CVPR 2024 Papers — Page 24
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2716 papers
StyleCineGAN: Landscape Cinemagraph Generation using a Pre-trained StyleGAN
Jongwoo Choi (KAIST), Junyong Noh (KAIST)
GenerationData SynthesisGenerative Adversarial NetworkImageVideo
🎯 What it does: Using the pre-trained StyleGAN deep feature space, high-resolution (1024×1024) landscape cinemagraphs are automatically generated through multi-scale deep feature warping.
StyLitGAN: Image-Based Relighting via Latent Control
Anand Bhattad (University of Illinois Urbana-Champaign), D.A. Forsyth (University of Illinois Urbana-Champaign)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes StyLitGAN, which utilizes the latent space vector control method of StyleGAN to achieve diverse lighting reshaping of indoor scenes without altering geometry and surface reflectance (albedo).
SubT-MRS Dataset: Pushing SLAM Towards All-weather Environments
Shibo Zhao (Carnegie Mellon University), Sebastian Scherer (Carnegie Mellon University)
Autonomous DrivingOptimizationRobotic IntelligenceSimultaneous Localization and MappingMultimodalityPoint CloudBenchmark
🎯 What it does: Developed the SubT-MRS dataset, which covers over 5 years of multi-sensor, multi-platform, and multi-degradation (visual/geometric) panoramic environment SLAM data, and proposed a robustness metric based on speed error along with a challenge.
SUGAR: Pre-training 3D Visual Representations for Robotics
Shizhe Chen (Inria), Cordelia Schmid (Inria)
Knowledge DistillationRepresentation LearningRobotic IntelligenceTransformerPoint Cloud
🎯 What it does: This paper proposes the SUGAR framework, which utilizes Transformers for pre-training on 3D point clouds, focusing on semantic, geometric, and grasping attributes.
SuGaR: Surface-Aligned Gaussian Splatting for Efficient 3D Mesh Reconstruction and High-Quality Mesh Rendering
Antoine Guédon (Ecole des Ponts), Vincent Lepetit (Ecole des Ponts)
OptimizationComputational EfficiencyGaussian SplattingMesh
🎯 What it does: The SuGaR method is proposed, which can quickly extract high-quality editable triangular meshes from 3D Gaussian Splatting models in a matter of minutes, and achieve detail reconstruction and editing of the mesh through Gaussian scattering rendering.
Summarize the Past to Predict the Future: Natural Language Descriptions of Context Boost Multimodal Object Interaction Anticipation
Razvan-George Pasca (ETH Zurich), Xi Wang (ETH Zurich)
RecognitionObject DetectionTransformerVideoTextMultimodality
🎯 What it does: A multi-modal fusion Transformer (TransFusion) is proposed, utilizing language to summarize past actions for short-term object interaction prediction.
Super-Resolution Reconstruction from Bayer-Pattern Spike Streams
Yanchen Dong (Peking University), Tiejun Huang (Peking University)
RestorationSuper ResolutionConvolutional Neural NetworkOptical FlowImageVideo
🎯 What it does: A super-resolution network CSCSR is proposed to recover high-resolution color images from low-resolution Bayer pattern pulse streams.
SuperNormal: Neural Surface Reconstruction via Multi-View Normal Integration
Xu Cao (CyberAgent), Takafumi Taketomi (CyberAgent)
GenerationComputational EfficiencyImageBenchmark
🎯 What it does: Through multi-view normal map fusion, combined with multi-resolution hash encoding, a neural SDF achieves fast and detail-rich 3D reconstruction.
SuperPrimitive: Scene Reconstruction at a Primitive Level
Kirill Mazur (Imperial College London), Andrew J. Davison (Imperial College London)
SegmentationDepth EstimationOptimizationSimultaneous Localization and MappingImageBenchmark
🎯 What it does: A 'SuperPrimitive' representation based on image segmentation and single-view normal prediction is proposed, utilizing local geometric priors for multi-view optimization to achieve depth completion, few-view SFM, and monocular visual odometry.
SuperSVG: Superpixel-based Scalable Vector Graphics Synthesis
Teng Hu (Shanghai Jiao Tong University), Yu-Kun Lai (Cardiff University)
GenerationData SynthesisTransformerImage
🎯 What it does: SuperSVG is proposed, a two-stage self-supervised framework based on superpixels, which first captures the main structure using a coarse model and then refines the details with a fine model, ultimately generating high-quality SVG vector graphics.
Supervised Anomaly Detection for Complex Industrial Images
Aimira Baitieva (Valeo), Olivier Bernard (Valeo)
Anomaly DetectionImage
🎯 What it does: This paper presents a new industrial defect detection dataset VAD and designs a segmentation-based supervised anomaly detection method SegAD for efficient anomaly detection in complex industrial images.
Suppress and Rebalance: Towards Generalized Multi-Modal Face Anti-Spoofing
Xun Lin (Beihang University), Alex Kot (Nanyang Technological University)
Domain AdaptationAnomaly DetectionTransformerImageMultimodalityBenchmark
🎯 What it does: A multi-modal domain generalization framework MMDG is proposed, which utilizes the U-Adapter to suppress unreliable information across modalities and adaptively adjusts gradients through ReGrad to address modality imbalance, significantly improving cross-domain facial deception detection performance.
SURE: SUrvey REcipes for building reliable and robust deep networks
Yuting Li (Intellindust), Xi Shen (Intellindust)
ClassificationOptimizationConvolutional Neural NetworkSupervised Fine-TuningImageReview/Survey Paper
🎯 What it does: This paper proposes a unified training strategy (SURE) that enhances the uncertainty estimation and robustness of deep networks by combining regularization, classifier improvements, and optimization techniques.
