CVPR 2026 Papers — Page 13
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 4071 papers
FFP-300K: Scaling First-Frame Propagation for Generalizable Video Editing
Xijie Huang (Fudan University), Yanwei Fu (Tencent YouTu Lab)
GenerationTransformerDiffusion modelVideo
🎯 What it does: Proposed an unguided video editing framework named FreeProp for First Frame Propagation (FFP), and constructed a large-scale high-quality dataset named FFP-300K
FG-Portrait: 3D Flow Guided Editable Portrait Animation
Yating Xu (National University of Singapore), Jifei Song (University of Surrey)
GenerationTransformerDiffusion modelFlow-based ModelImageVideoMesh
🎯 What it does: This paper proposes using 3D flow as a new motion guidance, combined with diffusion models to achieve facial animation, supporting user editing of expressions and head poses during the inference stage.
FHAvatar: Fast and High-Fidelity Reconstruction of Face-and-Hair Composable 3D Head Avatar from Few Casual Captures
Yujie Sun (Shanghai Jiao Tong University), Fan Wu (Shanghai Jiao Tong University)
GenerationTransformerGaussian SplattingImageMesh
🎯 What it does: Proposed an FHAvatar framework capable of rapidly generating high-fidelity, animatable 3D avatar models from a small number of phone-captured images, modeling the face and hair as separately composable Gaussian clouds.
FiDeSR: High-Fidelity and Detail-Preserving One-Step Diffusion Super-Resolution
Aro Kim (Kyungpook National University), Sang-hyo Park (Kyungpook National University)
Super ResolutionSupervised Fine-TuningDiffusion modelImage
🎯 What it does: Proposes FiDeSR—a one-step diffusion super-resolution model that addresses the challenges of high-fidelity reconstruction and detail recovery.
Fighting Hallucinations with Counterfactuals: Diffusion-Guided Perturbations for LVLM Hallucination Suppression
Hamidreza Dastmalchi (York University), Hamed Barzamini (Northern Illinois University)
Explainability and InterpretabilityAdversarial AttackLarge Language ModelVision Language ModelDiffusion modelMultimodality
🎯 What it does: Proposes CIPHER, a technique that eliminates visually induced hallucinations without requiring training during inference by generating adversarial 'forged' images through diffusion models on images.
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)
OptimizationComputational EfficiencyGaussian Splatting
🎯 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.
FILTR: Extracting Topological Features from Pretrained 3D Models
Louis Martinez (Institut Polytechnique de Paris), Maks Ovsjanikov (Institut Polytechnique de Paris)
Data SynthesisRepresentation LearningTransformerPoint CloudMesh
🎯 What it does: Investigate whether pre-trained 3D point cloud Transformer encoders can capture topological information, propose a synthetic dataset DONUT with topological labels, and design the FILTR model to directly predict persistence diagrams from frozen encoder features.
Finding Distributed Object-Centric Properties in Self-Supervised Transformers
Samyak Rawlekar (University of Illinois Urbana Champaign), Narendra Ahuja (University of Illinois Urbana Champaign)
Object DetectionRepresentation LearningTransformerVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: Analyze and mine object-centric information in self-supervised vision Transformers, and propose the training-free Object-DINO method to automatically identify object-centric attention heads distributed across layers, enhancing unsupervised object discovery and visual guidance effectiveness in multi-modal large language models.
Fine-Grained GRPO for Precise Preference Alignment in Flow Models
Yujie Zhou (Shanghai Jiao Tong University), Guangtao Zhai (Shanghai Jiao Tong University)
GenerationReinforcement Learning from Human FeedbackReinforcement LearningFlow-based ModelImageTextMultimodalityStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: Propose the Granular-GRPO (G RPO) framework, which aligns streaming generative models with human preferences using online reinforcement learning.
Fine-grained Image Aesthetic Assessment: Learning Discriminative Scores from Relative Ranks
Zhichao Yang (Xidian University), Leida Li (Xidian University)
ClassificationRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningImageText
🎯 What it does: Propose the fine-grained image aesthetics assessment (FG-IAA) task, construct the FGAesthetics large-scale similar image series dataset, and design the FGAesQ model to achieve discriminative aesthetic scores learning from relative ranking.
Fine-Grained Multi Image Object Hallucination Benchmark
Joonki Min (Seoul National University), Joonseok Lee (Seoul National University)
Vision Language ModelImageBenchmark
🎯 What it does: Proposes the MIOH benchmark for fine-grained evaluation of object hallucinations in multi-image scenarios;
Fine-Grained Post-Training Quantization for Large Vision Language Models with Quantization-Aware Integrated Gradients
Ziwei Xiang (CASIA), Xu-Yao Zhang (CASIA)
Computational EfficiencyTransformerVision Language ModelMultimodality
🎯 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)
Federated LearningTransformerSupervised Fine-TuningPrompt EngineeringVision Language ModelImage
🎯 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.
