CVPR 2023 Papers — Page 8
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2353 papers
Exploring the Relationship Between Architectural Design and Adversarially Robust Generalization
Aishan Liu (Beihang University), Dacheng Tao (JD Explore Academy)
Adversarial AttackConvolutional Neural NetworkTransformerImage
🎯 What it does: Conducted adversarial training on 20 types of CNN and Transformer architectures on CIFAR-10/Imagenette, evaluating their robust generalization performance against various ℓp attacks, and theoretically explained the relationship between weight sparsity and robustness through Rademacher complexity.
expOSE: Accurate Initialization-Free Projective Factorization Using Exponential Regularization
José Pedro Iglesias (Chalmers University of Technology), Carl Olsson (Lund University)
OptimizationSimultaneous Localization and MappingImage
🎯 What it does: This paper proposes an initialization-independent projective factorization method called expOSE, which utilizes exponential regularization instead of the original linear regularization. It forms a complete Structure-from-Motion processing pipeline from uncalibrated data with radial distortion by weighting the radial/tangential components of the Object Space Error (OSE).
Extracting Class Activation Maps From Non-Discriminative Features As Well
Zhaozheng Chen (Singapore Management University), Qianru Sun (Singapore Management University)
SegmentationImage
🎯 What it does: This paper proposes a new class activation map generation method called LPCAM, which can extract activation maps that fully cover the target object from classification models.
Extracting Motion and Appearance via Inter-Frame Attention for Efficient Video Frame Interpolation
Guozhen Zhang (Nanjing University), Limin Wang (Nanjing University)
Image TranslationRestorationComputational EfficiencyConvolutional Neural NetworkTransformerOptical FlowVideo
🎯 What it does: A framework is proposed that utilizes Inter-Frame Attention to unify the extraction of motion and appearance information in video frame interpolation.
F2-NeRF: Fast Neural Radiance Field Training With Free Camera Trajectories
Peng Wang (University of Hong Kong), Wenping Wang (Texas A&M University)
GenerationData SynthesisComputational EfficiencyNeural Radiance FieldImage
🎯 What it does: A fast neural radiance field training framework F²‑NeRF is proposed, capable of handling arbitrary free camera trajectories, significantly reducing training time and improving synthesis quality.
FAC: 3D Representation Learning via Foreground Aware Feature Contrast
Kangcheng Liu (Nanyang Technological University), Ling Shao (Nanyang Technological University)
Object DetectionSegmentationRepresentation LearningContrastive LearningPoint Cloud
🎯 What it does: Proposes a Foreground-Aware Contrast (FAC) framework that uses unsupervised contrastive learning for pre-training point clouds.
FaceLit: Neural 3D Relightable Faces
Anurag Ranjan (Apple), Oncel Tuzel (Apple)
GenerationData SynthesisPose EstimationGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a generative framework named FaceLit, which can learn decoupled 3D facial models from single-view unlabeled facial images and render realistic faces under arbitrary poses and lighting conditions.
Fair Federated Medical Image Segmentation via Client Contribution Estimation
Meirui Jiang (Chinese University of Hong Kong), Ziyue Xu (NVIDIA)
SegmentationFederated LearningImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A federated learning method called FedCE is proposed, which achieves collaborative fairness and performance fairness by estimating the contributions of clients in the gradient space and data space, using the estimated contributions as global aggregation weights.
Fair Scratch Tickets: Finding Fair Sparse Networks Without Weight Training
Pengwei Tang (Renmin University of China), Yong Liu (Renmin University of China)
ClassificationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a method to directly obtain fair and accurate computer vision models without weight training, by searching for sparse masks (Fair Scratch Tickets, FST) in randomly initialized dense networks.
Fake It Till You Make It: Learning Transferable Representations From Synthetic ImageNet Clones
Mert Bülent Sarıyıldız (NAVER LABS Europe), Yannis Kalantidis (NAVER LABS Europe)
ClassificationData SynthesisDomain AdaptationConvolutional Neural NetworkPrompt EngineeringDiffusion modelImage
🎯 What it does: The study uses an ImageNet clone dataset generated by Stable Diffusion for training image classification models from scratch, exploring whether synthetic data can replace real images.
FAME-ViL: Multi-Tasking Vision-Language Model for Heterogeneous Fashion Tasks
Xiao Han (University of Surrey), Tao Xiang (University of Surrey)
ClassificationGenerationRetrievalKnowledge DistillationTransformerVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes a multi-task vision-language model named FAME-ViL, which unifies the processing of four types of fashion tasks: cross-modal retrieval, text-guided retrieval, multi-modal classification, and image captioning.
Fantastic Breaks: A Dataset of Paired 3D Scans of Real-World Broken Objects and Their Complete Counterparts
Nikolas Lamb (Clarkson University), Natasha Kholgade Banerjee (Clarkson University)
RestorationPoint CloudMeshBenchmark
🎯 What it does: Created the Fantastic Breaks dataset, which includes 150 aligned damaged/intact 3D scans along with corresponding repair parts and crack annotations, and is used to evaluate automatic shape repair algorithms.