SurMo: Surface-based 4D Motion Modeling for Dynamic Human Rendering
Tao Hu (Nanyang Technological University), Ziwei Liu (Nanyang Technological University)
GenerationPose EstimationGenerative Adversarial NetworkVideoMesh
🎯 What it does: We propose SurMo, a four-dimensional motion encoding framework based on surface tri-planes, which jointly models the motion and appearance of dynamic human bodies to achieve high-quality time-varying rendering of multi-view videos.
SurroundSDF: Implicit 3D Scene Understanding Based on Signed Distance Field
Lizhe Liu (Xiaomi), Bing Wang (Xiaomi)
SegmentationDepth EstimationAutonomous DrivingConvolutional Neural NetworkPoint Cloud
🎯 What it does: A 3D scene understanding framework based on an omnidirectional camera has been developed, capable of continuously and accurately predicting obstacle surfaces and semantics.
SVDinsTN: A Tensor Network Paradigm for Efficient Structure Search from Regularized Modeling Perspective
Yu-Bang Zheng (Southwest Jiaotong University), Ting-Zhu Huang (University of Electronic Science and Technology of China)
OptimizationComputational EfficiencyVideo
🎯 What it does: This paper proposes a new tensor network decomposition paradigm called SVDinsTN, and based on it, presents an efficient tensor network structure search method from the perspective of regularization modeling.
SVDTree: Semantic Voxel Diffusion for Single Image Tree Reconstruction
Yuan Li (Institute of Automation, Chinese Academy of Sciences), Jianwei Guo (Institute of Automation, Chinese Academy of Sciences)
SegmentationGenerationDiffusion modelImage
🎯 What it does: This paper proposes a single-image tree model reconstruction framework based on semantic voxel diffusion, called SVDTree, which can generate high-fidelity three-dimensional tree geometry from a single tree photograph.
SVGDreamer: Text Guided SVG Generation with Diffusion Model
Ximing Xing (Beihang University), Qian Yu (Beihang University)
GenerationDiffusion modelImageText
🎯 What it does: A text-guided SVG generation framework called SVGDreamer has been developed, capable of generating editable, high-visual-quality, and diverse vector images.
SwiftBrush: One-Step Text-to-Image Diffusion Model with Variational Score Distillation
Thuan Hoang Nguyen (VinAI Research), Anh Tran (VinAI Research)
GenerationKnowledge DistillationDiffusion modelScore-based ModelImageText
🎯 What it does: A single-step text-to-image diffusion model distillation method named SwiftBrush has been developed, capable of distilling multi-step Stable Diffusion into a first-order generator without any image training data.
SwitchLight: Co-design of Physics-driven Architecture and Pre-training Framework for Human Portrait Relighting
Hoon Kim (Beeble AI), Sanghyun Woo (New York University)
Image TranslationRestorationGenerationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: A framework named SwitchLight for portrait relighting has been developed, which can decompose input portraits into surface normals, diffuse reflection, roughness, Fresnel reflection, and lighting information, and re-render to generate realistic images under target lighting.
Symphonize 3D Semantic Scene Completion with Contextual Instance Queries
Haoyi Jiang (Huazhong University of Science and Technology), Xinggang Wang (Huazhong University of Science and Technology)
SegmentationAutonomous DrivingTransformerPoint CloudBenchmark
🎯 What it does: Proposes the Symphonies method, which drives 2D-to-3D semantic scene completion through instance queries;
SyncMask: Synchronized Attentional Masking for Fashion-centric Vision-Language Pretraining
Chull Hwan Song (Dealicious Inc), Yeong Hyeon Gu (Sejong University)
RetrievalTransformerVision Language ModelContrastive LearningImageText
🎯 What it does: This paper proposes a synchronous attention masking (SyncMask) mechanism, which utilizes the cross-modal attention maps generated by a momentum teacher model to mask segments that co-occur in images and text. This improves the masked language model (MLM) and masked image model (MIM) tasks in visual-text pre-training in the fashion domain, and introduces a semi-hard negative sample strategy during batch sampling to reduce the interference of false negative samples on contrastive learning and matching tasks.
SyncTalk: The Devil is in the Synchronization for Talking Head Synthesis
Ziqiao Peng (Renmin University of China), Zhaoxin Fan (Renmin University of China)
GenerationData SynthesisNeural Radiance FieldSimultaneous Localization and MappingOptical FlowVideoAudio
🎯 What it does: A real-time talking head synthesis framework called SyncTalk based on NeRF is proposed, which achieves high-quality synchronization of lip movements, expressions, and head poses while maintaining identity consistency.
Synergistic Global-space Camera and Human Reconstruction from Videos
Yizhou Zhao (Carnegie Mellon University), Chun-Hao Paul Huang (Adobe Research)
Object TrackingPose EstimationDepth EstimationTransformerSimultaneous Localization and MappingVideoPoint CloudMesh
🎯 What it does: An end-to-end system named SynCHMR is proposed, capable of simultaneously recovering camera trajectories, global scale human SMPL meshes, and dense scene point clouds from monocular videos, unifying all results into the same global coordinate system.
SynFog: A Photo-realistic Synthetic Fog Dataset based on End-to-end Imaging Simulation for Advancing Real-World Defogging in Autonomous Driving
Yiming Xie (Tsinghua University), Xiangyang Ji (Tsinghua University)
RestorationData SynthesisAutonomous DrivingImage
🎯 What it does: An end-to-end fog image simulation pipeline is proposed, and based on this pipeline, the SynFog synthetic fog dataset is constructed for training and evaluating dehazing algorithms.