Fine-VAD: Towards Fine-Grained Video Anomaly Detection via Progressive Cross-Granularity Learning
Menghao Zhang (Beijing University of Posts and Telecommunications), Jingyu Wang (Beijing University of Posts and Telecommunications)
Anomaly DetectionTransformerVision Language ModelContrastive LearningVideo
🎯 What it does: Studied fine-grained video anomaly detection and proposed a cross-granularity hierarchical learning paradigm and the Fine-VAD framework.
FINE: Factorizing Knowledge for Initialization of Variable-sized Diffusion Models
Yucheng Xie (Southeast University), Xin Geng (Southeast University)
GenerationTransformerDiffusion modelImageMagnetic Resonance Imaging
🎯 What it does: Propose the FINE method, which decomposes the weights of diffusion models during pre-training to generate learngenes that can be shared across models of different scales, enabling direct initialization of models with varying sizes.
FINER: MLLMs Hallucinate under Fine-grained Negative Queries
Rui Xiao (Technical University of Munich), Stephan Alaniz (Institut Polytechnique de Paris)
Explainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningPrompt EngineeringVision Language ModelMultimodalityBenchmark
🎯 What it does: Proposes a fine-grained negation query (FINER) benchmark tailored for multi-object, multi-attribute, multi-relationship, and 'What-' type questions, and designs a FINER-Tuning alignment method based on this benchmark to mitigate hallucination issues in multimodal large language models.
FinPercep-RM: A Fine-grained Reward Model and Co-evolutionary Curriculum for RL-based Real-world Super-Resolution
Yidi Liu (University of Science and Technology of China), Zheng-jun Zha (University of Science and Technology of China)
Super ResolutionReinforcement Learning from Human FeedbackReinforcement LearningVision Language ModelImage
🎯 What it does: Propose FinPercep-RM, a fine-grained perceptual reward model, and Co-evolutionary Curriculum Learning (CCL) to improve real-world super-resolution based on RLHF;
FireScope: Wildfire Risk Raster Prediction With a Chain-of-Thought Oracle
Mario Markov (INSAIT Sofia University St Kliment Ohridski), Danda Pani Paudel (INSAIT Sofia University St Kliment Ohridski)
GenerationTransformerReinforcement LearningVision Language ModelImageMultimodalityTabularTime SeriesBenchmarkChain-of-Thought
🎯 What it does: Proposes the FireScope-Bench dataset and FireScope framework, leveraging Chain-of-Thought (CoT) vision-language models (VLMs) as 'Oracle,' and generating high-resolution continuous fire risk grids via FiLM-conditioned lightweight Encoder-Decoder architectures.
First Frame Is the Place to Go for Video Content Customization
Jingxi Chen (University of Maryland), Yiannis Aloimonos (University of Maryland)
SegmentationGenerationSupervised Fine-TuningVision Language ModelDiffusion modelVideo
🎯 What it does: Propose a lightweight plugin called FFGo, which utilizes the first frame of a pre-trained video generation model as a concept buffer to achieve mixed generation with multiple reference objects.
First Logit Boosting: Visual Grounding Method to Mitigate Object Hallucination in Large Vision-Language Models
Jiwoo Ha (DGIST), Jinhyun So (DGIST)
GenerationTransformerVision Language ModelMultimodality
🎯 What it does: Proposes a new method called First Logit Boosting (FLB), aimed at mitigating object hallucination in large vision-language models, particularly maintaining visual information stability during the generation process.
FisherPoser: Human Motion Estimation from Sparse Observations with Hierarchical Region-Wise Fisher-Matrix Uncertainty Modeling
Songpengcheng Xia (Shanghai Jiao Tong University), Ling Pei (Shanghai Jiao Tong University)
Pose EstimationTransformerTime SeriesSequential
🎯 What it does: This paper proposes a probabilistic framework based on the SO(3) matrix Fisher distribution for full-body motion estimation using sparse VR observations from only an HMD and two hand controllers, capable of simultaneously predicting the rotation of each joint and its uncertainty.
FISHuman: Fine-grained Single-image 3D Human Reconstruction via Multi-view 4D Remeshing
Hanxi Liu (Peking University), Zhouhui Lian (Peking University)
GenerationPose EstimationTransformerDiffusion modelImageMesh
🎯 What it does: Generate fine-grained, ultra-realistic 3D human mesh and texture from a single image.
Fixed Anchors Are Not Enough: Dynamic Retrieval and Persistent Homology for Dataset Distillation
Muquan Li (UESTC), Tao He (UESTC)
Knowledge DistillationImage
🎯 What it does: Proposes a framework named RETA for dataset distillation, enhancing synthetic data quality through dynamic retrieval and topological alignment.
FLARE: A Failure-Aware Framework for Autonomous Correction and Recovery in Visual-Language Robotic Manipulation
Ganlong Zhao (Chinese University of Hong Kong), Guanbin Li (Sun Yat-sen University)
Robotic IntelligenceLarge Language ModelVision-Language-Action ModelMultimodality
🎯 What it does: Propose the FLARE framework, which achieves self-correction and recovery from ID (intrinsic) and OOD (extrinsic) errors by introducing two loops—Retry and Reset—into the Vision-Language-Action model.
