arXivSub Start free trial

CVPR 2024 Papers — Page 23

IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2716 papers

Single-to-Dual-View Adaptation for Egocentric 3D Hand Pose Estimation

Ruicong Liu (University of Tokyo), Yoichi Sato (University of Tokyo)

Pose EstimationDomain AdaptationGenerative Adversarial NetworkImage

🎯 What it does: Transfer the pre-trained monocular hand pose estimation model to any dual-camera environment for unsupervised adaptation.

Single-View Scene Point Cloud Human Grasp Generation

Yan-Kang Wang (Sun Yat-sen University), Wei-Shi Zheng (Sun Yat-sen University)

GenerationPose EstimationRobotic IntelligenceTransformerDiffusion modelPoint Cloud

🎯 What it does: This paper proposes an end-to-end framework named S2HGrasp for generating physically constrained human grasp poses from single-view scene point clouds.

SingularTrajectory: Universal Trajectory Predictor Using Diffusion Model

Inhwan Bae (Gwangju Institute of Science and Technology), Hae-Gon Jeon (Gwangju Institute of Science and Technology)

Domain AdaptationOptimizationTransformerDiffusion modelTime Series

🎯 What it does: This paper proposes SingularTrajectory, a general trajectory prediction framework that can uniformly handle five types of trajectory prediction tasks (random, deterministic, short-term, domain transfer, and few-shot).

SinSR: Diffusion-Based Image Super-Resolution in a Single Step

Yufei Wang (Nanyang Technological University), Bihan Wen (Nanyang Technological University)

RestorationSuper ResolutionKnowledge DistillationDiffusion modelImage

🎯 What it does: A single-step diffusion-based image super-resolution method SinSR is proposed, compressing the inference steps of the diffusion model into one.

SIRA: Scalable Inter-frame Relation and Association for Radar Perception

Ryoma Yataka (Mitsubishi Electric Research Laboratories), Ryuhei Takahashi (Mitsubishi Electric Research Laboratories)

Object DetectionObject TrackingAutonomous DrivingTransformerReinforcement LearningPoint Cloud

🎯 What it does: The SIRA framework is proposed, which combines scalable cross-frame association and association modules, significantly improving the target detection and tracking performance of radar perception.

SiTH: Single-view Textured Human Reconstruction with Image-Conditioned Diffusion

Hsuan- I Ho, Otmar Hilliges (ETH Zurich)

GenerationData SynthesisDiffusion modelImageMesh

🎯 What it does: Generate complete, textured 3D human meshes from a single portrait image.

Situational Awareness Matters in 3D Vision Language Reasoning

Yunze Man (University of Illinois), Yu-Xiong Wang (University of Illinois)

Pose EstimationTransformerVision Language ModelTextPoint Cloud

🎯 What it does: This paper proposes an end-to-end model SIG3D for 3D vision-language reasoning, which can estimate self-pose based on natural language descriptions and perform question answering from that perspective.

Skeleton-in-Context: Unified Skeleton Sequence Modeling with In-Context Learning

Xinshun Wang (Sun Yat-sen University), Mengyuan Liu (Peking University)

Pose EstimationTransformerPrompt EngineeringSequentialBenchmark

🎯 What it does: Proposes the Skeleton-in-Context framework, which utilizes in-context learning to unify the handling of various skeletal sequence tasks.

SketchINR: A First Look into Sketches as Implicit Neural Representations

Hmrishav Bandyopadhyay (University of Surrey), Yi-Zhe Song (University of Surrey)

GenerationRetrievalCompressionAuto EncoderImageSequential

🎯 What it does: An implicit neural representation called SketchINR is proposed for high-fidelity compression and reconstruction of sequential vector sketches.

