CVPR 2026 Papers — Page 12
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 4071 papers
EvoID: Reinforced Evolution for Identity-Preserving Video Generation
Yiheng Zhang (HiDream.ai University of Science and Technology of China), Tao Mei (HiDream.ai University of Science and Technology of China)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningDiffusion modelImageVideoText
🎯 What it does: Reframe the identity-preserving video generation task as a self-evolving reinforcement learning (RL) process, guided by an Intrinsic Critic that dynamically evaluates video quality to steer the generation model;
Evolutionary Multimodal Reasoning via Hierarchical Semantic Representation for Intent Recognition
Qianrui Zhou (Tsinghua University), Hanlei Zhang (Tsinghua University)
ClassificationRecognitionLarge Language ModelSupervised Fine-TuningVision Language ModelMultimodalityChain-of-Thought
🎯 What it does: Propose the HIER model, which utilizes hierarchical semantic representation (modality-level Token → concept-level Cluster → relation-level Relation) combined with self-evolving reasoning to address the challenges of hierarchical semantic modeling and adaptive reasoning in multi-modal intent recognition.
Evolving Contextual Safety in Multi-Modal Large Language Models via Inference-Time Self-Reflective Memory
Ce Zhang (Carnegie Mellon University), Yaqi Xie (Carnegie Mellon University)
Safty and PrivacyTransformerLarge Language ModelPrompt EngineeringVision Language ModelMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: Propose the MM-SafetyBench++ evaluation benchmark and design the EchoSafe framework, enhancing the contextual safety of multimodal large language models through self-reflective memory retrieval during inference.
EW-DETR: Evolving World Object Detection via Incremental Low-Rank DEtection TRansformer
Munish Monga (Sony Research India), C.V. Jawahar (IIIT Hyderabad)
Object DetectionDomain AdaptationTransformerImage
🎯 What it does: Propose the EWOD (Evolving World Object Detection) paradigm and design the EW-DETR framework based on it, addressing three major challenges: incremental learning, domain adaptation, and unknown detection.
Exact-GS: Mathematically Rigorous and Accurate 3D Gaussian Splatting for 3D X-ray Reconstruction
Guangpu Yang, Sven Simon
RestorationGaussian SplattingBiomedical DataComputed Tomography
🎯 What it does: Propose an exact three-dimensional Gaussian expansion (Exact-GS) model free of approximation errors for X-ray CT reconstruction and novel view synthesis;
Exemplar-Free Class Incremental Learning via Preserving Class-Discriminative Structure
Xin Zhang (Shanxi University), Xian Yang (University of Manchester)
ClassificationRepresentation LearningTransformerSupervised Fine-TuningImage
🎯 What it does: Studied a method to achieve class-incremental learning using pre-trained models without preserving samples.
Exemplar-Free Continual Learning for State Space Models
Isaac Ning Lee (Monash University), Mehrtash Harandi (Monash University)
Representation LearningImage
🎯 What it does: This paper proposes a geometry-aware regularization framework called Inf-SSM, enabling state space models (SSM) to perform continuous learning (CL) in an exemplar-free setting by constraining the infinite observable subspace of SSM to mitigate catastrophic forgetting.
ExMesh: EXplicit Mesh Reconstruction with Topology Adaptation
Chuanjin Fan (University of Science and Technology of China), Tianzhu Zhang (University of Science and Technology of China)
OptimizationComputational EfficiencyImageMesh
🎯 What it does: This paper proposes the ExMesh framework, which directly optimizes explicit meshes by combining differentiable optimization with discrete topology updates, achieving high-precision mesh reconstruction from multi-view images.
EXOTIC: External Vision-driven Incomplete Multi-view Classification
Shilin Xu (Sichuan University), Yuan Sun (Sichuan University)
ClassificationTransformerVision Language ModelMultimodality
🎯 What it does: Propose the EXOTIC framework, which leverages external visual knowledge generated by pre-trained vision-language models to guide the completion and classification of missing multi-view data, thereby improving accuracy in incomplete multi-view learning.
Expand and Prune: Maximizing Trajectory Diversity for Effective GRPO in Generative Models
Shiran Ge (University of Chinese Academy of Sciences), Ran He (Chinese Academy of Sciences)
GenerationComputational EfficiencyReinforcement LearningDiffusion modelFlow-based ModelImageStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: Analyze the computational bottleneck of GRPO in text-to-image generation and propose the Pro-GRPO framework, which significantly reduces training costs while maintaining or enhancing performance through dynamic trajectory pruning and 'Expand-and-Prune' strategies.
Expanding mmWave Datasets for Human Pose Estimation with Unlabeled Data and LiDAR Datasets
Zhuoxuan Peng (Hong Kong University of Science and Technology), S.-H. Gary Chan (Hong Kong University of Science and Technology)
Data SynthesisPose EstimationTransformerFlow-based ModelPoint Cloud
🎯 What it does: This study proposes a data expansion method called EMDUL, which expands the millimeter-wave human pose estimation (HPE) dataset by leveraging unlabelled millimeter-wave point cloud data and existing LiDAR data.
Expanding Spatial and Temporal Context for Robotic Imitation Learning With Scene Graphs
Jianing Qian (University of Pennsylvania), Tarik Kelestemur (RAI Institute)
Object DetectionObject TrackingRobotic IntelligenceTransformerLarge Language ModelVision-Language-Action ModelDiffusion modelImageVideoTextGraph
🎯 What it does: Construct and maintain task-related dynamic scene graphs as a memory mechanism, using them as conditional inputs to diffusion imitation learning strategies to address long-horizon tasks in partially observable environments.