FashionSAP: Symbols and Attributes Prompt for Fine-Grained Fashion Vision-Language Pre-Training
Yunpeng Han (Harbin Institute of Technology), Zhao Cao (Huawei Technologies)
ClassificationRetrievalTransformerVision Language ModelContrastive LearningMultimodality
🎯 What it does: The FashionSAP model is proposed, achieving fine-grained visual language pre-training for fashion data through clothing symbols and attribute prompts.
Fast Contextual Scene Graph Generation With Unbiased Context Augmentation
Tianlei Jin (Zhejiang Lab), Wei Song (Zhejiang Lab)
Object DetectionGenerationGraph Neural NetworkGraph
🎯 What it does: A method for scene graph generation (SGG) is proposed that only uses contextual information of object categories and locations, and alleviates long-tail bias through context augmentation, subsequently constructing context-guided visual scene graph generation by combining visual information.
Fast Monocular Scene Reconstruction With Global-Sparse Local-Dense Grids
Wei Dong (Carnegie Mellon University), Anima Anandkumar (NVIDIA)
Depth EstimationOptimizationComputational EfficiencyImage
🎯 What it does: Utilizing a sparse global and dense local SDF voxel grid, rapid indoor scene reconstruction from a monocular image is achieved without the use of MLP.
Fast Point Cloud Generation With Straight Flows
Lemeng Wu (University of Texas at Austin), Qiang Liu (University of Texas at Austin)
GenerationData SynthesisKnowledge DistillationDiffusion modelFlow-based ModelNeural Radiance FieldPoint CloudOrdinary Differential Equation
🎯 What it does: A first-order point cloud generation model called Point Straight Flow (PSF) is proposed, which transforms the curved transport trajectory of diffusion models into a straight path and utilizes knowledge distillation to achieve high-quality point cloud generation in a single step.
FastInst: A Simple Query-Based Model for Real-Time Instance Segmentation
Junjie He (DAMO Academy, Alibaba Group), Xuansong Xie (DAMO Academy, Alibaba Group)
Object DetectionSegmentationTransformerImage
🎯 What it does: Proposes FastInst, a query-based real-time instance segmentation model;
FCC: Feature Clusters Compression for Long-Tailed Visual Recognition
Jian Li (Jilin University), Hao Xu (Jilin University)
RecognitionCompressionImage
🎯 What it does: Proposes the Feature Clusters Compression (FCC) method, which increases feature clustering density by scaling down backbone features, thereby enhancing long-tail visual recognition performance.
FeatER: An Efficient Network for Human Reconstruction via Feature Map-Based TransformER
Ce Zheng (University of Central Florida), Chen Chen (University of Central Florida)
Pose EstimationConvolutional Neural NetworkTransformerImage
🎯 What it does: A feature map-based Transformer network (FeatER) is proposed for human pose estimation and human shape reconstruction, which can directly process 2D feature maps while maintaining spatial structure in the Transformer.
Feature Aggregated Queries for Transformer-Based Video Object Detectors
Yiming Cui (University of Florida)
Object DetectionTransformerVideo
🎯 What it does: A Transformer-based video object detection method based on query aggregation is proposed, implemented in two forms: vanilla and dynamic.
Feature Alignment and Uniformity for Test Time Adaptation
Shuai Wang (Tsinghua University), Rui Li (Tsinghua University)
SegmentationDomain AdaptationKnowledge DistillationImageBiomedical Data
🎯 What it does: An online testing adaptive method based on feature unification and alignment is proposed to address the issue of model performance degradation under domain shift.
Feature Representation Learning With Adaptive Displacement Generation and Transformer Fusion for Micro-Expression Recognition
Zhijun Zhai (Wuhan University), Huijuan Zhao (Wuhan University)
RecognitionRepresentation LearningConvolutional Neural NetworkTransformerOptical FlowVideo
🎯 What it does: An end-to-end micro-expression recognition framework FRL-DGT is proposed, which utilizes a self-supervised Displacement Generation Module (DGM) to extract dynamic features between the start and peak frames, and classifies by fusing AU regions and full-face features through a multi-level Transformer.
Feature Separation and Recalibration for Adversarial Robustness
Woo Jae Kim (Korea Advanced Institute of Science and Technology), Sung-Eui Yoon (Korea Advanced Institute of Science and Technology)
Adversarial AttackConvolutional Neural NetworkImage
🎯 What it does: A Feature Separation and Recalibration (FSR) module is proposed, which enhances the model's adversarial robustness by separating intermediate features into robust and non-robust activations and recalibrating the non-robust activations.
Feature Shrinkage Pyramid for Camouflaged Object Detection With Transformers
Zhou Huang (Sichuan Changhong Electric Co., Ltd.), Huan Xiong (MBZUAI)
Object DetectionTransformerImage
🎯 What it does: This paper proposes a Feature Shrinkage Pyramid Network (FSPNet) based on Transformer for high-precision concealed object detection.
FeatureBooster: Boosting Feature Descriptors With a Lightweight Neural Network
Xinjiang Wang (Shanghai Jiao Tong University), Danping Zou (Shanghai Jiao Tong University)
Object DetectionRetrievalTransformerImage
🎯 What it does: A lightweight network is designed to enhance existing feature descriptors in images, generating more discriminative descriptors.