SkillDiffuser: Interpretable Hierarchical Planning via Skill Abstractions in Diffusion-Based Task Execution

Zhixuan Liang (University of Hong Kong), Ping Luo (University of Hong Kong)

Explainability and InterpretabilityRobotic IntelligenceDiffusion modelTextSequential

🎯 What it does: An end-to-end hierarchical planning framework called SkillDiffuser is proposed, which combines interpretable skill abstraction with a conditional diffusion model to generate multi-step trajectories from natural language instructions for robots.

SkySense: A Multi-Modal Remote Sensing Foundation Model Towards Universal Interpretation for Earth Observation Imagery

Xin Guo (Ant Group), Yansheng Li (Wuhan University)

Object DetectionSegmentationTransformerContrastive LearningMultimodalityTime Series

🎯 What it does: A multi-modal remote sensing foundation model, SkySense, has been constructed to perform unified feature learning on optical, SAR, and multi-spectral time series.

SleepVST: Sleep Staging from Near-Infrared Video Signals using Pre-Trained Transformers

Jonathan F. Carter (Institute of Biomedical Engineering, University of Oxford), Lionel Tarassenko (Institute of Biomedical Engineering, University of Oxford)

ClassificationRecognitionTransformerOptical FlowVideoTime Series

🎯 What it does: This study investigates a Transformer model called SleepVST that utilizes heart rate and respiratory waveforms for sleep stage classification, and transfers it to near-infrared video signals to achieve completely contactless sleep monitoring.

SLICE: Stabilized LIME for Consistent Explanations for Image Classification

Revoti Prasad Bora (Norwegian University of Science and Technology), Kiran Raja (Norwegian University of Science and Technology)

ClassificationExplainability and InterpretabilityConvolutional Neural NetworkImage

🎯 What it does: An improved LIME method called SLICE is proposed, which enhances the consistency and accuracy of image classification explanations through feature selection and adaptive Gaussian blur.

Slice3D: Multi-Slice Occlusion-Revealing Single View 3D Reconstruction

Yizhi Wang (Simon Fraser University), Hao Zhang (Simon Fraser University)

RestorationGenerationTransformerDiffusion modelMesh

🎯 What it does: This paper proposes a single-view 3D reconstruction method called Slice3D, which predicts multi-slice images from a single view and generates 3D models based on these slices.

SlowFormer: Adversarial Attack on Compute and Energy Consumption of Efficient Vision Transformers

K L Navaneet (Harvard University), Hamed Pirsiavash (University of California, Davis)

Computational EfficiencyAdversarial AttackTransformerImage

🎯 What it does: A universal adversarial patch (SlowFormer) is designed to significantly increase the model's FLOPs and power consumption by forcing adaptive efficient visual Transformers to revert to full computation by pasting a fixed patch in the input image.

Small Scale Data-Free Knowledge Distillation

He Liu (China Telecom Cloud Technology), Anbang Yao (Intel Labs China)

ClassificationSegmentationKnowledge DistillationReinforcement LearningGenerative Adversarial NetworkImage

🎯 What it does: A small-scale data-free knowledge distillation method SSD-KD is proposed, which utilizes a teacher network to inversely synthesize a minimal amount of high-quality synthetic samples, and employs prioritized sampling to accelerate training during the distillation process.

Small Steps and Level Sets: Fitting Neural Surface Models with Point Guidance

Chamin Hewa Koneputugodage (Australian National University), Stephen Gould (Australian National University)

OptimizationPoint Cloud

🎯 What it does: A point-guided neural SDF optimization method is proposed for reconstructing surfaces from raw point clouds without normals.

Smart Help: Strategic Opponent Modeling for Proactive and Adaptive Robot Assistance in Households

Zhihao Cao (Beijing Institute for General Artificial Intelligence), Lifeng Fan (University of California)

Robotic IntelligenceTransformerReinforcement LearningAgentic AIMultimodality

🎯 What it does: Designed and implemented the Smart Help challenge, constructing a multi-agent home task environment based on AI2-THOR, and proposed a help robot model that can actively adapt to user capabilities and goals.

