CVPR 2025 Papers — Page 10
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2871 papers
Feature-Preserving Mesh Decimation for Normal Integration
Moritz Heep (University of Bonn), Eduard Zell
CompressionOptimizationMesh
🎯 What it does: In the process of normal map triangulation and normal integration, a screen space adaptive sparse triangle mesh solver is proposed, which first sparsifies the mesh (edge collapse) and aligns the mesh to surface features through vertex and edge alignment, maintaining high precision 3D reconstruction at high compression rates.
Feature4X: Bridging Any Monocular Video to 4D Agentic AI with Versatile Gaussian Feature Fields
Shijie Zhou (University of California Los Angeles), Achuta Kadambi (University of California Los Angeles)
RecognitionObject DetectionSegmentationData SynthesisKnowledge DistillationAgentic AIGaussian SplattingVideo
🎯 What it does: Construct an interactive 4D Gaussian feature field from monocular video and transfer the functionalities of various 2D foundational models (such as SAM2, CLIP-LSeg, InternVideo2) to 4D space, achieving tasks like segmentation, editing, and VQA.
FedAWA: Adaptive Optimization of Aggregation Weights in Federated Learning Using Client Vectors
Changlong Shi (Jilin University), Yi Chang (Jilin University)
OptimizationFederated LearningImage
🎯 What it does: This paper studies a method for dynamically adjusting aggregation weights in federated learning through client vectors, called FedAWA.
FedBiP: Heterogeneous One-Shot Federated Learning with Personalized Latent Diffusion Models
Haokun Chen (Siemens Technology), Volker Tresp (Ludwig Maximilian University of Munich)
Data SynthesisFederated LearningDiffusion modelImageBiomedical Data
🎯 What it does: A heterogeneous one-round federated learning framework FedBiP based on a dual-layer personalized diffusion model is proposed to address data scarcity and feature heterogeneity issues.
FedCALM: Conflict-aware Layer-wise Mitigation for Selective Aggregation in Deeper Personalized Federated Learning
Hao Zheng (Central South University), Boyu Wang (University of Western Ontario)
ClassificationFederated LearningConvolutional Neural NetworkImage
🎯 What it does: A conflict-aware hierarchical mitigation method named FedCALM is proposed to address the server aggregation conflict issue in personalized federated learning (PFL) with deep base models.
FedCS: Coreset Selection for Federated Learning
Chenhe Hao (Xidian University), Yunsong Li (Xidian University)
ClassificationFederated LearningContrastive LearningImage
🎯 What it does: In the context of federated learning, we propose FedCS, a core sample selection method based on Distance Contrast (DC) metrics, which constructs an efficient and unbiased coreset using global class center aggregation and dual pruning.
Federated Learning with Domain Shift Eraser
Zheng Wang (Xiamen University), Cheng Wang (Xiamen University)
ClassificationDomain AdaptationFederated LearningConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper proposes a new framework to address domain shift issues in federated learning—Federated Domain Shift Eraser (FDSE).
FedMIA: An Effective Membership Inference Attack Exploiting "All for One" Principle in Federated Learning
Gongxi Zhu (Tsinghua University), Yuxing Han (Tsinghua University)
Federated LearningSafty and PrivacyAdversarial AttackImage
🎯 What it does: This paper proposes FedMIA, a Federated Learning membership inference attack that utilizes updates from all clients.
FedSPA: Generalizable Federated Graph Learning under Homophily Heterogeneity
Zihan Tan (Wuhan University), Mang Ye (Wuhan University)
Federated LearningGraph Neural NetworkGraph
🎯 What it does: This paper proposes the FedSPA framework, which achieves federated graph learning under homogeneity and heterogeneity through subgraph feature propagation separation and homogeneity bias-driven aggregation.
FeedEdit: Text-Based Image Editing with Dynamic Feedback Regulation
Fengyi Fu (University of Science and Technology of China), Zhendong Mao (University of Science and Technology of China)
Image TranslationGenerationDiffusion modelImageText
🎯 What it does: This paper proposes a training-free, feedback-regulated text-guided image editing framework called FeedEdit, which can dynamically adjust the editing intensity at each denoising step to achieve more precise text editing.
Ferret: An Efficient Online Continual Learning Framework under Varying Memory Constraints
Yuhao Zhou (Sichuan University), Jiancheng Lv (Sichuan University)
ClassificationComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: This paper proposes and implements the Ferret framework, aimed at enhancing the online accuracy of online continual learning (OCL) algorithms under variable memory constraints.
Few-shot Implicit Function Generation via Equivariance
Suizhi Huang (Shanghai Jiao Tong University), Xinchao Wang (National University of Singapore)
GenerationData SynthesisDiffusion modelContrastive LearningImagePoint Cloud
🎯 What it does: This paper proposes the Few-shot Implicit Function Generation task, which aims to train a generative model using only a few target category INRs to produce diverse and functionally consistent weight samples.
Few-shot Personalized Scanpath Prediction
Ruoyu Xue (Stony Brook University), Dimitris Samaras (Stony Brook University)
RecognitionGenerationTransformerContrastive LearningImage
🎯 What it does: A personalized gaze scanning path prediction method based on a small number of support samples (FS-PSP) is proposed.