Flash-DMD: Towards High-Fidelity Few-Step Image Generation with Efficient Distillation and Joint Reinforcement Learning
Guanjie Chen (Shanghai Jiao Tong University), Peng Chen (Tencent)
GenerationKnowledge DistillationReinforcement LearningDiffusion modelScore-based ModelGenerative Adversarial NetworkImage
🎯 What it does: Efficient distillation of multi-step diffusion models, combined with reinforcement learning during distillation to achieve higher quality few-step generation.
FlashCap: Millisecond-Accurate Human Motion Capture via Flashing LEDs and Event-Based Vision
Zekai Wu (Xiamen University), Cheng Wang (Xiamen University)
Pose EstimationSpiking Neural NetworkTransformerImageMultimodalityPoint CloudTime SeriesBenchmark
🎯 What it does: Proposed the FlashCap system, which utilizes flickering LEDs and event cameras to achieve millisecond-level human motion capture, and built the FlashMotion dataset based on this system; simultaneously proposed the ResPose baseline model, which refines RGB priors with high-frequency events to accomplish high temporal resolution pose estimation.
FlashDecoder: Real-Time Latent-to-Pixel Streaming Decoder with Transformers
Minguk Kang (Pika Labs), Suha Kwak (POSTECH)
GenerationCompressionTransformerAuto EncoderGenerative Adversarial NetworkImageVideoText
🎯 What it does: Propose FlashDecoder, a pure Transformer-based latent-to-pixel video decoder that enables frame-by-frame streaming decoding, maintaining constant latency and low memory usage.
FlashIn: Fast and Accurate Image Inversion for Real-time Image Editing
Guangzhi Wang (Tencent ARC Lab)
RestorationDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: Propose a fast and accurate image inversion method called FlashIn, which can complete high-quality real-time image editing within 1~4 steps.
FlashLips: 100-FPS Mask-Free Latent Lip-Sync using Reconstruction Instead of Diffusion or GANs
Andreas Zinonos (Imperial College London), Nikita Drobyshev (Cantina Labs)
GenerationTransformerFlow-based ModelAuto EncoderVideoAudio
🎯 What it does: Designed a two-stage, mask-free, real-time lip-sync system called FlashLips, which employs a first-order latent editor and an audio-to-lip pose flow-matching Transformer. It achieves high-quality lip-sync video generation at over 100 FPS on a single GPU without using GANs or diffusion models.
FlashMesh: Faster and Better Autoregressive Mesh Synthesis via Structured Speculation
Tingrui Shen (Singapore Management University), Shengfeng He (Singapore Management University)
Data SynthesisTransformerMesh
🎯 What it does: Proposed the FlashMesh framework, achieving fast and high-quality autoregressive mesh generation through a prediction-correction-validation three-stage process.
FlashMotion: Few-Step Controllable Video Generation with Trajectory Guidance
Quanhao Li (Fudan University), Zuxuan Wu (Fudan University)
GenerationConvolutional Neural NetworkTransformerDiffusion modelGenerative Adversarial NetworkVideoBenchmark
🎯 What it does: Propose a three-stage training framework FlashMotion for controllable video generation with few steps; first train a trajectory adapter on a multi-step generator, then distill the generator into a few-step model, finally fine-tune the adapter using diffusion + adversarial dual objectives, and create FlashBench, a long-sequence trajectory evaluation benchmark.
FlashPortrait: 6x Faster Infinite Portrait Animation with Adaptive Latent Prediction
Shuyuan Tu (Fudan University), Zuxuan Wu (Fudan University)
GenerationTransformerVision Language ModelDiffusion modelVideo
🎯 What it does: Propose FlashPortrait, an end-to-end video diffusion transformer capable of generating infinitely long, identity-preserving portrait animations at up to 6 times acceleration.
FlashVGGT: Efficient and Scalable Visual Geometry Transformers with Compressed Descriptor Attention
Zipeng Wang (Hong Kong University of Science and Technology), Dan Xu (Hong Kong University of Science and Technology)
Depth EstimationComputational EfficiencyRepresentation LearningTransformerContrastive LearningImage
🎯 What it does: Propose FlashVGGT, achieving efficient and scalable multi-view 3D reconstruction through compressed descriptor attention.
