CVPR 2024 Papers — Page 9
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2716 papers
Exploiting Inter-sample and Inter-feature Relations in Dataset Distillation
Wenxiao Deng (Nanjing University), Yang Gao
Data SynthesisKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a dataset distillation method based on distribution matching, introducing class centering constraints and covariance matching constraints to enhance the class distinguishability and distribution matching accuracy of synthetic data.
Exploiting Style Latent Flows for Generalizing Deepfake Video Detection
Jongwook Choi (Chung-Ang University), Jongwon Choi (Chung-Ang University)
ClassificationAnomaly DetectionRecurrent Neural NetworkTransformerGenerative Adversarial NetworkContrastive LearningVideo
🎯 What it does: A detection framework for Deepfake videos is constructed by extracting the temporal flow of StyleGAN style latent vectors from video frames, combined with StyleGRU, attention fusion, and Transformer encoding.
Exploring Efficient Asymmetric Blind-Spots for Self-Supervised Denoising in Real-World Scenarios
Shiyan Chen (Peking University), Tiejun Huang (Peking University)
RestorationKnowledge DistillationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A self-supervised denoising method AT-BSN is proposed, which learns denoising at the original resolution using adjustable asynchronous blind spots.
Exploring Orthogonality in Open World Object Detection
Zhicheng Sun (Peking University), Yadong Mu (Peking University)
Object DetectionImage
🎯 What it does: In the open-world object detection task, a multi-layer orthogonalization method (feature orthogonalization, prediction orthogonalization, and cross-task calibration) is proposed to achieve the decoupling of objectness and category recognition, supporting incremental learning.
Exploring Pose-Aware Human-Object Interaction via Hybrid Learning
Eastman Z Y Wu (Tsinghua University), Shengjin Wang (Tsinghua University)
Object DetectionPose EstimationTransformerSupervised Fine-TuningImage
🎯 What it does: Two methods, Pose-Aware Feature Refinement and Hybrid Learning, are proposed to improve the two-stage human-object interaction detection model.
Exploring Region-Word Alignment in Built-in Detector for Open-Vocabulary Object Detection
Heng Zhang (JD.com), Sulong Xu (JD.com)
Object DetectionTransformerVision Language ModelImageText
🎯 What it does: This paper proposes an end-to-end 'Built-IN Detector' named BIND, which learns region-word alignment in image-text pairs through a dual encoder, and then uses a DETR-style decoder to complete open vocabulary object detection, avoiding reliance on external detectors or knowledge transfer as in traditional methods.
Exploring Regional Clues in CLIP for Zero-Shot Semantic Segmentation
Yi Zhang (Beihang University), Shi-Min Hu (Tsinghua University)
SegmentationTransformerVision Language ModelContrastive LearningImage
🎯 What it does: This paper proposes a CLIP-based single-stage zero-shot semantic segmentation framework called CLIP-RC, which utilizes region-level bridging and a recovery decoder to achieve the transfer from image-level knowledge to pixel-level semantics.
Exploring the Potential of Large Foundation Models for Open-Vocabulary HOI Detection
Ting Lei (Peking University), Yang Liu (Peking University)
RecognitionObject DetectionTransformerLarge Language ModelVision Language ModelImageMultimodality
🎯 What it does: A human-computer interaction detection framework CMD-SE is proposed for open vocabulary, utilizing multi-layer visual feature conditional matching and fine-grained semantic enhancement of body part descriptions generated by LLM to achieve interaction recognition.
Exploring the Transferability of Visual Prompting for Multimodal Large Language Models
Yichi Zhang (Tsinghua University), Jun Zhu (Tsinghua University)
Large Language ModelPrompt EngineeringVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: This paper proposes Transferable Visual Prompting (TVP), which trains visual prompts on a single multimodal large model and transfers them to other models to enhance downstream task performance.
Exploring Vision Transformers for 3D Human Motion-Language Models with Motion Patches
Qing Yu (LY Corporation), Kent Fujiwara (LY Corporation)
RetrievalTransformerContrastive LearningVideoTextMultimodality
🎯 What it does: This paper proposes to transform 3D human motion sequences into 'motion patches' and uses a pre-trained Vision Transformer as a motion encoder to construct a cross-modal embedding space for motion and language.
ExtDM: Distribution Extrapolation Diffusion Model for Video Prediction
Zhicheng Zhang (Nankai University), Jufeng Yang (Nankai University)
GenerationComputational EfficiencyDiffusion modelAuto EncoderOptical FlowVideo
🎯 What it does: By decomposing the video into motion cues and appearance information, the diffusion model is used to extrapolate the motion distribution, and future frames are generated using a sparse spatiotemporal window U-Net;
Extend Your Own Correspondences: Unsupervised Distant Point Cloud Registration by Progressive Distance Extension
Quan Liu (Shanghai Jiao Tong University), Minyi Guo (Shanghai Jiao Tong University)
Autonomous DrivingContrastive LearningPoint Cloud
🎯 What it does: A completely unsupervised long-range point cloud registration method called EYOC is proposed, which utilizes continuous LiDAR sequences to adaptively learn features and achieve registration through self-labeling.
ExtraNeRF: Visibility-Aware View Extrapolation of Neural Radiance Fields with Diffusion Models
Meng-Li Shih (University of Washington), Janne Kontkanen (Google Research)
GenerationData SynthesisDiffusion modelNeural Radiance FieldImage
🎯 What it does: A perspective extrapolation method based on NeRF, called ExtraNeRF, is proposed, which achieves high-quality synthesis of discrete views through visibility awareness and diffusion models.