Experience Transfer for Multimodal LLM Agents in Minecraft Game
Chenghao Li (University of Electronic Science and Technology of China), Chaoning Zhang (Hong Kong Polytechnic University)
Explainability and InterpretabilityMeta LearningTransformerLarge Language ModelSupervised Fine-TuningAgentic AIMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: This paper proposes Echo, a multi-modal large language model agent that achieves rapid and efficient task transfer and planning in Minecraft games through an explicit experience transfer framework and structured contextual analogical learning.
Expert-Teacher-Student Collaborative Learning for Domain Adaptive Object Detection
Yiming Cui (Hangzhou Dianzi University), Chenggang Yan (Hangzhou Dianzi University)
Object DetectionDomain AdaptationKnowledge DistillationConvolutional Neural NetworkTransformerContrastive LearningImage
🎯 What it does: Achieve domain adaptive object detection by combining visual foundation models, teacher models, and student models in the absence of labeled data in the target domain.
Explaining CLIP Zero-shot Predictions Through Concepts
Onat Ozdemir (University of Edinburgh), Emre Akbas (Middle East Technical University)
ClassificationExplainability and InterpretabilityVision Language ModelContrastive LearningImage
🎯 What it does: Linearly project CLIP's image-text embeddings into an interpretable concept space to achieve explainability in zero-shot prediction.
Explaining Object Detectors via Collective Contribution of Pixels
Toshinori Yamauchi (Chiba University), Kazuhiko Kawamoto (Chiba University)
Object DetectionExplainability and InterpretabilityImage
🎯 What it does: Propose a visualization method called VX-CODE, which leverages Shapley values and interactions to capture the collective contributions of image pixels, thereby explaining the bounding box localization and category prediction of object detectors.
Explore with Long-term Memory: A Benchmark and Multimodal LLM-based Reinforcement Learning Framework for Embodied Exploration
Sen Wang (East China Normal University), Xin Tan (East China Normal University)
Robotic IntelligenceTransformerLarge Language ModelReinforcement LearningVision Language ModelImageTextMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: Proposed the Long-Term Memory Embodied Exploration (LMEE) framework and the LMEEBench benchmark, constructed a multi-objective navigation + memory-based question answering dataset, and developed MemoryExplorer based on reinforcement learning, which utilizes multimodal large language models to actively retrieve memories and make exploration decisions.
Exploring 6D Object Pose Estimation with Deformation
Zhiqiang Liu (State Key Laboratory of ISN, Xidian University), Yinlin Hu (MagicLeap Canonical)
Pose EstimationSimultaneous Localization and MappingOptical FlowImagePoint CloudMesh
🎯 What it does: Proposed the DeSOPE dataset, focusing on 6D pose estimation with multiple deformation states, and generated 665K high-quality pose annotations through precise 3D alignment and semi-automated labeling processes;
Exploring Adaptive Masked Reconstruction for Self-Supervised Skeleton-Based Action Recognition
Shengkai Sun (Hefei University of Technology), Meng Wang (Jilin University)
RecognitionRepresentation LearningTransformerContrastive LearningGraph
🎯 What it does: Propose the Adaptive Masked Reconstruction (AMR) framework to enhance the efficiency and effectiveness of self-supervised pre-training for skeletal action recognition.
Exploring Conditions for Diffusion Models in Robotic Control
Heeseong Shin (KAIST AI), Taekyung Kim (NAVER AI Lab)
Robotic IntelligencePrompt EngineeringDiffusion model
🎯 What it does: Studying how to generate task-adaptive visual representations using task and visual prompts without fine-tuning pre-trained text-to-image diffusion models, in order to improve robot control performance.
Exploring Spatial Intelligence from a Generative Perspective
Muzhi Zhu (Zhejiang University), Chunhua Shen (Zhejiang University)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelVision-Language-Action ModelImageMultimodalityPoint CloudBenchmark
🎯 What it does: This paper proposes a benchmark called GSI-Bench for evaluating spatial intelligence from a generative perspective, which includes high-quality real-world scene data (GSI-Real) and a large-scale controllable synthetic dataset (GSI-Syn). It measures the spatial compliance, accuracy, locality, and visual consistency of models in image editing through a unified evaluation protocol.
Exploring Spatiotemporal Feature Propagation for Video-Level Compressive Spectral Reconstruction: Dataset, Model and Benchmark
Lijing Cai (Nanjing University), Xun Cao (Nanjing University)
RestorationTransformerVideoBenchmark
🎯 What it does: This paper constructs a high-quality dynamic hyperspectral image dataset called DynaSpec, proposes a Transformer-based video-level compressed spectral reconstruction network named PG-SVRT, and achieves significantly improved reconstruction quality and spatiotemporal consistency on SCI systems such as DD-CASSI.
Exploring the Underwater World Segmentation without Extra Training
Bingyu Li (China Telecom), Xuelong Li (China Telecom)
SegmentationTransformerLarge Language ModelVision Language ModelImageTextBenchmark
🎯 What it does: Constructed a large fine-grained underwater segmentation dataset AquaOV255 and a unified evaluation benchmark UOVSBench, and proposed a training-free Earth2Ocean framework to achieve cross-domain underwater object segmentation.
Exploring Visual Pretraining for Learning Language Intelligence
Zhonghan Zhao (Shanghai AI Laboratory), Kai Chen (Shanghai AI Laboratory)
Representation LearningTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelAuto EncoderMultimodality
🎯 What it does: Studied pretraining large language models on visual corpora and demonstrated that they can perform as well as or better than text-only pretraining.
ExpoCM: Exposure-Aware One-Step Generative Single-Image HDR Reconstruction
Aoyu Liu (University of Electronic Science and Technology of China), Shuaicheng Liu (University of Electronic Science and Technology of China)
GenerationConvolutional Neural NetworkDiffusion modelFlow-based ModelImageOrdinary Differential Equation
🎯 What it does: Propose a single-image HDR reconstruction method called ExpoCM, which achieves high-quality HDR outputs through an exposure-aware one-step generation framework.
