ICCV 2023 Papers — Page 7
IEEE/CVF International Conference on Computer Vision · 2156 papers
EigenPlaces: Training Viewpoint Robust Models for Visual Place Recognition
Gabriele Berton (Politecnico di Torino), Carlo Masone (Politecnico di Torino)
RecognitionRetrievalConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: In the visual place recognition task, a training method is proposed that does not require additional annotations, utilizing map partitioning and principal component analysis to automatically construct multi-view classes of the same location, thereby training a global descriptor that is robust to viewpoint changes.
EigenTrajectory: Low-Rank Descriptors for Multi-Modal Trajectory Forecasting
Inhwan Bae (GIST AI Graduate School), Hae-Gon Jeon (GIST AI Graduate School)
Autonomous DrivingOptimizationGraph Neural NetworkMultimodalityTime Series
🎯 What it does: By performing low-rank approximation on pedestrian trajectories, an EigenTrajectory descriptor is constructed, projecting observed and predicted trajectories into a low-dimensional ET space, where multimodal prediction and trajectory anchor refinement are achieved.
ElasticViT: Conflict-aware Supernet Training for Deploying Fast Vision Transformer on Diverse Mobile Devices
Chen Tang (Tsinghua University), Mao Yang (Microsoft Research)
ClassificationObject DetectionOptimizationNeural Architecture SearchTransformerImage
🎯 What it does: This study proposes ElasticViT, a two-stage neural architecture search method designed for deploying high-accuracy, low-latency visual Transformer (ViT) models on various mobile devices.
ELFNet: Evidential Local-global Fusion for Stereo Matching
Jieming Lou (Institute for Infocomm Research), Jun Cheng (Institute for Infocomm Research)
Depth EstimationDomain AdaptationAutonomous DrivingTransformerSupervised Fine-TuningImage
🎯 What it does: This paper proposes ELFNet, which integrates two matching approaches: cost volume (PCWNet) and Transformer (STTR), and introduces evidence learning to estimate uncertainty in each branch, achieving trustworthy fusion through a two-stage process (multi-scale internal fusion + cross-model fusion).
ELITE: Encoding Visual Concepts into Textual Embeddings for Customized Text-to-Image Generation
Yuxiang Wei (Harbin Institute of Technology), Wangmeng Zuo (Harbin Institute of Technology)
GenerationData SynthesisTransformerDiffusion modelContrastive LearningImageText
🎯 What it does: This paper proposes a learning-based encoder called ELITE, which quickly and accurately maps visual concepts to text embedding space, enabling customized text-to-image generation.
EMDB: The Electromagnetic Database of Global 3D Human Pose and Shape in the Wild
Manuel Kaufmann (ETH Zurich), Otmar Hilliges (ETH Zurich)
Pose EstimationSimultaneous Localization and MappingVideoMultimodality
🎯 What it does: Designed and released the EMDB dataset and the EMP method, achieving precise acquisition of full-body SMPL pose, shape, and global trajectory in outdoor scenes through radio electromagnetic sensors, RGB-D, and camera poses.
EMMN: Emotional Motion Memory Network for Audio-driven Emotional Talking Face Generation
Shuai Tan (Shanghai Jiao Tong University), Ye Pan (Shanghai Jiao Tong University)
RecognitionGenerationData SynthesisImageVideoAudio
🎯 What it does: An Emotion Motion Memory Network (EMMN) is proposed, which can generate a full-face emotional talking face using only a reference face image and emotional audio.
EmoSet: A Large-scale Visual Emotion Dataset with Rich Attributes
Jingyuan Yang (Shenzhen University), Hui Huang (Shenzhen University)
ClassificationRecognitionConvolutional Neural NetworkImage
🎯 What it does: A visual emotion dataset called EmoSet has been constructed, containing 3.3 million images (118,000 manually annotated), with each image labeled with 8 emotion categories and 6 visual attributes (brightness, saturation, scene, object, facial expression, human action). An attribute auxiliary module was also designed to enhance emotion recognition.
EmoTalk: Speech-Driven Emotional Disentanglement for 3D Face Animation
Ziqiao Peng (Renmin University of China), Zhaoxin Fan (Psyche AI Inc.)
GenerationTransformerSupervised Fine-TuningAudio
🎯 What it does: This paper presents EmoTalk, a speech-driven 3D facial animation model that achieves more realistic emotional expression through emotion decoupling.
Emotional Listener Portrait: Neural Listener Head Generation with Emotion
Luchuan Song (University of Rochester), Chenliang Xu (University of Rochester)
GenerationRecurrent Neural NetworkVideoMultimodalityAudio
🎯 What it does: This paper proposes the Emotional Listening Portrait (ELP) framework, which automatically generates realistic listener head movements in dialogue based on the speaker's emotions and voice.
Empowering Low-Light Image Enhancer through Customized Learnable Priors
Naishan Zheng (University of Science and Technology of China), Feng Zhao (University of Science and Technology of China)
RestorationConvolutional Neural NetworkAuto EncoderImage
🎯 What it does: This paper proposes a customizable and learnable prior deep unrolling framework (CUE) for low-light image enhancement, utilizing Masked Autoencoder pre-trained illumination and noise priors, which are embedded in the proximal operations and regularization terms of the Retinex unrolling steps, thereby achieving a more transparent and interpretable enhancement model.
