ECCV 2024 Papers — Page 21
European Conference on Computer Vision · 2387 papers
Spatio-Temporal Proximity-Aware Dual-Path Model for Panoramic Activity Recognition
Sumin Lee (Korea Advanced Institute of Science and Technology), Changick Kim (Korea Advanced Institute of Science and Technology)
RecognitionConvolutional Neural NetworkTransformerVideo
🎯 What it does: Propose a dual-path network called SPDP-Net for panoramic activity recognition, which first encodes individual features through spatiotemporal proximity relationships, and then predicts individual actions, community actions, and global actions simultaneously via a dual-path transformer.
SpecFormer: Guarding Vision Transformer Robustness via Maximum Singular Value Penalization
Xixu Hu (City University of Hong Kong), Xing Xie (Microsoft Research Asia)
ClassificationAdversarial AttackTransformerImage
🎯 What it does: To address the vulnerability of Vision Transformers (ViT) to adversarial attacks, the authors propose the SpecFormer scheme, which introduces a maximum singular value penalty (MSVP) in each self-attention layer to control the local Lipschitz constant, thereby enhancing the model's adversarial robustness.
Spectral Subsurface Scattering for Material Classification
Haejoon Lee (Carnegie Mellon University), Aswin Sankaranarayanan
ClassificationImagePhysics Related
🎯 What it does: The study achieves material classification through single-frame measurements of spectral subsurface scattering (S4).
SpeedUpNet: A Plug-and-Play Adapter Network for Accelerating Text-to-Image Diffusion Models
Weilong Chai (Ant Group), Chenguang Ma (Ant Group)
GenerationComputational EfficiencyConvolutional Neural NetworkDiffusion modelImageText
🎯 What it does: Propose SpeedUpNet (SUN), a plug-and-play adapter network that can be trained-free, achieving acceleration of text-to-image diffusion models by learning positive and negative prompt offsets in the U-Net cross-attention layer, removing CFG, and enabling single-step inference.
SphereHead: Stable 3D Full-head Synthesis with Spherical Tri-plane Representation
Heyuan Li (Chinese University of Hong Kong Shenzhen), Xiaoguang Han (Chinese University of Hong Kong Shenzhen)
GenerationData SynthesisNeural Radiance FieldGenerative Adversarial NetworkImage
🎯 What it does: Proposed a 3D head synthesis framework called SphereHead, achieving full-view, artifact-free head generation through spherical tri-plane representation and view-image consistency loss.
Spherical Linear Interpolation and Text-Anchoring for Zero-shot Composed Image Retrieval
Young Kyun Jang (Meta AI), Ser-Nam Lim (University of Central Florida)
RetrievalSupervised Fine-TuningVision Language ModelContrastive LearningMultimodality
🎯 What it does: Propose a zero-shot compositional image retrieval (ZS-CIR) method based on spherical linear interpolation (Slerp), and introduce a text-anchored fine-tuning (TAT) technique to bridge the modality gap between images and text, achieving efficient retrieval without manual annotation.
Spherical World-Locking for Audio-Visual Localization in Egocentric Videos
Heeseung Yun (Seoul National University), Calvin Murdock (Reality Labs Research at Meta)
Pose EstimationTransformerVideoMultimodality
🎯 What it does: Proposed the Spherical World-Locking framework, mapping multimodal audio-visual information onto a head-centered sphere to mitigate interference caused by self-motion.
Spike-Temporal Latent Representation for Energy-Efficient Event-to-Video Reconstruction
Jianxiong Tang (Sun Yat-sen University), Xiaohua Xie (Sun Yat-sen University)
Image TranslationSpiking Neural NetworkVideo
🎯 What it does: Proposed a spiking neural network model called STLR based on spatiotemporal latent representations for reconstructing event streams from event cameras into grayscale videos.
Spiking Wavelet Transformer
Yuetong Fang (Hong Kong University of Science and Technology), Renjing Xu (Hong Kong University of Science and Technology)
ClassificationRecognitionSpiking Neural NetworkTransformerImage
🎯 What it does: Proposed an attention-free Spiking Wavelet Transformer (SWformer) that combines spiking neural networks with wavelet transforms to achieve event-driven learning of high-frequency features.
SPIN: Hierarchical Segmentation with Subpart Granularity in Natural Images
Josh David Myers-Dean (University of Colorado Boulder), Danna Gurari (University of Colorado Boulder)
SegmentationConvolutional Neural NetworkTransformerLarge Language ModelVision Language ModelImageBenchmark
🎯 What it does: Create the first natural image sub-part hierarchical semantic segmentation dataset SPIN, propose two evaluation metrics: spatial consistency and semantic consistency, and conduct quantitative benchmark evaluations of various modern models at the object, part, and sub-part levels.
SPIRE: Semantic Prompt-Driven Image Restoration
Chenyang QI, Hossein Talebi (Google Research)
RestorationConvolutional Neural NetworkPrompt EngineeringVision Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: Built SPIRE, a text-prompt-based image restoration framework that can simultaneously utilize semantic prompts, restoration strength prompts, and blind restoration instructions to achieve controllable editing of degradation types and restoration intensity.
SplatFields: Neural Gaussian Splats for Sparse 3D and 4D Reconstruction
Marko Mihajlovic (ETH Zürich), Edmond Boyer (Meta Reality Labs)
GenerationConvolutional Neural NetworkNeural Radiance FieldGaussian SplattingImage
🎯 What it does: Regularize 3D Gaussian Splatting by proposing SplatFields, which generates spatially correlated splat features through neural networks and extends the framework to 4D dynamic scenes.