FlashVSR: Towards Real-time Diffusion-Based Streaming Video Super Resolution
Junhao Zhuang (Tsinghua University), Tianfan Xue (CUHK MMLab)
Super ResolutionKnowledge DistillationTransformerDiffusion modelFlow-based ModelImageVideo
🎯 What it does: Developed FlashVSR, a real-time streaming diffusion-based video super-resolution framework that balances efficiency and high quality;
Flat-Pack Bench: Evaluating Spatio-Temporal Understanding in Large Vision-Language Models through Furniture Assembly
Aditya Chetan (Cornell University), Bharath Hariharan (Cornell University)
Agentic AIPrompt EngineeringVision Language ModelVideoMultimodalityBenchmark
🎯 What it does: Proposed the FLAT-PACK BENCH benchmark, using furniture assembly videos to evaluate the spatiotemporal fine-grained reasoning capabilities of large vision-language models
FlexAvatar: Flexible Large Reconstruction Model for Animatable Gaussian Head Avatars with Detailed Deformation
Cheng Peng (Tsinghua University), Yebin Liu (Tsinghua University)
GenerationData SynthesisTransformerGaussian SplattingImage
🎯 What it does: Propose FlexAvatar, a method that can generate real-time animatable 3D Gaussian head avatars from single or sparse images without camera pose/expression annotations;
FlexAvatar: Learning Complete 3D Head Avatars with Partial Supervision
Tobias Kirschstein (Technical University of Munich), Matthias Nießner (Technical University of Munich)
GenerationTransformerGenerative Adversarial NetworkGaussian SplattingImageVideo
🎯 What it does: Generate high-quality, animatable complete 3D head avatars from a single portrait image.
FlexiVideo: Variation-Aware Temporal Dynamics Modeling for Efficient Video Understanding
Da Peng (Xi'an Jiaotong University), Maosong Sun (Tsinghua University)
RecognitionComputational EfficiencyConvolutional Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelVideoTextMultimodality
🎯 What it does: Propose FlexiVideo, a mutation-aware temporal dynamic modeling framework that achieves efficient and accurate encoding in video understanding through adaptive time segmentation and dynamic spatiotemporal embeddings.
FlexTraj: Image-to-Video Generation with Flexible Point Trajectory Control
Zhiyuan Zhang (City University of Hong Kong), Jing Liao (City University of Hong Kong)
GenerationData SynthesisTransformerDiffusion modelAuto EncoderVideo
🎯 What it does: Propose the FlexTraj framework, achieving controllable point trajectory generation from images to videos, supporting multi-granularity control across dense to sparse, spatial/temporal sparse, and misaligned conditions.
FloodDiffusion: Tailored Diffusion Forcing for Streaming Motion Generation
Yiyi Cai (Shanda AI Research Tokyo), Haiyang Liu (University of Tokyo)
GenerationData SynthesisTransformerDiffusion modelFlow-based ModelTextTime SeriesSequential
🎯 What it does: Propose FloodDiffusion, a streaming framework for human motion generation based on diffusion forcing, capable of real-time generating coherent motion according to time-varying text prompts.
FloVerse: Floor Plan-Guided Multi-Modal Navigation
Weiqi Huang (Beijing Institute of Technology), Wei Liang (Beijing Institute of Technology)
Robotic IntelligenceVision-Language-Action ModelDiffusion modelImageMultimodality
🎯 What it does: Designed the FloVerse multi-modal navigation task and constructed the FloVerse 16K dataset, proposing a two-stage diffusion navigation model called ThreeDiff, which achieves unified point, object, and image goal navigation using floor plans and visual perception.
Flow Map Distillation Without Data
Shangyuan Tong (Massachusetts Institute of Technology), Tommi Jaakkola (Massachusetts Institute of Technology)
Knowledge DistillationScore-based ModelFlow-based ModelImage
🎯 What it does: Designed a prior-based flow mapping distillation framework called FreeFlow, which does not rely on external data and can achieve one-step sampling for the teacher flow model;
Flow Matching for Multimodal Distributions
Gaoxiang Luo (University of Minnesota Twin Cities), Ju Sun (University of Minnesota Twin Cities)
GenerationData SynthesisTransformerFlow-based ModelImage
🎯 What it does: This paper proposes using multimodal source distributions and mode-related couplings in flow matching models to accelerate training and sampling.
FLOW: Optimal Transport-Driven Feature Warping for Generalized Remote Physiological Measurement
Bo Zhao (Great Bay University), Zitong Yu (Great Bay University)
Domain AdaptationConvolutional Neural NetworkBiomedical Data
🎯 What it does: Proposes the FLOW framework, leveraging optimal transport-driven feature distortion to achieve end-to-end rPPG domain generalization.
Flow3r: Factored Flow Prediction for Scalable Visual Geometry Learning
Zhongxiao Cong (Carnegie Mellon University), Shubham Tulsiani (Carnegie Mellon University)
Pose EstimationTransformerContrastive LearningOptical FlowVideo
🎯 What it does: Propose Flow3r, a scalable visual geometry learning framework that utilizes 2D optical flow from unannotated videos as auxiliary supervision to learn 3D structure and camera motion from unannotated videos.
Flow4DGS-SLAM: Optical Flow-Guided 4D Gaussian Splatting SLAM
Yunsong Wang (National University of Singapore), Gim Hee Lee (National University of Singapore)
Gaussian SplattingSimultaneous Localization and MappingOptical FlowVideo
🎯 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.