Extreme Point Supervised Instance Segmentation
Hyeonjun Lee (Lunit Inc.), Suha Kwak (POSTECH)
Object DetectionSegmentationTransformerImage
🎯 What it does: This paper proposes a weakly supervised instance segmentation method using extreme points (the topmost, leftmost, bottommost, and rightmost points of the target). First, it trains a pseudo-label generator using extreme points, and then uses the generated high-quality pseudo-masks to train a conventional instance segmentation network.
Eyes Wide Shut? Exploring the Visual Shortcomings of Multimodal LLMs
Shengbang Tong (New York University), Saining Xie (New York University)
RecognitionRetrievalTransformerLarge Language ModelVision Language ModelContrastive LearningImageMultimodalityBenchmark
🎯 What it does: This paper systematically evaluates and reveals the shortcomings of multimodal large language models (MLLMs) in visual detail recognition by constructing CLIP-blind pairs and designing the MMVP benchmark.
F3Loc: Fusion and Filtering for Floorplan Localization
Changan Chen (ETH Zurich), Marc Pollefeys (Microsoft)
Pose EstimationDepth EstimationOptimizationConvolutional Neural NetworkSimultaneous Localization and MappingImage
🎯 What it does: This paper proposes a self-localization framework for indoor floor plans based on single-frame and multi-view depth estimation, and implements high-frequency sequential localization using an SE(2) histogram filter.
Face2Diffusion for Fast and Editable Face Personalization
Kaede Shiohara (University of Tokyo), Toshihiko Yamasaki (University of Tokyo)
Image TranslationGenerationDiffusion modelImageVideo
🎯 What it does: This paper proposes Face2Diffusion, which achieves high editability for facial personalization under the condition of fine-tuning without testing, using a multi-scale identity encoder, expression guidance, and class-guided denoising regularization.
FaceChain-ImagineID: Freely Crafting High-Fidelity Diverse Talking Faces from Disentangled Audio
Chao Xu (Alibaba Group), Baigui Sun (Alibaba Group)
GenerationData SynthesisTransformerDiffusion modelContrastive LearningVideoAudio
🎯 What it does: This paper proposes the Listening and Imagining framework, which generates diverse and coherent talking videos by progressively disentangling identity, content, and emotion from audio.
FaceChain-SuDe: Building Derived Class to Inherit Category Attributes for One-shot Subject-Driven Generation
Pengchong Qiao (Peking University), Jie Chen (Peking University)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: This paper proposes modeling the subject as a derived class of its semantic category, utilizing the inheritance concept from object-oriented programming to allow the subject to inherit the public attributes of the category while maintaining its private attributes, thereby enhancing the attribute-related text generation effect for a single example image.
FaceCom: Towards High-fidelity 3D Facial Shape Completion via Optimization and Inpainting Guidance
Yinglong Li (Beihang University), Aimin Hao (Beihang University)
RestorationOptimizationGraph Neural NetworkAuto EncoderImagePoint CloudMesh
🎯 What it does: Designed and implemented the FaceCom method for high-fidelity shape completion of incomplete 3D facial scans.
FaceLift: Semi-supervised 3D Facial Landmark Localization
David Ferman (Flawless AI), Gaurav Bharaj (Flawless AI)
GenerationPose EstimationTransformerGenerative Adversarial NetworkImageVideo
🎯 What it does: This paper proposes an unsupervised 3D facial keypoint localization method that achieves precise localization of 3D keypoints using semi-supervised learning with 2D human annotations and 3D-agnostic GAN priors.
Faces that Speak: Jointly Synthesising Talking Face and Speech from Text
Youngjoon Jang, Joon Son Chung
GenerationData SynthesisAuto EncoderGenerative Adversarial NetworkVideoTextMultimodalityAudio
🎯 What it does: A unified text-driven multimodal synthesis system has been constructed, capable of simultaneously generating natural speaking facial videos and corresponding audio.
FaceTalk: Audio-Driven Motion Diffusion for Neural Parametric Head Models
Shivangi Aneja (Technical University of Munich), Matthias Nießner (Technical University of Munich)
GenerationTransformerDiffusion modelVideoAudio
🎯 What it does: FaceTalk has been developed, a method for generating audio-driven neural parametric head model animations based on latent diffusion models, capable of synchronizing mouth movements and generating high-fidelity, temporally coherent 3D facial motion sequences.
Facial Identity Anonymization via Intrinsic and Extrinsic Attention Distraction
Zhenzhong Kuang (Hangzhou Dianzi University), Jun Yu (Harbin Institute of Technology)
RecognitionData SynthesisSafty and PrivacyGenerative Adversarial NetworkImage
🎯 What it does: A facial anonymization method based on identity attention suppression is proposed, utilizing attention dispersion in feature space and visual space to achieve identity information perturbation.
FACT: Frame-Action Cross-Attention Temporal Modeling for Efficient Action Segmentation
Zijia Lu (Northeastern University), Ehsan Elhamifar (Northeastern University)
SegmentationComputational EfficiencyConvolutional Neural NetworkTransformerVideo
🎯 What it does: This paper proposes a method for parallel temporal modeling at both frame-level and action-level in videos, achieving efficient and accurate action segmentation through bidirectional cross-attention for iterative feature updates.