FedDM: Iterative Distribution Matching for Communication-Efficient Federated Learning
Yuanhao Xiong (University of California, Los Angeles), Cho-Jui Hsieh (University of California, Los Angeles)
Federated LearningComputational EfficiencyKnowledge DistillationImage
🎯 What it does: The FedDM method is proposed, which constructs local proxy functions by generating synthetic data at each client, allowing the server to update the global model based on these proxy functions, thus achieving communication-efficient federated learning.
Federated Domain Generalization With Generalization Adjustment
Ruipeng Zhang (Shanghai Jiao Tong University), Yanfeng Wang (Shanghai Jiao Tong University)
Domain AdaptationFederated LearningConvolutional Neural NetworkImage
🎯 What it does: A Federated Domain Generalization (FedDG) global objective and Generalization Adjustment (GA) algorithm is proposed, which utilizes dynamic weight adjustment to reduce the variance of the generalization gap in the source domains, thereby enhancing cross-domain generalization ability.
Federated Incremental Semantic Segmentation
Jiahua Dong (Chinese Academy of Sciences), Dengxin Dai (ETH Zürich)
SegmentationFederated LearningKnowledge DistillationImage
🎯 What it does: The Federated Incremental Semantic Segmentation (FISS) problem is proposed, and the FBL (Forgetting-Balanced Learning) model is presented to implement federated incremental semantic segmentation.
Federated Learning With Data-Agnostic Distribution Fusion
Jian-hui Duan (Nanjing University), Sanglu Lu (Nanjing University)
Federated LearningAuto EncoderImage
🎯 What it does: A federated learning aggregation method called FedFusion is proposed, which infers the global data distribution using virtual distribution components and dynamically adjusts aggregation weights;
FedSeg: Class-Heterogeneous Federated Learning for Semantic Segmentation
Jiaxu Miao (Zhejiang University), Yi Yang (Zhejiang University)
SegmentationFederated LearningContrastive LearningImage
🎯 What it does: This paper proposes a semantic segmentation framework for category heterogeneous federated learning (FedSeg), addressing the issues of foreground-background inconsistency and local optimization drift in data from different clients.
FEND: A Future Enhanced Distribution-Aware Contrastive Learning Framework for Long-Tail Trajectory Prediction
Yuning Wang (Xi'an Jiaotong University), Jianru Xue (Xi'an Jiaotong University)
Autonomous DrivingRepresentation LearningRecurrent Neural NetworkContrastive LearningTime SeriesSequential
🎯 What it does: The FEND framework is proposed to address the long-tail problem in trajectory prediction through future-enhanced distribution-aware contrastive learning and hypernetwork decoders.
Few-Shot Class-Incremental Learning via Class-Aware Bilateral Distillation
Linglan Zhao (Shanghai Jiao Tong University), Xiangzhong Fang (Shanghai Jiao Tong University)
ClassificationKnowledge DistillationImage
🎯 What it does: This paper proposes a new framework based on class-aware bidirectional distillation and attention aggregation to address the issues of overfitting and catastrophic forgetting in few-shot incremental learning.
Few-Shot Geometry-Aware Keypoint Localization
Xingzhe He (University of British Columbia), Pablo Garrido (Flawless AI)
Object DetectionPose EstimationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A framework for geometric consistency keypoint localization is proposed, utilizing a small number of labeled images and a large number of unlabeled images, combining self-supervised reconstruction, 3D geometric constraints, and uncertainty modeling to achieve few-shot keypoint detection for various objects.
Few-Shot Learning With Visual Distribution Calibration and Cross-Modal Distribution Alignment
Runqi Wang (Beihang University), Baochang Zhang (Beihang University)
ClassificationAdversarial AttackMeta LearningTransformerPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes a few-shot learning framework SADA based on a pre-trained vision-language model, combining two modules: Selective Attack (SA) and Cross-Modal Distribution Alignment (CMDA), which significantly enhances few-shot classification performance.
Few-Shot Non-Line-of-Sight Imaging With Signal-Surface Collaborative Regularization
Xintong Liu (Tsinghua University), Zuoqiang Shi (Beijing Institute of Mathematical Sciences and Applications)
RestorationImageBenchmark
🎯 What it does: A signal-surface collaborative regularization (SSCR) framework is proposed, which can achieve non-line-of-sight image reconstruction with a very small number of measurements (only 5×5 confocal).
Few-Shot Referring Relationships in Videos
Yogesh Kumar (Indian Institute of Technology Jodhpur), Anand Mishra (Indian Institute of Technology Jodhpur)
Object DetectionObject TrackingMeta LearningVideo
🎯 What it does: This paper proposes the task of Few-Shot Referring Relationship in Videos and provides a solution framework based on T-partite random fields.
Few-Shot Semantic Image Synthesis With Class Affinity Transfer
Marlène Careil (Télécom Paris), Stéphane Lathuilière (Télécom Paris)
SegmentationGenerationData SynthesisDomain AdaptationDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: This study focuses on few-shot semantic image synthesis and proposes the CAT (Class Affinity Transfer) method, which transfers a pre-trained source model to the target dataset, requiring only tens to hundreds of labeled images for training.
FFCV: Accelerating Training by Removing Data Bottlenecks
Guillaume Leclerc (Massachusetts Institute of Technology), Aleksander Mądry (Massachusetts Institute of Technology)
Computational EfficiencyConvolutional Neural NetworkImageVideo
🎯 What it does: Proposes the FFC-Vision-Computer-Learning (FFC-V) library to eliminate data bottlenecks during training and accelerate model training.