Flowception: Temporally Expansive Flow Matching for Video Generation
Tariq Berrada Ifriqi (FAIR at Meta), Ricky T. Q. Chen (FAIR at Meta)
GenerationData SynthesisTransformerDiffusion modelFlow-based ModelVideoMultimodality
🎯 What it does: Propose a variable-length, non-autoregressive video generation framework called Flowception, which generates complete videos from a single frame by alternately performing frame denoising and insertion.
FlowComposer: Composable Flows for Compositional Zero-Shot Learning
Zhenqi He (Hong Kong University of Science and Technology), Long Chen (Hong Kong University of Science and Technology)
Representation LearningVision Language ModelFlow-based ModelRectified FlowMultimodality
🎯 What it does: This paper proposes FlowComposer, a plug-and-play framework based on flow matching, which explicitly constructs attribute-object combinations in the embedding space to improve compositional zero-shot learning (CZSL).
FlowDC: Flow-Based Decoupling-Decay for Complex Image Editing
Yilei Jiang (Zhejiang University), Long Chen (Harbin Institute of Technology)
Image TranslationLarge Language ModelPrompt EngineeringFlow-based ModelRectified FlowImageTextBenchmark
🎯 What it does: Propose a complex text editing method called FlowDC based on flow matching, which decomposes multi-objective editing into parallel sub-edits and performs orthogonalization and decay on the velocity field, balancing semantic alignment and source image consistency.
FlowDirector: Training-Free Flow Steering for Precise Text-to-Video Editing
Guangzhao Li (Westlake University), Chi Zhang (Westlake University)
GenerationTransformerDiffusion modelFlow-based ModelRectified FlowVideoTextOrdinary Differential Equation
🎯 What it does: Proposed a training-free and inversion-free text-driven video editing framework called FlowDirector, which directly edits videos by driving video evolution through ODEs in the data space.
FlowDIS: Language-Guided Dichotomous Image Segmentation with Flow Matching
Andranik Sargsyan (Picsart AI Research), Shant Navasardyan (Picsart AI Research)
SegmentationTransformerVision Language ModelFlow-based ModelAuto EncoderImageText
🎯 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.
FlowFixer: Towards Detail-Preserving Subject-Driven Generation
Jinyoung Jun (Amazon), Jungbeom Lee (Korea University)
Image TranslationImage HarmonizationGenerationData SynthesisPose EstimationDiffusion modelImageBenchmark
🎯 What it does: Propose a detail restoration framework called FlowFixer based on image-to-image translation to enhance the detail fidelity of subject-driven generation (SDG) images.
FlowFM: Advancing Dark Optical Flow Estimation with Flow Matching
Fengyuan Zuo (Xi'an University Of Technology), Yuerong Mu (Xi'an University Of Technology)
Flow-based ModelOptical FlowImage
🎯 What it does: Proposed a dark-light optical flow estimation model called FlowFM based on flow matching, achieving one-step denoising and obtaining accurate optical flow through explicit flow field regression.
FlowHijack: A Dynamics-Aware Backdoor Attack on Flow-Matching Vision-Language-Action Models
Xinyuan An (Zhejiang University), Dongxia Wang (Zhejiang University)
Adversarial AttackVision-Language-Action ModelFlow-based ModelMultimodality
🎯 What it does: Proposed a backdoor attack framework named FlowHijack targeting flow-matching Vision-Language-Action (VLA) models, achieving a dynamically perceptible backdoor implantable and activatable via semantic triggers;
FlowMotion: Training-Free Flow Guidance for Video Motion Transfer
Zhen Wang (Hong Kong University of Science and Technology), Long Chen (Hong Kong University of Science and Technology)
GenerationData SynthesisFlow-based ModelVideoText
🎯 What it does: Proposes a new training-agnostic video motion transfer framework called FlowMotion, which can efficiently and flexibly transfer motion patterns from source videos to target videos.
FlowPalm: Optical Flow Driven Non-Rigid Deformation for Geometrically Diverse Palmprint Generation
Yuchen Zou (Xi'an Jiaotong University), Dexing Zhong (Xi'an Jiaotong University)
GenerationData SynthesisConvolutional Neural NetworkDiffusion modelOptical FlowImage
🎯 What it does: Propose a flow-driven non-rigid deformation framework called FlowPalm for generating geometrically diverse palmprint images that maintain identity consistency.
FlowPortal: Residual-Corrected Flow for Training-Free Video Relighting and Background Replacement
Wenshuo Gao (Peking University), Shuai Yang (Peking University)
GenerationFlow-based ModelVideo
🎯 What it does: Proposed a no-training, flow-model-based 'FlowPortal' framework for efficient video relighting and background replacement;
FlowSteer: Guiding Few-Step Image Synthesis with Authentic Trajectories
Lei Ke (Tsinghua University), Jing Lyu (Tencent)
GenerationData SynthesisDiffusion modelScore-based ModelFlow-based ModelRectified FlowGenerative Adversarial NetworkImageOrdinary Differential Equation
🎯 What it does: Proposed the FlowSteer method, which utilizes real teacher-generated trajectories to guide student models for few-step image synthesis.