Spline-based Transformers
Prashanth Chandran (DisneyResearch|Studios), Moritz Bächer (Disney Research)
Representation LearningTransformerImageVideo
🎯 What it does: Proposed a class of Transformer models named Spline-based Transformers, which encode sequence information using B-spline hidden space trajectory representations, eliminating the need for traditional position encodings.
SPVLoc: Semantic Panoramic Viewport Matching for 6D Camera Localization in Unseen Environments
Niklas Gard (Fraunhofer Heinrich Hertz Institute), Peter Eisert (Fraunhofer Heinrich Hertz Institute)
Object DetectionPose EstimationDomain AdaptationConvolutional Neural NetworkNeural Radiance FieldSimultaneous Localization and MappingImageMesh
🎯 What it does: Propose a 6D indoor camera localization method SPVLoc based on semantic panoramic viewport matching, which can achieve precise localization in unseen scenes using only a simplified 3D model.
SQ-LLaVA: Self-Questioning for Large Vision-Language Assistant
Guohao Sun (Rochester Institute of Technology), Zhiqiang Tao (Rochester Institute of Technology)
Representation LearningTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: This paper proposes a self-questioning training method that trains visual-language models to proactively generate high-quality questions related to images after receiving them, thereby enhancing cross-modal alignment and understanding capabilities.
SRPose: Two-view Relative Pose Estimation with Sparse Keypoints
Rui Yin (Hanglok-Tech), Biao Jia (Hanglok-Tech)
Pose EstimationTransformerImage
🎯 What it does: This paper proposes SRPose, a two-view relative pose estimation framework based on sparse keypoints, capable of simultaneously handling two scenarios: camera-to-world and object-to-camera.
SSL-Cleanse: Trojan Detection and Mitigation in Self-Supervised Learning
Mengxin Zheng (University of Central Florida), Xiaofeng Wang (University of Central Florida)
Anomaly DetectionRepresentation LearningAdversarial AttackContrastive LearningImage
🎯 What it does: This paper proposes a complete framework named SSL-Cleanse for detecting and removing backdoors in self-supervised learning (SSL) encoders without relying on downstream labels or training sets.
ST-LDM: A Universal Framework for Text-Grounded Object Generation in Real Images
Xiangtian Xue (Southeast University), Huazhong Shu (Southeast University)
GenerationTransformerVision Language ModelDiffusion modelAuto EncoderImageTextMultimodality
🎯 What it does: Proposed a general framework ST-LDM based on Swin-Transformer, which can generate new objects (Text-Grounded Object Generation, TOG) in real images according to text descriptions, and achieve spatial control of LDM through training-free region backward guidance.
ST-LLM: Large Language Models Are Effective Temporal Learners
Ruyang Liu (Peking University), Ge Li (Peking University)
RecognitionTransformerLarge Language ModelVideoText
🎯 What it does: Propose the ST-LLM model, which directly inputs all spatial-temporal visual tokens into LLM, achieving efficient video understanding through dynamic masking and global-local input.
Stable Preference: Redefining training paradigm of human preference model for Text-to-Image Synthesis
Hanting Li (University of Science and Technology of China), Feng Zhao (University of Science and Technology of China)
GenerationReinforcement Learning from Human FeedbackSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: Trained a human preference model for text-to-image generation, proposing the Stable Preference two-step training method.
Stable Video Portraits
Mirela Ostrek (Max Planck Institute for Intelligent Systems), Justus Thies (Max Planck Institute for Intelligent Systems)
GenerationDiffusion modelVideoText
🎯 What it does: Proposes Stable Video Portraits, a hybrid 2D/3D monocular head avatar generation method that integrates 2D Stable Diffusion with 3DMM, capable of generating high-fidelity, temporally coherent video avatars and transforming avatars into celebrity faces via text input without additional fine-tuning during testing.
StableDrag: Stable Dragging for Point-based Image Editing
Yutao Cui (Nanjing University), Limin Wang (Nanjing University)
GenerationConvolutional Neural NetworkSupervised Fine-TuningDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: Propose the StableDrag framework, improving point tracking and motion supervision to achieve more stable point-based image editing, and build StableDrag-GAN and StableDrag-Diff based on GAN and diffusion models.
STAG4D: Spatial-Temporal Anchored Generative 4D Gaussians
Yifei Zeng (Nanjing University), Yao Yao (CASIA)
GenerationData SynthesisDiffusion modelScore-based ModelGaussian SplattingImageVideoTextMultimodality
🎯 What it does: Propose the STAG4D framework, combining pre-trained diffusion models with 4D Gaussian splatting to achieve high-fidelity 4D content generation from video, text, or image inputs.
STAMP: Outlier-Aware Test-Time Adaptation with Stable Memory Replay
Yu Yongcan (Chinese Academy of Sciences), Jian Liang (Chinese Academy of Sciences)
Domain AdaptationAnomaly DetectionOptimizationImage
🎯 What it does: This study proposes a framework called STAMP for test-time adaptation in the presence of unknown classes, achieving identification and anomaly detection through reliable class-balanced memory and self-weighted entropy minimization.
Statewide Visual Geolocalization in the Wild
Florian Fervers (Fraunhofer Institute of Optronics, System Technologies and Image Exploitation), Rainer Stiefelhagen (Karlsruhe Institute of Technology)
Pose EstimationRetrievalConvolutional Neural NetworkTransformerContrastive LearningImage
🎯 What it does: Proposes a statewide-level street-level image geolocation method based on multi-scale aerial map matching.