FFF: Fragment-Guided Flexible Fitting for Building Complete Protein Structures
Weijie Chen (DP Technology), Yuhang Wang (DP Technology)
Protein Structure PredictionConvolutional Neural NetworkBiomedical Data
🎯 What it does: A protein construction method FFF based on fragment recognition and flexible registration has been developed.
FFHQ-UV: Normalized Facial UV-Texture Dataset for 3D Face Reconstruction
Haoran Bai (Nanjing University of Science and Technology), Linchao Bao (Tencent AI Lab)
RestorationGenerationGenerative Adversarial NetworkImage
🎯 What it does: A large-scale high-quality facial UV texture dataset FFHQ-UV is proposed, along with a complete automated generation pipeline and a GAN-based texture decoder, further enhancing the accuracy and texture quality of 3D facial reconstruction from a single image.
FIANCEE: Faster Inference of Adversarial Networks via Conditional Early Exits
Polina Karpikova (Samsung AI Center), Aleksei Ivakhnenko (Samsung AI Center)
GenerationComputational EfficiencyGenerative Adversarial NetworkImage
🎯 What it does: This study proposes the insertion of variable-depth early exit branches in the generator of Generative Adversarial Networks (GANs) to dynamically select computation paths based on image difficulty, thereby accelerating the generation process.
Filtering, Distillation, and Hard Negatives for Vision-Language Pre-Training
Filip Radenovic (Meta AI), Dhruv Mahajan (Meta AI)
RetrievalKnowledge DistillationRepresentation LearningTransformerContrastive LearningImageMultimodality
🎯 What it does: This paper proposes an improved visual-language pre-training framework DiHT, which significantly enhances the performance of zero-shot tasks through three techniques: CAT filtering, concept distillation, and hard negative sample contrastive loss.
Finding Geometric Models by Clustering in the Consensus Space
Daniel Barath (ETH Zurich), Jiri Matas
OptimizationComputational EfficiencyPoint Cloud
🎯 What it does: A multi-instance geometric model fitting algorithm is proposed that does not rely on rigid point-model assignments, by finding dominant model instances through clustering in the consensus space.
Fine-Grained Audible Video Description
Xuyang Shen (Shanghai Artificial Intelligence Laboratory), Yiran Zhong (Shanghai Artificial Intelligence Laboratory)
TransformerVideoTextMultimodalityBenchmarkAudio
🎯 What it does: This paper proposes and implements the Fine-grained Audible Video Description (FAVD) task and constructs the first dataset for this task, FAVDBench.
Fine-Grained Classification With Noisy Labels
Qi Wei (Shandong University), Yilong Yin (Shandong University)
ClassificationRepresentation LearningContrastive LearningImage
🎯 What it does: A learning framework SNSCL is proposed to address the issue of noisy labels in fine-grained classification, mitigating the negative impact of noisy labels on representation learning.
Fine-Grained Face Swapping via Regional GAN Inversion
Zhian Liu (South China University of Technology), Yongwei Nie (South China University of Technology)
Image TranslationGenerationGenerative Adversarial NetworkImage
🎯 What it does: Proposes an E4S framework based on regional GAN inversion to achieve high-resolution, fine-grained facial swapping and editing.
Fine-Grained Image-Text Matching by Cross-Modal Hard Aligning Network
Zhengxin Pan (Zhejiang University), Bailing Zhang (NingboTech University)
RetrievalConvolutional Neural NetworkRecurrent Neural NetworkContrastive LearningImageTextMultimodality
🎯 What it does: A hard allocation coding-based image-text matching network CHAN is proposed, improving traditional cross-modal alignment methods;
Fine-Tuned CLIP Models Are Efficient Video Learners
Hanoona Rasheed (Mohamed bin Zayed University of AI), Fahad Shahbaz Khan (Linkoping University)
RecognitionComputational EfficiencyTransformerSupervised Fine-TuningPrompt EngineeringContrastive LearningVideo
🎯 What it does: Fine-tune the pre-trained CLIP model in the video domain to form the ViFi-CLIP baseline.
Finetune Like You Pretrain: Improved Finetuning of Zero-Shot Vision Models
Sachin Goyal (Carnegie Mellon University), Aditi Raghunathan (Carnegie Mellon University)
ClassificationDomain AdaptationPrompt EngineeringContrastive LearningImageText
🎯 What it does: This paper proposes a fine-tuning strategy called FLYP for pre-trained image-text models (such as CLIP), which directly uses the same contrastive loss as in pre-training and provides supervision through category text prompts.
FitMe: Deep Photorealistic 3D Morphable Model Avatars
Alexandros Lattas (Imperial College London), Stefanos Zafeiriou (Imperial College London)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: The FitMe system is proposed, which can reconstruct directly renderable high-fidelity 3D facial avatars from single or multiple 'wild' facial images.
Fix the Noise: Disentangling Source Feature for Controllable Domain Translation
Dongyeun Lee (Korea Advanced Institute of Science and Technology), Junmo Kim (Korea Advanced Institute of Science and Technology)
Image TranslationGenerationGenerative Adversarial NetworkImage
🎯 What it does: A FixNoise training strategy is proposed, which utilizes a fixed subspace of noise input in StyleGAN2 to maintain source domain features and achieve smooth control of cross-domain features within a single model.