EMQ: Evolving Training-free Proxies for Automated Mixed Precision Quantization
Peijie Dong (National University of Defense Technology), Hengyue Pan (National University of Defense Technology)
OptimizationComputational EfficiencyConvolutional Neural NetworkReinforcement LearningImageBenchmark
🎯 What it does: This paper proposes a training-free Mixed Precision Quantization Proxy (MQ Proxy), which can quickly evaluate the quantization accuracy of different bit-width configurations without requiring additional training, and achieves efficient quantization bit-width allocation on ImageNet.
EMR-MSF: Self-Supervised Recurrent Monocular Scene Flow Exploiting Ego-Motion Rigidity
Zijie Jiang (Tokyo Institute of Technology), Masatoshi Okutomi (Tokyo Institute of Technology)
Depth EstimationAutonomous DrivingRecurrent Neural NetworkOptical FlowImage
🎯 What it does: This paper proposes a self-supervised monocular scene flow estimation framework EMR-MSF, which jointly predicts depth, dense SE3 motion fields, and camera motion, and derives scene flow from these.
Encyclopedic VQA: Visual Questions About Detailed Properties of Fine-Grained Categories
Thomas Mensink (Google Research), Vittorio Ferrari (Google Research)
RetrievalTransformerVision Language ModelTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: The Encyclopedic-VQA dataset is proposed, focusing on detailed attribute question answering for fine-grained categories and instances, and providing a traceable knowledge base.
End-to-end 3D Tracking with Decoupled Queries
Yanwei Li (Chinese University of Hong Kong), Jose Alvarez (NVIDIA)
Object TrackingAutonomous DrivingTransformerPoint Cloud
🎯 What it does: An end-to-end 3D multi-object tracking framework called DQTrack is proposed, which decouples object queries from trajectory queries to address task conflicts associated with single queries.
End-to-End Diffusion Latent Optimization Improves Classifier Guidance
Bram Wallace (Salesforce AI), Nikhil Naik (Salesforce AI)
GenerationOptimizationDiffusion modelImageMultimodality
🎯 What it does: By directly performing gradient optimization on the noise vector of the diffusion model (DOODL), precise guidance is applied to the final generated images using pre-trained discriminative networks (such as CLIP, FGVC classifiers, and aesthetic scorers), achieving multimodal, personalized, and aesthetic optimization without the need to retrain noise-aware classifiers.
End2End Multi-View Feature Matching with Differentiable Pose Optimization
Barbara Roessle (Technical University of Munich), Matthias Nießner (Technical University of Munich)
Pose EstimationOptimizationGraph Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper proposes a differentiable end-to-end training framework that integrates multi-view feature matching and pose optimization through a graph attention network, capable of completing multi-frame feature matching and pose estimation in a single inference, and automatically learning and down-weighting the confidence of erroneous matches using the gradient backpropagation of pose optimization.
Energy-based Self-Training and Normalization for Unsupervised Domain Adaptation
Samitha Herath (Monash University), Gholamreza Haffari (Monash University)
Domain AdaptationTransformerImage
🎯 What it does: This paper proposes an unsupervised domain adaptation method based on an energy model, achieving cross-domain feature alignment through energy self-training and energy normalization.
Enhanced Meta Label Correction for Coping with Label Corruption
Mitchell Keren Taraday (Technion Israel Institute of Technology), Chaim Baskin (Technion Israel Institute of Technology)
OptimizationMeta LearningImage
🎯 What it does: An improved meta-label correction framework EMLC is proposed to enhance learning performance on noisy label data.
Enhanced Soft Label for Semi-Supervised Semantic Segmentation
Jie Ma (Sun Yat-sen University), Guanbin Li (Sun Yat-sen University)
SegmentationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes a semi-supervised semantic segmentation framework based on enhanced soft labels (ESL), which combines dynamic soft labels (DSL) with pixel-to-part contrastive learning to fully utilize high-entropy pseudo-label information and improve category boundary recognition capabilities.
Enhancing Adversarial Robustness in Low-Label Regime via Adaptively Weighted Regularization and Knowledge Distillation
Dongyoon Yang (Seoul National University), Yongdai Kim (Seoul National University)
Knowledge DistillationAdversarial AttackImage
🎯 What it does: In the scenario of limited labeled data, a semi-supervised adversarial training method is proposed that combines adaptive weighted regularization with semi-supervised knowledge distillation;
Enhancing Fine-Tuning Based Backdoor Defense with Sharpness-Aware Minimization
Mingli Zhu (Chinese University of Hong Kong), Baoyuan Wu (Chinese University of Hong Kong)
ClassificationOptimizationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A fine-tuning strategy based on Sharpness-Aware Minimization (SAM) called FT-SAM is proposed to more effectively eliminate implanted backdoors when only a small number of clean samples are available.
Enhancing Generalization of Universal Adversarial Perturbation through Gradient Aggregation
Xuannan Liu (Beijing University of Posts and Telecommunications), Weihong Deng (Beijing University of Posts and Telecommunications)
OptimizationAdversarial AttackImage
🎯 What it does: A general adversarial perturbation generation method based on Stochastic Gradient Aggregation (SGA) is proposed, significantly enhancing the generalization ability of UAP.
Enhancing Modality-Agnostic Representations via Meta-Learning for Brain Tumor Segmentation
Aishik Konwer (Stony Brook University), Prateek Prasanna (Stony Brook University)
SegmentationMeta LearningTransformerGenerative Adversarial NetworkImageMultimodalityBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: A modality-independent brain tumor segmentation framework based on meta-learning and adversarial learning is proposed, capable of training and inference with only partial modality samples and a small number of complete modality samples.