Few-Shot Recognition via Stage-Wise Retrieval-Augmented Finetuning
Tian Liu (Texas A&M University), Shu Kong (University of Macau)
ClassificationRecognitionRetrievalTransformerSupervised Fine-TuningVision Language ModelImageRetrieval-Augmented Generation
🎯 What it does: A two-stage retrieval-enhanced fine-tuning method (SWAT) is proposed, utilizing the pre-training data of VLM for retrieval tasks with a small number of labeled samples, and fine-tuning the visual encoder and classifier separately.
FFaceNeRF: Few-shot Face Editing in Neural Radiance Fields
Kwan Yun (Korea Advanced Institute of Science and Technology), Junyong Noh (Korea Advanced Institute of Science and Technology)
SegmentationGenerationNeural Radiance FieldImage
🎯 What it does: Proposes FFaceNeRF, which utilizes a geometric adapter and feature injection to achieve customizable mask layouts for 3D face editing with very few samples;
FFR: Frequency Feature Rectification for Weakly Supervised Semantic Segmentation
Ziqian Yang (XJTLU University of Liverpool), Jimin Xiao (XJTLU University of Liverpool)
SegmentationTransformerImage
🎯 What it does: Proposes the Frequency Feature Rectification (FFR) framework to correct the boundary mis-segmentation caused by the attenuation of high-frequency features in Vision Transformer-based weakly supervised semantic segmentation, thereby improving pseudo-label and segmentation accuracy.
FG^2: Fine-Grained Cross-View Localization by Fine-Grained Feature Matching
Zimin Xia (Ecole Polytechnique Federale de Lausanne), Alexandre Alahi (Ecole Polytechnique Federale de Lausanne)
Object DetectionPose EstimationContrastive LearningSimultaneous Localization and MappingImagePoint Cloud
🎯 What it does: This paper proposes a fine-grained cross-view localization method that maps ground image features to 3D point clouds and learns feature selection in the height dimension, thereby generating a BEV plane corresponding to aerial images and matching it to estimate the 3DoF pose of the ground camera.
FIction: 4D Future Interaction Prediction from Video
Kumar Ashutosh (University of Texas at Austin), Kristen Grauman (University of Texas at Austin)
Pose EstimationTransformerSimultaneous Localization and MappingVideoMultimodality
🎯 What it does: Predict the locations and poses of future human interactions with objects in 3D environments.
FIFA: Fine-grained Inter-frame Attention for Driver's Video Gaze Estimation
Daosong Hu (Sun Yat-sen University), Kai Huang (Sun Yat-sen University)
RecognitionPose EstimationAutonomous DrivingConvolutional Neural NetworkTransformerVideo
🎯 What it does: This paper proposes a FIFA fine-grained cross-frame attention mechanism and a parallel CNN+Transformer framework for estimating driver gaze in videos; it captures pupil displacement across adjacent frames through chunking and cross-attention, while adjusting weights to retain head pose information.
FilmComposer: LLM-Driven Music Production for Silent Film Clips
Zhifeng Xie (Shanghai University), Mengtian Li (Shanghai University)
GenerationTransformerLarge Language ModelAgentic AIVideoMultimodalityAudio
🎯 What it does: Developed the FilmComposer framework, which utilizes LLM to generate high-quality soundtracks synchronized with film editing through a multi-stage process of visual-rhythm-music.
Filter Images First, Generate Instructions Later: Pre-Instruction Data Selection for Visual Instruction Tuning
Bardia Safaei (Johns Hopkins University), Shao-Yuan Lo (Honda Research Institute USA)
GenerationComputational EfficiencyTransformerLarge Language ModelVision Language ModelImageText
🎯 What it does: This paper proposes PreSel, a framework that pre-selects images before generating visual instructions, aiming to reduce the training time of VIT and the cost of instruction generation.
FIMA-Q: Post-Training Quantization for Vision Transformers by Fisher Information Matrix Approximation
Zhuguanyu Wu (Beihang University), Yunhong Wang (Beihang University)
ClassificationObject DetectionSegmentationOptimizationTransformerImage
🎯 What it does: A post-training quantization method based on Fisher Information Matrix (FIM) approximation, called FIMA-Q, is proposed, specifically targeting block-level reconstruction quantization for Vision Transformers.
Finding Local Diffusion Schrodinger Bridge using Kolmogorov-Arnold Network
Xingyu Qiu (Harbin Institute of Technology), Shuo Li (Harbin Institute of Technology)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: This paper studies a framework for finding Local Schrödinger Bridges (LDSB) in the diffusion path subspace, significantly improving image generation quality and sampling efficiency by optimizing the diffusion path weights (fA(t), fB(t)) using a pre-trained denoising network.
Fine-Grained Erasure in Text-to-Image Diffusion-based Foundation Models
Kartik Thakral (Indian Institute of Technology Jodhpur), Richa Singh (Indian Institute of Technology Jodhpur)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: A method called FADE is proposed to achieve fine-grained, adjacency-aware concept elimination in text-to-image diffusion models.
Fine-Grained Image-Text Correspondence with Cost Aggregation for Open-Vocabulary Part Segmentation
Jiho Choi (Korea Advanced Institute of Science and Technology), Hyunjung Shim (Korea Advanced Institute of Science and Technology)
Object DetectionSegmentationTransformerContrastive LearningImage
🎯 What it does: A cost aggregation-based open vocabulary part segmentation framework, PartCATSeg, is proposed, achieving dual image-text correspondence at both object and part levels, and completing fine-grained segmentation of unseen categories during the inference phase.