FJMP: Factorized Joint Multi-Agent Motion Prediction Over Learned Directed Acyclic Interaction Graphs
Luke Rowe (University of Waterloo), Krzysztof Czarnecki (University of Waterloo)
Autonomous DrivingGraph Neural NetworkGraph
🎯 What it does: A factorization-based joint trajectory prediction framework named FJMP is proposed, which can generate scene-consistent multimodal future trajectories in multi-agent interactive driving scenarios.
FLAG3D: A 3D Fitness Activity Dataset With Language Instruction
Yansong Tang (Tsinghua University), Xiu Li (Tsinghua University)
RecognitionGenerationPose EstimationVideoTextMultimodality
🎯 What it does: The FLAG3D dataset is proposed, containing 180K segments of 60 types of daily fitness actions, equipped with high-precision MoCap 3D poses, SMPL parameters, synthetic videos, real camera videos, and fine-grained language descriptions.
FlatFormer: Flattened Window Attention for Efficient Point Cloud Transformer
Zhijian Liu (Massachusetts Institute of Technology), Song Han (Tsinghua University)
Object DetectionAutonomous DrivingComputational EfficiencyTransformerPoint Cloud
🎯 What it does: FlatFormer is proposed, an efficient point cloud Transformer that achieves low-latency 3D object detection through equal-sized grouping and window sorting.
FLEX: Full-Body Grasping Without Full-Body Grasps
Purva Tendulkar (Columbia University), Carl Vondrick (Columbia University)
Pose EstimationOptimizationRobotic IntelligenceMesh
🎯 What it does: A framework named FLEX is proposed, which generates full-body grasping poses for interacting with objects in complex scenes by combining hand grasping and body posture priors without using any full-body grasping data.
Flexible-Cm GAN: Towards Precise 3D Dose Prediction in Radiotherapy
Riqiang Gao (Siemens Healthineers), Ali Kamen (Siemens Healthineers)
GenerationGenerative Adversarial NetworkBiomedical DataStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: A conditional generative adversarial network (FCGAN) capable of handling multi-condition missing data is proposed for three-dimensional radiotherapy dose prediction.
FlexiViT: One Model for All Patch Sizes
Lucas Beyer (Google Research), Filip Pavetic (Google Research)
ClassificationRecognitionObject DetectionSegmentationKnowledge DistillationRepresentation LearningTransformerImage
🎯 What it does: During the training phase, different image patch sizes are randomly sampled, and the Patch Embedding and positional encoding are adjusted in size as needed during forward propagation, resulting in a single ViT model that maintains high performance across various patch sizes.
FlexNeRF: Photorealistic Free-Viewpoint Rendering of Moving Humans From Sparse Views
Vinoj Jayasundara (University of Maryland), Larry S. Davis (University of Maryland)
GenerationPose EstimationNeural Radiance FieldVideo
🎯 What it does: This paper proposes FlexNeRF, which utilizes sparse viewpoints in monocular videos to generate lighting rendering of moving human bodies.
Flow Supervision for Deformable NeRF
Chaoyang Wang (Carnegie Mellon University), Simon Lucey (University of Adelaide)
Data SynthesisNeural Radiance FieldOptical FlowVideoOrdinary Differential Equation
🎯 What it does: Proposes a Deformable NeRF method supervised by optical flow, directly using optical flow to constrain the deformation field, avoiding the need to invert the backward deformation field;
FlowFormer++: Masked Cost Volume Autoencoding for Pretraining Optical Flow Estimation
Xiaoyu Shi (Chinese University of Hong Kong), Hongsheng Li (Chinese University of Hong Kong)
TransformerAuto EncoderOptical FlowVideo
🎯 What it does: This paper proposes the Masked Cost Volume Autoencoding (MCVA) scheme based on FlowFormer, performing self-supervised pre-training on the cost volume convolutional encoder to enhance optical flow estimation performance.
FlowGrad: Controlling the Output of Generative ODEs With Gradients
Xingchao Liu (University of Texas at Austin), Qiang Liu (University of Texas at Austin)
GenerationOptimizationComputational EfficiencyFlow-based ModelImageTextOrdinary Differential Equation
🎯 What it does: This paper proposes an efficient framework called FlowGrad, which enables controllable generation through gradient optimization in pre-trained ODE generative models, particularly for text-guided image editing.
Focus on Details: Online Multi-Object Tracking With Diverse Fine-Grained Representation
Hao Ren (Huazhong University of Science and Technology), Faquan Wang (Huazhong University of Science and Technology)
Object TrackingVideo
🎯 What it does: This paper proposes the FineTrack framework, which utilizes multi-scale feature alignment and multi-head part mask generation to construct fine-grained appearance representations, enhancing identity recognition accuracy in multi-object tracking.
Focused and Collaborative Feedback Integration for Interactive Image Segmentation
Qiaoqiao Wei (Tsinghua University), Jun-Hai Yong (Tsinghua University)
SegmentationImageVideo
🎯 What it does: This paper proposes a click-interaction-based image segmentation framework called FCFI, which uses the segmentation results from the previous interaction as feedback to guide subsequent clicks.