SynSP: Synergy of Smoothness and Precision in Pose Sequences Refinement
Tao Wang (Beijing University of Posts and Telecommunications), Jian Zhao (Northwestern Polytechnical University)
Pose EstimationOptimizationTransformerVideo
🎯 What it does: This paper proposes a posture sequence optimization network named SynSP, which aims to improve the accuracy and smoothness of posture estimation while maintaining low latency.
Synthesize Diagnose and Optimize: Towards Fine-Grained Vision-Language Understanding
Wujian Peng (Fudan University), Zuxuan Wu (Fudan University)
ClassificationSegmentationGenerationRetrievalOptimizationTransformerVision Language ModelDiffusion modelContrastive LearningImageTextMultimodalityBenchmark
🎯 What it does: This paper proposes a step-by-step generation of a candidate set of images that differ only in a specific attribute, and based on this, constructs the SPEC fine-grained visual-language understanding benchmark. Subsequently, this benchmark is used to diagnose the performance of mainstream VLMs, and CLIP is fine-tuned by adding hard negative samples to enhance fine-grained understanding capabilities.
Synthesize Step-by-Step: Tools Templates and LLMs as Data Generators for Reasoning-Based Chart VQA
Zhuowan Li (Johns Hopkins University), Shabnam Ghadar (AWS AI Labs)
GenerationData SynthesisTransformerLarge Language ModelImageTextChain-of-Thought
🎯 What it does: A step-by-step synthesis data generation method based on large language models is proposed to automatically generate chart questions and answers that include reasoning steps, thereby enhancing the reasoning ability of chart visual question answering (Chart VQA) models.
Systematic Comparison of Semi-supervised and Self-supervised Learning for Medical Image Classification
Zhe Huang (Tufts University), Michael C. Hughes (Tufts University)
ClassificationConvolutional Neural NetworkContrastive LearningBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper systematically evaluates the performance of semi-supervised learning and self-supervised learning on four medical image classification tasks with limited labeled data, proposing a unified experimental and hyperparameter tuning process.
T-VSL: Text-Guided Visual Sound Source Localization in Mixtures
Tanvir Mahmud (University of Texas at Austin), Diana Marculescu (University of Texas at Austin)
RecognitionObject DetectionSegmentationTransformerVision Language ModelContrastive LearningVideoTextMultimodalityAudio
🎯 What it does: This paper proposes a text-guided multi-source visual sound source localization framework T-VSL, which utilizes the tri-modal joint embedding space of AudioCLIP to use text features as an intermediary for semantic decoupling of mixed audio, achieving accurate localization of multi-source visual sound sources.
T4P: Test-Time Training of Trajectory Prediction via Masked Autoencoder and Actor-specific Token Memory
Daehee Park (KAIST), Kuk-Jin Yoon (KAIST)
Domain AdaptationAutonomous DrivingTransformerAuto EncoderTime Series
🎯 What it does: Dynamically train the trajectory prediction model during the testing phase to adapt to changes in different data distributions;
Tackling the Singularities at the Endpoints of Time Intervals in Diffusion Models
Pengze Zhang (Sun Yat-sen University), Xiaohua Xie (Sun Yat-sen University)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: The paper conducts a theoretical analysis of the singularities at the time endpoints of diffusion models and proposes the SingDiffusion plugin for sampling at the initial singularity moment, thereby eliminating average brightness imbalance and enhancing image quality.
TACO: Benchmarking Generalizable Bimanual Tool-ACtion-Object Understanding
Yun Liu (Tsinghua University), Li Yi (Tsinghua University)
Robotic IntelligenceVideoMeshBenchmark
🎯 What it does: This paper presents a large-scale dual-hand-dual-object interaction dataset TACO, covering various tool-action-object combinations, and establishes three evaluation benchmarks for generalizable action recognition, motion prediction, and grasp synthesis based on this dataset.
Tactile-Augmented Radiance Fields
Yiming Dou (University of Michigan), Andrew Owens (University of Michigan)
RecognitionGenerationData SynthesisDiffusion modelNeural Radiance FieldContrastive LearningImageMultimodality
🎯 What it does: This paper proposes the Tactile-Augmented Radiance Field (TaRF) scene representation, which integrates visual and tactile information into a unified 3D space and infers tactile signals at un-sampled locations through a generative model.
Tailored Visions: Enhancing Text-to-Image Generation with Personalized Prompt Rewriting
Zijie Chen (Zhejiang University), Zhenzhong Lan (Westlake University)
GenerationRetrievalTransformerLarge Language ModelPrompt EngineeringDiffusion modelImageTextRetrieval-Augmented Generation
🎯 What it does: This paper proposes a personalized prompt rewriting method based on user historical interactions and constructs a PIP dataset containing 300,237 prompts to enhance the personalization effect of text-to-image generation.
Taming Mode Collapse in Score Distillation for Text-to-3D Generation
Peihao Wang (University of Texas at Austin), Vikas Chandra (Meta Reality Labs)
GenerationData SynthesisDiffusion modelScore-based ModelNeural Radiance FieldImageText
🎯 What it does: Proposes the Entropic Score Distillation (ESD) method, which utilizes entropy regularization to improve score distillation in text-to-3D generation, thereby alleviating the Janus multi-view forgery problem.