SmartEdit: Exploring Complex Instruction-based Image Editing with Multimodal Large Language Models

Yuzhou Huang (Chinese University of Hong Kong), Ying Shan (Tencent)

GenerationData SynthesisTransformerLarge Language ModelDiffusion modelImageMultimodality

🎯 What it does: A directive image editing framework named SmartEdit is proposed, which can understand and execute complex instructions that include attributes such as position, size, color, mirror relationships, and require world knowledge reasoning.

SmartMask: Context Aware High-Fidelity Mask Generation for Fine-grained Object Insertion and Layout Control

Jaskirat Singh (Adobe Research), Liang Zheng (Australian National University)

SegmentationGenerationDiffusion modelImage

🎯 What it does: We propose SmartMask, which enables novice users to automatically generate high-fidelity object masks through semantic descriptions and optional coarse guidance (such as boxes, doodles, or no guidance), and combines with ControlNet-Inpaint to achieve fine-grained object insertion and layout control; it can also generate complex scene layouts through multi-step iterative planning.

SmartRefine: A Scenario-Adaptive Refinement Framework for Efficient Motion Prediction

Yang Zhou (SenseTime Research), Yu Liu (Shanghai Artificial Intelligence Laboratory)

Autonomous DrivingComputational EfficiencyTime Series

🎯 What it does: A scene-adaptive refinement framework called SmartRefine is proposed, which enhances the prediction accuracy of motion prediction models by utilizing adaptive anchor points, retrieval radius, context encoding, and quality scoring without significantly increasing computational load.

Smooth Diffusion: Crafting Smooth Latent Spaces in Diffusion Models

Jiayi Guo (Tsinghua University), Humphrey Shi (Georgia Tech)

GenerationDiffusion modelImage

🎯 What it does: This paper proposes the Smooth Diffusion model, which enhances the performance of downstream tasks such as image interpolation, inversion, and editing by incorporating Step-wise Variation Regularization during training to make the latent space of the diffusion model smoother.

SnAG: Scalable and Accurate Video Grounding

Fangzhou Mu (University of Wisconsin-Madison), Yin Li (University of Wisconsin-Madison)

RecognitionRetrievalComputational EfficiencyTransformerVideoText

🎯 What it does: A spatiotemporal video grounding framework called SnAG is proposed for multi-query long videos, aiming to address the scalability issues of traditional methods in scenarios involving long videos and numerous queries.

Snap Video: Scaled Spatiotemporal Transformers for Text-to-Video Synthesis

Willi Menapace (Snap Inc.), Sergey Tulyakov (Snap Inc.)

GenerationData SynthesisTransformerDiffusion modelVideoText

🎯 What it does: This paper presents Snap Video, a high-resolution model for text-to-video generation, and implements a complete inference process from noise to video.

Snapshot Lidar: Fourier Embedding of Amplitude and Phase for Single-Image Depth Reconstruction

Sarah Friday (Dartmouth), Adithya Pediredla (University of California)

Depth EstimationImage

🎯 What it does: This paper proposes a snapshot Lidar system based on Continuous Wave Time of Flight (CW-ToF) cameras, utilizing a rolling shutter to achieve spatial linear variation of the phase, thereby obtaining both light intensity (amplitude) and depth (phase) in a single capture.

SNED: Superposition Network Architecture Search for Efficient Video Diffusion Model

Zhengang Li (Northeastern University), Yanzhi Wang (Northeastern University)

GenerationData SynthesisComputational EfficiencyNeural Architecture SearchDiffusion modelVideoText

🎯 What it does: Construct and train a video diffusion super network (SNED) that can dynamically switch model scale and resolution.

SNI-SLAM: Semantic Neural Implicit SLAM

Siting Zhu (Shanghai Jiao Tong University), Hesheng Wang (Shanghai Jiao Tong University)

Neural Radiance FieldSimultaneous Localization and MappingPoint Cloud

🎯 What it does: SNI-SLAM is a real-time semantic SLAM system based on NeRF that can simultaneously perform semantic mapping, surface reconstruction, and camera tracking.