FADES: Fair Disentanglement with Sensitive Relevance
Taeuk Jang (Purdue University), Xiaoqian Wang (Purdue University)
ClassificationGenerationData SynthesisAuto EncoderImage
🎯 What it does: Proposes the FADES method, which utilizes sensitive correlated codes and conditional mutual information to achieve fair disentanglement;
Fair Federated Learning under Domain Skew with Local Consistency and Domain Diversity
Yuhang Chen (Wuhan University), Mang Ye (Wuhan University)
Federated LearningConvolutional Neural NetworkImage
🎯 What it does: The study addresses the fairness issue in federated learning scenarios with domain skew, proposing the FedHEAL framework to alleviate performance unfairness by filtering out unimportant parameters and adjusting aggregation weights.
Fair-VPT: Fair Visual Prompt Tuning for Image Classification
Sungho Park (Yonsei University), Hyeran Byun (Yonsei University)
ClassificationTransformerPrompt EngineeringContrastive LearningImage
🎯 What it does: To address the fairness issues arising from Visual Prompt Tuning (VPT) in downstream classification tasks, we propose the Fair Visual Prompt Tuning (Fair-VPT) method, which first divides the prompts into target prompts and clean prompts, and then achieves the decoupling and debiasing of sensitive attribute information through masked multi-head self-attention dual-branch encoding and contrastive hashing loss.
FairCLIP: Harnessing Fairness in Vision-Language Learning
Yan Luo (Harvard University), Mengyu Wang
ClassificationRecognitionData-Centric LearningTransformerVision Language ModelContrastive LearningImageTextBiomedical Data
🎯 What it does: This study first created the first audiovisual medical dataset, Harvard‑FairVLMed, which includes detailed demographic attributes, clinical texts, and high-quality labels, and conducted a fairness evaluation of the two major vision-language models, CLIP and BLIP2, on this dataset. Subsequently, a FairCLIP method based on optimal transport was proposed, achieving a good balance between performance and fairness by minimizing the Sinkhorn distance.
FairDeDup: Detecting and Mitigating Vision-Language Fairness Disparities in Semantic Dataset Deduplication
Eric Slyman (Oregon State University), Kushal Kafle (Adobe)
Data-Centric LearningTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: To address the deduplication problem in large-scale vision-language datasets, we propose FairDeDup, a deduplication algorithm that incorporates fairness constraints based on SemDeDup.
FairRAG: Fair Human Generation via Fair Retrieval Augmentation
Robik Shrestha (AWS AI Labs), Siqi Deng (Amazon AGI)
GenerationRetrievalDiffusion modelImageRetrieval-Augmented Generation
🎯 What it does: Proposes the FairRAG framework, which enhances the fairness of pre-trained text-to-image models using external reference images.
Fairy: Fast Parallelized Instruction-Guided Video-to-Video Synthesis
Bichen Wu (Meta), Peter Vajda (Meta)
GenerationData SynthesisComputational EfficiencyDiffusion modelVideo
🎯 What it does: This paper presents Fairy, a video-to-video editing framework based on anchor-based cross-frame attention, which achieves efficient parallel generation while maintaining high quality.
FakeInversion: Learning to Detect Images from Unseen Text-to-Image Models by Inverting Stable Diffusion
George Cazenavette (Massachusetts Institute of Technology), Ben Usman (Google Research)
ClassificationObject DetectionRetrievalConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: This paper proposes a synthetic image detector that uses noise images and reconstructed images obtained from the DDIM inversion of Stable Diffusion, along with the original images as input.
Fantastic Animals and Where to Find Them: Segment Any Marine Animal with Dual SAM
Pingping Zhang (Dalian University of Technology), Huchuan Lu (Dalian University of Technology)
SegmentationDomain AdaptationTransformerSupervised Fine-TuningImage
🎯 What it does: This paper proposes the Dual-SAM framework, designed specifically for marine animal segmentation tasks, using a dual-branch SAM encoder, cross-layer adapters, multi-layer coupled prompts, dilated fusion attention, and cross-connected prediction.
FAR: Flexible Accurate and Robust 6DoF Relative Camera Pose Estimation
Chris Rockwell (University of Michigan), David F. Fouhey (New York University)
Pose EstimationTransformerImage
🎯 What it does: Combining deep learning with traditional solvers to achieve accurate and robust 6DoF camera pose estimation under varying matching quality;
Fast Adaptation for Human Pose Estimation via Meta-Optimization
Shengxiang Hu (Nanjing University of Science and Technology), Jianfeng Lu (Tianjin AiForward Science and Technology Co., Ltd.)
Pose EstimationDomain AdaptationMeta LearningImageVideo
🎯 What it does: A fast adaptive method for testing based on meta-assisted learning, MeTTA, is proposed, utilizing body-specific image inpainting as a self-supervised auxiliary task, performing a small number of gradient updates on human pose estimation models during inference to address domain transfer.
Fast ODE-based Sampling for Diffusion Models in Around 5 Steps
Zhenyu Zhou (Zhejiang University), Chun Chen (Zhejiang University)
GenerationData SynthesisComputational EfficiencyKnowledge DistillationDiffusion modelImageOrdinary Differential Equation
🎯 What it does: A single-step ODE solver (AMED-Solver) and its plugin are proposed, utilizing the geometric properties of sampling trajectories that almost lie in a two-dimensional subspace to eliminate truncation errors by learning the mean direction, thus achieving high-quality image generation with only about 5 function evaluations (NFE).