ExPose: Reinforcing Video Generation Models for Extreme Pose Estimation
Youngho Yoon (KAIST), Kuk-Jin Yoon (NAVER LABS)
GenerationPose EstimationSupervised Fine-TuningReinforcement LearningDiffusion modelVideoPoint Cloud
🎯 What it does: Propose the ExPose framework, which combines video generation models with extreme perspective pose estimation, enhancing geometric consistency by generating intermediate frames.
Exposing and Evaluating Hallucinations for GUI Grounding
Zicheng Zhang (Shanghai AI Lab), Guangtao Zhai (Shanghai AI Lab)
Anomaly DetectionTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextMultimodalityBenchmark
🎯 What it does: This paper investigates the hallucination phenomenon in graphical user interface (GUI) localization tasks, proposing two categories of hallucinations: confusion hallucinations and fabrication hallucinations, and constructing a specialized benchmark (GUI-HalluBench) to evaluate these hallucinations.
Exposing Functional Fusion: A New Class of Strategic Backdoor in Dynamic Prompt Architectures
Zeyao Liu (Institute of Information Engineering, Chinese Academy of Sciences), Xiaoshuang Ji (Institute of Information Engineering, Chinese Academy of Sciences)
Adversarial AttackTransformerPrompt EngineeringImage
🎯 What it does: Propose VIPER, a ViT backdoor attack framework based on a dynamic visual prompt generator, which utilizes functional fusion to share sparse parameter kernels between malicious logic and the forward task.
ExpPortrait: Expressive Portrait Generation via Personalized Representation
Junyi Wang (University of Science and Technology of China), Juyong Zhang (University of Science and Technology of China)
GenerationRepresentation LearningTransformerDiffusion modelNeural Radiance FieldVideo
🎯 What it does: Proposed a high-fidelity personalized head representation for generating animated videos with high identity consistency and expression details, and trained a Diffusion Transformer generator based on this representation
Extend3D: Town-Scale 3D Generation
Seungwoo Yoon (Seoul National University), Jaesik Park (Seoul National University)
GenerationData SynthesisDepth EstimationOptimizationDiffusion modelScore-based ModelFlow-based ModelImagePoint CloudStochastic Differential Equation
🎯 What it does: Proposes Extend3D, an untrained 3D scene generation pipeline capable of generating large-scale 3D scenes from a single image.
Extending Embodied Question Answering from Perception to Decision
Xicheng Gong (Peking University), Yadong Mu (Peking University)
Robotic IntelligenceLarge Language ModelSupervised Fine-TuningReinforcement LearningVision Language ModelImageVideoTextMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Proposes a large-scale embodied question-answering dataset called EQA-Decision, covering four dimensions: static scene construction, spatial understanding, task dynamic reasoning, and immediate decision-making, and builds a unified evaluation benchmark based on this dataset;
Extending One-Step Image Generation from Class Labels to Text via Discriminative Text Representation
Chenxi Zhao (Nankai University), Jufeng Yang (Nankai University)
GenerationTransformerLarge Language ModelVision Language ModelFlow-based ModelImageTextMultimodality
🎯 What it does: The study extends MeanFlow from category-tag-based single-step generation to text-based generation.
ExtrinSplat: Decoupling Geometry and Semantics for Open-Vocabulary Understanding in 3D Gaussian Splatting
Jiayu Ding (Peking University), Ge Li (Peking University)
RecognitionSegmentationGaussian SplattingPoint CloudMesh
🎯 What it does: This paper proposes the ExtrinSplat framework, achieving decoupling of geometry and semantics in 3D Gaussian Splatting scenarios, supporting open-vocabulary point-level 3D recognition and segmentation.
F^2HDR: Two-Stage HDR Video Reconstruction via Flow Adapter and Physical Motion Modeling
Huanjing Yue (Tianjin University), Jingyu Yang (Tianjin University)
RestorationConvolutional Neural NetworkOptical FlowVideo
🎯 What it does: Propose a two-stage HDR video reconstruction framework F HDR 2, which achieves robust optical flow alignment and detail recovery across exposure scenarios by utilizing Flow Adapter and physical motion modeling.
F$^2$-Assist: Multi-Phase Fetal Growth Forecast and Report Generation from Ultrasound Examination
Bin Pu (Hunan University), Jiawei Ma (City University of Hong Kong)
TransformerLarge Language ModelVision Language ModelImageBiomedical DataUltrasound
🎯 What it does: This paper proposes a multi-stage fetal growth prediction and report generation task, and constructs the GrowthFetus large-scale multimodal fetal ultrasound dataset. A unified multimodal large language model, F-Assist, is utilized to achieve fetal growth trajectory prediction and corresponding report generation.
F2Net: A Frequency-Fused Network for Ultra-High Resolution Remote Sensing Segmentation
Hengzhi Chen (University of Sydney), Kun Hu (Edith Cowan University)
SegmentationConvolutional Neural NetworkTransformerImage
🎯 What it does: Propose a frequency-aware network called F2Net, which decomposes ultra-high-resolution remote sensing images into high- and low-frequency branches to preserve details and global semantics, respectively, achieving semantic segmentation.
FAAR: Efficient Frequency-Aware Multi-Task Fine-Tuning via Automatic Rank Selection
Maxime Fontana (King's College London), Miaojing Shi (Tongji University)
SegmentationComputational EfficiencyRepresentation LearningTransformerSupervised Fine-TuningImage
🎯 What it does: This paper proposes a parameter-efficient multi-task fine-tuning method called FAAR, combining a frequency-aware decoder and automatic rank selection to significantly reduce trainable parameters and enhance multi-task performance.