SNIDA: Unlocking Few-Shot Object Detection with Non-linear Semantic Decoupling Augmentation

Yanjie Wang (Huazhong University of Science and Technology), Jiahuan Zhou (Peking University)

Object DetectionAuto EncoderImage

🎯 What it does: A semantic-guided nonlinear instance-level data augmentation method is proposed, which enhances data diversity and semantic perception for few-shot object detection through foreground/background decoupling and instance reconstruction.

SNIFFER: Multimodal Large Language Model for Explainable Out-of-Context Misinformation Detection

Peng Qi (National University of Singapore), Mong Li Lee (National University of Singapore)

Explainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality

🎯 What it does: Designed and trained a multimodal large language model named SNIFFER to detect and explain 'out-of-context' misinformation between images and text.

SOAC: Spatio-Temporal Overlap-Aware Multi-Sensor Calibration using Neural Radiance Fields

Quentin Herau (Huawei), Cédric Demonceaux (Bourgogne)

Autonomous DrivingOptimizationNeural Radiance FieldPoint Cloud

🎯 What it does: This paper proposes SOAC, a target-independent, self-supervised spatiotemporal multi-sensor calibration method using NeRF corresponding to multiple cameras.

SocialCircle: Learning the Angle-based Social Interaction Representation for Pedestrian Trajectory Prediction

Conghao Wong (Huazhong University of Science and Technology), Xinge You (Huazhong University of Science and Technology)

Representation LearningGraph Neural NetworkTransformerTime SeriesSequential

🎯 What it does: This paper proposes an angle-based social circle (SocialCircle) representation to capture social interactions in pedestrian trajectory prediction and integrates it into various baseline models.

SocialCounterfactuals: Probing and Mitigating Intersectional Social Biases in Vision-Language Models with Counterfactual Examples

Phillip Howard (Intel Labs), Vasudev Lal (Intel Labs)

GenerationData SynthesisRetrievalTransformerVision Language ModelDiffusion modelImageTextMultimodality

🎯 What it does: This paper automatically generates adversarial image-text pairs through a text-to-image diffusion model, constructing a large-scale counterfactual dataset of cross-social attributes (such as gender, race, and physical characteristics) called SocialCounterfactuals, aimed at detecting and mitigating cross-social biases in visual-language models.

SODA: Bottleneck Diffusion Models for Representation Learning

Drew A. Hudson (Google DeepMind), Alexander Lerchner (Google DeepMind)

GenerationRepresentation LearningConvolutional Neural NetworkDiffusion modelContrastive LearningImage

🎯 What it does: This paper proposes a self-supervised diffusion model called SODA, which combines an image encoder and a denoising decoder, utilizing novel view synthesis as a self-supervised objective to learn compact latent representations that can be used for downstream classification, reconstruction, and new view generation.

Soften to Defend: Towards Adversarial Robustness via Self-Guided Label Refinement

Zhuorong Li (Hangzhou City University), Sixian Chan (Zhejiang University of Technology)

OptimizationAdversarial AttackConvolutional Neural NetworkSupervised Fine-TuningImage

🎯 What it does: This paper proposes a Self-Guided Label Refinement (SGLR) method, which utilizes the model's own prediction distribution to generate soft labels and dynamically calibrates adversarial training to alleviate the problem of robust overfitting.