FastMAC: Stochastic Spectral Sampling of Correspondence Graph
Yifei Zhang (University of Chinese Academy of Sciences), Siheng Chen (Shanghai Jiao Tong University)
Autonomous DrivingOptimizationComputational EfficiencyGraph Neural NetworkPoint Cloud
🎯 What it does: Using graph signal processing and random spectral sampling on 3D correspondence graphs, the FastMAC algorithm is proposed, significantly accelerating the registration of the maximum clique (MAC) to real-time levels while maintaining a high registration success rate.
FC-GNN: Recovering Reliable and Accurate Correspondences from Interferences
Haobo Xu (Shanghai Jiao Tong University), Cunyan Li (Shanghai Jiao Tong University)
RecognitionImage TranslationPose EstimationGraph Neural NetworkImage
🎯 What it does: To address the issues of noise and keypoint errors in sparse feature matching, we propose FC-GNN, a graph neural network capable of jointly filtering out anomalous matches and calibrating the precision of the remaining matches.
FCS: Feature Calibration and Separation for Non-Exemplar Class Incremental Learning
Qiwei Li (Peking University), Jiahuan Zhou (Peking University)
ClassificationKnowledge DistillationContrastive LearningImage
🎯 What it does: The Feature Calibration and Separation (FCS) method is proposed for No-Example Class Incremental Learning (NECIL), which alleviates catastrophic forgetting by calibrating old class prototypes and enhancing feature separation between old and new classes.
Feature 3DGS: Supercharging 3D Gaussian Splatting to Enable Distilled Feature Fields
Shijie Zhou (University of California), Achuta Kadambi (University of California)
SegmentationKnowledge DistillationRepresentation LearningGaussian SplattingPoint Cloud
🎯 What it does: Explicit modeling of the scene using 3D Gaussian splatting, and feature field distillation through 2D large foundation models (such as SAM, CLIP‑LSeg), achieving various functions from view synthesis to semantic segmentation, language-guided editing, and prompt-based instance segmentation.
Feature Re-Embedding: Towards Foundation Model-Level Performance in Computational Pathology
Wenhao Tang (Chongqing University), Bo Liu (Walmart Global Tech)
ClassificationRecognitionTransformerBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper proposes an online instance feature re-embedding framework that utilizes the Re-embedded Regional Transformer (R2 Transformer) to re-embed offline extracted instance features in the multi-instance learning (MIL) process, significantly improving the performance of computational pathology tasks.
FedAS: Bridging Inconsistency in Personalized Federated Learning
Xiyuan Yang (Wuhan University), Mang Ye (Wuhan University)
ClassificationFederated LearningImage
🎯 What it does: Proposes the FedAS framework to address the intra-client and inter-client inconsistency issues in personalized federated learning, enhancing the collaborative training of localized shared parameters and personalized heads.
Federated Generalized Category Discovery
Nan Pu (University of Trento), Zhun Zhong (University of Nottingham)
Federated LearningContrastive LearningImageBenchmark
🎯 What it does: Proposed the Fed-GCD task and designed the AGCL framework, combining federated learning with learnable GMM to achieve universal category discovery without shared data.
Federated Online Adaptation for Deep Stereo
Matteo Poggi (University of Bologna), Fabio Tosi (University of Bologna)
Depth EstimationDomain AdaptationAutonomous DrivingFederated LearningConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A deep stereo matching online adaptive framework based on federated learning is proposed, enabling resource-constrained devices to collaboratively achieve adaptation through the cloud, thereby improving accuracy and maintaining real-time performance in harsh environments.
FedHCA2: Towards Hetero-Client Federated Multi-Task Learning
Yuxiang Lu (Shanghai Jiao Tong University), Hongtao Lu (Shanghai Jiao Tong University)
SegmentationFederated LearningTransformerImage
🎯 What it does: A new scenario of Heterogeneous Client Federated Multi-Task Learning (HC-FMTL) is proposed, and the FedHCA2 framework is designed to achieve personalized model federated training for clients under different task settings.
FedMef: Towards Memory-efficient Federated Dynamic Pruning
Hong Huang (City University of Hong Kong), Lingjuan Lyu (Sony AI)
Federated LearningComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: This paper studies a framework called FedMef for achieving memory-efficient dynamic pruning in cross-device federated learning.
FedSelect: Personalized Federated Learning with Customized Selection of Parameters for Fine-Tuning
Rishub Tamirisa (University of Illinois Urbana-Champaign), Aviv Shamsian (Bar-Ilan University)
Federated LearningImage
🎯 What it does: An algorithm called FedSelect is proposed for adaptive selection of personalized sub-networks in federated learning, which can gradually expand the personalized parameters of each client while maintaining global sharing.
FedSOL: Stabilized Orthogonal Learning with Proximal Restrictions in Federated Learning
Gihun Lee (KAIST), Se-Young Yun (KAIST)
ClassificationFederated LearningImage
🎯 What it does: This paper proposes a federated learning method called FedSOL, which promotes local learning while ensuring global knowledge retention by perturbing model parameters to align with the proximal target gradient during local training.
FedUV: Uniformity and Variance for Heterogeneous Federated Learning
Ha Min Son (University of California), Xin Liu (University of California)
Federated LearningImage
🎯 What it does: This paper proposes a FedUV method that approximates IID distribution in federated learning by regularizing the variance of the final classifier and the spherical uniformity of feature encoding.