Foundation Model Drives Weakly Incremental Learning for Semantic Segmentation
Chaohui Yu (Alibaba Group), Fan Wang (Alibaba Group)
SegmentationConvolutional Neural NetworkVision Language ModelContrastive LearningImage
🎯 What it does: This paper proposes the FMWISS framework, which collaboratively generates pseudo-pixel labels in weak incremental semantic segmentation tasks through a pre-trained vision-language model and a self-supervised model, and optimizes using a teacher-student structure and dense contrastive loss during incremental training;
Four-View Geometry With Unknown Radial Distortion
Petr Hruby (ETH Zurich), Viktor Larsson (Lund University)
Pose EstimationComputational EfficiencySimultaneous Localization and MappingImage
🎯 What it does: This paper proposes a minimal solver for the relative pose of a four-view camera (P3→P1) under unknown radial distortion, achieving solutions for 13 points (uncalibrated, calibrated, and upright scenes) and 7 points (upright scenes).
Frame Flexible Network
Yitian Zhang (Northeastern University), Yun Fu (Northeastern University)
RecognitionOptimizationComputational EfficiencyKnowledge DistillationVideo
🎯 What it does: A video recognition framework capable of maintaining high performance at different frame rates (Frame Flexible Network, FFN) is proposed.
Frame Interpolation Transformer and Uncertainty Guidance
Markus Plack (University of Bonn), Christopher Schroers (Disney Research Studios)
RestorationData SynthesisTransformerOptical FlowVideo
🎯 What it does: A video frame interpolation network based on Transformer is proposed, which simultaneously predicts the error map of the interpolation results, useful for rendering patch selection and quality inspection.
Frame-Event Alignment and Fusion Network for High Frame Rate Tracking
Jiqing Zhang (Dalian University of Technology), Xin Yang (Dalian University of Technology)
Object TrackingConvolutional Neural NetworkMultimodality
🎯 What it does: A high frame rate single object tracking method that utilizes the fusion of event cameras and RGB frames is proposed.
FREDOM: Fairness Domain Adaptation Approach to Semantic Scene Understanding
Thanh-Dat Truong (University of Arkansas), Khoa Luu (University of Arkansas)
SegmentationDomain AdaptationTransformerImage
🎯 What it does: This paper proposes a fair domain adaptation method for semantic scene segmentation called FREDOM, aimed at addressing the fairness issues caused by class imbalance during domain transfer.
FreeNeRF: Improving Few-Shot Neural Rendering With Free Frequency Regularization
Jiawei Yang (University of California Los Angeles), Yue Wang (Nvidia Research)
GenerationData SynthesisNeural Radiance FieldImage
🎯 What it does: This paper proposes a concise baseline method called FreeNeRF to address the few-shot neural rendering problem under sparse input.
FreeSeg: Unified, Universal and Open-Vocabulary Image Segmentation
Jie Qin (Institute of Automation, Chinese Academy of Sciences), Xingang Wang (Institute of Automation, Chinese Academy of Sciences)
SegmentationTransformerPrompt EngineeringVision Language ModelImage
🎯 What it does: We propose FreeSeg, a unified open-source vocabulary segmentation framework that can perform semantic, instance, and panoptic segmentation tasks under the same model and inference parameters.
Freestyle Layout-to-Image Synthesis
Han Xue (Shanghai Jiao Tong University), Wenjun Zhang (Shanghai Jiao Tong University)
Image TranslationGenerationData SynthesisDiffusion modelImageText
🎯 What it does: This paper proposes the task of Freestyle Layout to Image Synthesis (FLIS) and designs FreestyleNet, which combines a pre-trained text-to-image diffusion model (Stable Diffusion) with a new Rectified Cross-Attention (RCA) module to achieve high-fidelity image generation based on semantic masks and text.
Frequency-Modulated Point Cloud Rendering With Easy Editing
Yi Zhang, Wenjun Zhang
GenerationOptimizationComputational EfficiencyNeural Radiance FieldPoint Cloud
🎯 What it does: A point cloud rendering pipeline based on frequency modulation is proposed, capable of achieving high-fidelity detail reconstruction, real-time rendering, and user-friendly editing.
Fresnel Microfacet BRDF: Unification of Polari-Radiometric Surface-Body Reflection
Tomoki Ichikawa (Kyoto University), Ko Nishino (Kyoto University)
OptimizationImage
🎯 What it does: A BRDF model based on Fresnel microfacets (FMBRDF) is proposed, which can simultaneously describe surface reflection, volume reflection, Fresnel transmission, and polarization phenomena within the same microfacet set, and achieve model parameter inversion through a single known illuminated polarized image.
From Images to Textual Prompts: Zero-Shot Visual Question Answering With Frozen Large Language Models
Jiaxian Guo (University of Sydney), Steven Hoi (Salesforce Research)
RecognitionGenerationTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: A Plug-and-Play module called Img2LLM is proposed, which utilizes existing visual models to generate question-answer pairs and descriptions related to images, directly injecting them as prompts into any large language model (LLM) to accomplish zero-shot visual question answering (VQA) tasks.