SOK-Bench: A Situated Video Reasoning Benchmark with Aligned Open-World Knowledge

Andong Wang (University of Hong Kong), Chuang Gan (University of Massachusetts Amherst)

GenerationRetrievalRecommendation SystemTransformerLarge Language ModelPrompt EngineeringVision Language ModelVideoMultimodalityBenchmark

🎯 What it does: Created SOK-Bench, a common sense reasoning benchmark containing 44K question-answer pairs and 10K video scenes, and automatically generated situational knowledge graphs, general knowledge graphs, and situational common sense graphs, supporting multi-step reasoning;

Solving Masked Jigsaw Puzzles with Diffusion Vision Transformers

Jinyang Liu (Northeastern University), Octavia Camps (Northeastern University)

GenerationData SynthesisTransformerDiffusion modelImageVideo

🎯 What it does: A JPDVT method based on diffusion visual Transformer is proposed, which can simultaneously solve the image and video jigsaw puzzle (including missing fragments) problem.

Solving the Catastrophic Forgetting Problem in Generalized Category Discovery

Xinzi Cao (Sun Yat-sen University), Yonghong Tian (Peking University)

ClassificationRepresentation LearningTransformerContrastive LearningImage

🎯 What it does: The LegoGCD framework is proposed, combining SimGCD with local entropy regularization (LER) and dual-view KL constraints (DKL) to alleviate the catastrophic forgetting problem of known categories in Generalized Category Discovery.

SonicVisionLM: Playing Sound with Vision Language Models

Zhifeng Xie (Shanghai University), Mengtian Li (Shanghai University)

GenerationData SynthesisTransformerVision Language ModelDiffusion modelVideoTextMultimodalityAudio

🎯 What it does: Proposes the SonicVisionLM framework, which first identifies events in videos and provides textual descriptions using a visual language model, and then generates visually synchronized sound through a time-controllable text-to-audio diffusion model, allowing users to edit off-screen sound effects.

SoundingActions: Learning How Actions Sound from Narrated Egocentric Videos

Changan Chen (University of Texas at Austin), Kristen Grauman (University of Texas at Austin)

RetrievalRepresentation LearningTransformerContrastive LearningVideoTextMultimodalityAudio

🎯 What it does: This paper proposes a self-supervised multimodal embedding (MC3) that learns the sounds produced by human actions through narrative first-person videos.

Source-Free Domain Adaptation with Frozen Multimodal Foundation Model

Song Tang (University of Shanghai for Science and Technology), Xiatian Zhu (University of Surrey)

Domain AdaptationKnowledge DistillationTransformerPrompt EngineeringVision Language ModelContrastive LearningMultimodality

🎯 What it does: This paper studies the source-free unsupervised domain adaptation task and proposes the DIFO framework, which first learns task-specific prompts for CLIP and then distills its knowledge to the target model, achieving domain transfer without source data.

Space-Time Diffusion Features for Zero-Shot Text-Driven Motion Transfer

Danah Yatim (Weizmann Institute of Science), Tali Dekel (Weizmann Institute of Science)

GenerationData SynthesisDiffusion modelVideoText

🎯 What it does: A zero-shot text-driven motion transfer method is proposed, utilizing a pre-trained text-to-video diffusion model to transfer motion to target objects of different forms based on text prompts while preserving the original video's motion and scene layout.

Spacetime Gaussian Feature Splatting for Real-Time Dynamic View Synthesis

Zhan Li (OPPO US Research Center), Yi Xu (Portland State University)

GenerationData SynthesisGaussian SplattingVideo

🎯 What it does: Proposes a Spacetime Gaussian Feature Splatting method for real-time high-resolution dynamic view synthesis of multi-view videos.

SPAD: Spatially Aware Multi-View Diffusers

Yash Kant (University of Toronto), Igor Gilitschenski (Snap Research)

GenerationData SynthesisTransformerDiffusion modelImage

🎯 What it does: Train a multi-view diffusion model SPAD that can generate high-quality images consistent across multiple views based on text or a single image.