Feedback-Guided Autonomous Driving
Jimuyang Zhang (Boston University), Eshed Ohn-Bar (Boston University)
Autonomous DrivingKnowledge DistillationLarge Language ModelVision Language ModelMultimodality
🎯 What it does: FeD is proposed, a feedback-guided end-to-end perception-driven strategy utilizing large-scale multimodal language models, achieving complete driving decision-making under single camera input;
Few-shot Learner Parameterization by Diffusion Time-steps
Zhongqi Yue (Nanyang Technological University), Qianru Sun (Singapore Management University)
ClassificationMeta LearningDiffusion modelImage
🎯 What it does: A few-shot learning method based on diffusion model time steps is proposed, training class-specific low-rank adapters to reconstruct noisy images, thereby achieving debiased classification.
Few-Shot Object Detection with Foundation Models
Guangxing Han (Columbia University), Ser-Nam Lim (University of Central Florida)
Object DetectionTransformerLarge Language ModelImage
🎯 What it does: A few-shot object detection framework based on foundational models (FM-FSOD) is proposed, which combines a frozen DINOv2 visual feature extractor with a large language model (Vicuna) to classify proposals using contextual information.
FFF: Fixing Flawed Foundations in Contrastive Pre-Training Results in Very Strong Vision-Language Models
Adrian Bulat (Samsung AI Center Cambridge), Georgios Tzimiropoulos (Queen Mary University of London)
ClassificationRetrievalTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: A method is proposed to address the issues of incorrect negative sample allocation and low-quality/lack of diversity in explanations in contrastive audiovisual pre-training, significantly improving model performance through multiple positive sample training.
Finding Lottery Tickets in Vision Models via Data-driven Spectral Foresight Pruning
Leonardo Iurada (Politecnico di Torino), Tatiana Tommasi (Politecnico di Torino)
ClassificationSegmentationCompressionConvolutional Neural NetworkImage
🎯 What it does: A Path eXclusion (PX) method based on a path perspective is proposed, which retains the paths that have the greatest impact on training dynamics by calculating the upper bound of the Neural Tangent Kernel (NTK) trajectory, thus achieving model compression at high sparsity.
Fine-Grained Bipartite Concept Factorization for Clustering
Chong Peng (Qingdao University), Qiang Cheng (University of Kentucky)
OptimizationRepresentation LearningGraph Neural NetworkImageText
🎯 What it does: A fine-grained bilateral concept decomposition method (Figer-CF) is proposed, utilizing cross-order adjacency information for clustering and data representation.
Fine-grained Prototypical Voting with Heterogeneous Mixup for Semi-supervised 2D-3D Cross-modal Retrieval
Fan Zhang (Georgia Tech Shenzhen Institute Tianjin University), Xiao Luo (University of California Los Angeles)
RetrievalConvolutional Neural NetworkGraph Neural NetworkContrastive LearningImageMultimodalityPoint Cloud
🎯 What it does: This paper proposes a semi-supervised 2D-3D cross-modal retrieval method called FIVE, which generates robust pseudo-labels through fine-grained prototype voting and achieves cross-modal alignment in a shared embedding space using heterogeneous Mixup.
FineParser: A Fine-grained Spatio-temporal Action Parser for Human-centric Action Quality Assessment
Jinglin Xu (University of Science and Technology Beijing), Yuxin Peng (Peking University)
Convolutional Neural NetworkTransformerVideo
🎯 What it does: A FineParser is proposed, capable of fine-grained analysis of human subject actions in the spatiotemporal dimension, and utilizes this analysis for action quality assessment (AQA).
FinePOSE: Fine-Grained Prompt-Driven 3D Human Pose Estimation via Diffusion Models
Jinglin Xu (University of Science and Technology Beijing), Yuxin Peng (Peking University)
Pose EstimationTransformerPrompt EngineeringDiffusion modelImage
🎯 What it does: This paper proposes a FinePOSE framework based on diffusion models, utilizing a fine-grained prompt-driven denoiser to achieve 3D human pose estimation for both single and multiple persons.
FINER: Flexible Spectral-bias Tuning in Implicit NEural Representation by Variable-periodic Activation Functions
Zhen Liu (Nanjing University), Xun Cao (Nanjing University)
Neural Radiance FieldImage
🎯 What it does: An implicit neural representation method named FINER is proposed, utilizing a variable-period activation function to achieve adjustable spectral bias.
FineSports: A Multi-person Hierarchical Sports Video Dataset for Fine-grained Action Understanding
Jinglin Xu (University of Science and Technology Beijing), Yuxin Peng (Peking University)
RecognitionObject DetectionTransformerPrompt EngineeringVideoText
🎯 What it does: A multi-person basketball video dataset FineSports covering 10,000 NBA games has been constructed, and a prompt-based spatiotemporal action localization method called PoSTAL has been proposed.
Finsler-Laplace-Beltrami Operators with Application to Shape Analysis
Simon Weber (Technical University of Munich), Daniel Cremers (Technical University of Munich)
SegmentationConvolutional Neural NetworkMesh
🎯 What it does: This paper proposes a Finsler-Laplace-Beltrami operator (FLBO) based on Finsler manifolds and validates its effectiveness in shape matching tasks.