Taming Self-Training for Open-Vocabulary Object Detection
Shiyu Zhao (Rutgers University), Dimitris N. Metaxas (Rutgers University)
Object DetectionVision Language ModelImage
🎯 What it does: A self-supervised training-based open vocabulary object detection framework SAS-Det is designed, utilizing CLIP to generate pseudo-labels and enhancing detection performance through a branch detection head and periodic teacher updates.
Taming Stable Diffusion for Text to 360 Panorama Image Generation
Cheng Zhang (Monash University), Jianfei Cai (Monash University)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: A dual-branch diffusion model called PanFusion is proposed, capable of generating 360° panoramic images from a single text prompt, balancing global perspectives with local detail consistency.
Taming the Tail in Class-Conditional GANs: Knowledge Sharing via Unconditional Training at Lower Resolutions
Saeed Khorram (Oregon State University), Li Fuxin (McGill University)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: Introduce a 'low-resolution unconditional training' mechanism in the training of conditional generative adversarial networks (cGAN) under long-tail distribution to share head class knowledge with tail classes;
TAMM: TriAdapter Multi-Modal Learning for 3D Shape Understanding
Zhihao Zhang (Xi'an Jiaotong University), Yu-Xiong Wang (University of Illinois Urbana-Champaign)
RecognitionDomain AdaptationRepresentation LearningContrastive LearningMultimodalityPoint Cloud
🎯 What it does: A Three-Adapter Multimodal Learning (TAMM) framework is proposed, which first fine-tunes the CLIP image adapter and then uses visual and semantic adapters to decompose 3D representations, achieving pre-training of 3D shapes.
Targeted Representation Alignment for Open-World Semi-Supervised Learning
Ruixuan Xiao (Zhejiang University), Haobo Wang (Zhejiang University)
ClassificationRepresentation LearningImage
🎯 What it does: This paper proposes the TRAILER framework, which achieves targeted alignment of features in open-world semi-supervised learning, thereby simultaneously recognizing known categories and discovering unknown categories.
TASeg: Temporal Aggregation Network for LiDAR Semantic Segmentation
Xiaopei Wu (Zhejiang University), Wanli Ouyang (Zhejiang University)
SegmentationAutonomous DrivingKnowledge DistillationConvolutional Neural NetworkMultimodalityPoint Cloud
🎯 What it does: The TASeg framework is proposed, combining Temporal LiDAR Aggregation and Distillation (TLAD), Temporal Image Aggregation and Fusion (TIAF), and Static-Moving Switch Augmentation (SMSA) to achieve semantic segmentation of multi-frame LiDAR and multi-temporal images.
Task-Adaptive Saliency Guidance for Exemplar-free Class Incremental Learning
Xialei Liu (Nankai University), Ming-Ming Cheng (Nankai University)
ClassificationObject DetectionConvolutional Neural NetworkImage
🎯 What it does: For sample-free incremental learning, the Task-Adaptive Saliency Supervision (TASS) method combines boundary guidance, low-level task assistance, and saliency noise injection to suppress saliency drift, enhancing the model's adaptability to new tasks and memory retention.
Task-aligned Part-aware Panoptic Segmentation through Joint Object-Part Representations
Daan de Geus (Eindhoven University of Technology), Gijs Dubbelman (Eindhoven University of Technology)
Object DetectionSegmentationTransformerImage
🎯 What it does: A task-aligned object and part joint segmentation network TAPPS is proposed, which can simultaneously predict object-level and part-level segmentation on the same query;
Task-Aware Encoder Control for Deep Video Compression
Xingtong Ge (Beijing Institute of Technology), Hongwei Qin (SenseTime Research)
RecognitionObject DetectionObject TrackingCompressionOptical FlowVideo
🎯 What it does: A controllable deep video compression coding framework is proposed, utilizing existing pre-trained decoders to adapt to different machine vision tasks (detection, tracking, action recognition) during the encoding phase by controlling mode prediction and GOP structure.
Task-Conditioned Adaptation of Visual Features in Multi-Task Policy Learning
Pierre Marza (INSA Lyon), Christian Wolf (Naver Labs Europe)
Robotic IntelligenceTransformerReinforcement LearningImageBenchmark
🎯 What it does: This paper proposes a task-conditioned visual adapter in multi-task policy learning, which modulates visual features using a pre-trained large visual Transformer (MAE) while keeping the weights unchanged, enabling a single policy to accomplish multiple robotic tasks simultaneously, and achieving rapid adaptation to new tasks with few samples through optimizing task embeddings.
Task-Customized Mixture of Adapters for General Image Fusion
Pengfei Zhu (Tianjin University), Qinghua Hu (Tianjin University)
Image TranslationRestorationPrompt EngineeringMixture of ExpertsImageMultimodalityMagnetic Resonance Imaging
🎯 What it does: A task-customized mixed adapter (TC-MoA) is designed to adaptively handle multi-modal, multi-exposure, and multi-focal image fusion tasks within the same base model.
Task-Driven Exploration: Decoupling and Inter-Task Feedback for Joint Moment Retrieval and Highlight Detection
Jin Yang (Xi'an Jiaotong University), Ziyang Ren (Xi'an Jiaotong University)
RecognitionRetrievalConvolutional Neural NetworkTransformerVideoTextMultimodality
🎯 What it does: This paper proposes TaskWeave, a task-driven upstream framework for joint video moment retrieval and highlight detection.