Enhancing NeRF akin to Enhancing LLMs: Generalizable NeRF Transformer with Mixture-of-View-Experts
Wenyan Cong (University of Texas at Austin), Zhangyang Wang (University of Texas at Austin)
GenerationData SynthesisTransformerMixture of ExpertsNeural Radiance FieldImage
🎯 What it does: A general NeRF model GNT-MOVE has been constructed that can directly synthesize new views in unseen scenes.
Enhancing Non-line-of-sight Imaging via Learnable Inverse Kernel and Attention Mechanisms
Yanhua Yu (ShanghaiTech University), Shiying Li (ShanghaiTech University)
RestorationDepth EstimationConvolutional Neural NetworkImage
🎯 What it does: An end-to-end deep learning framework is proposed, which combines a learnable frequency domain inverse kernel with self-attention and cross-attention to achieve high-frequency detail reconstruction for non-line-of-sight (NLOS) imaging.
Enhancing Privacy Preservation in Federated Learning via Learning Rate Perturbation
Guangnian Wan (Beijing University of Posts and Telecommunications), Jie Xu (Beijing University of Posts and Telecommunications)
Federated LearningSafty and PrivacyImage
🎯 What it does: This paper proposes a random perturbation of the learning rate for each client in federated learning (Learning Rate Perturbation, LRP) and provides an adaptive version (Ada-LRP) for classification tasks, thereby enhancing privacy protection against gradient inversion attacks without significantly affecting model accuracy.
Enhancing Sample Utilization through Sample Adaptive Augmentation in Semi-Supervised Learning
Guan Gui (Nanjing University), Yinghuan Shi (Nanjing University)
ClassificationData-Centric LearningContrastive LearningImage
🎯 What it does: A Sample Adaptive Augmentation (SAA) framework is proposed to identify and apply more diverse augmentations to 'naive' samples in semi-supervised learning;
ENTL: Embodied Navigation Trajectory Learner
Klemen Kotar (Stanford University), Roozbeh Mottaghi (Meta AI)
Robotic IntelligenceTransformerGenerative Adversarial NetworkSimultaneous Localization and MappingVideoSequential
🎯 What it does: This paper proposes the Embodied Navigation Trajectory Learner (ENTL), a framework that unifies world modeling, localization, and imitation learning into a self-supervised sequence prediction task for learning long sequences of navigation representations.
ENVIDR: Implicit Differentiable Renderer with Neural Environment Lighting
Ruofan Liang (University of Toronto), Nandita Vijaykumar (University of Toronto)
GenerationData SynthesisNeural Radiance FieldImage
🎯 What it does: The ENVIDR framework is proposed, which combines a neural renderer with a decomposed rendering component and an SDF-based neural surface model, capable of high-quality reconstruction and rendering of glossy and specular objects, while supporting environment relighting and material editing.
Environment Agnostic Representation for Visual Reinforcement Learning
Hyesong Choi (Ewha Womans University), Dongbo Min (Ewha Womans University)
Robotic IntelligenceReinforcement LearningImageVideo
🎯 What it does: An Environment-Agnostic Reinforcement Learning (EAR) framework is proposed, which utilizes feature decomposition, reconstruction, and action-based state shift self-supervised objectives to extract environment-agnostic features for visual RL policy learning.
Environment-Invariant Curriculum Relation Learning for Fine-Grained Scene Graph Generation
Yukuan Min (Xidian University), Cheng Deng (Xidian University)
Object DetectionGenerationGraph Neural NetworkSupervised Fine-TuningImageGraph
🎯 What it does: A framework named EICR is proposed, providing a unified solution to the issues of predicate category imbalance and subject-object pair context imbalance in scene graph generation.
eP-ALM: Efficient Perceptual Augmentation of Language Models
Mustafa Shukor (Sorbonne University), Matthieu Cord (Valeo)
OptimizationComputational EfficiencyTransformerLarge Language ModelPrompt EngineeringVision Language ModelVideoTextMultimodalityAudio
🎯 What it does: We propose eP-ALM, a multimodal adaptation method that efficiently integrates large language models with visual/video/audio encoders using only a few linear projection layers and soft prompts, while nearly freezing all parameters.
EP2P-Loc: End-to-End 3D Point to 2D Pixel Localization for Large-Scale Visual Localization
Minjung Kim (Seoul National University), Gunhee Kim (Seoul National University)
Pose EstimationRetrievalTransformerSimultaneous Localization and MappingImagePoint Cloud
🎯 What it does: This paper proposes an end-to-end 3D point to 2D pixel localization method EP2P-Loc for large-scale visual localization;
EPiC: Ensemble of Partial Point Clouds for Robust Classification
Meir Yossef Levi (Technion Israel Institute of Technology), Guy Gilboa (Technion Israel Institute of Technology)
ClassificationGraph Neural NetworkPoint Cloud
🎯 What it does: Proposes a point cloud classification ensemble framework EPiC based on local and global sampling, training each sub-network using only a portion of the point cloud, thereby significantly enhancing robustness against different OOD damages.
EQ-Net: Elastic Quantization Neural Networks
Ke Xu (Anhui University), Xingyi Zhang (Anhui University)
CompressionOptimizationKnowledge DistillationHyperparameter SearchConvolutional Neural NetworkImage
🎯 What it does: A flexible quantization neural network (EQ-Net) is proposed, which can generate quantized sub-networks of different bit widths, symmetric/asymmetric, and adjustable granularity through a single training session.