FINECAPTION: Compositional Image Captioning Focusing on Wherever You Want at Any Granularity
Hang Hua (University of Rochester), Jiebo Luo (University of Rochester)
SegmentationGenerationConvolutional Neural NetworkLarge Language ModelSupervised Fine-TuningVision Language ModelImage
🎯 What it does: Proposes the FINECAPTION model, which supports multi-granularity image compositional descriptions with mask-aware referencing, and constructs the COMPOSITIONCAP dataset.
FineLIP: Extending CLIP's Reach via Fine-Grained Alignment with Longer Text Inputs
Mothilal Asokan (Technology Innovation Institute), Fatima Albreiki (Technology Innovation Institute)
GenerationRetrievalTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: The FineLIP model is proposed, which extends CLIP to handle long texts and improves zero-shot performance in image-text retrieval and text-to-image generation through token-level fine-grained alignment.
FinePhys: Fine-grained Human Action Generation by Explicitly Incorporating Physical Laws for Effective Skeletal Guidance
Dian Shao (Northwestern Polytechnical University), Binglu Wang (Northwestern Polytechnical University)
GenerationData SynthesisPose EstimationDiffusion modelVideoPhysics Related
🎯 What it does: Proposes the FinePhys framework to achieve fine-grained human action video generation.
Finer-CAM: Spotting the Difference Reveals Finer Details for Visual Explanation
Ziheng Zhang (Ohio State University), Wei-Lun Chao (Ohio State University)
Explainability and InterpretabilityImageMultimodality
🎯 What it does: The Finer-CAM method is proposed, which generates fine-grained saliency maps by comparing the target class with similar classes, thereby improving the detail localization ability of visual explanations and being compatible with existing CAM methods such as Grad-CAM and Score-CAM; this idea is also extended to multimodal zero-shot models.
FineVQ: Fine-Grained User Generated Content Video Quality Assessment
Huiyu Duan (Shanghai Jiao Tong University), Guangtao Zhai (Bilibili Inc.)
Recommendation SystemTransformerLarge Language ModelSupervised Fine-TuningVideo
🎯 What it does: This paper constructs the FineVD database and proposes the FineVQ model for multi-dimensional fine-grained quality assessment of UGC videos.
Fingerprinting Denoising Diffusion Probabilistic Models
Huan Teng (South China University of Technology), Hui Ji (National University of Singapore)
RecognitionGenerationDiffusion modelImage
🎯 What it does: This paper proposes the first non-invasive fingerprint recognition framework for DDPM (Denoising Diffusion Probabilistic Models) called FingerInv, which generates trigger signals using specific QR code images to complete model ownership verification.
Finsler Multi-Dimensional Scaling: Manifold Learning for Asymmetric Dimensionality Reduction and Embedding
Thomas Dagès (Technion), Ron Kimmel (Technion)
Graph
🎯 What it does: Proposes the Finsler Multidimensional Scaling (Finsler MDS) method, which uses Finsler geometry (specifically Canonical Randers space) for low-dimensional embedding of asymmetric distance data;
FiRe: Fixed-points of Restoration Priors for Solving Inverse Problems
Matthieu Terris (University of Paris-Saclay), Thomas Moreau (University of Paris-Saclay)
RestorationSuper ResolutionImageStochastic Differential Equation
🎯 What it does: This paper proposes the FiRe framework, which combines a general restoration model with its degradation operations during training, using its fixed points as implicit priors to solve linear inverse problems.
FIRE: Robust Detection of Diffusion-Generated Images via Frequency-Guided Reconstruction Error
Beilin Chu (Beijing University of Posts and Telecommunications), Linna Zhou (Beijing University of Posts and Telecommunications)
GenerationAnomaly DetectionDiffusion modelAuto EncoderImage
🎯 What it does: This study proposes a detection method based on frequency-guided reconstruction error, called FIRE, to distinguish between real images and images generated by diffusion models.
FireEdit: Fine-grained Instruction-based Image Editing via Region-aware Vision Language Model
Jun Zhou (Sun Yat-sen University), Xiaodan Liang (Tencent)
Image TranslationGenerationTransformerVision Language ModelDiffusion modelImage
🎯 What it does: We propose FireEdit, a method for fine-grained instruction-driven image editing achieved through a region-aware visual language model.
FirePlace: Geometric Refinements of LLM Common Sense Reasoning for 3D Object Placement
Ian Huang (Stanford University), Alireza Fathi (Google DeepMind)
GenerationOptimizationExplainability and InterpretabilityTransformerLarge Language ModelVision Language ModelMultimodalityPoint Cloud
🎯 What it does: The FirePlace framework is proposed, which combines multimodal large language models with fine-grained geometric constraints to achieve 3D object placement based on natural language instructions.
Fish-Vista: A Multi-Purpose Dataset for Understanding & Identification of Traits from Images
Kazi Sajeed Mehrab (Virginia Tech), Anuj Karpatne (University of California)
ClassificationRecognitionSegmentationConvolutional Neural NetworkTransformerImageBenchmark
🎯 What it does: The Fish-Vista dataset is proposed, containing 69,269 images and 4,316 species of fish, providing tasks for species classification, visual feature recognition, and segmentation.
FisherTune: Fisher-Guided Robust Tuning of Vision Foundation Models for Domain Generalized Segmentation
Dong Zhao (Xidian University), Zhun Zhong (Xidian University)
SegmentationDomain AdaptationAutonomous DrivingTransformerSupervised Fine-TuningImage
🎯 What it does: This paper proposes FisherTune, a robust fine-tuning method for Vision Foundation Models, aimed at improving the performance of domain generalization semantic segmentation while maintaining cross-domain generalization capabilities.