FISBe: A Real-World Benchmark Dataset for Instance Segmentation of Long-Range Thin Filamentous Structures
Lisa Mais (Max Delbrueck Center for Molecular Medicine in the Helmholtz Association), Dagmar Kainmueller (Max Delbrueck Center for Molecular Medicine in the Helmholtz Association)
Object DetectionSegmentationImageBenchmark
🎯 What it does: This study released the FISBe dataset, which provides the first batch of instance segmentation data for real-world sparse fiber long-range structures.
Fitting Flats to Flats
Gabriel Dogadov (TU Berlin), Marc Alexa (TU Berlin)
OptimizationPoint Cloud
🎯 What it does: This paper proposes a method to solve the least squares problem of 'fitting a set of affine planes of different dimensions to another plane' using squared distance field representation.
Fixed Point Diffusion Models
Xingjian Bai (University of Oxford), Luke Melas-Kyriazi (University of Oxford)
GenerationData SynthesisComputational EfficiencyTransformerDiffusion modelImage
🎯 What it does: Designed the Fixed Point Diffusion Model (FPDM), which introduces implicit fixed point layers into the diffusion model to achieve more efficient image generation.
FlashAvatar: High-fidelity Head Avatar with Efficient Gaussian Embedding
Jun Xiang (University of Science and Technology of China), Juyong Zhang (University of Science and Technology of China)
GenerationData SynthesisComputational EfficiencyGaussian SplattingVideo
🎯 What it does: Quickly construct high-fidelity animated digital head portraits from monocular video, with training taking only a few minutes.
FlashEval: Towards Fast and Accurate Evaluation of Text-to-image Diffusion Generative Models
Lin Zhao (Tsinghua University), Yu Wang (Tsinghua University)
GenerationData SynthesisComputational EfficiencyPrompt EngineeringDiffusion modelImageText
🎯 What it does: In response to the high evaluation cost of text-to-image diffusion models, this paper proposes the FlashEval method, which selects a representative subset from the original prompt set through iterative search and frequency filtering, significantly reducing evaluation time while maintaining high ranking relevance.
Flatten Long-Range Loss Landscapes for Cross-Domain Few-Shot Learning
Yixiong Zou (Huazhong University of Science and Technology), Ruixuan Li (Huazhong University of Science and Technology)
Domain AdaptationMeta LearningImage
🎯 What it does: This paper proposes a method to achieve long-range loss flattening for cross-domain few-shot learning by performing random interpolation on features with different normalization methods in the representation space, thereby enhancing model transfer and fine-tuning performance.
Flattening the Parent Bias: Hierarchical Semantic Segmentation in the Poincare Ball
Simon Weber (Technical University of Munich), Daniel Cremers (Technical University of Munich)
SegmentationDomain AdaptationAutonomous DrivingImage
🎯 What it does: This study investigates the parent bias problem in hierarchical semantic segmentation and verifies the advantages of planar classifiers in cross-domain scenarios. It proposes the use of hyperbolic space (Poincaré ball model) to balance inter-class distances, enhancing the accuracy and calibration quality of parent class predictions.
Flexible Biometrics Recognition: Bridging the Multimodality Gap through Attention Alignment and Prompt Tuning
Leslie Ching Ow Tiong (Samsung Electronics), Andrew Beng Jin Teoh (Yonsei University)
RecognitionTransformerPrompt EngineeringContrastive LearningImageMultimodality
🎯 What it does: A flexible biometric recognition framework (FBR) is proposed, which supports both cross-modal and single-modal recognition of facial, periocular, and soft biometric features simultaneously.
Flexible Depth Completion for Sparse and Varying Point Densities
Jinhyung Park (Carnegie Mellon University), Kris Kitani (Carnegie Mellon University)
Depth EstimationAutonomous DrivingTransformerPoint Cloud
🎯 What it does: An Adaptive Shift Correction (ASC) module based on pixel and depth point affinity is proposed to improve depth completion under sparse and variable point density conditions.
FLHetBench: Benchmarking Device and State Heterogeneity in Federated Learning
Junyuan Zhang (Beihang University), Liangqiong Qu (University of Hong Kong)
Federated LearningImageBenchmark
🎯 What it does: Proposes the FLHetBench benchmark to evaluate FL under device and state heterogeneity;
Florence-2: Advancing a Unified Representation for a Variety of Vision Tasks
Bin Xiao (Microsoft), Lu Yuan (Microsoft)
ClassificationObject DetectionSegmentationTransformerPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: The unified visual foundation model Florence-2 is proposed, capable of outputting results in text form for various tasks such as image classification, detection, segmentation, and annotation through text prompts.
Flow-Guided Online Stereo Rectification for Wide Baseline Stereo
Anush Kumar (Torc Robotics), Felix Heide (Princeton University)
Depth EstimationAutonomous DrivingTransformerOptical FlowImageVideo
🎯 What it does: A method for online real-time calibration of wide baseline stereo cameras is proposed, utilizing cross-attention features and phase correlation to predict the relative rotation of the cameras, and achieving unsupervised calibration through self-supervised vertical optical flow constraints.
FlowDiffuser: Advancing Optical Flow Estimation with Diffusion Models
Ao Luo (Southwest Jiaotong University), Shuaicheng Liu (University of Electronic Science and Technology of China)
GenerationRecurrent Neural NetworkTransformerDiffusion modelOptical FlowImageVideo
🎯 What it does: Transform the optical flow estimation task into a conditional generation task, using a diffusion model to gradually denoise from random noise to obtain the optical flow field.