Equivariant Similarity for Vision-Language Foundation Models
Tan Wang (Nanyang Technological University), Lijuan Wang (Microsoft)
RetrievalTransformerVision Language ModelContrastive LearningImageVideoTextMultimodalityBenchmark
🎯 What it does: This paper proposes and implements Equivariant Similarity Learning (EQSIM) in Vision-Language Models (VLM) and constructs a benchmark focused on visual minor changes (EQBEN), aiming to enhance the model's robustness under fine-grained semantic variations.
Erasing Concepts from Diffusion Models
Rohit Gandikota (Northeastern University), David Bau (Northeastern University)
GenerationData SynthesisDiffusion modelScore-based ModelImageText
🎯 What it does: A method is proposed for text-to-image diffusion models that allows for the erasure of specified visual concepts using only concept names without any additional data on pre-trained model weights.
ESSAformer: Efficient Transformer for Hyperspectral Image Super-resolution
Mingjin Zhang (Xidian University), Jing Zhang (Xidian University)
RestorationSuper ResolutionTransformerImage
🎯 What it does: Proposes the ESSAformer Transformer network for single-image hyperspectral image super-resolution tasks.
Essential Matrix Estimation using Convex Relaxations in Orthogonal Space
Arman Karimian (Boston University), Roberto Tron (Boston University)
Pose EstimationOptimizationSimultaneous Localization and MappingImage
🎯 What it does: This paper proposes a new two-view structure from motion (SfM) method that uses a four-dimensional rotation matrix embedding to estimate the essential matrix, achieving a globally optimal solution through semidefinite relaxation and Riemannian gradient descent.
ESTextSpotter: Towards Better Scene Text Spotting with Explicit Synergy in Transformer
Mingxin Huang (South China University of Technology), Lianwen Jin (South China University of Technology)
RecognitionObject DetectionTransformerVision Language ModelImage
🎯 What it does: This paper proposes an explicit collaborative scene text detection and recognition framework based on Transformer, called ESTextSpotter, which employs task-aware queries and a visual-language communication module to achieve explicit collaboration between detection and recognition within a single decoder.
Estimator Meets Equilibrium Perspective: A Rectified Straight Through Estimator for Binary Neural Networks Training
Xiao-Ming Wu (Sun Yat-sen University), Wei-Shi Zheng (Sun Yat-sen University)
OptimizationConvolutional Neural NetworkImage
🎯 What it does: A new binary network training method called ReSTE is proposed, which improves STE to balance estimation error and gradient stability, achieving efficient binary training.
ETran: Energy-Based Transferability Estimation
Mohsen Gholami (Huawei Technologies Canada), Yong Zhang (Huawei Technologies Canada)
ClassificationObject DetectionImage
🎯 What it does: This paper proposes an energy-based transferability estimation method, ETran, for quickly assessing the transferability of pre-trained models on target classification or detection tasks, and provides corresponding rankings.
Eulerian Single-Photon Vision
Shantanu Gupta (University of Wisconsin Madison), Mohit Gupta (University of Wisconsin Madison)
Object DetectionObject TrackingComputational EfficiencyOptical FlowVideo
🎯 What it does: A single-photon vision method based on the Eulerian approach is proposed, which performs phase-related lightweight filtering directly on the original binary photon data to achieve edge detection and motion estimation without the need for image reconstruction.
Evaluating Data Attribution for Text-to-Image Models
Sheng-Yu Wang (Carnegie Mellon University), Richard Zhang (Adobe Research)
GenerationData SynthesisRetrievalDiffusion modelContrastive LearningImage
🎯 What it does: This paper constructs a 'plus one' attribution dataset generated based on Custom Diffusion, utilizing this dataset to evaluate and improve the training data attribution methods for text-to-image models.
Evaluation and Improvement of Interpretability for Self-Explainable Part-Prototype Networks
Qihan Huang (Zhejiang University), Mingli Song (Zhejiang University)
Explainability and InterpretabilityConvolutional Neural NetworkImage
🎯 What it does: Proposed quantifiable metrics for evaluating the interpretability of part prototype networks, and based on this, improved ProtoPNet to enhance its prototype consistency and robustness.
Event Camera Data Pre-training
Yan Yang (Australian National University), Liu Liu (Huawei)
Object DetectionSegmentationRepresentation LearningTransformerContrastive LearningOptical FlowImageMultimodality
🎯 What it does: Using a self-supervised learning framework, pre-training is conducted on event camera data and corresponding RGB images to obtain a transferable event network.
Event-based Temporally Dense Optical Flow Estimation with Sequential Learning
Wachirawit Ponghiran (Purdue University), Kaushik Roy (Purdue University)
Autonomous DrivingRecurrent Neural NetworkSpiking Neural NetworkOptical FlowSequential
🎯 What it does: This paper designs two event camera optical flow estimation networks, LSTM-FlowNet and EfficientSpike-FlowNet, achieving temporally dense optical flow inference at a frequency of 100 Hz using the event count sequence of each pixel.
Event-Guided Procedure Planning from Instructional Videos with Text Supervision
An-Lan Wang (Sun Yat-sen University), Wei-Shi Zheng (Sun Yat-sen University)
TransformerPrompt EngineeringVideoText
🎯 What it does: This paper proposes an event-driven program planning method that infers intermediate action sequences from initial and target visual states, and completes learning using text supervision.