Stepping Stones: A Progressive Training Strategy for Audio-Visual Semantic Segmentation
Juncheng Ma (University of Chinese Academy of Sciences), Di Hu (Renmin University of China)
SegmentationConvolutional Neural NetworkTransformerMultimodality
🎯 What it does: This paper presents two contributions: a two-stage progressive training strategy (Stepping Stones) for audio-visual semantic segmentation (AVSS), and a novel audio-visual segmentation framework (Adaptive Audio Visual Segmentation, AAVS).
Stepwise Multi-grained Boundary Detector for Point-supervised Temporal Action Localization
Mengnan Liu (Xi'an Jiaotong University), Gang Hua (Dolby Laboratories)
RecognitionVideo
🎯 What it does: This paper proposes a temporal action localization framework based on point-level supervision called Stepwise Multi-grained Boundary Detector (SMBD), which first generates reliable background anchors (Background Anchor Generator) and then utilizes a dual boundary detector (Dual Boundary Detector) to identify fine-grained and coarse-grained action boundaries between background and action segments, thereby assigning pseudo labels to every frame in the video to achieve learning of complete action semantics;
StereoGlue: Joint Feature Matching and Robust Estimation
Daniel Barath (ETH Zurich), Marc Pollefeys (ETH Zurich)
Pose EstimationImagePoint Cloud
🎯 What it does: Propose a joint feature matching and robust estimation framework called StereoGlue, which generates candidate matches, estimates models, and guides matching through an iterative process using a single-point minimal solver, thereby achieving consistent one-to-one correspondences and model scores.
Stitched ViTs are Flexible Vision Backbones
Zizheng Pan (Monash University), Bohan Zhuang (Monash University)
Object DetectionSegmentationDepth EstimationTransformerImage
🎯 What it does: Proposes the SN-Netv2 framework, which can flexibly adapt large-scale pre-trained ViT to downstream dense prediction tasks without retraining multi-scale ViT;
StoryImager: A Unified and Efficient Framework for Coherent Story Visualization and Completion
Ming Tao (Nanjing University of Posts and Telecommunications), Changsheng Xu (Peng Cheng Laboratory)
GenerationTransformerSupervised Fine-TuningDiffusion modelImageText
🎯 What it does: Proposes a unified and efficient framework for story visualization and completion called StoryImager, supporting bidirectional generation and addressing both story visualization and story completion tasks.
Straightforward Layer-wise Pruning for More Efficient Visual Adaptation
Ruizi Han (Northwest A&F University), Jinglei Tang (Northwest A&F University)
ClassificationObject DetectionSegmentationComputational EfficiencyTransformerImageBenchmark
🎯 What it does: Propose a pruning method called SLS based on hierarchical feature clustering, which prunes redundant layers caused by frozen parameters in PETL models during cross-domain tasks;
Stream Query Denoising for Vectorized HD-Map Construction
Shuo Wang (University of Science and Technology of China), Feng Zhao (University of Science and Technology of China)
Autonomous DrivingTransformerMultimodality
🎯 What it does: Proposed and implemented the Stream Query Denoising (SQD) method, which leverages denoised queries to learn temporal consistency in high-precision HD-Maps.
Street Gaussians: Modeling Dynamic Urban Scenes with Gaussian Splatting
Yunzhi Yan (Zhejiang University), Sida Peng (Zhejiang University)
Autonomous DrivingGaussian SplattingPoint Cloud
🎯 What it does: Designed an explicit scene representation based on 3D Gaussian point clouds called Street Gaussians for fast modeling and real-time rendering of dynamic urban street scenes.
Strengthening Multimodal Large Language Model with Bootstrapped Preference Optimization
Renjie Pi (Hong Kong University of Science and Technology), Tong Zhang (University of Illinois at Urbana-Champaign)
OptimizationReinforcement Learning from Human FeedbackLarge Language ModelVision Language ModelMultimodality
🎯 What it does: Propose Bootstrapped Preference Optimization (BPO), which conducts preference learning for multi-modal large language models by automatically generating negative samples, thereby mitigating pre-training bias and improving the accuracy of visual information perception and generation.
Strike a Balance in Continual Panoptic Segmentation
Jinpeng Chen (City University of Hong Kong), Sam Kwong (Lingnan University)
SegmentationKnowledge DistillationTransformerImage
🎯 What it does: This paper proposes a new framework called BalConpas for Continual Panoptic Segmentation (CPS), aiming to address core challenges in continual learning, including knowledge retention and adaptation to new knowledge, class distribution imbalance, and misleading effects caused by incomplete annotations of replay samples.
Stripe Observation Guided Inference Cost-free Attention Mechanism
Zhongzhan Huang (Sun Yat-sen University), Liang Lin (Sun Yat-sen University)
ClassificationObject DetectionSegmentationComputational EfficiencyConvolutional Neural NetworkTransformerImage
🎯 What it does: Proposes an Attention-like Structural Re-parameterization (ASR) based on Stripe Observation, which uses learnable vectors as inputs to the attention module, making attention values converge to constants after training, thereby achieving structural re-parameterization without additional inference costs.
StructLDM: Structured Latent Diffusion for 3D Human Generation
Tao Hu (Nanyang Technological University), Ziwei Liu (Nanyang Technological University)
GenerationPose EstimationConvolutional Neural NetworkDiffusion modelNeural Radiance FieldGenerative Adversarial NetworkImageVideoMesh
🎯 What it does: This paper proposes a 3D human generation framework based on a structured latent diffusion model (StructLDM), which utilizes a 2D structured latent representation in a dense UV space to construct a self-decoder combining local NeRF and a global style mixer. A diffusion model is trained in this latent space to achieve unconditional and controllable high-quality 3D human generation and editing across multiple poses and viewpoints.