FabricGen: Microstructure-Aware Woven Fabric Generation
Yingjie Tang (Nankai University), Beibei Wang (Nanjing University)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringDiffusion modelImageText
🎯 What it does: Proposed the FabricGen framework, enabling the generation of high-quality fabric materials from text/image prompts;
Face-Guided Sentiment Boundary Enhancement for Weakly-Supervised Temporal Sentiment Localization
Cailing Han (Hefei University Of Technology), Dan Guo (Hefei University Of Technology)
ClassificationRecognitionConvolutional Neural NetworkTransformerContrastive LearningVideoMultimodality
🎯 What it does: Propose FSENet, a point-level weakly supervised temporal emotion localization framework, which utilizes facial features to guide multimodal interaction, and further enhances emotion boundary detection through point-aware semantic contrast and boundary smoothing pseudo labels.
FACE: A Face-based Autoregressive Representation for High-Fidelity and Efficient Mesh Generation
Hanxiao Wang (Institute of Automation, Chinese Academy of Sciences), Dong-Ming Yan (Institute of Automation, Chinese Academy of Sciences)
GenerationTransformerDiffusion modelAuto EncoderPoint CloudMesh
🎯 What it does: Proposed a face-level autoregressive autoencoder (FACE) based on the 'one face as one token' principle, achieving efficient and accurate generation and reconstruction of triangular meshes by significantly shortening the sequence length
Face2Scene: Using Facial Degradation as an Oracle for Diffusion-Based Scene Restoration
Amirhossein Kazerouni (University of Toronto), Alex Levinshtein (AI Center-Toronto, Samsung Electronics)
RestorationDiffusion modelImage
🎯 What it does: This paper proposes the Face2Scene framework, which utilizes a reference face restoration model to extract facial degradation information as a prior for full-scene recovery;
FaceCam: Portrait Video Camera Control via Scale-Aware Conditioning
Weijie Lyu (University of California Merced), Zhixin Shu (Adobe Research)
GenerationData SynthesisTransformerDiffusion modelFlow-based ModelAuto EncoderGaussian SplattingVideoPoint Cloud
🎯 What it does: FaceCam provides a portrait video generation system based on a single video and target camera trajectory, capable of precisely controlling camera position and motion while preserving subject identity and dynamic expressions.
Factorize, Reconstruct, Enhance: A Unified Framework for Multimodal Sentiment Analysis
Zhilu Yang, Mingcheng Li (Imperial College London)
ClassificationRepresentation LearningRecurrent Neural NetworkTransformerAuto EncoderContrastive LearningVideoTextMultimodalityAudio
🎯 What it does: Proposes a unified framework named FUSE-Net for multi-modal sentiment analysis, which first decomposes each modality into shared, specific, and noise subspaces, then preserves semantic integrity through information bottleneck-driven reconstruction channels, and adaptively aggregates different subspaces using a multi-factor dynamic fusion module.
Factorized Context Aggregation for Robust Cancer Risk Estimation via Soft Re-Ranked Retrieval and Hierarchical Anchors
Puria Azadi Moghadam, Ali Bashashati (University Of British Columbia)
ClassificationRetrievalKnowledge DistillationRepresentation LearningMultimodalityBiomedical DataRetrieval-Augmented Generation
🎯 What it does: Propose a framework based on pre-trained models, soft re-ranking retrieval, and hierarchical anchors, using pathology slides as the main benchmark. During training, missing modality information supplementation is utilized to achieve cancer risk prediction under multi-modal missing scenarios.
Failure Modes for Deep Learning-Based Online Mapping: How to Measure and Address Them
Michael Hubbertz (University of Wuppertal), Tobias Meisen (University of Wuppertal)
Autonomous DrivingTransformerPoint Cloud
🎯 What it does: This paper proposes an evaluation framework to distinguish between two failure modes of deep learning online map models: geolocation memory and map geometric overfitting, and constructs a fine-grained evaluation metric based on geometric similarity and discrete Fréchet distance.
FailureAtlas: Mapping the Failure Landscape of T2I Models via Active Exploration
Muxi Chen (Chinese University of Hong Kong), Qiang Xu (Chinese University of Hong Kong)
GenerationExplainability and InterpretabilityData-Centric LearningTransformerLarge Language ModelDiffusion modelImageTextBenchmark
🎯 What it does: Proposed the FAILUREATLAS framework for actively exploring and automating the discovery of failure slices in text-to-image models, and constructed a large-scale entity-attribute corpus to structure the search space.
FairLLaVA: Fairness-Aware Parameter-Efficient Fine-Tuning for Large Vision-Language Assistants
Mahesh Bhosale (University at Buffalo), Xuan Gong (Harvard Medical School)
Safty and PrivacyTransformerSupervised Fine-TuningVision Language ModelBiomedical Data
🎯 What it does: Proposes FairLLaVA, a parameter-efficient fine-tuning method for medical multimodal large language models, leveraging mutual information minimization to eliminate potential sensitive attribute shortcuts in images;
Faithful Contouring: Near-Lossless 3D Voxel Representation Free from Iso-surface
Yihao Luo (Imperial College London), ChoonHwai Yap (Imperial College London)
CompressionRepresentation LearningConvolutional Neural NetworkAuto EncoderPoint CloudMesh
🎯 What it does: Propose a sparse voxelization representation called Faithful Contouring, which can directly convert any triangle mesh into a near-lossless voxel representation and is applicable to high-resolution (2048+) reconstruction and compression.
FaithFusion: Harmonizing Reconstruction and Generation via Pixel-wise Information Gain
YuAn Wang (Baidu Inc), Jun Wang (Baidu Inc)
Autonomous DrivingDiffusion modelGaussian SplattingImageVideo
🎯 What it does: Propose the FaithFusion framework, achieving pixel-level information gain fusion between 3D Gaussian Splatting and diffusion models for controllable driving scene reconstruction and generation.