Task-Driven Wavelets using Constrained Empirical Risk Minimization
Eric Marcus (Netherlands Cancer Institute), Jonas Teuwen (Netherlands Cancer Institute)
SegmentationOptimizationConvolutional Neural NetworkImageBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: A framework called CERM is proposed for training deep networks with strict constraints (such as wavelet filters) by performing gradient descent on constrained subspaces;
Task2Box: Box Embeddings for Modeling Asymmetric Task Relationships
Rangel Daroya (University of Massachusetts Amherst), Subhransu Maji (University of Massachusetts Amherst)
Explainability and InterpretabilityRepresentation LearningContrastive LearningImage
🎯 What it does: This paper proposes the TASK2BOX framework, which uses axis-aligned box embeddings to represent asymmetric relationships (such as hierarchy and transferability) between tasks/datasets, and provides interpretable visualizations.
TCP:Textual-based Class-aware Prompt tuning for Visual-Language Model
Hantao Yao (Chinese Academy of Sciences), Changsheng Xu (Chinese Academy of Sciences)
ClassificationRecognitionTransformerPrompt EngineeringVision Language ModelContrastive LearningImageText
🎯 What it does: This paper proposes a text-level category-aware prompt tuning (TCP) method, which maps the class-level textual knowledge of CLIP into category-aware prompts and inserts them into the text encoder to generate a more discriminative text classifier for downstream tasks of visual-language models.
TE-TAD: Towards Full End-to-End Temporal Action Detection via Time-Aligned Coordinate Expression
Ho-Joong Kim (Korea University), Seong-Whan Lee (Korea University)
RecognitionObject DetectionTransformerVideo
🎯 What it does: This paper proposes an end-to-end temporal action detection Transformer (TE-TAD) that achieves detection without relying on sliding windows and NMS through time-aligned coordinate representation.
TEA: Test-time Energy Adaptation
Yige Yuan (Chinese Academy of Sciences), Xueqi Cheng (Chinese Academy of Sciences)
Domain AdaptationContrastive LearningImageStochastic Differential Equation
🎯 What it does: Enhancing the model's generalization ability under distribution shift during the testing phase through the Energy Adaptation (TEA) method.
Teeth-SEG: An Efficient Instance Segmentation Framework for Orthodontic Treatment based on Multi-Scale Aggregation and Anthropic Prior Knowledge
Bo Zou (Tsinghua University), Youjian Zhao (Tsinghua University)
SegmentationTransformerImage
🎯 What it does: This paper proposes a Vision Transformer-based dental instance segmentation framework called TeethSEG and releases the first large-scale 2D dental image dataset IO150K for training and evaluation.
Telling Left from Right: Identifying Geometry-Aware Semantic Correspondence
Junyi Zhang (Shanghai Jiao Tong University), Ming-Hsuan Yang (UC Merced)
RecognitionPose EstimationDiffusion modelImage
🎯 What it does: This study investigates the semantic correspondence problem in geometric perception and proposes a series of improvement methods: pose alignment during testing, dense training objectives, pose transformation augmentation, and window soft-argmax, to enhance the recognition ability for geometric ambiguities such as left/right and front/back.
TeMO: Towards Text-Driven 3D Stylization for Multi-Object Meshes
Xuying Zhang (Nankai University), Ming-Ming Cheng (Shenzhen Futian)
GenerationData SynthesisTransformerContrastive LearningMesh
🎯 What it does: A text-driven stylization method for multi-object 3D meshes, TeMO, is proposed, achieving precise parsing and style transfer for multi-object scenes.
Template Free Reconstruction of Human-object Interaction with Procedural Interaction Generation
Xianghui Xie (University of Tübingen), Gerard Pons-Moll (University of Tübingen)
GenerationData SynthesisPose EstimationDiffusion modelAuto EncoderImagePoint Cloud
🎯 What it does: A large-scale human-machine interaction data synthesis method based on ProciGen is proposed, utilizing a template-free hierarchical diffusion model (HDM) to reconstruct the 3D shapes of humans and objects from a single RGB image.
Temporally Consistent Unbalanced Optimal Transport for Unsupervised Action Segmentation
Ming Xu (Australian National University), Stephen Gould (Australian National University)
SegmentationOptimizationVideo
🎯 What it does: Proposes an unsupervised action segmentation method based on unbalanced Gromov-Wasserstein optimal transport.
Test-Time Adaptation for Depth Completion
Hyoungseob Park (Yale University), Alex Wong (Yale University)
Depth EstimationDomain AdaptationConvolutional Neural NetworkContrastive LearningPoint Cloud
🎯 What it does: This paper proposes a testing time adaptation method called ProxyTTA for the depth completion task, which completes the adaptation in a single transmission by aligning sparse depth feature maps with RGB features, guiding the model to quickly adapt to the target domain.
Test-Time Domain Generalization for Face Anti-Spoofing
Qianyu Zhou (Shanghai Jiao Tong University), Lizhuang Ma (Shanghai Jiao Tong University)
RecognitionDomain AdaptationConvolutional Neural NetworkTransformerContrastive LearningImage
🎯 What it does: A testing domain generalization framework is proposed, which projects test data into the training domain without model updates to enhance facial anti-spoofing performance.
Test-Time Linear Out-of-Distribution Detection
Ke Fan (Fudan University), Xingqun Jiang (BOE Technology Group)
Anomaly DetectionConvolutional Neural NetworkSupervised Fine-TuningImageBenchmark
🎯 What it does: This study investigates the linear relationship between the OOD scores generated by current OOD detection methods and network features, and based on this, proposes a robust testing-time linear correction method (RTL, RTL++) and its online version.
Test-Time Zero-Shot Temporal Action Localization
Benedetta Liberatori (University of Trento), Elisa Ricci (University of Trento)
RecognitionObject DetectionTransformerVision Language ModelContrastive LearningVideoText
🎯 What it does: In the absence of labeled training data, a test-time adaptive zero-shot temporal action localization method called T3AL is proposed, which utilizes a pre-trained vision-language model to achieve video-level pseudo-label generation, prediction refinement based on self-supervised learning, and subtitle-guided region suppression.