Structured-NeRF: Hierarchical Scene Graph with Neural Representation
Zhide Zhong (Hong Kong University Of Science And Technology), Zike Yan (Tsinghua University)
GenerationOptimizationConvolutional Neural NetworkGraph Neural NetworkLarge Language ModelVision Language ModelNeural Radiance FieldImageStochastic Differential Equation
🎯 What it does: By constructing a hierarchical scene graph, the NeRF model is decomposed into editable object nodes, enabling pose optimization and shadow rendering based on semantic and physical constraints, thus achieving indoor scene reconstruction and interactive editing.
STSP: Spatial-Temporal Subspace Projection for Video Class-incremental Learning
Hao Cheng (Nanyang Technological University), Bihan Wen (Nanyang Technological University)
ClassificationRecognitionSafty and PrivacyRepresentation LearningVideoBenchmark
🎯 What it does: This paper proposes a sample-free memory, subspace projection-based framework named STSP for video category incremental learning (VCIL), addressing the problem of forgetting old categories when new categories are introduced.
Style-Extracting Diffusion Models for Semi-Supervised Histopathology Segmentation
Mathias Öttl (Friedrich-Alexander-Universität), Katharina Breininger (Friedrich-Alexander-Universität)
SegmentationGenerationTransformerDiffusion modelAuto EncoderImageBiomedical Data
🎯 What it does: Designed and implemented Style-Extracting Diffusion Models (STEDM), which can generate diverse images by leveraging style information from unseen images given content (e.g., semantic layouts), and applied these synthetic images to semi-supervised histopathological segmentation tasks.
StyleCity: Large-Scale 3D Urban Scenes Stylization
Yingshu Chen (Hong Kong University of Science and Technology), Sai-Kit Yeung (Hong Kong University of Science and Technology)
GenerationVision Language ModelDiffusion modelImageTextMesh
🎯 What it does: Perform visual and text-driven texture style transfer on large-scale 3D urban scenes, generating panoramic sky backgrounds consistent with the style.
StyleTokenizer: Defining Image Style by a Single Instance for Controlling Diffusion Models
Wen Li (Ant Group), Ming Yang (Ant Group)
GenerationTransformerDiffusion modelContrastive LearningImageTextMultimodality
🎯 What it does: Propose a StyleTokenizer method that maps the style of a single reference image into the text embedding space, enabling single-image style control without training in Stable Diffusion;
Subspace Prototype Guidance for Mitigating Class Imbalance in Point Cloud Semantic Segmentation
Jiawei Han (Beijing Institute of Technology), Guangzhi Chen (Beijing Institute of Technology)
SegmentationContrastive LearningPoint Cloud
🎯 What it does: Propose the Subspace Prototype Guidance (SPG) method, which generates class prototypes through an auxiliary branch in point cloud semantic segmentation and achieves mutual supervision with the main network to alleviate the class imbalance problem.
SUMix: Mixup with Semantic and Uncertain Information
Huafeng Qin (Chongqing Technology and Business University), Xinbo Gao (Chongqing University of Posts and Telecommunications)
ClassificationConvolutional Neural NetworkTransformerImage
🎯 What it does: Propose the SUMix method, which learns mixing proportions and models uncertainty in mixed samples to address the label mismatch problem during the Mixup process.
SUP-NeRF: A Streamlined Unification of Pose Estimation and NeRF for Monocular 3D Object Reconstruction
Yuliang Guo (Bosch Research North America), Liu Ren (Bosch Research North America)
GenerationPose EstimationDepth EstimationAutonomous DrivingConvolutional Neural NetworkNeural Radiance FieldImage
🎯 What it does: Propose a unified network framework called SUP-NeRF, which can simultaneously predict the target's pose, shape, and texture from a single image, and achieve high-quality monocular 3D reconstruction without relying on external 3D detectors;
SuperFedNAS: Cost-Efficient Federated Neural Architecture Search for On-Device Inference
Alind Khare (Georgia Institute of Technology), Alexey Tumanov (Cisco Research)
Federated LearningNeural Architecture SearchImageText
🎯 What it does: Proposes the SuperFedNAS method, combining federated learning with super networks to achieve local NAS that requires no additional training after a single training session, enabling rapid fulfillment of diverse device inference objectives.
SuperGaussian: Repurposing Video Models for 3D Super Resolution
Yuan Shen (University of Illinois at Urbana-Champaign), Anna Fruehstueck
Super ResolutionDomain AdaptationSupervised Fine-TuningGaussian SplattingVideoPoint Cloud
🎯 What it does: By transferring a pre-trained video super-resolution model to the 3D super-resolution task, a general framework is constructed to upgrade coarse 3D models (such as low-resolution Gaussian Splats, NeRF, etc.) into high-resolution Gaussian Splat representations.
Superpixel-informed Implicit Neural Representation for Multi-Dimensional Data
Jia-Yi Li, Min Wang (Jiangxi University of Finance and Economics)
RestorationRepresentation LearningTransformerImageMultimodalityPoint Cloud
🎯 What it does: Proposes using generalized superpixels as the fundamental units of implicit neural representations (INR), and achieves recovery and enhancement of multi-dimensional data through specialized attention MLP and shared dictionary matrices.
Sur^2f: A Hybrid Representation for High-Quality and Efficient Surface Reconstruction from Multi-view Images
Zhangjin Huang (South China University of Technology), Kui Jia (Chinese University of Hong Kong)
ImageMesh
🎯 What it does: Propose a new hybrid representation, Surf2f, for reconstructing high-quality and efficient 3D surfaces from multi-view images.