Fitted Neural Lossless Image Compression
Zhe Zhang (Wuhan University), Shan Liu (Tencent Media Lab)
CompressionImage
🎯 What it does: Lossless image compression is achieved using an overfitting latent variable model for each image and a pre-fitted autoregressive model.
FLAIR: VLM with Fine-grained Language-informed Image Representations
Rui Xiao (Technical University of Munich), Stephan Alaniz (Technical University of Munich)
SegmentationRetrievalTransformerLarge Language ModelVision Language ModelContrastive LearningImageText
🎯 What it does: Designed and trained a visual-language model named FLAIR, which learns fine-grained image representations on local sub-regions of images through text-conditioned attention pooling, achieving more precise image-text alignment.
FLAME: Frozen Large Language Models Enable Data-Efficient Language-Image Pre-training
Anjia Cao (Xi'an Jiaotong University), Zhiheng Ma (Chinese Academy of Sciences)
RetrievalComputational EfficiencyKnowledge DistillationRepresentation LearningTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: The paper proposes the FLAME framework, which utilizes a frozen large language model (LLM) as a text encoder to directly process long texts and achieve efficient language-image pre-training.
FLARE: Feed-forward Geometry, Appearance and Camera Estimation from Uncalibrated Sparse Views
Shangzhan Zhang (Zhejiang University), Gordon Wetzstein (Stanford University)
Pose EstimationDepth EstimationTransformerGaussian SplattingImage
🎯 What it does: We propose FLARE, an end-to-end forward network that can recover high-quality camera poses, 3D geometry, and appearance from 2-8 uncalibrated sparse views in less than 0.5 seconds.
Flash-Split: 2D Reflection Removal with Flash Cues and Latent Diffusion Separation
Tianfu Wang (University of Maryland), Christopher A. Metzler (University of Maryland)
RestorationDiffusion modelAuto EncoderImage
🎯 What it does: A 2D reflection removal framework called Flash-Split is proposed, which utilizes iterative diffusion in the latent space to separate illumination and reflection from pairs of flash/no-flash images, and recovers high-frequency details through cross-latent decoding.
Flash3D: Super-scaling Point Transformers through Joint Hardware-Geometry Locality
Liyan Chen (University of Texas), Paul Vernaza (Cruise)
SegmentationAutonomous DrivingComputational EfficiencyTransformerPoint Cloud
🎯 What it does: In the 3D point cloud Transformer, the authors propose the Flash3D Transformer, which unifies geometric locality with GPU memory locality. By utilizing Perfect Spatial Hashing (PSH) to map points to contiguous memory and implementing Bucket-and-Swin window shifting without additional cost, it significantly enhances computational efficiency.
FlashGS: Efficient 3D Gaussian Splatting for Large-scale and High-resolution Rendering
Guofeng Feng (Institute of Computing Technology, Chinese Academy of Sciences), Bo Dai (University of Hong Kong)
Computational EfficiencyGaussian SplattingPoint Cloud
🎯 What it does: This paper presents FlashGS, a high-performance 3D Gaussian splatting rendering framework that significantly enhances the real-time rendering speed of large-scale high-resolution scenes.
FlashSloth : Lightning Multimodal Large Language Models via Embedded Visual Compression
Bo Tong (Xiamen University), Rongrong Ji (Xiamen University)
CompressionComputational EfficiencyTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: Designed and implemented a lightweight high-speed multimodal large language model FlashSloth, integrating an embedded visual compression module (Spatial Attention Pooling SAP and Embedded Query EmbQ) to reduce the number of visual tokens and enhance inference efficiency.
FLAVC: Learned Video Compression with Feature Level Attention
Chun Zhang (Waseda University), Jiro Katto (Waseda University)
CompressionTransformerVideo
🎯 What it does: A Transformer-based video compression framework called FLAVC is proposed, which incorporates a feature-level attention module, a dense overlapping patcher, and a Transformer-CNN hybrid encoder.
FlexDrive: Toward Trajectory Flexibility in Driving Scene Gaussian Splatting Reconstruction and Rendering
Jingqiu Zhou (Chinese University of Hong Kong), Hongsheng Li (Chinese University of Hong Kong)
GenerationAutonomous DrivingGaussian SplattingPoint CloudBenchmark
🎯 What it does: This study investigates how to enhance the quality of out-of-path view reconstruction and rendering of 3D Gaussian scattering in driving scenes using inverse view reprojection and depth bootstrapping techniques.
FlexGS: Train Once, Deploy Everywhere with Many-in-One Flexible 3D Gaussian Splatting
Hengyu Liu (Chinese University of Hong Kong), Zhangyang Wang (Chinese University of Hong Kong)
CompressionGaussian SplattingPoint Cloud
🎯 What it does: This paper proposes FlexGS, which utilizes a learnable Gaussian selector and transformation module to achieve elastic inference of 3D Gaussian Splatting within a single model, allowing for dynamic rendering at any compression ratio without the need for post-tuning.
Flexible Frame Selection for Efficient Video Reasoning
Shyamal Buch (Google DeepMind), Cordelia Schmid (Google DeepMind)
Computational EfficiencyRepresentation LearningTransformerVision Language ModelVideoMultimodality
🎯 What it does: This paper proposes the Flexible Frame Selector (FFS), a learnable frame selection strategy designed to efficiently select the most informative frames for video question-answering models.
