classifier guidance.

FlowDPS : Flow-Driven Posterior Sampling for Inverse Problems

Jeongsol Kim (Korea Advanced Institute of Science and Technology), Jong Chul Ye (Korea Advanced Institute of Science and Technology)

RestorationSuper ResolutionTransformerDiffusion modelFlow-based ModelImageOrdinary Differential Equation

🎯 What it does: This paper proposes the FlowDPS (Flow-Driven Posterior Sampling) method, which decomposes the flow ODE into image purification and noise estimation components within the flow model framework, and injects the likelihood gradient into the purification component to achieve posterior sampling without additional training, thereby addressing various linear inverse problems.

FlowEdit: Inversion-Free Text-Based Editing Using Pre-Trained Flow Models

Vladimir Kulikov (Technion - Israel Institute of Technology), Tomer Michaeli (Technion - Israel Institute of Technology)

Image TranslationData SynthesisFlow-based ModelImageTextOrdinary Differential Equation

🎯 What it does: A model-agnostic text editing method based on pre-trained flow models, called FlowEdit, is proposed without inversion, optimization, or model dependency.

FlowR: Flowing from Sparse to Dense 3D Reconstructions

Tobias Fischer (ETH Zurich), Peter Kontschieder (Meta Reality Labs)

GenerationData SynthesisDepth EstimationTransformerDiffusion modelFlow-based ModelAuto EncoderGaussian SplattingSimultaneous Localization and MappingImage

🎯 What it does: The FlowR method is proposed, which uses flow matching technology to directly map the views rendered from sparse reconstruction to the target distribution of dense reconstruction, thereby improving the quality of new perspective synthesis in 3D Gaussian splatting.

FlowSeek: Optical Flow Made Easier with Depth Foundation Models and Motion Bases

Matteo Poggi (University of Bologna), Fabio Tosi (University of Bologna)

Depth EstimationOptical FlowImageVideo

🎯 What it does: This paper presents FlowSeek, a lightweight optical flow framework that combines a single-image deep pre-trained model with motion bases, capable of being trained using only a single consumer-grade GPU while maintaining high accuracy across multi-domain scenarios.

FlowStyler: Artistic Video Stylization via Transformation Fields Transports

Yuning Gong (Sichuan University), Yanci Zhang (Sichuan University)

Image TranslationOptical FlowVideo

🎯 What it does: We propose FlowStyler, a physics-driven non-generative video style transfer framework that achieves temporally consistent and artistic video stylization by utilizing a geometric style velocity field and an orthogonal constrained color transformation field.

FlowTok: Flowing Seamlessly Across Text and Image Tokens

Ju He (ByteDance), Liang-Chieh Chen (Johns Hopkins University)

GenerationData SynthesisTransformerFlow-based ModelContrastive LearningImageTextMultimodality

🎯 What it does: This paper proposes the FlowTok framework, which enables the direct flow of text and images in a unified 1D token low-dimensional space, allowing for simultaneous generation from text to image and from image to text.

FLSeg: Enhancing Privacy and Robustness in Federated Learning under Heterogeneous Data via Model Segmentation

Zichun Su (Huazhong University of Science and Technology), Songfeng Lu (Shenzhen Huazhong University of Science and Technology Research Institute)

Federated LearningSafty and PrivacyDiffusion modelImage

🎯 What it does: The FLSeg framework is proposed, utilizing model segment exchange and aggregation to achieve privacy protection and robustness in federated learning;

Focal Plane Visual Feature Generation and Matching on a Pixel Processor Array

Hongyi Zhang (University of Bristol), Walterio Mayol-Cuevas (Amazon)

Object DetectionAutonomous DrivingComputational EfficiencySimultaneous Localization and MappingImage

🎯 What it does: The generation, storage, and matching of binary descriptors are implemented on a pixel processing array (PPA), allowing feature matching to be completed without leaving the sensor.

FOLDER: Accelerating Multi-Modal Large Language Models with Enhanced Performance

Haicheng Wang (Shanghai Jiao Tong University), Enzo Tartaglione (Telecom Paris)

CompressionComputational EfficiencyTransformerLarge Language ModelVision Language ModelImageMultimodality

🎯 What it does: The FOLDER module is proposed to achieve pluggable visual token compression for visual encoders in multimodal large language models, significantly reducing the number of visual tokens while maintaining or enhancing performance.