Surf-D: Generating High-Quality Surfaces of Arbitrary Topologies Using Diffusion Models
Zhengming Yu (Texas Aamp;M University), Wenping Wang (Texas Aamp;M University)
GenerationData SynthesisDiffusion modelAuto EncoderImageTextMultimodalityPoint CloudMesh
🎯 What it does: This paper proposes Surf-D, a framework that generates high-quality, arbitrary topology 3D surfaces using diffusion models.
Surface Reconstruction for 3D Gaussian Splatting via Local Structural Hints
Qianyi Wu (Monash University), Jianfei Cai (Monash University)
GenerationDepth EstimationOptimizationGaussian SplattingImageMesh
🎯 What it does: Propose a method called GSrec for surface mesh reconstruction using 3D Gaussian Splatting, which optimizes Gaussian atoms with local structural hints to achieve more accurate 3D surfaces.
Surface-Centric Modeling for High-Fidelity Generalizable Neural Surface Reconstruction
Rui Peng (Peking University), Ronggang Wang (University of Birmingham)
Convolutional Neural NetworkNeural Radiance FieldImageBenchmark
🎯 What it does: Proposes SuRF, a neural surface reconstruction framework based on surface centers, which achieves high-fidelity surface reconstruction under sparse multi-view images.
SV3D: Novel Multi-view Synthesis and 3D Generation from a Single Image using Latent Video Diffusion
Vikram Voleti, Varun Jampani
Image TranslationGenerationData SynthesisVision Language ModelDiffusion modelScore-based ModelNeural Radiance FieldGaussian SplattingImageMesh
🎯 What it does: Propose SV3D, a single-image multi-view synthesis and 3D generation framework based on video diffusion models, capable of generating consistent, controllable multi-view images from a single image and optimizing them into high-quality 3D meshes.
SWAG: Splatting in the Wild images with Appearance-conditioned Gaussians
Hiba Dahmani (Huawei Paris Research Center), Dzmitry Tsishkou (Huawei Paris Research Center)
RestorationGenerationGaussian SplattingImage
🎯 What it does: Extend 3D Gaussian Splatting to unconstrained outdoor image collections, achieving scene reconstruction, variable appearance synthesis, and transient object removal.
SwapAnything: Enabling Arbitrary Object Swapping in Personalized Image Editing
Jing Gu (University of California, Santa Cruz), Xin Eric Wang (University of California, Santa Cruz)
Image TranslationGenerationConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: SwapAnything achieves personalized replacement of any target object in images while maintaining the background unchanged by performing targeted variable swapping in the U-Net of Stable Diffusion.
SweepNet: Unsupervised Learning Shape Abstraction via Neural Sweepers
Mingrui Zhao (Simon Fraser University), Ali Mahdavi-Amiri (Simon Fraser University)
Representation LearningConvolutional Neural NetworkAuto EncoderMesh
🎯 What it does: Investigated an unsupervised shape abstraction method that exploits learnable sweeping surfaces to approximate 3D shapes.
SwiftBrush v2: Make Your One-step Diffusion Model Better Than Its Teacher
Trung Tuan Dao (VinAI Research), Anh T Tran (VinAI Research)
GenerationData SynthesisKnowledge DistillationDiffusion modelImageText
🎯 What it does: This paper improves SwiftBrush by using SD Turbo weight initialization, introducing clamped CLIP loss, and adopting two efficient training schemes (full training and LoRA+TinyVAE), ultimately training a first-order text-image diffusion model with FID-30K 8.14, which outperforms multi-step Stable Diffusion.
SWinGS: Sliding Windows for Dynamic 3D Gaussian Splatting
Richard Shaw, Eduardo Pérez Pellitero (Huawei)
GenerationGaussian SplattingOptical FlowVideoPoint Cloud
🎯 What it does: Propose a dynamic 3D Gaussian Splatting (SWinGS) framework based on a sliding window, capable of reconstructing dynamic scenes from multi-view videos and achieving real-time view synthesis.
Switch Diffusion Transformer: Synergizing Denoising Tasks with Sparse Mixture-of-Experts
Byeongjun Park (KAIST), Changick Kim (KAIST)
GenerationData SynthesisTransformerMixture of ExpertsDiffusion modelImage
🎯 What it does: This paper proposes Switch-DiT, which combines sparse mixture of experts (SMoE) and time-step driven gating networks to better model the interrelationships and parameter isolation among different denoising tasks.
Syn-to-Real Domain Adaptation for Point Cloud Completion via Part-based Approach
Yunseo Yang (KAIST), Kuk-Jin Yoon (KAIST)
RestorationDomain AdaptationTransformerGenerative Adversarial NetworkPoint Cloud
🎯 What it does: A part-based point cloud completion framework is studied, which leverages complete synthetic point clouds and incomplete real point clouds for domain adaptation, enabling the model to complete real-world point clouds without requiring complete real annotations.
Sync from the Sea: Retrieving Alignable Videos from Large-Scale Datasets
Ishan Rajendrakumar Dave (University of Central Florida), Simon Jenni (Adobe Research)
RetrievalContrastive LearningVideo
🎯 What it does: Propose the Aligned Retrievable Video (AVR) task, construct a three-stage pipeline of retrieval-re-ranking-alignment, evaluate video alignability using the DRAQ metric, and retrieve and align videos on large-scale datasets.