TetraSphere: A Neural Descriptor for O(3)-Invariant Point Cloud Analysis
Pavlo Melnyk (Linköping University), Mårten Wadenbäck (Linköping University)
ClassificationRecognitionPoint Cloud
🎯 What it does: A point cloud descriptor TetraSphere based on a tunable TetraTransform layer and vector neural networks is proposed, which remains invariant under arbitrary rotations and reflections.
TeTriRF: Temporal Tri-Plane Radiance Fields for Efficient Free-Viewpoint Video
Minye Wu, Tinne Tuytelaars
Data SynthesisCompressionComputational EfficiencyNeural Radiance FieldVideo
🎯 What it does: This paper proposes Temporal Tri-Plane Radiance Fields (TeTriRF), which utilizes a hybrid representation of three planes and voxel grids to achieve low storage, low training costs, and real-time rendering for free-viewpoint video synthesis.
TexOct: Generating Textures of 3D Models with Octree-based Diffusion
Jialun Liu (Baidu Inc.), Errui Ding (Baidu Inc.)
GenerationData SynthesisDiffusion modelPoint Cloud
🎯 What it does: This paper presents TexOct, a diffusion model based on octrees that can directly generate high-quality, complete surface textures in 3D space.
Text Grouping Adapter: Adapting Pre-trained Text Detector for Layout Analysis
Tianci Bi (Xi'an Jiaotong University), Nanning Zheng (Xi'an Jiaotong University)
Object DetectionSegmentationTransformerText
🎯 What it does: This paper proposes a general module called Text Grouping Adapter (TGA) that enables trained text detectors to directly perform text layout analysis (clustering detected text instances into paragraphs or lines).
Text Is MASS: Modeling as Stochastic Embedding for Text-Video Retrieval
Jiamian Wang (Rochester Institute of Technology), Zhiqiang Tao (Rochester Institute of Technology)
RetrievalTransformerContrastive LearningVideoText
🎯 What it does: This paper proposes a method for modeling text as a random distribution (text quality) - T-MASS, to enhance the semantic alignment effect of text-video retrieval.
Text Prompt with Normality Guidance for Weakly Supervised Video Anomaly Detection
Zhiwei Yang (Xidian University), Peng Wu (Northwestern Polytechnical University)
Anomaly DetectionTransformerPrompt EngineeringContrastive LearningVideoText
🎯 What it does: Proposes the TPWNG framework, which combines learnable text prompts and normality visual prompts with CLIP for text-image alignment, generating high-quality pseudo-labels and self-training the classifier to achieve weakly supervised video anomaly detection.
Text-conditional Attribute Alignment across Latent Spaces for 3D Controllable Face Image Synthesis
Feifan Xu (South China University of Technology), Hau San Wong (City University of Hong Kong)
GenerationData SynthesisGenerative Adversarial NetworkImageText
🎯 What it does: A 3D controllable facial image synthesis method based on text conditional attribute alignment, TcALign, is proposed, which enables fine-grained control of multiple attributes such as facial expressions, poses, and lighting through text.
Text-Conditioned Generative Model of 3D Strand-based Human Hairstyles
Vanessa Sklyarova (Max Planck Institute for Intelligent Systems), Justus Thies (Technical University of Darmstadt)
GenerationDiffusion modelAuto EncoderMesh
🎯 What it does: This paper presents HAAR, a text-conditioned 3D hair bundle generation model that can be directly used for rendering and physical animation.
Text-Driven Image Editing via Learnable Regions
Yuanze Lin (University of Oxford), Ming-Hsuan Yang (University of California Merced)
Image TranslationGenerationTransformerDiffusion modelContrastive LearningImageText
🎯 What it does: A mask-free local image editing method based on text descriptions is proposed, which can automatically identify the editing area and complete the edits while maintaining the context of the original image.
Text-Enhanced Data-free Approach for Federated Class-Incremental Learning
Minh-Tuan Tran (Monash University), Dinh Phung (Monash University)
Federated LearningKnowledge DistillationTransformerLarge Language ModelImage
🎯 What it does: In Federated Class Incremental Learning (FCIL), the LANDER method is proposed, utilizing label text embeddings (LTE) generated by pre-trained language models as feature anchors, and achieving data-independent knowledge transfer through Bounding Loss and Learnable Data Stats.
Text-Guided 3D Face Synthesis - From Generation to Editing
Yunjie Wu (Netease Fuxi AI Lab), Xin Yu (Netease Fuxi AI Lab)
GenerationData SynthesisDiffusion modelScore-based ModelImage
🎯 What it does: A comprehensive generation and editing framework from text to 3D faces (FaceG2E) is proposed, achieving high-fidelity geometric and texture separation generation as well as recursive fine-grained editing.
Text-guided Explorable Image Super-resolution
Kanchana Vaishnavi Gandikota (Institute for Vision and Graphics), Paramanand Chandramouli (Institute for Vision and Graphics)
RestorationSuper ResolutionDiffusion modelImage
🎯 What it does: This paper proposes a zero-shot text-guided extreme super-resolution method that can explore multiple high-resolution reconstruction results consistent with low-resolution input and semantically coherent through natural language prompts.
Text-Guided Variational Image Generation for Industrial Anomaly Detection and Segmentation
Mingyu Lee (Chung-Ang University), Jongwon Choi (Chung-Ang University)
SegmentationGenerationAnomaly DetectionGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: This paper proposes a text-guided variational image generation framework that uses generated defect-free images to augment training data for industrial defect detection, thereby improving defect detection and segmentation performance.