From Node Interaction To Hop Interaction: New Effective and Scalable Graph Learning Paradigm
Jie Chen (Fudan University), Jian Pu (Fudan University)
Graph Neural NetworkContrastive LearningGraph
🎯 What it does: A new graph learning paradigm is proposed, shifting from node interaction to hop interaction, and the HopGNN framework is designed.
Frustratingly Easy Regularization on Representation Can Boost Deep Reinforcement Learning
Qiang He (Institute of Automation), Xinwen Hou (Institute of Automation)
Representation LearningReinforcement LearningSequentialBenchmark
🎯 What it does: The PEER regularization method is proposed, which explicitly constrains the internal representations of the Q network and its target network to maintain distinguishability, thereby enhancing the performance and sample efficiency of deep reinforcement learning.
FrustumFormer: Adaptive Instance-Aware Resampling for Multi-View 3D Detection
Yuqi Wang (Institute of Automation, Chinese Academy of Sciences), Zhaoxiang Zhang (Institute of Automation, Chinese Academy of Sciences)
Object DetectionAutonomous DrivingTransformerPoint Cloud
🎯 What it does: In the multi-view 3D object detection task, the authors propose the FrustumFormer framework, which utilizes instance frustums and Adaptive Instance-aware Resampling (AIR) to achieve more accurate BEV representations.
Full or Weak Annotations? An Adaptive Strategy for Budget-Constrained Annotation Campaigns
Javier Gamazo Tejero (University of Bern), Pablo Márquez-Neila
SegmentationOptimizationConvolutional Neural NetworkImage
🎯 What it does: An adaptive labeling strategy based on budget constraints is proposed to determine the ratio of collecting complete pixel annotations and image-level weak annotations during the construction of semantic segmentation datasets, aiming to achieve optimal model performance.
Fully Self-Supervised Depth Estimation From Defocus Clue
Haozhe Si (Shanghai AI Laboratory), Xuelong Li (Northwestern Polytechnical University)
Depth EstimationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a completely self-supervised depth estimation framework that utilizes sparse focus stacks without requiring true labels for depth or panoramic focus images.
Fusing Pre-Trained Language Models With Multimodal Prompts Through Reinforcement Learning
Youngjae Yu (Allen Institute for Artificial Intelligence), Yejin Choi (OpenAI)
GenerationTransformerLarge Language ModelReinforcement LearningVision Language ModelImageTextMultimodalityAudio
🎯 What it does: Proposes the ESPER framework, which utilizes reinforcement learning to extend pre-trained language models into text generators capable of handling multimodal inputs such as images and audio.
Fuzzy Positive Learning for Semi-Supervised Semantic Segmentation
Pengchong Qiao (Peking University), Jie Chen (Peking University)
SegmentationImage
🎯 What it does: This paper proposes a Fuzzy Positive Learning (FPL) framework that utilizes multiple fuzzy positive class labels for each pixel to perform semi-supervised semantic segmentation, avoiding the misguidance of the model by a single pseudo-label.
G-MSM: Unsupervised Multi-Shape Matching With Graph-Based Affinity Priors
Marvin Eisenberger (Technical University of Munich), Daniel Cremers (Technical University of Munich)
Graph Neural NetworkDiffusion modelMesh
🎯 What it does: This paper proposes an unsupervised multi-shape matching framework G-MSM, which utilizes shape graphs to learn the topological priors of shape collections and propagates multiple correspondences through shortest paths, significantly enhancing matching robustness.
GaitGCI: Generative Counterfactual Intervention for Gait Recognition
Huanzhang Dou (Zhejiang University), Xi Li (Zhejiang University)
RecognitionConvolutional Neural NetworkGenerative Adversarial NetworkVideo
🎯 What it does: This paper proposes a generative counterfactual intervention framework GaitGCI, aimed at suppressing confounding factors in gait recognition through counterfactual learning and diversity-constrained dynamic convolution, allowing the model to focus on interpretable key information related to body boundaries.
Galactic: Scaling End-to-End Reinforcement Learning for Rearrangement at 100k Steps-per-Second
Vincent-Pierre Berges (Meta AI), Eric Undersander (Meta AI)
Robotic IntelligenceRecurrent Neural NetworkReinforcement LearningTabularBenchmark
🎯 What it does: A Galactic framework is proposed, combining batch physical simulation with GPU rendering, supporting large-scale training of Fetch robots for movement control and object rearrangement tasks in indoor environments.
GALIP: Generative Adversarial CLIPs for Text-to-Image Synthesis
Ming Tao (Nanjing University of Posts and Telecommunications), Changsheng Xu (Chinese Academy of Sciences)
GenerationData SynthesisPrompt EngineeringGenerative Adversarial NetworkImageText
🎯 What it does: A CLIP-based Generative Adversarial Network (GALIP) has been developed for high-quality text-to-image synthesis.
GamutMLP: A Lightweight MLP for Color Loss Recovery
Hoang M. Le (York University), Michael S. Brown (Adobe Research)
RestorationMeta LearningImage
🎯 What it does: Recovering lost wide-gamut ProPhoto color values in sRGB images using a lightweight MLP specifically optimized for each image, with the model embedded in the image metadata.