Synchronization is All You Need: Exocentric-to-Egocentric Transfer for Temporal Action Segmentation with Unlabeled Synchronized Video Pairs
Camillo Quattrocchi (University of Catania), Giovanni Maria Farinella (University of Catania)
SegmentationDomain AdaptationKnowledge DistillationRepresentation LearningConvolutional Neural NetworkTransformerVideoSequential
🎯 What it does: Studied a method to adapt temporal action segmentation models using synchronized unlabeled exo-ego video pairs.
Synchronization of Projective Transformations
Rakshith Madhavan, Federica Arrigoni
Computational EfficiencyData-Centric LearningConvolutional Neural NetworkRecurrent Neural NetworkImage
🎯 What it does: The paper explores a new algorithm to improve the training efficiency and accuracy of deep learning models.
Synchronous Diffusion for Unsupervised Smooth Non-Rigid 3D Shape Matching
Dongliang Cao (University of Bonn), Florian Bernard (University of Bonn)
OptimizationDiffusion modelMesh
🎯 What it does: Proposed an unsupervised synchronized diffusion regularization method to enhance spatial smoothness in non-rigid 3D shape matching.
Synergy of Sight and Semantics: Visual Intention Understanding with CLIP
Qu Yang (National Engineering Research Center for Multimedia Software, Wuhan University), Dacheng Tao (Nanyang Technological University)
ClassificationRecognitionConvolutional Neural NetworkTransformerPrompt EngineeringVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: Designed and implemented the IntCLIP framework, integrating CLIP's visual knowledge with multi-label intent understanding tasks, achieving efficient image intent recognition through dual-branch encoders, hierarchical category integration, and vision-knowledge-assisted aggregation.
Synthesizing Environment-Specific People in Photographs
Mirela Ostrek (Max Planck Institute for Intelligent Systems), Justus Thies (Max Planck Institute for Intelligent Systems)
Image TranslationImage HarmonizationSegmentationGenerationData SynthesisPose EstimationSuper ResolutionDiffusion modelAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: Propose the ESP method, which generates full-body characters that fit the scene and performs seamless illustration by utilizing 2D pose and environmental image information.
Synthesizing Time-varying BRDFs via Latent Space
Takuto Narumoto (Osaka University), Fumio Okura (Osaka University)
GenerationData SynthesisRecurrent Neural NetworkAuto EncoderImageTime Series
🎯 What it does: Propose a neural model that learns temporal variations in a latent space for synthesizing time-varying BRDFs.
T-CorresNet: Template Guided 3D Point Cloud Completion with Correspondence Pooling Query Generation Strategy
Fan Duan (Tsinghua University), Li Chen (Tsinghua University)
RestorationGraph Neural NetworkTransformerPoint Cloud
🎯 What it does: Proposes a 3D point cloud completion method based on Gaussian spherical template-guided coarse-to-fine template generation and correspondence-pooling query generator.
T-MAE: Temporal Masked Autoencoders for Point Cloud Representation Learning
Weijie Wei (University of Amsterdam), Martin R. Oswald (University of Amsterdam)
Autonomous DrivingRepresentation LearningTransformerAuto EncoderPoint Cloud
🎯 What it does: Proposes a Temporal Masked Autoencoders (T-MAE) self-supervised pre-training framework that leverages the temporal relationship between two LiDAR point cloud frames. By occluding a high proportion of the current frame and using the complete information from historical frames to reconstruct the current frame, it learns powerful representations and temporal modeling for sparse point clouds.
T-Rex2: Towards Generic Object Detection via Text-Visual Prompt Synergy
Qing Jiang (South China University of Technology), Lei Zhang (South China University of Technology)
Object DetectionTransformerPrompt EngineeringContrastive LearningMultimodality
🎯 What it does: Proposed and implemented T-Rex2, an open-set object detection model that can collaborate through text and visual prompts, supporting multiple prompt methods and achieving zero-shot object detection within a single framework.
T2IShield: Defending Against Backdoors on Text-to-Image Diffusion Models
Zhongqi Wang (Key Laboratory of AI Safety, Institute of Computing Technology, Chinese Academy of Sciences), Xilin Chen (Key Laboratory of AI Safety, Institute of Computing Technology, Chinese Academy of Sciences)
Safty and PrivacyAdversarial AttackDiffusion modelContrastive LearningImageText
🎯 What it does: Designed T2IShield for detecting, locating, and eliminating backdoor attacks in text-to-image diffusion models.
Tackling Structural Hallucination in Image Translation with Local Diffusion
Seunghoi Kim (University College London), Daniel Alexander (AstraZeneca)
Image TranslationAnomaly DetectionDiffusion modelImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: By incorporating a local diffusion process into pre-trained diffusion models, first using PatchCore for OOD region detection, then performing parallel generation (branching) separately on OOD and IND regions, and finally fusing predictions (fusion), thus reducing structural hallucinations in image translation.
TAG: Text Prompt Augmentation for Zero-Shot Out-of-Distribution Detection
Xixi Liu (Chalmers University of Technology), Christopher Zach (Chalmers University of Technology)
Anomaly DetectionPrompt EngineeringMultimodality
🎯 What it does: This paper proposes a zero-shot OOD detection method based on Text Prompt Augmentation (TAG), leveraging CLIP's multimodal features to enhance the separation between ID and OOD samples.
Take A Step Back: Rethinking the Two Stages in Visual Reasoning
Mingyu Zhang (Shanghai Jiao Tong University), Yong-Lu Li (Shanghai Jiao Tong University)
Convolutional Neural NetworkGraph Neural NetworkTransformerImageTextMultimodalityBenchmark
🎯 What it does: Propose decomposing visual reasoning into two stages: symbolization and reasoning, and experimentally verify that a shared reasoner can achieve cross-domain generalization.