FALCON: False-Negative Aware Learning of Contrastive Negatives in Vision-Language Alignment
Myunsoo Kim (Korea University), Byung-Jun Lee (Korea University)
RetrievalRepresentation LearningData-Centric LearningTransformerContrastive LearningMultimodality
🎯 What it does: Propose a learning-based mini-batch construction strategy called FALCON, which adaptively balances the ratio of hard negative samples and false negative samples in visual-language pretraining, reducing adversarial learning conflicts caused by false negatives;
FantasyVLN: Unified Multimodal Chain-of-Thought Reasoning for Vision-and-Language Navigation
Jing Zuo (Beijing University of Posts and Telecommunications), Yonggang Qi (Beijing University of Posts and Telecommunications)
Autonomous DrivingComputational EfficiencyRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelVision-Language-Action ModelMultimodalityChain-of-Thought
🎯 What it does: Propose FantasyVLN, a unified multi-modal chain-of-thought reasoning framework that integrates text, vision, and multi-modal chain-of-thought reasoning, while avoiding generating a large number of tokens during inference to maintain real-time navigation;
FAPE-IR: Frequency-Aware Planning and Execution Framework for All-in-One Image Restoration
Jingren Liu (Tianjin University), Zhong Ji (Tianjin University)
RestorationSuper ResolutionLarge Language ModelMixture of ExpertsVision Language ModelDiffusion modelGenerative Adversarial NetworkImageMultimodality
🎯 What it does: Proposes a frequency-aware planning and execution framework, FAPE-IR, capable of uniformly handling multiple image restoration tasks including denoising, dehazing, deraining, desnowing, deblurring, low-light enhancement, and super-resolution.
FARMER: Flow AutoRegressive Transformer over Pixels
Guangting Zheng (University Of Science And Technology Of China), Rui Zhu (ByteDance Seed China)
GenerationTransformerFlow-based ModelImage
🎯 What it does: Proposes an end-to-end pixel-level generation framework called FARMER that combines normalizing flows (NF) with autoregressive (AR) models.
Fast Markov Random Field Optimisation for Topologically Noisy 3D Shape Matching
Paul Roetzer (University of Bonn), Florian Bernard (University of Bonn)
OptimizationComputational EfficiencyMesh
🎯 What it does: Propose a 3D shape matching framework based on Markov Random Fields (MRF), specifically addressing shape correspondence problems under topological noise (e.g., changes in genus).
Fast Reasoning Segmentation for Images and Videos
Yiqing Shen (Johns Hopkins University), Mathias Unberath (Johns Hopkins University)
SegmentationComputational EfficiencyKnowledge DistillationRepresentation LearningTransformerLarge Language ModelReinforcement LearningImageVideo
🎯 What it does: Propose FastReasonSeg, a two-stage distillation framework that performs image and video reasoning segmentation through digital twin representations, significantly reducing model parameters and achieving real-time inference;
Fast SceneScript: Fast and Accurate Language-Based 3D Scene Understanding via Multi-Token Prediction
Ruihong Yin (Qualcomm XR Labs), Theo Gevers (University of Amsterdam)
Object DetectionSegmentationTransformerVision Language ModelPoint CloudMesh
🎯 What it does: This paper proposes Fast SceneScript, an efficient 3D scene understanding framework implemented using multi-word prediction (MTP);
Fast Spatial Tracking with Visual Geometry Transformer
Chengjie Huang (Huawei Noah's Ark Lab), Bingbing Liu (Huawei Technologies)
Object TrackingTransformerVideoPoint CloudBenchmark
🎯 What it does: Real-time prediction of 3D trajectories for arbitrary query points in monocular videos, avoiding the use of depth or camera pose priors
Fast-FoundationStereo: Real-Time Zero-Shot Stereo Matching
Bowen Wen, Stan Birchfield
Depth EstimationComputational EfficiencyKnowledge DistillationNeural Architecture SearchImageVideoBenchmark
🎯 What it does: Propose Fast-FoundationStereo, which integrates zero-shot generalization based on foundation models with a block-based acceleration framework for real-time inference
Fast-ThinkAct: Efficient Vision-Language-Action Reasoning via Verbalizable Latent Planning
Chi-Pin Huang (NVIDIA), Fu-En Yang (NVIDIA)
Computational EfficiencyKnowledge DistillationRobotic IntelligenceTransformerLarge Language ModelReinforcement LearningVision-Language-Action ModelDiffusion modelImageTextMultimodalityChain-of-Thought
🎯 What it does: Propose Fast-ThinkAct, an efficient vision-language-action reasoning framework that leverages verbalizable latent planning.
FAST: Topology-Aware Frequency-Domain Distribution Matching for Coreset Selection
Jin Cui (Xi'an Jiaotong University), Pengju Ren (Xi'an Jiaotong University)
Data-Centric LearningImageText
🎯 What it does: This paper proposes a DNN-free core subset selection method called FAST, which achieves complete distribution alignment with the original dataset through frequency domain distribution matching.
Fast3Dcache: Training-free 3D Geometry Synthesis Acceleration
Mengyu Yang (Westlake University), Chi Zhang (Westlake University)
GenerationData SynthesisComputational EfficiencyDiffusion modelFlow-based Model
🎯 What it does: Training-free, geometry-aware cache acceleration for 3D diffusion models
Faster-GS: Analyzing and Improving Gaussian Splatting Optimization
Florian Hahlbohm (TU Braunschweig), Marcus Magnor (University of New Mexico)
OptimizationComputational EfficiencyGaussian Splatting
🎯 What it does: Propose Faster-GS, an efficient training framework based on 3D Gaussian splatting, significantly accelerating optimization and reducing VRAM usage.