Flexible Group Count Enables Hassle-Free Structured Pruning
Jiamu Zhang (Rice University), Vipin Chaudhary (Case Western Reserve University)
CompressionOptimizationConvolutional Neural NetworkImage
🎯 What it does: A post-training, one-shot, data-independent structured pruning method named LeanFlex-GKP is proposed, which achieves flexible pruning by utilizing a variable number of convolution groups.
FlexiDiT: Your Diffusion Transformer Can Easily Generate High-Quality Samples with Less Compute
Sotiris Anagnostidis (Meta GenAI), Edgar Schönfeld (Meta GenAI)
GenerationComputational EfficiencyTransformerDiffusion modelImageVideoText
🎯 What it does: This paper proposes FlexiDiT, a Diffusion Transformer that uses variable patch sizes in different denoising steps, allowing for dynamic allocation of computational resources while significantly reducing FLOPs without compromising image quality.
FlexUOD: The Answer to Real-world Unsupervised Image Outlier Detection
Zhonghang Liu (Singapore Management University), Jiangbo Lu (SmartMore Corporation)
Anomaly DetectionImageTabularMagnetic Resonance Imaging
🎯 What it does: The FlexUOD framework is proposed to automatically estimate contamination factors during unsupervised image anomaly detection and select appropriate anomaly detection methods based on the signal-to-noise ratio.
FlipSketch: Flipping Static Drawings to Text-Guided Sketch Animations
Hmrishav Bandyopadhyay (University of Surrey), Yi-Zhe Song (University of Surrey)
GenerationData SynthesisSupervised Fine-TuningDiffusion modelImageVideoText
🎯 What it does: Proposes the FlipSketch system, which utilizes a text-guided T2V generation model to transform static sketches into free animations.
Floating No More: Object-Ground Reconstruction from a Single Image
Yunze Man (University of Illinois Urbana-Champaign), Yu-Xiong Wang (University of Illinois Urbana-Champaign)
Object DetectionDepth EstimationTransformerImagePoint Cloud
🎯 What it does: Reconstruct the 3D shape of an object, its relative position to the ground, and the camera parameters simultaneously from a single image.
Florence-VL: Enhancing Vision-Language Models with Generative Vision Encoder and Depth-Breadth Fusion
Jiuhai Chen (University of Maryland), Bin Xiao (Microsoft Research)
RecognitionGenerationRetrievalTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: This paper presents Florence-VL, a novel multimodal large language model that uses the Florence-2 generative visual foundation model as a visual encoder and projects the fused visual features from multiple levels and different task prompts to the LLM through Depth-Breadth Fusion (DBFusion).
FloVD: Optical Flow Meets Video Diffusion Model for Enhanced Camera-Controlled Video Synthesis
Wonjoon Jin (Microsoft Research), Sunghyun Cho (Microsoft Research)
GenerationData SynthesisDiffusion modelOptical FlowVideo
🎯 What it does: This paper proposes FloVD, a two-stage video diffusion framework that utilizes optical flow for controllable camera video generation.
Flow-NeRF: Joint Learning of Geometry, Poses, and Dense Flow within Unified Neural Representations
Xunzhi Zheng (Hong Kong University of Science and Technology), Dan Xu (Hong Kong University of Science and Technology)
GenerationPose EstimationDepth EstimationNeural Radiance FieldOptical FlowVideo
🎯 What it does: This paper proposes Flow-NeRF, which can simultaneously learn scene geometry, camera poses, and dense optical flow without pose priors, and can infer long-range optical flow between any two frames.
Flowing from Words to Pixels: A Noise-Free Framework for Cross-Modality Evolution
Qihao Liu (Meta), Mannat Singh
Image TranslationGenerationData SynthesisTransformerFlow-based ModelImageTextMultimodality
🎯 What it does: The CrossFlow framework is proposed, which directly maps one modality to another (e.g., text to image) without the need for noise or cross-attention conditioning;
FlowRAM: Grounding Flow Matching Policy with Region-Aware Mamba Framework for Robotic Manipulation
Sen Wang (Xi'an Jiaotong University), Wei Tang (University of Illinois at Chicago)
Robotic IntelligenceReinforcement LearningPoint Cloud
🎯 What it does: Proposes the FlowRAM framework, which combines 3D region awareness and conditional flow matching to achieve high-precision robotic manipulation;
Floxels: Fast Unsupervised Voxel Based Scene Flow Estimation
David T. Hoffmann (Robert Bosch GmbH), Martin Meinke (Robert Bosch GmbH)
Autonomous DrivingOptimizationOptical FlowPoint Cloud
🎯 What it does: An unsupervised scene flow estimation method based on voxel grids (Floxels) is proposed for dynamic scene flow inference from LiDAR point cloud sequences.
FluidNexus: 3D Fluid Reconstruction and Prediction from a Single Video
Yue Gao (Stanford University), Jiajun Wu (Stanford University)
GenerationData SynthesisOptimizationDiffusion modelVideo
🎯 What it does: From a single video input, reconstruct and predict the appearance and velocity of three-dimensional fluid by generating multi-view videos and combining them with differentiable physical simulation.
FluxSpace: Disentangled Semantic Editing in Rectified Flow Models
Yusuf Dalva (Virginia Tech), Pinar Yanardag (Virginia Tech)
GenerationData SynthesisTransformerFlow-based ModelRectified FlowImageText
🎯 What it does: This paper proposes FluxSpace, a method for text-guided fine-grained and coarse-grained image editing that utilizes the linear output of the attention layer in the Rectified Flow Transformer (Flux) without requiring additional training.