TalkingGaussian: Structure-Persistent 3D Talking Head Synthesis via Gaussian Splatting
Jiahe Li (Beihang University), Lin Gu (RIKEN AIP)
GenerationData SynthesisGaussian SplattingVideoPoint Cloud
🎯 What it does: This paper proposes a deformation-driven 3D talking head avatar synthesis framework based on 3D Gaussian scattering—TalkingGaussian—which can generate realistic talking head videos by applying smooth deformations to persistent Gaussian primitives.
Taming CLIP for Fine-grained and Structured Visual Understanding of Museum Exhibits
Ada-Astrid Balauca (INSAIT, Sofia University), Luc Van Gool (ETH Zurich)
RecognitionTransformerVision Language ModelContrastive LearningImageTabular
🎯 What it does: Studies how to utilize the CLIP model combined with a parsing network to generate structured tabular information from museum exhibit images.
Taming Latent Diffusion Model for Neural Radiance Field Inpainting
Chieh Hubert Lin, Hung-Yu Tseng
RestorationDiffusion modelNeural Radiance FieldImage
🎯 What it does: This paper studies how to enhance view rendering quality within the NeRF (Neural Radiance Fields) framework by using the Masked LPIPS loss, further comparing the performance of different NeRF models;
Taming Lookup Tables for Efficient Image Retouching
Sidi Yang (Tsinghua University), Yujiu Yang (University Of Hong Kong)
Image HarmonizationComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: Designed and implemented an image enhancement model called ICELUT, which is entirely based on lookup tables (LUTs), eliminating CNN computations to achieve ultra-low-latency inference.
TAPTR: Tracking Any Point with Transformers as Detection
Hongyang Li (South China University of Technology), Lei Zhang (South China University of Technology)
Object TrackingTransformerVideo
🎯 What it does: Developed a TAPTR model based on the DETR transformer for tracking arbitrary points in videos.
Task-Driven Uncertainty Quantification in Inverse Problems via Conformal Prediction
Jeffrey Wen (Ohio State University), Phillip Schniter
ClassificationRestorationFlow-based ModelBiomedical Data
🎯 What it does: This paper proposes a task-driven uncertainty quantification framework based on conformal prediction, for evaluating uncertainty in downstream tasks (e.g., soft output classification) from images recovered with limited measurements.
TC4D: Trajectory-Conditioned Text-to-4D Generation
Sherwin Bahmani (University of Toronto), David B Lindell
GenerationPrompt EngineeringDiffusion modelScore-based ModelNeural Radiance FieldVideoTextMesh
🎯 What it does: Propose a trajectory-based text-driven 4D generation framework (TC4D) that can synthesize large-scale, controllable global motion and detailed local animations while maintaining 3D geometric integrity.
TCAN: Animating Human Images with Temporally Consistent Pose Guidance using Diffusion Models
Jeongho Kim (KAIST), Jaegul Choo (KAIST)
GenerationPose EstimationConvolutional Neural NetworkTransformerDiffusion modelImageVideo
🎯 What it does: Achieve pose and appearance feature alignment using a frozen pre-trained ControlNet and LoRA, and propose the TCAN framework to realize temporally consistent animation under driven video poses by introducing Temporal ControlNet and Pose-Driven Temperature Map.
TCC-Det: Temporarily consistent cues for weakly-supervised 3D detection
Jan Skvrna (Czech Technical University), Lukáš Neumann
Autonomous DrivingConvolutional Neural NetworkImagePoint Cloud
🎯 What it does: Propose a method that utilizes off-the-shelf 2D detectors and temporal consistency for training a 3D object detector without manual annotations.
TCLC-GS: Tightly Coupled LiDAR-Camera Gaussian Splatting for Autonomous Driving
Cheng Zhao (Bosch Research North America Bosch Center for Artificial Intelligence), Liu Ren (Bosch Research North America Bosch Center for Artificial Intelligence)
Autonomous DrivingGaussian SplattingImageMultimodalityPoint Cloud
🎯 What it does: A 3D Gaussian distribution model called TCLC-GS, which integrates LiDAR and camera data, is constructed to rapidly and high-quality reconstruct and real-time render scenes in autonomous driving environments.
Teach CLIP to Develop a Number Sense for Ordinal Regression
Yao DU, Xiaomeng Li (Hong Kong University of Science and Technology)
Vision Language ModelContrastive LearningMultimodalityBenchmark
🎯 What it does: Propose NumCLIP, which leverages CLIP's cross-modal knowledge to learn numerical perception and improve ordinal regression performance
Teaching Tailored to Talent: Adverse Weather Restoration via Prompt Pool and Depth-Anything Constraint
Sixiang Chen, Lei Zhu (Hong Kong University of Science and Technology)
RestorationDepth EstimationPrompt EngineeringDiffusion modelContrastive LearningImage
🎯 What it does: Proposed the T-DiffWeather 3 model, utilizing a prompt pool and Depth-Anything constraints to achieve efficient removal of adverse weather conditions such as rain, fog, and snow, and restore background images under multiple weather scenarios.
Teddy: Efficient Large-Scale Dataset Distillation via Taylor-Approximated Matching
Ruonan Yu (National University of Singapore), Xinchao Wang (National University of Singapore)
Computational EfficiencyKnowledge DistillationImage
🎯 What it does: This paper proposes the Teddy framework, which decouples bi-level optimization through Taylor approximation and pre-caches weak teacher models to achieve efficient training for large-scale dataset distillation.