FastEventDGS: Deformable Gaussian Splatting for Fast Dynamic Scenes from a Single Event Camera
Zijia Dai (ShanghaiTech University), Laurent Kneip (ShanghaiTech University)
GenerationDepth EstimationGaussian Splatting
🎯 What it does: Propose the FastEventDGS framework, achieving high-frame-rate 4D dynamic scene reconstruction using a single-camera event stream.
FastGaMer: Efficient GainMap Learning for Practical Inverse Tone Mapping
Yuanshen Guan (University of Science and Technology of China), Zhiwei Xiong (University of Science and Technology of China)
RestorationImage
🎯 What it does: Propose FastGaMer, an efficient, resolution-agnostic inverse tone mapping (ITM) framework that directly learns a three-channel color gain map instead of HDR values.
FastGS: Training 3D Gaussian Splatting in 100 Seconds
Shiwei Ren (Nankai University), Biao Lu (Nankai University)
OptimizationComputational EfficiencyNeural Radiance FieldGaussian SplattingImage
🎯 What it does: Propose the FastGS framework, which accelerates 3D Gaussian splatting training through multi-view consistency-based density and pruning.
FastHybrid: Accelerating Hybrid Autoregressive Image Generation with Lookahead and Guided Decoding
Zhengguo Jiang (University Of Science And Technology Of China), Linli Xu (University Of Science And Technology Of China)
GenerationComputational EfficiencyTransformerDiffusion modelImage
🎯 What it does: Propose an accelerated hybrid autoregressive (AR) image generation framework that combines the Lookahead decoding strategy with Guided Diffusion Sampling to generate multiple frames in parallel and correct them in the AR branch, significantly improving inference speed.
FastLightGen: Fast and Light Video Generation with Fewer Steps and Parameters
Shitong Shao (Hong Kong University of Science and Technology), Zeke Xie (Hong Kong University of Science and Technology)
GenerationKnowledge DistillationTransformerDiffusion modelFlow-based ModelVideo
🎯 What it does: Propose FastLightGen, a three-stage joint distillation framework, achieving video generation with fewer steps and parameter compression through identifying irrelevant layers, dynamic pruning, and distribution matching.
FastRef: Fast Prototype Refinement for Few-shot Industrial Anomaly Detection
Yufei Li (Xidian University), Xiyang Liu (Xidian University)
Anomaly DetectionOptimizationMeta LearningConvolutional Neural NetworkVision Language ModelImageBenchmark
🎯 What it does: Proposes a fast prototype refinement module called FastRef, which enhances the representativeness of prototypes in few-shot industrial defect detection by iteratively updating transformation matrices and transport probabilities.
FAVE: A Structured Benchmark for Fine-Grained Audio-Visual Temporal Evaluation in Multimodal LLMs
Weiheng Lu (Peking University), Ming-Ching Chang (State University of New York at Albany)
TransformerLarge Language ModelVision Language ModelVideoTextMultimodalityBenchmarkAudio
🎯 What it does: Proposes the FAVE benchmark to systematically evaluate the performance of multimodal large language models in fine-grained temporal reasoning on audio-visual content.
FB-CLIP: Fine-Grained Zero-Shot Anomaly Detection with Foreground-Background Disentanglement
Ming Hu (Xi'an Institute of Optics and Precision Mechanics Chinese Academy of Sciences), Quan Wang (Xi'an Institute of Optics and Precision Mechanics Chinese Academy of Sciences)
Anomaly DetectionTransformerPrompt EngineeringVision Language ModelContrastive LearningImageBiomedical Data
🎯 What it does: In the unannotated anomaly detection task, the FB-CLIP framework is proposed, significantly enhancing CLIP's zero-shot performance in fine-grained anomaly localization through multi-strategy text feature fusion, foreground-background separation, and background suppression.
FBTA: Enabling Single-GPU End-to-End Gigapixel WSI Classification with Feature Bridging and Translation Alignment
Jiuyang Dong (Harbin Institute of Technology), Yongbing Zhang (Harbin Institute of Technology)
ClassificationContrastive LearningImageBiomedical Data
🎯 What it does: Achieve end-to-end multiple instance learning (MIL) training for gigapixel pathology slides on a single 24GB GPU, and provide practical inference solutions.
FE2E: From Editor to Dense Geometry Estimator
Jiyuan Wang (Beijing Jiaotong University), Xiangxiang Chu (Alibaba Group)
Depth EstimationTransformerDiffusion modelAuto EncoderImage
🎯 What it does: Proposed a framework named FE2E that transforms an image editing model (Step1X-Edit) into a dense geometric estimator for depth and normals.
FEAST: Fully Connected Expressive Attention for Spatial Transcriptomics
Taejin Jeong (Yonsei University), Seong Jae Hwang (Yonsei University)
Computational EfficiencyRepresentation LearningTransformerBiomedical Data
🎯 What it does: Proposed a spatial transcriptomics prediction framework FEAST based on fully connected attention, enhancing model performance through negative attention and discrete sampling.
FEAT: Fashion Editing and Try-On from Any Design
Soye Kwon (Kookmin University), Jaekoo Lee (Kookmin University)
Image TranslationGenerationPrompt EngineeringVision Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: Propose a full clothing and accessory editing and try-on method FEAT based on diffusion models, which can freely reshape and try on clothing and accessories using any design source (image, text, sketch);
FeatureFool: Zero-Query Fooling of Video Models via Feature Map
Duoxun Tang (Tsinghua University), Jiyao Wang (HKUST)
Adversarial AttackOptical FlowVideo
🎯 What it does: Proposes a zero-query black-box video adversarial attack called FeatureFool, which perturbs videos by extracting gradient feature maps from the maximum optical flow frames, successfully deceiving traditional video classifiers and video LLMs.