FontAnimate: High Quality Few-shot Font Generation via Animating Font Transfer Process

Bin Fu (Shenzhen Institutes of Advanced Technology), Peng Gao (Shenzhen Institutes of Advanced Technology)

GenerationData SynthesisTransformerDiffusion modelVideo

🎯 What it does: This paper proposes a video generation framework based on diffusion models called FontAnimate, aimed at few-shot font generation tasks.

FonTS: Text Rendering With Typography and Style Controls

Wenda Shi (Hong Kong Polytechnic University), Xingxing Zou (Hong Kong Polytechnic University)

GenerationData SynthesisTransformerSupervised Fine-TuningDiffusion modelImageText

🎯 What it does: A two-stage Diffusion Transformer (DiT) based text rendering framework is proposed, capable of achieving typographic control at the word level (font, bold, italic, underline) while maintaining the accuracy of the text content and achieving artistic style control.

ForCenNet: Foreground-Centric Network for Document Image Rectification

Peng Cai (Qihoo Technology), Jiankang Deng (Imperial College London)

RestorationSegmentationTransformerSupervised Fine-TuningImage

🎯 What it does: This paper proposes ForCenNet, a method for geometric correction of document images that generates labels using foreground elements and guides attention and curvature consistency loss through a foreground mask.

Forecasting Continuous Non-Conservative Dynamical Systems in SO(3)

Lennart Bastian (Technical University of Munich), Tolga Birdal (Imperial College London)

Pose EstimationRobotic IntelligenceRecurrent Neural NetworkTime SeriesSequentialStochastic Differential EquationOrdinary Differential Equation

🎯 What it does: A neural control differential equation based on Savitzky-Golay filtering control paths (SG-nCDE) is proposed for long-term prediction of rotational trajectories on SO(3), capable of achieving robust attitude inference and compensation under conditions of unknown inertia, external forces, and noisy observations.

Forensic-MoE: Exploring Comprehensive Synthetic Image Detection Traces with Mixture of Experts

Mingqi Fang (University of Science and Technology of China), Yongdong Zhang (University of Science and Technology of China)

Data SynthesisAnomaly DetectionKnowledge DistillationTransformerMixture of ExpertsGenerative Adversarial NetworkImage

🎯 What it does: This paper proposes Forensic-MoE, a synthetic image detection framework based on Mixture of Experts, which integrates multiple experts through an Adapter-Backbone structure to capture multi-perspective detection traces.

Foresight in Motion: Reinforcing Trajectory Prediction with Reward Heuristics

Muleilan Pei (Hong Kong University of Science and Technology), Shaojie Shen (Hong Kong University of Science and Technology)

Autonomous DrivingTransformerReinforcement LearningMultimodalityTime Series

🎯 What it does: A 'reasoning first, prediction later' trajectory prediction framework is proposed, combining reward-driven intent reasoning with multimodal trajectory generation.

ForeSight: Multi-View Streaming Joint Object Detection and Trajectory Forecasting

Sandro Papais (University of Toronto), Steven L. Waslander (University of Toronto)

Object DetectionObject TrackingAutonomous DrivingTransformerImageMultimodality

🎯 What it does: ForeSight is a joint detection and trajectory prediction framework based on multi-view cameras, achieving bidirectional streaming learning of detection and prediction through shared query memory.

ForestFormer3D: A Unified Framework for End-to-End Segmentation of Forest LiDAR 3D Point Clouds

Binbin Xiang (Norwegian Institute of Bioeconomy Research), Rasmus Astrup (Norwegian Institute of Bioeconomy Research)

SegmentationTransformerPoint Cloud

🎯 What it does: Proposes ForestFormer3D, a unified end-to-end framework for individual tree and semantic segmentation of forest LiDAR 3D point clouds;

ForgeLens: Data-Efficient Forgery Focus for Generalizable Forgery Image Detection

Yingjian Chen (Henan University), Yakun Niu (Henan University)

ClassificationAnomaly DetectionTransformerGenerative Adversarial NetworkImage

🎯 What it does: Proposes the ForgeLens framework, which utilizes a frozen CLIP-ViT along with a lightweight Weight-Shared Guidance Module (WSGM) and a multi-stage feature fusion module FAFormer for efficient and general detection of forged images.