Focal Split: Untethered Snapshot Depth from Differential Defocus
Junjie Luo (Purdue University), Qi Guo (Purdue University)
Depth EstimationImage
🎯 What it does: A handheld wireless depth camera named Focal Split is proposed, which utilizes dual sensors and a beam splitter to simultaneously capture two differently focused images within a single shutter and quickly estimates a sparse depth map using differential defocus theory.
Focus-N-Fix: Region-Aware Fine-Tuning for Text-to-Image Generation
Xiaoying Xing (Google Research), Deepak Ramachandran (Google DeepMind)
GenerationReinforcement LearningDiffusion modelImageText
🎯 What it does: A region-aware fine-tuning method called Focus-N-Fix is proposed to correct only the previously problematic areas of images in text-to-image (T2I) generation models.
FOCUS: Knowledge-enhanced Adaptive Visual Compression for Few-shot Whole Slide Image Classification
Zhengrui Guo (Hong Kong University of Science and Technology), Hao Chen (Hong Kong University of Science and Technology)
ClassificationCompressionLarge Language ModelImageBiomedical Data
🎯 What it does: FOCUS is proposed, a knowledge-enhanced adaptive visual compression framework for few-shot whole-slide image classification.
Focusing on Tracks for Online Multi-Object Tracking
Kyujin Shim (Korea Advanced Institute of Science and Technology), Changick Kim (Korea Advanced Institute of Science and Technology)
Object TrackingVideo
🎯 What it does: An online multi-object tracking framework based on 'TrackTrack' is proposed, primarily addressing data association and trajectory initialization issues, which can better utilize all detection boxes (high confidence, low confidence, and NMS deleted boxes) while maintaining trajectory ID continuity.
Foley-Flow: Coordinated Video-to-Audio Generation with Masked Audio-Visual Alignment and Dynamic Conditional Flows
Shentong Mo (Carnegie Mellon University), Yibing Song (Alibaba Group)
GenerationData SynthesisFlow-based ModelAuto EncoderVideoMultimodalityAudio
🎯 What it does: Proposed the Foley-Flow framework to achieve collaborative generation from video to audio.
Font-Agent: Enhancing Font Understanding with Large Language Models
Yingxin Lai (Xiamen University), Shaozi Li (Xiamen University)
RecognitionOptimizationExplainability and InterpretabilityConvolutional Neural NetworkTransformerLarge Language ModelVision Language ModelImageMultimodality
🎯 What it does: This paper proposes Font-Agent, a font understanding framework based on a visual-language model, capable of quality assessment and interpretable question answering for generated fonts.
Forensic Self-Descriptions Are All You Need for Zero-Shot Detection, Open-Set Source Attribution, and Clustering of AI-generated Images
Tai D. Nguyen (Drexel University), Matthew C. Stamm (Drexel University)
Data SynthesisAnomaly DetectionGenerative Adversarial NetworkImage
🎯 What it does: Developed a self-supervised learning-based forensic self-description technique that enables zero-shot detection of AI-generated images, open-set source attribution, and unsupervised clustering without having seen generator samples.
Forensics Adapter: Adapting CLIP for Generalizable Face Forgery Detection
Xinjie Cui (Ocean University of China), Junyu Dong (Ocean University of China)
ClassificationRecognitionTransformerContrastive LearningImage
🎯 What it does: This paper proposes the Forensics Adapter, which combines CLIP with a lightweight adapter tailored for specific task objectives to enhance the generalization ability of facial forgery detection.
Forensics-Bench: A Comprehensive Forgery Detection Benchmark Suite for Large Vision Language Models
Jin Wang (University of Hong Kong), Ping Luo (Shanghai AI Laboratory)
TransformerVision Language ModelGenerative Adversarial NetworkImageVideoTextMultimodalityBenchmark
🎯 What it does: A Forensics-Bench benchmark has been constructed with a scale of 63k samples, covering 112 types of forgery detection, to evaluate the capabilities of large visual language models (LVLM) in forgery detection.
ForestLPR: LiDAR Place Recognition in Forests Attentioning Multiple BEV Density Images
Yanqing Shen (Xi'an Jiaotong University), Nanning Zheng (Xi'an Jiaotong University)
RecognitionRetrievalTransformerPoint Cloud
🎯 What it does: A LiDAR forest scene pose recognition method called ForestLPR is proposed, which is based on multi-layer BEV density maps and Transformers. It can adaptively focus on features at different heights, achieving rotation-invariant global descriptors.
Forming Auxiliary High-confident Instance-level Loss to Promote Learning from Label Proportions
Tianhao Ma (Jilin University), Ximing Li (Jilin University)
ClassificationImage
🎯 What it does: This paper studies how to effectively utilize pseudo-labels for instance-level learning in label-proportion weakly supervised learning (LLP) and proposes a high-confidence instance-level loss method to enhance classification performance.
Fortifying Federated Learning Towards Trustworthiness via Auditable Data Valuation and Verifiable Client Contribution
K Naveen Kumar (Indian Institute of Technology Hyderabad), Ravindra Babu Tallamraju (Sahaj AI Software Private Limited)
Federated LearningSafty and PrivacyBiomedical Data
🎯 What it does: The FAVD framework is proposed, utilizing local and global data density functions to achieve auditable data value assessment and verifiable client contributions in federated learning.