FlowerFormer: Empowering Neural Architecture Encoding using a Flow-aware Graph Transformer
Dongyeong Hwang (Korea Advanced Institute of Science and Technology), Kijung Shin (Korea Advanced Institute of Science and Technology)
Neural Architecture SearchGraph Neural NetworkTransformerFlow-based ModelGraph
🎯 What it does: A graph transformer named FlowerFormer is proposed for efficiently encoding neural network architectures and predicting their performance.
FlowIE: Efficient Image Enhancement via Rectified Flow
Yixuan Zhu (Tsinghua University), Jiwen Lu (Tsinghua University)
RestorationGenerationSuper ResolutionDiffusion modelRectified FlowImage
🎯 What it does: This paper proposes FlowIE, an efficient image enhancement framework based on rectified flow, which utilizes the generative prior of a pre-trained diffusion model to construct a direct path from noise to clear images, achieving various enhancement tasks in less than 5 steps.
FlowTrack: Revisiting Optical Flow for Long-Range Dense Tracking
Seokju Cho (Korea University), Joon-Young Lee (Adobe Research)
Object TrackingTransformerOptical FlowVideo
🎯 What it does: This paper proposes a long-term dense tracking framework called FlowTrack, which utilizes optical flow chaining and automatically triggers an error compensation module for correction when drift or occlusion occurs.
FlowVid: Taming Imperfect Optical Flows for Consistent Video-to-Video Synthesis
Feng Liang (University of Texas at Austin), Diana Marculescu (Meta GenAI)
GenerationData SynthesisDiffusion modelOptical FlowVideoText
🎯 What it does: A Video-to-Video synthesis framework called FlowVid is proposed, which can stylize, swap objects, and perform local editing on input videos based on given text prompts while maintaining temporal consistency.
FlowVQTalker: High-Quality Emotional Talking Face Generation through Normalizing Flow and Quantization
Shuai Tan (Shanghai Jiao Tong University), Ye Pan (Shanghai Jiao Tong University)
GenerationData SynthesisTransformerFlow-based ModelVideoMultimodalityAudio
🎯 What it does: This paper proposes FlowVQTalker, which utilizes regularized flows and vector quantization techniques to achieve one-to-many emotion-driven facial video generation based on speech.
FMA-Net: Flow-Guided Dynamic Filtering and Iterative Feature Refinement with Multi-Attention for Joint Video Super-Resolution and Deblurring
Geunhyuk Youk (KAIST), Munchurl Kim (KAIST)
RestorationSuper ResolutionTransformerOptical FlowVideo
🎯 What it does: This paper studies a framework called FMA-Net that combines video super-resolution and deblurring, achieving high-quality recovery through flow-guided dynamic filtering and multi-attention iterative feature refinement.
FocSAM: Delving Deeply into Focused Objects in Segmenting Anything
You Huang (Xiamen University), Rongrong Ji (Xiamen University)
Object DetectionSegmentationTransformerImageVideo
🎯 What it does: This paper proposes an improved interactive segmentation framework called FocSAM, which incorporates dynamic window multi-head self-attention and pixel-level dynamic ReLU based on SAM, achieving a focus on target objects and deeper integration of interactive information.
Focus on Hiders: Exploring Hidden Threats for Enhancing Adversarial Training
Qian Li (Shanghai Jiao Tong University), Yuntian Chen (Tsinghua University)
OptimizationAdversarial AttackConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes and implements the concept of 'hider' and develops an adversarial training framework called HFAT, based on auxiliary models and iterative evolutionary optimization, to simultaneously enhance the robustness and accuracy of models.
Focus on Your Instruction: Fine-grained and Multi-instruction Image Editing by Attention Modulation
Qin Guo (Peking University), Tianwei Lin (Horizon Robotics)
GenerationData SynthesisTransformerDiffusion modelImage
🎯 What it does: Proposes the FoI method, achieving precise multi-instruction image editing without additional training based on IP2P;
FocusMAE: Gallbladder Cancer Detection from Ultrasound Videos with Focused Masked Autoencoders
Soumen Basu (Indian Institute of Technology Delhi), Chetan Arora (Postgraduate Institute of Medical Education and Research)
ClassificationObject DetectionTransformerAuto EncoderVideoBiomedical DataUltrasound
🎯 What it does: This study proposes a video-based gallbladder cancer (GBC) detection method called FocusMAE, which utilizes a Masked Autoencoder (MAE) to learn representations of malignant lesions in ultrasound videos, breaking through the limitations of traditional single-frame image methods.
Fooling Polarization-Based Vision using Locally Controllable Polarizing Projection
Zhuoxiao Li (University of Tokyo), Yinqiang Zheng (Kyoto University)
SegmentationAdversarial AttackImage
🎯 What it does: A physical attack method utilizing locally controllable polarization projection has been designed and implemented, targeting polarization-based visual algorithms.
Forecasting of 3D Whole-body Human Poses with Grasping Objects
Haitao Yan (Fudan University), Shijie Guo (Fudan University)
Pose EstimationGraph Neural NetworkPoint Cloud
🎯 What it does: This paper proposes a cross-context cross-modal integration framework called C HOST for predicting 3D full-body (including hands) poses while considering grasping interactions with objects.