Eventful Transformers: Leveraging Temporal Redundancy in Vision Transformers
Matthew Dutson (University of Wisconsin Madison), Mohit Gupta (University of Wisconsin Madison)
Object DetectionComputational EfficiencyTransformerVideo
🎯 What it does: This paper proposes Eventful Transformers, which utilize temporal redundancy in video Transformer inference through token gating and delta mechanisms, significantly reducing computational load by only recalculating tokens with significant changes.
EverLight: Indoor-Outdoor Editable HDR Lighting Estimation
Mohammad Reza Karimi Dastjerdi (Universite Laval), Jean-François Lalonde (Universite Laval)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: EverLight predicts editable HDR 360° environment lighting from a single image while simultaneously generating high-quality reflection textures.
Examining Autoexposure for Challenging Scenes
SaiKiran Tedla (York University), Michael S. Brown (York University)
OptimizationImage
🎯 What it does: This paper constructs a four-dimensional exposure dataset and develops an AE evaluation platform to test and visualize the performance of different automatic exposure algorithms in challenging lighting scenarios.
ExBluRF: Efficient Radiance Fields for Extreme Motion Blurred Images
Dongwoo Lee (Seoul National University), Kyoung Mu Lee (Seoul National University)
RestorationNeural Radiance FieldImage
🎯 What it does: The study focuses on how to recover clear 3D scenes from extremely motion-blurred multi-view images and generate sharp view synthesis results.
Exemplar-Free Continual Transformer with Convolutions
Anurag Roy (Indian Institute of Technology Kharagpur), Abir Das (Indian Institute of Technology Kharagpur)
ClassificationRecognitionTransformerImage
🎯 What it does: A continuous learning framework named ConTraCon is designed, which utilizes convolution to learnably reweight the self-attention weights of the Transformer and achieves sample-free memory and continuous learning without task IDs through a learnable skip-gate for parameter reuse.
Explaining Adversarial Robustness of Neural Networks from Clustering Effect Perspective
Yulin Jin (Xidian University), Xiaofeng Chen (Xidian University)
Adversarial AttackConvolutional Neural NetworkImage
🎯 What it does: This paper proposes and experimentally verifies the clustering effect present in the forward propagation of neural networks, demonstrating its connection to information bottleneck theory. Based on this, it designs Sufficient Adversarial Training (SAT), which incorporates regularization of the perturbation magnitude in intermediate layers during the internal maximization of adversarial training, thereby expanding the perturbation search space and enhancing the model's robustness against Output Layer Attacks (OLA) and Intermediate Layer Attacks (ILA).
Explicit Motion Disentangling for Efficient Optical Flow Estimation
Changxing Deng (University of Macau), Shuaicheng Liu (Megvii Technology)
Computational EfficiencyTransformerOptical FlowImageVideo
🎯 What it does: An Explicit Motion Disentangling (EMD-Flow) framework is proposed, which explicitly separates global motion learning from local refinement, and achieves efficient optical flow estimation through two lightweight modules: Multi-scale Motion Aggregation (MMA) and Confidence-induced Flow Propagation (CFP).
Exploiting Proximity-Aware Tasks for Embodied Social Navigation
Enrico Cancelli (University of Padova), Lamberto Ballan (University of Padova)
Robotic IntelligenceRecurrent Neural NetworkReinforcement LearningPoint Cloud
🎯 What it does: This paper proposes an end-to-end social navigation architecture that injects common-sense cognition of human-machine interaction into reinforcement learning strategies through two types of 'proximity perception tasks'—Risk Estimation and Proximity Compass—achieving safe and efficient robot navigation in human environments.
Explore and Tell: Embodied Visual Captioning in 3D Environments
Anwen Hu (Renmin University of China), Qin Jin (Renmin University of China)
GenerationRobotic IntelligenceTransformerVision-Language-Action ModelImageTextPoint Cloud
🎯 What it does: This paper proposes the Embodied Captioning task, which requires an agent to navigate from random viewpoints in a 3D environment, collect multi-view information, and generate complete paragraph descriptions.
Exploring Group Video Captioning with Efficient Relational Approximation
Wang Lin (Zhejiang University), Zhou Zhao (Zhejiang University)
GenerationTransformerContrastive LearningVideoText
🎯 What it does: This paper proposes and implements a subtitle generation task for video groups, capable of generating descriptions given a target video group and a reference video group.
Exploring Lightweight Hierarchical Vision Transformers for Efficient Visual Tracking
Ben Kang (Dalian University of Technology), Huchuan Lu (Dalian University of Technology)
Object TrackingTransformerVideo
🎯 What it does: A lightweight hierarchical Vision Transformer tracking framework HiT is proposed, which utilizes a Bridge Module to fuse deep semantic features with shallow detail features, and enhances the relationship between the search and template through dual-image position encoding, constructing an efficient real-time visual tracker.
Exploring Model Transferability through the Lens of Potential Energy
Xiaotong Li (Peking University), Ling-Yu Duan (Peking University)
ClassificationDomain AdaptationRepresentation LearningContrastive LearningImagePhysics Related
🎯 What it does: A physics-driven method based on potential energy descent is proposed to simulate the representation dynamics in the process of transfer learning, thereby improving the transferability assessment of pre-trained models.
Exploring Object-Centric Temporal Modeling for Efficient Multi-View 3D Object Detection
Shihao Wang (Beijing Institute of Technology), Xiangyu Zhang (MEGVII Technology)
Object DetectionAutonomous DrivingTransformerVideoPoint Cloud
🎯 What it does: The StreamPETR framework is proposed, achieving online long-sequence 3D detection based on object queries.