GANHead: Towards Generative Animatable Neural Head Avatars
Sijing Wu (Shanghai Jiao Tong University), Guangtao Zhai (Shanghai Jiao Tong University)
GenerationGenerative Adversarial NetworkMesh
🎯 What it does: This paper proposes a generative animated neural head avatar model called GANHead, which can generate complete, realistic head avatars in 3D space that can be animated using FLAME parameters.
GANmouflage: 3D Object Nondetection With Texture Fields
Rui Guo (University of Michigan), Andrew Owens (XMotors.ai)
Object DetectionGenerationData SynthesisConvolutional Neural NetworkGenerative Adversarial NetworkImagePoint Cloud
🎯 What it does: Given the shape, position, and possible viewpoint distribution of an object, we learn to generate texture fields that make the object difficult to detect from multiple viewpoints, achieving concealment of three-dimensional objects.
GAPartNet: Cross-Category Domain-Generalizable Object Perception and Manipulation via Generalizable and Actionable Parts
Haoran Geng (Peking University), He Wang (Peking University)
SegmentationPose EstimationDomain AdaptationRobotic IntelligenceReinforcement LearningPoint Cloud
🎯 What it does: This paper proposes the concept of Generalizable and Operable Parts (GAPart), constructs a large-scale part-level interaction dataset GAPartNet, and trains a domain generalization model for 3D part segmentation and pose estimation based on this dataset, subsequently achieving cross-category object manipulation using the model.
GarmentTracking: Category-Level Garment Pose Tracking
Han Xue (Shanghai Qi Zhi Institute), Cewu Lu (Shanghai Jiao Tong University)
Object TrackingPose EstimationConvolutional Neural NetworkTransformerVideo
🎯 What it does: This study investigates category-level garment pose tracking and proposes a real-time online tracking framework called GarmentTracking. A large-scale garment deformation dataset, VR-Folding, was collected through a VR system.
Gated Multi-Resolution Transfer Network for Burst Restoration and Enhancement
Nancy Mehta, Fahad Shahbaz Khan
Restoration
🎯 What it does: Unable to access the content of the paper
Gated Stereo: Joint Depth Estimation From Gated and Wide-Baseline Active Stereo Cues
Stefanie Walz (Mercedes-Benz), Felix Heide (Algolux)
Depth EstimationAutonomous DrivingConvolutional Neural NetworkImage
🎯 What it does: A long-range high-resolution depth estimation method called Gated Stereo is proposed, which utilizes synchronized wide-baseline dual cameras to achieve dense depth prediction on active gated and passive HDR images.
Gaussian Label Distribution Learning for Spherical Image Object Detection
Hang Xu (Hangzhou Dianzi University), Feng Dai (Institute of Computing Technology, Chinese Academy of Sciences)
Object DetectionGaussian SplattingImage
🎯 What it does: This paper proposes a Gaussian Label Distribution Learning (GLDL)-based spherical bounding box regression loss and a corresponding dynamic sample selection strategy (GLDL-ATSS), achieving improvements on existing single and double-stage spherical detectors.
Gazeformer: Scalable, Effective and Fast Prediction of Goal-Directed Human Attention
Sounak Mondal (Stony Brook University), Minh Hoai (Stony Brook University)
Object DetectionComputational EfficiencyTransformerLarge Language ModelImage
🎯 What it does: A scalable, fast, and accurate visual search attention prediction model called Gazeformer is proposed, and the ZeroGaze zero-shot attention prediction task is introduced for the first time.
GazeNeRF: 3D-Aware Gaze Redirection With Neural Radiance Fields
Alessandro Ruzzi (ETH Zurich), Otmar Hilliges (ETH Zurich)
GenerationData SynthesisPose EstimationNeural Radiance FieldImage
🎯 What it does: A 3D gaze redirection method based on neural radiance fields, called GazeNeRF, is proposed.
GCFAgg: Global and Cross-View Feature Aggregation for Multi-View Clustering
Weiqing Yan (Yantai University), Weisi Lin (Shenzhen University)
OptimizationTransformerAuto EncoderContrastive LearningMultimodality
🎯 What it does: A multi-view clustering framework named GCFAggMVC is proposed, which obtains consensus representation through global and cross-view feature aggregation, and enhances clustering performance using structure-guided contrastive learning.
GD-MAE: Generative Decoder for MAE Pre-Training on LiDAR Point Clouds
Honghui Yang (Zhejiang University), Wanli Ouyang (Shanghai AI Laboratory)
Object DetectionAutonomous DrivingTransformerGenerative Adversarial NetworkPoint Cloud
🎯 What it does: This paper proposes GD-MAE, a simplified MAE pre-training framework that utilizes a generative decoder to achieve multi-scale self-supervised learning on LiDAR point clouds.
GEN: Pushing the Limits of Softmax-Based Out-of-Distribution Detection
Xixi Liu (Chalmers University of Technology), Christopher Zach (Chalmers University of Technology)
Anomaly DetectionImage
🎯 What it does: A universal entropy score (GEN) is proposed for OOD detection, which only utilizes the pre-trained softmax output.
GeneCIS: A Benchmark for General Conditional Image Similarity
Sagar Vaze (Meta AI), Ishan Misra (VGG)
RetrievalTransformerVision Language ModelContrastive LearningImageTextBenchmark
🎯 What it does: This paper proposes the GeneCIS benchmark for evaluating zero-shot image similarity judgment of visual models under different text conditions.