Forgery-aware Adaptive Transformer for Generalizable Synthetic Image Detection
Huan Liu (Beijing Jiaotong University), Yao Zhao (Beijing Jiaotong University)
ClassificationAnomaly DetectionTransformerDiffusion modelGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: This paper proposes an adaptive Transformer model called FatFormer for detecting forged images generated by different models (GANs and diffusion models).
FoundationPose: Unified 6D Pose Estimation and Tracking of Novel Objects
Bowen Wen (NVIDIA), Stan Birchfield (NVIDIA)
Object TrackingPose EstimationTransformerLarge Language ModelContrastive LearningImageVideo
🎯 What it does: A unified FoundationPose framework is proposed, capable of simultaneously completing 6D object pose estimation and tracking in both model-based and model-free settings.
Fourier Priors-Guided Diffusion for Zero-Shot Joint Low-Light Enhancement and Deblurring
Xiaoqian Lv (Harbin Institute of Technology), Liqiang Nie (Harbin Institute of Technology)
RestorationDiffusion modelImage
🎯 What it does: A zero-shot joint low-light enhancement and deblurring method called FourierDiff is proposed, which utilizes a pre-trained diffusion model combined with frequency domain priors to achieve brightness enhancement and structural recovery.
Fourier-basis Functions to Bridge Augmentation Gap: Rethinking Frequency Augmentation in Image Classification
Puru Vaish (University of Twente), Nicola Strisciuglio (University of Twente)
ClassificationDomain AdaptationComputational EfficiencyAdversarial AttackConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a frequency domain augmentation method based on Fourier basis functions (Auxiliary Fourier-basis Augmentation, AFA), which generates adversarial samples by adding plane waves with adjustable amplitudes to the image spectrum, and adapts to distribution shifts with the help of an auxiliary branch.
FREE: Faster and Better Data-Free Meta-Learning
Yongxian Wei (Tsinghua University), Dacheng Tao (Nanyang Technological University)
Knowledge DistillationMeta LearningImage
🎯 What it does: A joint framework named FREE is proposed, which includes the FIVE rapid task recovery generator and the BELL gradient alignment meta-learner, addressing the slow data recovery and heterogeneous pre-trained model issues in data-free meta-learning.
Free3D: Consistent Novel View Synthesis without 3D Representation
Chuanxia Zheng (Visual Geometry Group University of Oxford), Andrea Vedaldi (Visual Geometry Group University of Oxford)
GenerationData SynthesisPose EstimationSupervised Fine-TuningDiffusion modelImage
🎯 What it does: We propose Free3D, a single-view novel view synthesis method based on a pre-trained 2D diffusion model without 3D representation, capable of generating pose-accurate and consistent images within a 360° view.
FreeControl: Training-Free Spatial Control of Any Text-to-Image Diffusion Model with Any Condition
Sicheng Mo (University of California Los Angeles), Bolei Zhou (University of Wisconsin Madison)
Image TranslationGenerationData SynthesisDiffusion modelImage
🎯 What it does: FreeControl is proposed, a training-free spatial control method that enables precise generation under arbitrary conditions on any text-to-image diffusion model.
FreeCustom: Tuning-Free Customized Image Generation for Multi-Concept Composition
Ganggui Ding (Zhejiang University), Chunhua Shen (Zhejiang University)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: FreeCustom is proposed, a parameter-free multi-concept custom image generation method;
FreeDrag: Feature Dragging for Reliable Point-based Image Editing
Pengyang Ling (University of Science and Technology of China), Jinjin Zheng (University of Science and Technology of China)
Image TranslationGenerationDiffusion modelGenerative Adversarial NetworkImageBenchmark
🎯 What it does: A point-to-point image editing method called FreeDrag based on feature dragging is proposed to address the issues of point loss and mis-tracking present in existing methods like DragGAN.
FreeKD: Knowledge Distillation via Semantic Frequency Prompt
Yuan Zhang (Peking University), Shanghang Zhang (Peking University)
Object DetectionSegmentationKnowledge DistillationPrompt EngineeringImage
🎯 What it does: This paper proposes FreeKD, a frequency-domain based knowledge distillation framework that obtains high and low-frequency pixel attention points by inserting frequency prompts into the teacher model, and uses position-aware relational loss to achieve adaptive frequency distillation, enhancing the performance of dense prediction tasks (object detection and semantic segmentation).
FreeMan: Towards Benchmarking 3D Human Pose Estimation under Real-World Conditions
Jiong Wang (Chinese University of Hong Kong), Ruimao Zhang (Chinese University of Hong Kong)
Pose EstimationDomain AdaptationDiffusion modelVideoBenchmark
🎯 What it does: The first large-scale multi-view real scene 3D human pose dataset, FreeMan, has been proposed and released, along with a semi-automated annotation and error detection workflow, providing a multi-task benchmark.
FreePoint: Unsupervised Point Cloud Instance Segmentation
Zhikai Zhang (Wuhan University), Guisong Xia (Huawei Zurich Research Center)
SegmentationTransformerPoint Cloud
🎯 What it does: This paper studies unsupervised point cloud instance segmentation and proposes the FreePoint framework.
FreeU: Free Lunch in Diffusion U-Net
Chenyang Si (Nanyang Technological University), Ziwei Liu (Nanyang Technological University)
GenerationData SynthesisDiffusion modelImageVideoText
🎯 What it does: The FreeU method is proposed in the U-Net structure of diffusion models, enhancing generation quality by adjustable scaling of features from the backbone network and skip connections during the inference phase.