Exploring Open-Vocabulary Semantic Segmentation from CLIP Vision Encoder Distillation Only
Jun Chen (King Abdullah University of Science and Technology), Mohamed Elhoseiny (King Abdullah University of Science and Technology)
SegmentationKnowledge DistillationTransformerAuto EncoderContrastive LearningImage
🎯 What it does: We propose ZeroSeg, a zero-shot, open-vocabulary semantic segmentation model that distills knowledge solely from a pre-trained CLIP visual encoder to achieve pixel-level segmentation without the need for pixel-level or text annotations.
Exploring Positional Characteristics of Dual-Pixel Data for Camera Autofocus
Myungsub Choi (Samsung Advanced Institute of Technology), Hyong-euk Lee (Samsung Advanced Institute of Technology)
OptimizationComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: An automatic focusing model based on a dual-pixel camera is proposed, which learns spatially correlated errors through a position encoding method to enhance focusing accuracy.
Exploring Predicate Visual Context in Detecting of Human-Object Interactions
Frederic Z Zhang, Stephen Gould (Australian National University)
ClassificationObject DetectionTransformerImage
🎯 What it does: This paper proposes an improved two-stage Transformer structure for detecting human-object interactions (HOI), enhancing predicate classification by incorporating box-pair-based positional information and richer image context into the cross-attention mechanism.
Exploring Temporal Concurrency for Video-Language Representation Learning
Heng Zhang (Renmin University of China), Dacheng Tao (University of Sydney)
RetrievalRepresentation LearningTransformerContrastive LearningVideoTextMultimodality
🎯 What it does: A modeling method is proposed that treats video and text as Temporal Concurrent Processes (TCP), utilizing soft Dynamic Time Warping (soft-DTW) for cross-modal alignment, and maintaining internal temporal dynamic consistency across modalities through Brownian bridge regularization.
Exploring Temporal Frequency Spectrum in Deep Video Deblurring
Qi Zhu (University of Science and Technology of China), Feng Zhao (University of Science and Technology of China)
RestorationConvolutional Neural NetworkVideo
🎯 What it does: A deep video deblurring network based on frequency domain priors is proposed, which utilizes discrete Fourier transform to extract spectral features and introduces spectral information in four stages: feature extraction, alignment, fusion, and optimization.
Exploring the Benefits of Visual Prompting in Differential Privacy
Yizhe Li (Xi'an Jiaotong University), Xuebin Ren (IBM Research)
ClassificationSafty and PrivacyTransformerPrompt EngineeringImage
🎯 What it does: Utilizing visual prompting to reshape pre-trained models, using them as teacher models, and combining with the PATE framework for differential privacy training to construct a high-accuracy differential privacy image classifier.
Exploring the Sim2Real Gap Using Digital Twins
Sruthi Sudhakar (Columbia University), Vibhav Vineet (Microsoft Research)
Object DetectionSegmentationConvolutional Neural NetworkImage
🎯 What it does: Constructed two digital twin datasets, YCB-Real and YCB-Synthetic, systematically introducing five types of defects (noise, holes, texture blur, baked lighting, ambient lighting) into 3D models, and evaluated the impact of these defects on model performance in object detection and instance segmentation tasks, while providing a cost-benefit analysis of artist repair time and model accuracy.
Exploring Transformers for Open-world Instance Segmentation
Jiannan Wu (University of Hong Kong), Ping Luo (University of Hong Kong)
Object DetectionSegmentationTransformerContrastive LearningImage
🎯 What it does: This paper proposes a Transformer-based open-world instance segmentation framework called SWORD, which utilizes techniques such as stop gradient, IoU head, and contrastive learning to achieve detection and segmentation of unknown categories.
Exploring Video Quality Assessment on User Generated Contents from Aesthetic and Technical Perspectives
Haoning Wu (Nanyang Technological University), Weisi Lin (Nanyang Technological University)
Recommendation SystemTransformerSupervised Fine-TuningVideo
🎯 What it does: This study constructed the DIVIDE-3k user-generated content video quality database and proposed two no-reference video quality assessment methods, DOVER and DOVER++, based on the separation of aesthetic and technical perspectives.
ExposureDiffusion: Learning to Expose for Low-light Image Enhancement
Yufei Wang (Nanyang Technological University), Bihan Wen (Nanyang Technological University)
RestorationDiffusion modelImage
🎯 What it does: This paper proposes a diffusion-based low-light image enhancement method (ExposureDiffusion) based on a physical exposure model, which gradually simulates the exposure process through a shared network and iterates directly from low-light images to normal exposure during inference, achieving highlight fidelity with an adaptive residual layer.
Expressive Text-to-Image Generation with Rich Text
Songwei Ge (University of Maryland), Jia-Bin Huang (Carnegie Mellon University)
GenerationDiffusion modelImageText
🎯 What it does: A text-to-image generation method based on a rich text editor is proposed, utilizing attributes such as font color, style, size, and footnotes to achieve control over local color, style, detail, and importance.
Extensible and Efficient Proxy for Neural Architecture Search
Yuhong Li (University of Illinois), Deming Chen (University of Illinois)
Neural Architecture SearchImage
🎯 What it does: This paper proposes a scalable low-cost proxy (Eproxy) and discrete proxy search (DPS) for rapid evaluation of network architectures across different NAS search spaces and multimodal settings.