Text-IF: Leveraging Semantic Text Guidance for Degradation-Aware and Interactive Image Fusion
Xunpeng Yi (Wuhan University), Jiayi Ma (Wuhan University)
RestorationObject DetectionTransformerPrompt EngineeringImageMultimodality
🎯 What it does: A semantic text-guided image fusion framework (Text-IF) is proposed, achieving adaptive processing of degraded images and supporting user interactive generation of fusion results.
Text-Image Alignment for Diffusion-Based Perception
Neehar Kondapaneni (California Institute of Technology), Pietro Perona (California Institute of Technology)
Object DetectionSegmentationDepth EstimationDomain AdaptationVision Language ModelDiffusion modelImage
🎯 What it does: This paper proposes a framework that aligns automatic image captioning with diffusion models (TADP), enhancing the performance of visual perception tasks by ensuring the text input of the diffusion model is consistent with the image content.
Text-to-3D Generation with Bidirectional Diffusion using both 2D and 3D priors
Lihe Ding (Chinese University of Hong Kong), Tianfan Xue (Chinese University of Hong Kong)
GenerationData SynthesisDiffusion modelNeural Radiance FieldPoint CloudMesh
🎯 What it does: The BiDiff model is proposed to achieve text-to-3D generation by integrating 2D and 3D reverse diffusion.
Text-to-3D using Gaussian Splatting
Zilong Chen (Tsinghua University), Huaping Liu (Tsinghua University)
GenerationData SynthesisOptimizationDiffusion modelGaussian SplattingTextPoint Cloud
🎯 What it does: The GSGEN method is proposed, utilizing 3D Gaussian splatting for text-to-3D generation, and achieving high-quality, geometrically consistent 3D assets through a two-stage process (geometric optimization + appearance refinement).
Text-to-Image Diffusion Models are Great Sketch-Photo Matchmakers
Subhadeep Koley (University of Surrey), Yi-Zhe Song (University of Surrey)
RetrievalDiffusion modelImageMultimodality
🎯 What it does: This paper proposes using a frozen Stable Diffusion text-to-image diffusion model as a backbone feature extractor, combined with visual prompts and learnable text prompts, to achieve zero-shot sketch-to-image retrieval (ZS-SBIR) and cross-category fine-grained retrieval (ZS-FG-SBIR).
Text2HOI: Text-guided 3D Motion Generation for Hand-Object Interaction
Junuk Cha (Ulsan National Institute of Science and Technology), Seungryul Baek (Ulsan National Institute of Science and Technology)
GenerationData SynthesisTransformerDiffusion modelAuto EncoderTextMesh
🎯 What it does: This paper proposes a complete framework for generating 3D hand-object interaction actions from text prompts and object meshes, capable of producing diverse and physically feasible interaction sequences.
Text2Loc: 3D Point Cloud Localization from Natural Language
Yan Xia (Technical University of Munich), Daniel Cremers (Technical University of Munich)
Pose EstimationRetrievalAutonomous DrivingTransformerLarge Language ModelContrastive LearningTextPoint Cloud
🎯 What it does: A 3D point cloud localization method based on natural language description, Text2Loc, is proposed, which can accurately locate target points on city-scale maps.
Text2QR: Harmonizing Aesthetic Customization and Scanning Robustness for Text-Guided QR Code Generation
Guangyang Wu (Shanghai Jiao Tong University), Guangtao Zhai (Shanghai Jiao Tong University)
GenerationOptimizationDiffusion modelImage
🎯 What it does: Proposes the Text2QR method, which uses Stable Diffusion to generate QR codes that align with user aesthetics, and ensures scannability through subsequent optimization.
TextCraftor: Your Text Encoder Can be Image Quality Controller
Yanyu Li (Snap Inc), Jian Ren (Snap Inc)
GenerationData SynthesisReinforcement LearningDiffusion modelImageText
🎯 What it does: Fine-tuning the text encoder of Stable Diffusion improves the quality of generated images and text-image alignment, and can be combined with UNet fine-tuning.
TexTile: A Differentiable Metric for Texture Tileability
Carlos Rodriguez-Pardo (Universidad Rey Juan Carlos), Jorge Lopez-Moreno (Universidad Rey Juan Carlos)
GenerationData SynthesisConvolutional Neural NetworkImageBenchmark
🎯 What it does: Proposed and implemented a differentiable texture tileability metric, TexTile, using it as a no-reference evaluation metric, loss function, and for benchmarking texture synthesis algorithms.
TextNeRF: A Novel Scene-Text Image Synthesis Method based on Neural Radiance Fields
Jialei Cui (Peking University), Zhouhui Lian (Peking University)
GenerationData SynthesisDiffusion modelNeural Radiance FieldImage
🎯 What it does: Utilize NeRF for geometric modeling of real scenes, and insert and edit text in three-dimensional space to achieve controllable scene text image synthesis;
Texture-Preserving Diffusion Models for High-Fidelity Virtual Try-On
Xu Yang (South China University of Technology), Xiangmin Xu (South China University of Technology)
Image TranslationGenerationDiffusion modelImage
🎯 What it does: Proposes the Texture-Preserving Diffusion (TPD) model, which implements texture transfer in diffusion models using self-attention without the need for additional image encoders, and achieves high-fidelity virtual try-on by combining decoupled mask prediction.