FluidGaussian: Propagating Simulation-Based Uncertainty Toward Functionally-Intelligent 3D Reconstruction
Yuqiu Liu (Simon Fraser University), Michael W. Mahoney (University Of California Berkeley)
GenerationGaussian SplattingImagePhysics Related
🎯 What it does: Proposed FluidGaussian, a plugin that integrates fluid-structure interaction information into 3D Gaussian Splatting reconstruction, achieving both visual and physical consistency.
FluoCLIP: Stain-Aware Focus Quality Assessment in Fluorescence Microscopy
Hyejin Park (Ewha Womans University), Dongbo Min (Ewha Womans University)
ClassificationTransformerPrompt EngineeringVision Language ModelImageMultimodalityBiomedical DataBenchmark
🎯 What it does: Propose FluoCLIP, a two-stage audio-visual language framework for stain-dependent focus quality assessment in fluorescence microscopy, and construct the FluoMix dataset covering multi-tissue and multi-stain samples.
FluxMem: Adaptive Hierarchical Memory for Streaming Video Understanding
Yiweng Xie (Fudan University), Zuxuan Wu (Fudan University)
RecognitionTransformerLarge Language ModelVision Language ModelVideo
🎯 What it does: Propose FluxMem, a training-free, hierarchical memory framework adapted for multi-modal large language models, designed for streaming video understanding.
FM-Steer: Enhance Generalist Policies with Value-Guided Cascaded Denoising
Haoming Song (Shanghai Jiao Tong University), Xuelong Li (Shanghai AI Laboratory)
Robotic IntelligenceVision-Language-Action ModelFlow-based ModelMultimodality
🎯 What it does: This paper designs the FM-Steer framework, leveraging value-guided multi-candidate sampling and cascading denoising to enhance the robotic manipulation performance of streaming vision-language-action models during testing.
FMPose3D: monocular 3D pose estimation via flow matching
Ti Wang (École Polytechnique Fédérale de Lausanne), Mackenzie Weygandt Mathis (École Polytechnique Fédérale de Lausanne)
Pose EstimationGraph Neural NetworkTransformerFlow-based ModelImageOrdinary Differential Equation
🎯 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.
fMRI-LM: Towards a Universal Foundation Model for Language-Aligned fMRI Understanding
Yuxiang Wei (TReNDS Center (Georgia Institute of Technology, Georgia State University, Emory)), Vince D. Calhoun (TReNDS Center (Georgia Institute of Technology, Georgia State University, Emory))
ClassificationRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningContrastive LearningTextBiomedical DataMagnetic Resonance ImagingAlzheimer's Disease
🎯 What it does: Designed and trained fMRI-LM, a general foundation model that aligns functional magnetic resonance imaging signals with large language models, achieving language-based understanding and reasoning of brain images through multi-stage pre-training and instruction fine-tuning.
Focal-General Diffusion Model with Semantic Consistent Guidance for Sign Language Production
Yiheng Yu (Zhejiang University of Technology), Min Xu (Zhejiang University of Technology)
GenerationGraph Neural NetworkTransformerDiffusion modelVideoMultimodality
🎯 What it does: Propose a two-stage diffusion model FGDM for the pose generation task in sign language production.
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)
SegmentationGraph Neural NetworkTransformerPrompt EngineeringContrastive LearningImageBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 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)
ClassificationRecognitionObject DetectionSegmentationRepresentation LearningTransformerContrastive LearningVideoBiomedical Data
🎯 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)
SegmentationComputational EfficiencyTransformerVision Language ModelContrastive LearningImageText
🎯 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.
FocusUI: Efficient UI Grounding via Position-Preserving Visual Token Selection
Mingyu Ouyang (National University of Singapore), Hwee Tou Ng (National University of Singapore)
Computational EfficiencyTransformerVision Language ModelImage
🎯 What it does: Proposes FOCUSUI, a framework that selects instruction-related visual tokens while maintaining positional continuity to achieve efficient UI localization.
FoleyDesigner: Immersive Stereo Foley Generation with Precise Spatio-Temporal Alignment for Film Clips
Mengtian Li (Shanghai Engineering Research Center of Motion Picture Special Effects), Zhifeng Xie (Shanghai Engineering Research Center of Motion Picture Special Effects)
GenerationData SynthesisTransformerLarge Language ModelAgentic AIDiffusion modelVideoMultimodalityChain-of-ThoughtAudio
🎯 What it does: Built a complete automated Foley design framework called FoleyDesigner, capable of generating spatiotemporally precise stereo Foley sounds from silent movie clips and directly outputting 5.1 surround sound.