Temporal As a Plugin: Unsupervised Video Denoising with Pre-Trained Image Denoisers
Zixuan Fu (Nanyang Technological University), Bihan Wen (Nanyang Technological University)
RestorationConvolutional Neural NetworkAuto EncoderVideo
🎯 What it does: Propose an unsupervised video denoising framework called TAP that leverages a pre-trained image denoiser to remove noise from videos without requiring noisy-clean video pairs;
Temporal Event Stereo via Joint Learning with Stereoscopic Flow
Hoonhee Cho (Korea Advanced Institute of Science and Technology), Kuk-Jin Yoon (Korea Advanced Institute of Science and Technology)
Depth EstimationAutonomous DrivingConvolutional Neural NetworkContrastive LearningOptical FlowTime Series
🎯 What it does: This paper proposes a temporal event stereo matching framework that significantly improves the stereo matching accuracy of event cameras by jointly learning stereo flow and stereo matching networks, using features and cost volumes from previous time steps for Warping and fusion.
Temporal Residual Guided Diffusion Framework for Event-Driven Video Reconstruction
Lin Zhu (Beijing Institute of Technology), Hua Huang (Beijing Normal University)
RestorationGenerationRecurrent Neural NetworkTransformerDiffusion modelTime SeriesSequential
🎯 What it does: The paper proposes a time residual guided diffusion framework for reconstructing high-quality videos from event streams.
Temporal Residual Jacobians for Rig-free Motion Transfer
Sanjeev Muralikrishnan (University College London), Niloy Mitra
GenerationMeshSequentialOrdinary Differential Equation
🎯 What it does: Achieve realistic motion transfer for unbound skeletons on 3D meshes by learning local deformations in space and time through Temporal Residual Jacobians
Temporal-Mapping Photography for Event Cameras
Yuhan Bao (Zhejiang University), Kaiwei Wang (Zhejiang University)
RestorationSuper ResolutionTransformerImageTime Series
🎯 What it does: Proposed a time-mapping photography method based on an event camera (EvTemMap), which gradually increases light transmittance using a variable aperture in static scenes, records the timestamp of the first positive event (IPE) for each pixel, and converts sparse event streams into dense grayscale images.
Temporally Consistent Stereo Matching
Jiaxi Zeng (Beijing Institute of Technology), Yunde Jia (Guangdong Laboratory of Machine Perception and Intelligent Computing)
Depth EstimationRecurrent Neural NetworkContrastive LearningVideo
🎯 What it does: Proposed a stereo matching method called TC-Stereo based on temporal consistency, which utilizes semi-dense disparity projection from previous frames for completion, state fusion, and bidirectional iterative refinement in dual spaces (disparity and disparity gradient).
Tendency-driven Mutual Exclusivity for Weakly Supervised Incremental Semantic Segmentation
Chongjie Si (Shanghai Jiao Tong University), Wei Shen (Shanghai Jiao Tong University)
SegmentationOptimizationDiffusion modelImage
🎯 What it does: Proposed the Teddy framework for weakly supervised incremental semantic segmentation, addressing conflicts between old and new class predictions;
Tensorial template matching for fast cross-correlation with rotations and its application for tomography
Antonio Martinez-Sanchez (École Polytechnique Fédérale de Lausanne), Harold Phelippeau (Stanford University)
Pose EstimationOptimizationPoint CloudComputed Tomography
🎯 What it does: The paper proposes an optimization method based on rotational distance to find the optimal rotation transformation under a given reference pose;
Test-time Model Adaptation for Image Reconstruction Using Self-supervised Adaptive Layers
Yutian Zhao (National University of Singapore), Hui Ji (National University of Singapore)
RestorationConvolutional Neural NetworkBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: A self-supervised test-time model adaptation method called AdaptNet is proposed for medical image reconstruction tasks (CS-MRI and SV-CT). This method inserts a lightweight AdaptBlock into a pre-trained unrolling network and adaptively updates it during inference to achieve better reconstruction quality.
Test-Time Stain Adaptation with Diffusion Models for Histopathology Image Classification
Cheng-Chang Tsai (Academia Sinica), Chun-Shien Lu (Academia Sinica)
ClassificationDomain AdaptationDiffusion modelImageBiomedical Data
🎯 What it does: Designed and validated a test-time staining adaptation method called TT-SaD based on diffusion models to address distribution shift problems caused by staining differences in pathological images.
TetraDiffusion: Tetrahedral Diffusion Models for 3D Shape Generation
Nikolai Kalischek (ETH Zurich), Konrad Schindler (University of Zurich)
GenerationConvolutional Neural NetworkDiffusion modelMesh
🎯 What it does: This paper proposes a denoising diffusion model based on tetrahedral meshes, TetraDiffusion, which can generate high-resolution and texture-mappable 3D shapes within a few seconds.
TexDreamer: Towards Zero-Shot High-Fidelity 3D Human Texture Generation
Yufei Liu (Shanghai University), Dongjin Huang (Shanghai University)
GenerationTransformerSupervised Fine-TuningDiffusion modelImageTextMultimodality
🎯 What it does: This paper proposes TexDreamer, a zero-shot, multimodal high-fidelity 3D human texture generation method that can simultaneously accept text and image inputs;
TexGen: Text-Guided 3D Texture Generation with Multi-view Sampling and Resampling
Dong Huo (University of Alberta), Yee-Hong Yang (University of Alberta)
GenerationDiffusion modelAuto EncoderTextMesh
🎯 What it does: Proposes a multi-view sampling and resampling framework (TexGen) based on pre-trained text-to-image diffusion models, for automatically generating high-quality, view-consistent textures from 3D meshes that conform to arbitrary text descriptions.