F&F Attack: Adversarial Attack against Multiple Object Trackers by Inducing False Negatives and False Positives
Tao Zhou (Zhejiang University), Jiming Chen (Zhejiang University)
Object DetectionObject TrackingAdversarial AttackVideo
🎯 What it does: A novel adversarial attack mechanism for multi-object trackers is proposed—F&F attack, which induces identity switching by erasing original detections and injecting misleading false alarms in a few frames.
Face Clustering via Graph Convolutional Networks with Confidence Edges
Yang Wu (Institute of Automation, Chinese Academy of Sciences), Sulong Xu (JD.COM)
ClassificationRecognitionGraph Neural NetworkImage
🎯 What it does: This paper proposes a confidence edge-based graph convolutional network method to improve face clustering, person re-identification, and fashion image clustering.
FaceCLIPNeRF: Text-driven 3D Face Manipulation using Deformable Neural Radiance Fields
Sungwon Hwang (Korea Advanced Institute of Science and Technology), Jaegul Choo (Korea Advanced Institute of Science and Technology)
GenerationData SynthesisNeural Radiance FieldContrastive LearningVideoText
🎯 What it does: This paper presents FaceCLIPNeRF, a text-driven 3D avatar NeRF deformation and rendering framework that enables automated editing of facial expressions, emotions, and other attributes while maintaining facial identity and high-quality rendering.
FACET: Fairness in Computer Vision Evaluation Benchmark
Laura Gustafson (Meta AI Research), Candace Ross (Meta AI Research)
ClassificationObject DetectionSegmentationImageBenchmark
🎯 What it does: Created the FACET benchmark, which includes 32k images and 50k individuals, with manually annotated fine-grained attributes such as gender presentation, skin color, age, and 52 human categories, used to evaluate the fairness of classification, detection, segmentation, and visual localization models.
Factorized Inverse Path Tracing for Efficient and Accurate Material-Lighting Estimation
Liwen Wu (University of California San Diego), Ravi Ramamoorthi (University of California San Diego)
Computational EfficiencyImage
🎯 What it does: This paper proposes a new inverse path tracing method—Factorized Inverse Path Tracing (FIPT)—for efficiently and accurately estimating material properties and light distribution in indoor scenes.
FACTS: First Amplify Correlations and Then Slice to Discover Bias
Sriram Yenamandra (Georgia Institute of Technology), Judy Hoffman (Georgia Institute of Technology)
ClassificationConvolutional Neural NetworkSupervised Fine-TuningContrastive LearningImage
🎯 What it does: This paper proposes a method for automatically identifying subsets of bias conflicts in visual datasets caused by spurious correlations, called FACTS;
Fan-Beam Binarization Difference Projection (FB-BDP): A Novel Local Object Descriptor for Fine-Grained Leaf Image Retrieval
Xin Chen (Griffith University), Yongsheng Gao (Griffith University)
RetrievalImageAgriculture Related
🎯 What it does: This paper proposes a new local descriptor - Fan Beam Binary Differential Projection (FB-BDP) for fine-grained leaf image retrieval;
Fantasia3D: Disentangling Geometry and Appearance for High-quality Text-to-3D Content Creation
Rui Chen (South China University of Technology), Kui Jia (South China University of Technology)
GenerationDiffusion modelScore-based ModelMesh
🎯 What it does: We propose Fantasia3D, a text-to-3D content generation framework based on DMTET, which can decouple geometry and appearance, learning them separately through rendering normal maps and variable BRDFs, thus generating intricate geometry and high-quality materials for 3D assets.
FashionNTM: Multi-turn Fashion Image Retrieval via Cascaded Memory
Anwesan Pal (UC San Diego), Henrik I. Christensen (UC San Diego)
RetrievalRecurrent Neural NetworkVision Language ModelImageText
🎯 What it does: Designed and implemented the FashionNTM system for multi-turn text feedback fashion image retrieval, capable of retrieving the most matching products based on current and historical information in each round.
Fast Adversarial Training with Smooth Convergence
Mengnan Zhao (Dalian University of Technology), Baocai Yin (Dalian University of Technology)
OptimizationAdversarial AttackImage
🎯 What it does: This paper proposes a fast adversarial training method based on smooth convergence, called ConvergeSmooth, to eliminate catastrophic overfitting under large perturbation budgets.
Fast and Accurate Transferability Measurement by Evaluating Intra-class Feature Variance
Huiwen Xu (Seoul National University), U Kang (Seoul National University)
ClassificationRepresentation LearningContrastive LearningImage
🎯 What it does: This paper proposes TMI (Transferability Measurement with Intra-Class Feature Variance), a framework for quickly and accurately assessing the transferability of pre-trained models to downstream tasks through unidirectional forward propagation and intra-class variance (conditional entropy).
Fast Full-frame Video Stabilization with Iterative Optimization
Weiyue Zhao (Huazhong University of Science and Technology), Zhiguo Cao (Huazhong University of Science and Technology)
RestorationOptimizationOptical FlowVideo
🎯 What it does: A full-frame video stabilization method based on iterative optimization is proposed, which includes two main modules: motion trajectory smoothing and full-frame outpainting.
Fast Globally Optimal Surface Normal Estimation from an Affine Correspondence
Levente Hajder (Eotvos Lorand University), Daniel Barath (ETH Zurich)
OptimizationImage
🎯 What it does: This paper studies a new linear global optimal solver for directly estimating 3D surface normals from a single affine correspondence (with known camera calibration).