FoleyDirector: Fine-Grained Temporal Steering for Video-to-Audio Generation via Structured Scripts
You Li, Yi Yang
GenerationData SynthesisTransformerVision Language ModelVideoMultimodalityAudio
🎯 What it does: Propose the FoleyDirector framework, which utilizes structured temporal scripts to achieve fine-grained temporal control for video-to-audio generation;
Follow the Saliency: Supervised Saliency for Retrieval-augmented Dense Video Captioning
Seung hee Choi (Hanyang University), Dong-Jin Kim (Hanyang University)
GenerationRetrievalTransformerPrompt EngineeringVision Language ModelVideoTextRetrieval-Augmented Generation
🎯 What it does: Directly learn frame-level saliency using annotated event boundaries, and use this saliency to drive retrieval segmentation and caption generation, achieving temporal consistency between retrieval and generation;
FontCrafter: High-Fidelity Element-Driven Artistic Font Creation with Visual In-Context Generation
Wuyang Luo (Dalian University of Technology), Yongjiu Ma (Fudan University)
GenerationData-Centric LearningSupervised Fine-TuningDiffusion modelImage
🎯 What it does: Propose a novel element-driven artistic font generation framework called FontCrafter, which generates high-fidelity glyphs using visual context generation technology under the condition of given glyph masks and element images.
FORCE: Transferable Visual Jailbreaking Attacks via Feature Over-Reliance CorrEction
Runqi Lin (University of Sydney), Tongliang Liu (University of Sydney)
Adversarial AttackTransformerImageTextBenchmark
🎯 What it does: Studied the limitations of optimization-based visual jailbreak attacks in cross-model transfer, analyzed their dependencies on loss landscapes, layer features, and frequency domain features, and proposed the Feature Over-Reliance CorrEction (FORCE) method to correct these over-reliances, significantly enhancing the cross-model transferability of the attacks.
ForceVLA2: Unleashing Hybrid Force-Position Control with Force Awareness for Contact-Rich Manipulation
Yang Li (Tongji University), Jiangmiao Pang (Shanghai AI Laboratory)
Robotic IntelligenceMixture of ExpertsVision-Language-Action ModelFlow-based ModelImageMultimodality
🎯 What it does: Propose ForceVLA2, an end-to-end hybrid force-position control framework that integrates force perception with vision-language-action for contact-rich manipulation;
ForeAct: Steering Your VLA with Efficient Visual Foresight Planning
Zhuoyang Zhang (MIT), Song Han (MIT)
Robotic IntelligenceTransformerLarge Language ModelSupervised Fine-TuningVision-Language-Action ModelDiffusion modelWorld ModelImageTextMultimodality
🎯 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)
GenerationTransformerDiffusion modelImageVideo
🎯 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)
RecognitionTransformerContrastive LearningSimultaneous Localization and MappingVideoMultimodalityPoint Cloud
🎯 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)
GenerationData SynthesisPose EstimationTransformerDiffusion modelFlow-based ModelVideoMesh
🎯 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)
GenerationAdversarial AttackDiffusion modelImageTextMultimodality
🎯 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.
Forging a Dynamic Memory: Retrieval-Guided Continual Learning for Generalist Medical Foundation Models
Zizhi Chen (Fudan University), Lihua Zhang (Fudan University)
Vision Language ModelContrastive LearningMultimodalityBiomedical DataRetrieval-Augmented Generation
🎯 What it does: Proposed PRIMED, a retrieval-enhanced continuous learning framework in the medical field, for the continuous training of a general medical foundation model.
FoSS: Modeling Long-Range Dependencies and Multimodal Uncertainty in Trajectory Prediction via Fourier-State Space Integration
Yizhou Huang (Brunel University of London), Kezhi Wang (Brunel University of London)
Autonomous DrivingTime SeriesSequential
🎯 What it does: Propose the FoSS dual-branch architecture, which simultaneously models trajectory signals in the frequency domain and time domain, leveraging Fourier transform and linear time series models to achieve long-range dependency and multi-modal uncertainty inference.
Foundation Encoders Are All You Need for Preference-Aware Personalization
Hyungjin Kim (Inha University), Young-Duk Seo (Inha University)
GenerationTransformerVision Language ModelImageTextRetrieval-Augmented Generation
🎯 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)
Object DetectionDomain AdaptationTransformerImage
🎯 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)
RestorationSuper ResolutionMixture of ExpertsVision Language ModelDiffusion modelAuto EncoderImageText
🎯 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.
Foundry: Distilling 3D Foundation Models for the Edge
Guillaume Letellier (Normandy University), Gaurav Sharma (IIT Kanpur)
Computational EfficiencyKnowledge DistillationRepresentation LearningTransformerAuto EncoderPoint Cloud
🎯 What it does: Proposed a framework named Foundry, which distills large self-supervised 3D point cloud foundation models into compact, efficient proxy models suitable for edge devices through compression and reconstruction mechanisms.