Foundations of the Theory of Performance-Based Ranking
Sébastien Piérard (Montefiore Institute, University of Liège), Marc Van Droogenbroeck (Montefiore Institute, University of Liège)
🎯 What it does: A unified theoretical framework and axiomatic definition are proposed for evaluating and ranking the performance of algorithms, devices, and other entities.
FoundationStereo: Zero-Shot Stereo Matching
Bowen Wen (NVIDIA), Stan Birchfield (NVIDIA)
Depth EstimationDomain AdaptationConvolutional Neural NetworkTransformerImage
🎯 What it does: A FoundationStereo model for zero-shot stereo matching is proposed, capable of directly inferring depth maps across different domains without fine-tuning.
FoundHand: Large-Scale Domain-Specific Learning for Controllable Hand Image Generation
Kefan Chen (Brown University), Srinath Sridhar (Brown University)
GenerationPose EstimationDomain AdaptationDiffusion modelImageVideo
🎯 What it does: A controllable image generation model for hands, FoundHand, has been developed, which can achieve hand pose and viewpoint control based on 2D keypoint heatmaps, and supports functions such as pose transformation, domain transfer, NVS, and video generation.
Foveated Instance Segmentation
Hongyi Zeng (New York University), Sai Qian Zhang (New York University)
Object DetectionSegmentationComputational EfficiencyConvolutional Neural NetworkGaussian SplattingImage
🎯 What it does: This work proposes the FovealSeg framework based on real-time user gaze information, performing instance segmentation only on the interest instances that the user is gazing at, significantly reducing computational load.
Fractal Calibration for Long-tailed Object Detection
Konstantinos Panagiotis Alexandridis (Huawei Noah's Ark Lab), Shan Luo (King's College London)
Object DetectionImage
🎯 What it does: This paper proposes a post-processing calibration method called FRACAL for long-tail object detection.
FRAME: Floor-aligned Representation for Avatar Motion from Egocentric Video
Andrea Boscolo Camiletto (Max Planck Institute for Informatics), Christian Theobalt (Max Planck Institute for Informatics)
Pose EstimationTransformerSimultaneous Localization and MappingVideo
🎯 What it does: A low-latency full-body motion capture method that combines VR headset pose tracking with a binocular self-view camera is proposed.
FRAMES-VQA: Benchmarking Fine-Tuning Robustness across Multi-Modal Shifts in Visual Question Answering
Chengyue Huang (Georgia Institute of Technology), Zsolt Kira (Georgia Institute of Technology)
Domain AdaptationRepresentation LearningTransformerSupervised Fine-TuningVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: For the visual question answering (VQA) task, we propose the FRAMES-VQA benchmark system, which conducts multi-modal distribution shift labeling on ten existing VQA datasets and systematically evaluates the performance of different robust fine-tuning methods.
Free Lunch Enhancements for Multi-modal Crowd Counting
Haoliang Meng (Harbin Institute of Technology), Miao Shang (Harbin Institute of Technology)
RecognitionObject DetectionTransformerContrastive LearningMultimodality
🎯 What it does: In the multimodal crowd counting task, the authors propose an enhancement strategy that does not require additional data, parameters, or inference time consumption, improving existing models through post-pretraining cross-modal alignment (PPCA) and regional density supervision (RDS).
Free on the Fly: Enhancing Flexibility in Test-Time Adaptation with Online EM
Qiyuan Dai (ShanghaiTech University), Sibei Yang (Sun Yat-sen University)
ClassificationDomain AdaptationTransformerVision Language ModelImage
🎯 What it does: This paper proposes FreeTTA, a completely untrained and history-free online EM method to adaptively adjust the target distribution of visual-language models during testing and improve classification performance.
Free-viewpoint Human Animation with Pose-correlated Reference Selection
Fa-Ting Hong (Hong Kong University of Science and Technology), Dan Xu (Hong Kong University of Science and Technology)
GenerationData SynthesisPose EstimationTransformerDiffusion modelVideo
🎯 What it does: This paper proposes an adaptive reference selection diffusion network that utilizes multi-view reference images combined with pose correlation guidance to achieve human animation synthesis from free viewpoints.
Free360: Layered Gaussian Splatting for Unbounded 360-Degree View Synthesis from Extremely Sparse and Unposed Views
Chong Bao (Zhejiang University), Zhaopeng Cui
RestorationGenerationData SynthesisDiffusion modelGaussian SplattingImageVideo
🎯 What it does: A reconstruction and view synthesis framework suitable for extremely sparse, unposed 360° scenes is proposed.
FreeCloth: Free-form Generation Enhances Challenging Clothed Human Modeling
Hang Ye (Peking University), Yizhou Wang (Peking University)
SegmentationGenerationPoint Cloud
🎯 What it does: A hybrid framework named FreeCloth is proposed, combining LBS (Linear Blend Skinning) deformation with free-form generation, modeling for close-to-body, loose clothing, and exposed areas, thereby achieving high-quality animation of complex garments.
FreeGave: 3D Physics Learning from Dynamic Videos by Gaussian Velocity
Jinxi Li (Hong Kong Polytechnic University), Bo Yang (Hong Kong Polytechnic University)
Gaussian SplattingVideoPhysics Related
🎯 What it does: We propose FreeGave, a framework that directly learns the geometry, appearance, and physical motion (velocity field) of dynamic 3D scenes from multi-view videos.