Fast Inference and Update of Probabilistic Density Estimation on Trajectory Prediction
Takahiro Maeda (Toyota Technological Institute), Norimichi Ukita (Toyota Technological Institute)
Autonomous DrivingOptimizationComputational EfficiencyFlow-based ModelGenerative Adversarial NetworkTime SeriesSequential
🎯 What it does: A trajectory prediction network called FlowChain based on Continuous Index Flow (CIF) is proposed, which can quickly and accurately provide probability density estimates at each prediction moment and supports millisecond-level updates.
Fast Neural Scene Flow
Xueqian Li (University of Adelaide), Simon Lucey (University of Adelaide)
Autonomous DrivingOptimizationComputational EfficiencyNeural Radiance FieldOptical FlowPoint Cloud
🎯 What it does: A runtime optimization method for neural scene flow estimation based on distance transformation is implemented, which achieves real-time performance on dense point clouds without the need for training.
FastRecon: Few-shot Industrial Anomaly Detection via Fast Feature Reconstruction
Zheng Fang (XJTLU), Jimin Xiao (XJTLU)
Anomaly DetectionConvolutional Neural NetworkImageBenchmark
🎯 What it does: A training-free, few-shot industrial defect detection method called FastRecon is proposed, which achieves anomaly localization by quickly reconstructing the features of query samples and aligning them with the original features.
FastViT: A Fast Hybrid Vision Transformer Using Structural Reparameterization
Pavan Kumar Anasosalu Vasu (Apple), Anurag Ranjan (Apple)
ClassificationObject DetectionSegmentationTransformerSupervised Fine-TuningImage
🎯 What it does: Proposes FastViT, a fast visual network that integrates convolution and Transformer;
FateZero: Fusing Attentions for Zero-shot Text-based Video Editing
Chenyang QI, Qifeng Chen
GenerationData SynthesisDiffusion modelVideoText
🎯 What it does: Using a pre-trained diffusion model to achieve zero-shot text-driven video editing, supporting shape-aware local object editing and video style editing;
FB-BEV: BEV Representation from Forward-Backward View Transformations
Zhiqi Li (Nanjing University), Jose M. Alvarez (NVIDIA)
Object DetectionAutonomous DrivingConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A front-and-back projection joint BEV transformation module FB-BEV is proposed to generate denser and higher-quality bird's-eye view features for multi-camera 3D detection.
FBLNet: FeedBack Loop Network for Driver Attention Prediction
Yilong Chen (Chongqing University), Tao Xiang (Chongqing University)
Autonomous DrivingConvolutional Neural NetworkTransformerImage
🎯 What it does: Designed and implemented FBLNet, which uses a feedback loop mechanism to simulate the accumulation of driving experience to predict the driver's attention heatmap.
Fcaformer: Forward Cross Attention in Hybrid Vision Transformer
Haokui Zhang (Intellifusion), Xiaoyu Wang (The Hong Kong University of Science and Technology)
ClassificationObject DetectionSegmentationTransformerImage
🎯 What it does: This paper proposes a Forward Cross Attention (FCA) module, which uses tokens from the previous layer or several preceding layers at the same stage as additional inputs to more densely capture cross-layer semantic interactions, thereby constructing a new FcaFormer visual Transformer.
FCCNs: Fully Complex-valued Convolutional Networks using Complex-valued Color Model and Loss Function
Saurabh Yadav (Indraprastha Institute of Information Technology Delhi), Koteswar Rao Jerripothula (Indraprastha Institute of Information Technology Delhi)
ClassificationConvolutional Neural NetworkImage
🎯 What it does: A Full Complex-valued Convolutional Network (FCCN) is proposed, achieving a complete flow of complex numerical information from input to output.
FDViT: Improve the Hierarchical Architecture of Vision Transformer
Yixing Xu (Advanced Micro Devices), Ashish Sirasao (Advanced Micro Devices)
ClassificationObject DetectionSegmentationTransformerAuto EncoderImage
🎯 What it does: This paper proposes a new hierarchical visual Transformer structure called FDViT, which introduces flexible downsampling layers (FD layers) and masked autoencoders to reduce redundant computations in self-attention while retaining more information.
FeatEnHancer: Enhancing Hierarchical Features for Object Detection and Beyond Under Low-Light Vision
Khurram Azeem Hashmi (German Research Center for Artificial Intelligence), Muhammad Zeshan Afzal (German Research Center for Artificial Intelligence)
Object DetectionSegmentationConvolutional Neural NetworkImageVideo
🎯 What it does: Proposed and implemented the FeatEnHancer module, which enhances multi-scale features and scale-aware attention aggregation to improve the performance of downstream tasks (detection, segmentation, video detection) in low-light images.
Feature Modulation Transformer: Cross-Refinement of Global Representation via High-Frequency Prior for Image Super-Resolution
Ao Li (University of Electronic Science and Technology of China), Ce Zhu (University of Electronic Science and Technology of China)
RestorationSuper ResolutionConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper studies and proposes a high-frequency feature modulation Transformer (CRAFT) that combines convolution and Transformer for single image super-resolution tasks.
Feature Prediction Diffusion Model for Video Anomaly Detection
Cheng Yan (Tianjin University), Wenjun Wang (Singapore Management University)
Anomaly DetectionDiffusion modelVideo
🎯 What it does: This paper proposes a feature prediction method based on diffusion models (FPDM) for unsupervised video anomaly detection; it learns the distribution of normal frame features during the training phase and predicts and compares frame features during the inference phase to determine anomalies.