Fourier Angle Alignment for Oriented Object Detection in Remote Sensing
Changyu Gu (Beijing Institute of Technology), Ying Fu (Beijing Institute of Technology)
Object DetectionImageBenchmark
🎯 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)
Computational EfficiencyRepresentation LearningConvolutional Neural NetworkGraph Neural NetworkMeshGraph
🎯 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)
Domain AdaptationRepresentation LearningTransformerPrompt EngineeringImage
🎯 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.
FPS-Bench: A Benchmark for High Frame-Rate Video Understanding
Rohan Choudhury (Carnegie Mellon University), László A. Jeni (Carnegie Mellon University)
Vision Language ModelVideoBenchmark
🎯 What it does: Proposed FPS-Bench, a new multiple-choice question answering benchmark specifically designed to evaluate the fine-grained temporal reasoning capabilities of video language models at high frame rates;
Fractal Camouflage: A Bio-Inspired Approach for Multi-Scale Adversarial Attacks in the Infrared Domain
Chengyin Hu (China University of Petroleum-Beijing), Yiwei Wei (China University of Petroleum-Beijing)
Adversarial AttackImage
🎯 What it does: Propose a black-box physical adversarial attack called AdvFractal based on H-shaped fractal geometry, achieving stealthy attacks in infrared images under multi-scale and multi-physical conditions.
Frame2Freq: Spectral Adapters for Fine-Grained Video Understanding
Thinesh Thiyakesan Ponbagavathi (University of Stuttgart), Alina Roitberg
RecognitionTransformerVideo
🎯 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;
FRAMER: Frequency-Aligned Self-Distillation with Adaptive Modulation Leveraging Diffusion Priors for Real-World Image Super-Resolution
Seungho Choi (Chung-Ang University), Jihyong Oh (Chung-Ang University)
RestorationSuper ResolutionConvolutional Neural NetworkTransformerDiffusion modelContrastive LearningImage
🎯 What it does: This paper proposes a self-distillation based FRAMER training framework to enhance the reconstruction quality of real image super-resolution (Real-ISR) models under unknown degradation conditions.
Franca: Nested Matryoshka Clustering for Scalable Visual Representation Learning
Shashanka Venkataramanan (Qualcomm AI Research), Yuki M Asano
ClassificationObject DetectionSegmentationRepresentation LearningTransformerContrastive LearningImage
🎯 What it does: Trained and released a fully open-source visual foundation model named Franca, using public data and open-source code, providing model weights, training procedures, and intermediate checkpoints.
FrankenMotion: Part-level Human Motion Generation and Composition
Chuqiao Li (University of Tübingen), Gerard Pons-Moll (University of Tübingen)
GenerationTransformerLarge Language ModelPrompt EngineeringVision Language ModelDiffusion modelTextTime Series
🎯 What it does: This paper proposes a text-based part-level human motion generation framework called FrankenMotion, and constructs a new dataset with part-level temporal annotations named FrankenStein.
Free-Grained Hierarchical Visual Recognition
Seulki Park (University of Michigan), Stella X. Yu (University of Michigan)
ClassificationRecognitionLarge Language ModelContrastive LearningImageTextBenchmark
🎯 What it does: Propose a free-grained hierarchical visual recognition task, allowing training samples to be annotated at any level and adaptively selecting prediction depth during inference.
Free-Lunch Long Video Generation via Layer-Adaptive O.O.D Correction
Jiahao Tian, Chi Zhang
GenerationTransformerDiffusion modelVideoBenchmark
🎯 What it does: Achieve long video generation on pre-trained DiT video diffusion models using the training-agnostic FreeLOC framework.
FreeArtGS: Articulated Gaussian Splatting Under Free-moving Scenario
Hang Dai (Peking University), Hao Dong (Peking University)
GenerationPose EstimationDepth EstimationOptimizationGaussian SplattingVideoPoint Cloud
🎯 What it does: Propose FreeArtGS, a monocular RGB-D video-based framework for 3D reconstruction of freely moving articulated objects, capable of simultaneously recovering object geometry, texture, and joint parameters;
FreeForm: Reduced-Order Deformable Simulation from Particle-Based Skinning Eigenmodes
Donglai Xiang (NVIDIA), David I.W. Levin (NVIDIA)
Gaussian SplattingPoint CloudMeshPhysics Related
🎯 What it does: This paper proposes a mesh-free reduced-order elastic simulation method based on the Reproducing Kernel Particle Method (RKPM), utilizing skinning features to describe deformations of elastic objects, applicable to modern geometric representations such as Gaussian Splats.
FreeScale: Scaling 3D Scenes via Certainty-Aware Free-View Generation
Chenhan Jiang (Hong Kong University Of Science And Technology), Jiankang Deng (Imperial College London)
GenerationData SynthesisDiffusion modelGaussian SplattingImage
🎯 What it does: Reconstruct the scene using sparse real images through 3D Gaussian expansion, then generate a large number of high-quality free-viewpoint images in the reconstructed scene using a confidence-aware sampling strategy, which can be used as training data or to enhance existing scene reconstructions.