TextureDreamer: Image-Guided Texture Synthesis Through Geometry-Aware Diffusion
Yu-Ying Yeh (University of California), Zhengqin Li (Meta Reality Lab)
GenerationData SynthesisDiffusion modelScore-based ModelImageMesh
🎯 What it does: Using a small number (3-5 images) of texture information, the texture is transferred to any target 3D mesh after fine-tuning with a diffusion model, generating high-quality relightable textures.
TexVocab: Texture Vocabulary-conditioned Human Avatars
Yuxiao Liu (Shenzhen International Graduate School, Tsinghua University), Haoqian Wang (Shenzhen International Graduate School, Tsinghua University)
GenerationData SynthesisPose EstimationNeural Radiance FieldVideo
🎯 What it does: A texture dictionary TexVocab is constructed using multi-view RGB videos, combined with body part encoding to achieve animatable high-quality human avatars.
TFMQ-DM: Temporal Feature Maintenance Quantization for Diffusion Models
Yushi Huang (Beihang University), Xianglong Liu (Beihang University)
GenerationData SynthesisCompressionOptimizationDiffusion modelImage
🎯 What it does: A post-training quantization framework specifically designed for diffusion models, TFMQ-DM, is proposed to maintain temporal features to reduce the impact of quantization on generation quality.
The Audio-Visual Conversational Graph: From an Egocentric-Exocentric Perspective
Wenqi Jia (Georgia Tech), Ruohan Gao (Meta)
ClassificationRecognitionGraph Neural NetworkSupervised Fine-TuningVideoMultimodalityGraphAudio
🎯 What it does: A model has been constructed to predict panoramic conversational behavior graphs from first-person videos.
The Devil is in the Details: StyleFeatureEditor for Detail-Rich StyleGAN Inversion and High Quality Image Editing
Denis Bobkov (Higher School of Economics), Dmitry Vetrov (Constructor University)
RestorationGenerationGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes StyleFeatureEditor, which achieves high-quality image reconstruction and editing by encoding and editing in the high-dimensional feature space Fk of StyleGAN.
The Devil is in the Fine-Grained Details: Evaluating Open-Vocabulary Object Detectors for Fine-Grained Understanding
Lorenzo Bianchi (Italian National Research Council), Fabrizio Falchi (Italian National Research Council)
Object DetectionTransformerLarge Language ModelImageBenchmark
🎯 What it does: This paper proposes the fine-grained open vocabulary detection (FG-OVD) task and its evaluation protocol, and constructs a fine-grained evaluation benchmark based on a dynamic vocabulary.
The Manga Whisperer: Automatically Generating Transcriptions for Comics
Ragav Sachdeva (University of Oxford), Andrew Zisserman (University of Oxford)
Object DetectionSegmentationGenerationTransformerSupervised Fine-TuningImageBenchmark
🎯 What it does: This study addresses the accessibility issues of comics by proposing a unified end-to-end model called Magi, which can perform panel, text box, and character detection within a comic page, conduct character clustering (merging identical characters), and establish text-character associations, ultimately generating dialogue transcriptions in reading order.
The Mirrored Influence Hypothesis: Efficient Data Influence Estimation by Harnessing Forward Passes
Myeongseob Ko (Virginia Tech), Ruoxi Jia (Virginia Tech)
Computational EfficiencyData-Centric LearningConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: The 'Mirror Effect Hypothesis' is proposed, and based on this, the Forward-INF algorithm is introduced to estimate the influence of training samples on test predictions through forward propagation.
The More You See in 2D the More You Perceive in 3D
Xinyang Han (University of California Berkeley), Yossi Gandelsman (University of California Berkeley)
GenerationPose EstimationDiffusion modelScore-based ModelNeural Radiance FieldImage
🎯 What it does: A system called SAP3D based on a perspective-conditioned diffusion model is proposed, which can adaptively refine models from any number of uncalibrated images and achieve 3D reconstruction and novel view synthesis.
The Neglected Tails in Vision-Language Models
Shubham Parashar (Texas A&M University), Shu Kong (University of Macau)
ClassificationRetrievalTransformerLarge Language ModelReinforcement LearningPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: This paper studies the performance imbalance issue in zero-shot recognition caused by the long-tail distribution of training data in visual-language models, and proposes a concept frequency estimation and retrieval-enhanced learning framework (REAL) based on LLM to alleviate this problem.
The STVchrono Dataset: Towards Continuous Change Recognition in Time
Yanjun Sun (National Institute of Advanced Industrial Science and Technology), Kenji Iwata (National Institute of Advanced Industrial Science and Technology)
RecognitionSegmentationTransformerLarge Language ModelImageMultimodality
🎯 What it does: The STVchrono dataset is proposed and released, containing 71,900 street view images from 50 cities over an 18-year span, and baseline experiments on continuous change description and instance segmentation tasks are conducted on this dataset.
The Unreasonable Effectiveness of Pre-Trained Features for Camera Pose Refinement
Gabriele Trivigno (Politecnico di Torino), Torsten Sattler (Czech Technical University in Prague)
Pose EstimationOptimizationConvolutional Neural NetworkContrastive LearningGaussian SplattingSimultaneous Localization and MappingImageMesh
🎯 What it does: A simple pose refinement method utilizing pre-trained features and particle filters is proposed, which can iteratively improve the initial pose without the need for any scene-specific training.
Theoretically Achieving Continuous Representation of Oriented Bounding Boxes
Zikai Xiao, Shimin Hu
Object DetectionImage
🎯 What it does: A continuous oriented bounding box (COBB) representation is proposed to address the discontinuity issues of rotation and aspect ratio in traditional OOB representations.