FreePCA: Integrating Consistency Information across Long-short Frames in Training-free Long Video Generation via Principal Component Analysis
Jiangtong Tan (University of Science and Technology of China), Feng Zhao (University of Science and Technology of China)
GenerationData SynthesisDiffusion modelVideo
🎯 What it does: FreePCA is proposed, a training-free long video generation framework that decouples temporal feature dimensions through PCA, separating consistent appearance features from motion intensity features, and gradually fusing them to balance consistency and quality.
FreeScene: Mixed Graph Diffusion for 3D Scene Synthesis from Free Prompts
Tongyuan Bai (Jilin University), Rui Ma (Jilin University)
GenerationData SynthesisGraph Neural NetworkTransformerLarge Language ModelVision Language ModelDiffusion modelMultimodalityGraphChain-of-Thought
🎯 What it does: This study investigates how users can control indoor scene generation through free text or image input, proposing the FreeScene framework, which includes the Graph Designer and the MG-DiT model, to achieve multimodal controllable 3D scene synthesis.
FreeSim: Toward Free-viewpoint Camera Simulation in Driving Scenes
Lue Fan (Chinese Academy of Sciences), Zhaoxiang Zhang (Chinese Academy of Sciences)
GenerationData SynthesisAutonomous DrivingDiffusion modelGaussian SplattingPoint Cloud
🎯 What it does: This paper proposes FreeSim, which combines 3D Gaussian Splatting and diffusion models to achieve free-viewpoint camera simulation in driving scenes, capable of generating high-quality views that are more than 3 meters away from the original driving trajectory.
FreeTimeGS: Free Gaussian Primitives at Anytime Anywhere for Dynamic Scene Reconstruction
Yifan Wang (Zhejiang University), Xiaowei Zhou (Zhejiang University)
Gaussian SplattingVideo
🎯 What it does: A 4D Gaussian primitive representation that can appear at any time and location (FreeTimeGS) is proposed, with each primitive assigned a learnable motion function and temporal transparency function to achieve high-quality real-time view synthesis for complex dynamic scenes.
FreeUV: Ground-Truth-Free Realistic Facial UV Texture Recovery via Cross-Assembly Inference Strategy
Xingchao Yang (CyberAgent), Yoshihiro Kanamori (University of Tsukuba)
RestorationGenerationDiffusion modelImage
🎯 What it does: This paper proposes a method to recover high-quality 3D UV textures from a single facial image without the need for UV ground truth supervision.
FreqDebias: Towards Generalizable Deepfake Detection via Consistency-Driven Frequency Debiasing
Hossein Kashiani (Clemson University), Fatemeh Afghah (Clemson University)
ClassificationAnomaly DetectionConvolutional Neural NetworkGenerative Adversarial NetworkImageVideo
🎯 What it does: This paper proposes a frequency domain debiasing framework named FreqDebias, aimed at addressing the 'spectral bias' of deepfake detection models in the frequency domain, thereby enhancing the model's generalization ability on unknown types of forgery.
Frequency Dynamic Convolution for Dense Image Prediction
Linwei Chen (Beijing Institute of Technology), Ying Fu (Beijing Institute of Technology)
Object DetectionSegmentationConvolutional Neural NetworkImage
🎯 What it does: A frequency dynamic convolution (FDConv) module is designed and implemented to enhance the adaptability of visual tasks to different frequency information while keeping the parameter overhead low.
Frequency-Biased Synergistic Design for Image Compression and Compensation
Jiaming Liu (Fudan University), Yibo Fan (Fudan University)
RestorationObject DetectionCompressionConvolutional Neural NetworkImage
🎯 What it does: To address the distortion of compressed images, a 'Low-High Frequency Collaborative' framework is proposed, which modifies the quantization process during the compression phase to prioritize low-frequency compression, and implements dual-step compensation using a low-frequency regression network (combined with a basic attention block) and a high-frequency generation network during the compensation phase.
FRESA: Feedforward Reconstruction of Personalized Skinned Avatars from Few Images
Rong Wang (Australian National University), Yaser Sheikh (Meta Reality Labs Research)
SegmentationGenerationPose EstimationConvolutional Neural NetworkImageMesh
🎯 What it does: The FRESA method is proposed, which utilizes a small number of snapshot images to achieve real-time reconstruction and animation inference of personalized 3D skinned avatars.
From Alexnet to Transformers: Measuring the Non-linearity of Deep Neural Networks with Affine Optimal Transport
Quentin Bouniot (Telecom Paris), Samuel Kaski (Aalto University)
ClassificationRecognitionConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes a nonlinear metric based on optimal transport—Affinity Score—and uses it to define the nonlinear signature of deep networks, studying the nonlinear evolution of different computer vision models.
From Elements to Design: A Layered Approach for Automatic Graphic Design Composition
Jiawei Lin (Xi'an Jiaotong University), Jiang Bian (Microsoft Research)
GenerationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelMultimodality
🎯 What it does: A framework for automatic graphic design generation called LaDeCo is proposed, which can plan and generate layouts and attributes layer by layer according to the semantic hierarchy of multimodal elements provided by users, ultimately synthesizing high-quality designs.
From Faces to Voices: Learning Hierarchical Representations for High-quality Video-to-Speech
Ji-Hoon Kim (Korea Advanced Institute of Science and Technology), Joon Son Chung (Korea Advanced Institute of Science and Technology)
GenerationData SynthesisRepresentation LearningTransformerFlow-based ModelVideoAudio
🎯 What it does: This study investigates a system that generates high-quality speech from silent talking videos, utilizing hierarchical visual encoding and flow-matching decoders for audio generation.