Featurising Pixels from Dynamic 3D Scenes with Linear In-Context Learners
Nikita Araslanov (Google), Federico Tombari
SegmentationTransformerOptical FlowVideo
🎯 What it does: This paper proposes a linear context learning framework called LILA, which can learn pixel-level semantic and geometric features in unsupervised videos.
Fed-ADE: Adaptive Learning Rate for Federated Post-adaptation under Distribution Shift
Heewon Park (Sungkyunkwan University), Minhae Kwon (Sungkyunkwan University)
Domain AdaptationFederated LearningImageText
🎯 What it does: Propose an unsupervised federated post-deployment adaptation framework Fed-ADE, which dynamically adjusts the learning rate in a multi-client distribution drift environment to improve model performance on unlabeled data streams;
FedAdamom: Adaptive Momentum for Improved Generalization in Federated Optimization
Wenjie Hou (Beihang University), Wei Yang Bryan Lim (Nanyang Technological University)
OptimizationFederated LearningDiffusion modelImageTextBenchmarkStochastic Differential Equation
🎯 What it does: This paper proposes a new federated learning adaptive optimizer, FedAdamom, which enhances the generalization performance of federated models by adaptively adjusting the global momentum at the parameter level to escape saddle points quickly while favoring flat minima.
FedAFD: Multimodal Federated Learning via Adversarial Fusion and Distillation
Min Tan (Hangzhou Dianzi University), Zhou Yu (Hangzhou Dianzi University)
Federated LearningKnowledge DistillationAdversarial AttackConvolutional Neural NetworkRecurrent Neural NetworkTransformerImageTextMultimodality
🎯 What it does: Proposes the FedAFD framework, achieving cross-modal and cross-task collaborative training in multi-modal federated learning while balancing client personalization and server global performance.
FedAlign: Differentially Private Distribution Alignment for Non-IID Federated Learning
Peng Wu (Hunan University), Zhuo Tang (Hunan University)
Federated LearningSafty and PrivacyConvolutional Neural NetworkImage
🎯 What it does: Propose the FedAlign framework, which utilizes local statistical matrices (mean, variance, skewness, kurtosis) under differential privacy to align client data distributions, thereby accelerating federated learning convergence and improving global model accuracy.
FedARA: Resource-adaptive Low-rank Personalized Federated Learning via Anchor-driven Representation Alignment on Heterogeneous Edge Devices
Ruonan Zhao (Zhengzhou University), Laurence Tianruo Yang (Zhengzhou University)
Federated LearningComputational EfficiencyRepresentation LearningConvolutional Neural NetworkImage
🎯 What it does: Propose a personalized federated learning framework named FedARA, combining low-rank decomposition for adaptive model complexity and anchor-driven feature consistency learning, to achieve data-level and model-level personalization on heterogeneous edge devices.
FedBPrompt: Federated Domain Generalization Person Re-Identification via Body Distribution Aware Visual Prompts
Xin Xu (Wuhan University of Science and Technology), Kui Jiang (Harbin Institute of Technology Zhengzhou Research Institute)
RecognitionRetrievalDomain AdaptationFederated LearningTransformerSupervised Fine-TuningPrompt EngineeringImage
🎯 What it does: This paper proposes the FedBPrompt framework, which inserts learnable Body Distribution Aware Visual Prompts (BAPM) into ViT to achieve background suppression and view alignment in federated domain generalization for person Re-ID, and designs a Prompt-based Fine-Tuning Strategy (PFTS) to significantly reduce communication costs.
FedCART: Tackling Long-Tailed Distributions in Federated Adversarial Training via Classifier Refinement
Yuchen Qin (Dalian University of Technology), Heng Qi (Dalian University of Technology)
ClassificationFederated LearningContrastive LearningImage
🎯 What it does: In federated learning, the FedCART framework is proposed to achieve adversarial training for long-tailed data distributions.
FedDAP: Domain-Aware Prototype Learning for Federated Learning under Domain Shift
Huy Q. Le (Kyung Hee University), Choong Seon Hong (Kyung Hee University)
Domain AdaptationFederated LearningContrastive LearningImage
🎯 What it does: To address the domain shift problem in federated learning, the FedDAP framework is proposed, leveraging domain-aware prototype learning to enhance the generalization and convergence speed of the global model.
Federated Active Learning Under Extreme Non-IID and Global Class Imbalance
Chen-Chen Zong (Nanjing University of Aeronautics and Astronautics), Sheng-Jun Huang (Nanjing University of Aeronautics and Astronautics)
ClassificationFederated LearningImageBiomedical DataBenchmark
🎯 What it does: Propose a federated active learning framework named FairFAL to efficiently reduce annotation costs under extreme non-independent and identically distributed (non-IID) and globally imbalanced class distribution scenarios.
FedHarmony: Harmonizing Heterogeneous Label Correlations in Federated Multi-Label Learning
Zhiqiang Kou (Southeast University), Qiang Yang (Hong Kong Polytechnic University)
ClassificationFederated LearningImage
🎯 What it does: Analyze the label correlation drift problem in federated multi-label learning and propose the FedHarmony framework to synchronize label correlations across different clients.
FedMOP: Achieving Enhanced Privacy and Performance in Federated Learning via Momentum Orthogonal Projection
Yunlong Zhao (Central South University), Xiu Su (Central South University)
Federated LearningSafty and PrivacyImage
🎯 What it does: Propose the FedMOP method, achieving privacy protection and performance improvement in federated learning by using momentum-based orthogonal projection initialization offset before client training.
FedMPT: Federated Multi-Label Prompt Tuning of Vision-Language Models
Xucong Wang (University of Science and Technology of China), Yang Wang (University of Science and Technology of China)
RecognitionFederated LearningLarge Language ModelPrompt EngineeringMixture of ExpertsVision Language ModelMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Designed and implemented the FedMPT framework, leveraging conditional templates generated by LLMs, optimal transport alignment, and gating mechanisms to perform prompt tuning for multi-label recognition tasks in federated learning environments, reducing pseudo-relevance caused by local data bias.
FedRAC: Rolling Submodel Allocation for Collaborative Fairness in Federated Learning
Zihui Wang (Harbin Engineering University), Zheng Wang (Shanghai Innovation Institution)
Federated LearningConvolutional Neural NetworkImage
🎯 What it does: Propose the FedRAC framework, which achieves collaborative fairness and model performance in federated learning through dynamic reputation calculation and rolling sub-model allocation.
FedRE: A Representation Entanglement Framework for Model-Heterogeneous Federated Learning
Yuan Yao (Teleinfo, CAICT), Xiaoxiao Li (University of British Columbia)
Federated LearningSafty and PrivacyComputational EfficiencyRepresentation LearningImage
🎯 What it does: Proposed a federated learning framework named FedRE, utilizing 'entangled representation' as a lightweight and privacy-friendly method for uploading client knowledge to address the aggregation problem in model-heterogeneous FL.
FedRG: Unleashing the Representation Geometry for Federated Learning with Noisy Clients
Tian Wen (Chinese Academy of Sciences), Bo Han (Hong Kong Baptist University)
ClassificationFederated LearningContrastive LearningImage
🎯 What it does: Propose the FedRG method, which identifies noisy labels on unlabeled spherical representations by leveraging the representation geometry priority principle, and achieves robust optimization through a personalized noise absorption matrix.
FedSDR: Federated Graph Learning with Structural Noise Detection and Reconstruction
Jiaqi Liu (Wuhan University), Mang Ye (Wuhan University)
Federated LearningGraph Neural NetworkGraph
🎯 What it does: This paper proposes FedSDR, a federated graph learning framework designed for high structural noise scenarios. It can detect structural noise in client graphs and perform structural reconstruction, thereby enhancing the collaborative effectiveness between the global model and damaged clients.
FedSST: Rethinking Fair Federated Graph Learning under Structural Shift
Dingyi Zhao (Wuhan University)
Domain AdaptationFederated LearningGraph Neural NetworkGraph
🎯 What it does: Propose FedSST to address the structural heterogeneity problem in graph federated learning by constructing a structure-aware adaptive framework, achieving fair aggregation and local training allocation.
Feed-forward Gaussian Registration for Head Avatar Creation and Editing
Malte Prinzler (ETH Zurich), Timo Bolkart (Google)
GenerationTransformerGaussian SplattingImageMesh
🎯 What it does: Directly predict high-quality head Gaussian splat textures using multi-view inputs, and achieve dense semantic correspondence for all Gaussians, enabling rapid generation of editable and animatable head avatars.
Feed-Forward One-Shot Animatable Textured Mesh Avatar Reconstruction
Yisheng He (Tongyi Lab, Alibaba Group)
GenerationRecurrent Neural NetworkTransformerGaussian SplattingImageMesh
🎯 What it does: Propose MeshLAM, a plug-and-play framework for reconstructing animatable, textured mesh avatars from a single image based on Transformer architecture.
Few-for-Many Personalized Federated Learning
Ping Guo (City University of Hong Kong), Qingfu Zhang (Hong Kong Metropolitan University)
OptimizationFederated LearningImageTextBiomedical Data
🎯 What it does: Proposed the Few-for-Many personalized federated learning framework FedFew, which uses K groups of server models to serve M large-scale heterogeneous clients
Few-shot Acoustic Synthesis with Multimodal Flow Matching
Amandine Brunetto (Mines Paris PSL University)
GenerationData SynthesisTransformerDiffusion modelRectified FlowAuto EncoderContrastive LearningMultimodalityAudio
🎯 What it does: Propose FLAC, a flow-matching based generative model that can synthesize room impulse responses (RIR) matching the scene geometry under very few observations
Few-Shot Hybrid Incremental Learning:Continually Learning under Data Scarcity and Task Uncertainty
Yan Li (Northwestern Polytechnical University), Junwei Han (Chongqing University of Posts and Telecommunications)
ClassificationDomain AdaptationComputational EfficiencyRepresentation LearningData-Centric LearningMeta LearningTransformerMixture of ExpertsImage
🎯 What it does: Aiming at uncertainty scenarios combining few-shot incremental learning and domain migration, a new Few-Shot Hybrid Incremental Learning (FSHIL) paradigm is proposed, along with two core modules: Conditional Meta-Expanding Mixture-of-Experts (CME-MoE) and Self-Expanding Prototype Classifier (SEPC). These modules achieve a balance between feature-level stability and plasticity, as well as adaptive expansion of classifier-level multi-distribution decision boundaries.
Few-Shot Incremental 3D Object Detection in Dynamic Indoor Environments
Yun Zhu (Nanjing University of Science and Technology), Na Zhao (Nanjing University)
Object DetectionVision Language ModelImagePoint Cloud
🎯 What it does: Proposed a few-shot incremental 3D object detection framework called FI3Det, which can complete detection using only a small number of new category samples in dynamic indoor environments while maintaining performance on existing categories.
Few-Step Diffusion Sampling Through Instance-Aware Discretizations
Liangyu Yuan (Westlake University), Chi Zhang (Westlake University)
GenerationDiffusion modelImageVideoStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: This paper proposes an instance-aware time discretization method (INDIS), which enhances the few-step sampling quality of diffusion models by adaptively assigning step sizes to each sampling starting point (noise) and condition.