Spanning Training Progress: Temporal Dual-Depth Scoring (TDDS) for Enhanced Dataset Pruning

Xin Zhang (XiDian University), Joey Tianyi Zhou (Centre for Frontier AI Research, Agency for Science, Technology and Research)

ClassificationData-Centric LearningConvolutional Neural NetworkImage

🎯 What it does: A dataset pruning method based on Temporal Dual-Depth Scoring (TDDS) is proposed;

Sparse Global Matching for Video Frame Interpolation with Large Motion

Chunxu Liu (Nanjing University), Limin Wang (Nanjing University)

Image TranslationRestorationConvolutional Neural NetworkTransformerOptical FlowVideo

🎯 What it does: A Sparse Global Matching pipeline is proposed for video frame interpolation (VFI) in large motion scenes.

Sparse Semi-DETR: Sparse Learnable Queries for Semi-Supervised Object Detection

Tahira Shehzadi (DFKI), Muhammad Zeshan Afzal (DFKI)

Object DetectionTransformerImage

🎯 What it does: This paper proposes an end-to-end semi-supervised object detection framework named Sparse Semi-DETR.

Sparse Views Near Light: A Practical Paradigm for Uncalibrated Point-light Photometric Stereo

Mohammed Brahimi (Technische Universität München), Daniel Cremers (Technische Universität München)

RestorationDepth EstimationPoint CloudMesh

🎯 What it does: A multi-view non-calibrated point light source photometric stereo reconstruction framework is proposed, which combines voxel rendering with Disney BRDF, achieving high-precision 3D reconstruction under sparse views and typical indoor lighting.

SparseOcc: Rethinking Sparse Latent Representation for Vision-Based Semantic Occupancy Prediction

Pin Tang (Shanghai Jiao Tong University), Chao Ma (Huawei)

SegmentationAutonomous DrivingComputational EfficiencyTransformerPoint Cloud

🎯 What it does: This paper proposes a completely sparse 3D semantic occupancy prediction framework called SparseOcc, exploring the feasibility of using pure sparse representations in vision-driven autonomous driving perception.

Spatial-Aware Regression for Keypoint Localization

Dongkai Wang (Peking University), Shiliang Zhang (Peking University)

Pose EstimationConvolutional Neural NetworkTransformerImage

🎯 What it does: Proposes Spatial-Aware Regression (SAR), which incorporates spatial location information into regression-based keypoint localization to achieve efficient and robust keypoint detection.

SpatialTracker: Tracking Any 2D Pixels in 3D Space

Yuxi Xiao (Zhejiang University), Xiaowei Zhou (Zhejiang University)

Object TrackingDepth EstimationTransformerOptical FlowVideoPoint Cloud

🎯 What it does: We propose SpatialTracker, which achieves long-term dense pixel tracking by projecting 2D pixels into 3D space and iteratively updating using triplane representation and Transformer.

SpatialVLM: Endowing Vision-Language Models with Spatial Reasoning Capabilities

Boyuan Chen (Google DeepMind), Fei Xia (Google Research)

Object DetectionSegmentationData SynthesisDepth EstimationRobotic IntelligenceTransformerLarge Language ModelVision Language ModelImageTextPoint Cloud

🎯 What it does: This paper proposes SpatialVLM, which significantly enhances spatial reasoning capabilities by automatically generating a massive amount of 3D spatial reasoning VQA data and pre-training the VLM.

Spatio-Temporal Turbulence Mitigation: A Translational Perspective

Xingguang Zhang (Purdue University), Stanley H. Chan (Purdue University)

RestorationConvolutional Neural NetworkImageVideo

🎯 What it does: A deep multi-frame atmospheric turbulence suppression network (DATUM) is proposed, which achieves end-to-end recovery by combining classical multi-frame turbulence suppression steps.

SPECAT: SPatial-spEctral Cumulative-Attention Transformer for High-Resolution Hyperspectral Image Reconstruction

Zhiyang Yao (Tsinghua University), Lu Fang (Tsinghua University)

RestorationTransformerImage

🎯 What it does: A Transformer model specifically designed for high-resolution hyperspectral image reconstruction, called SPECAT, has been developed and implemented.

SpecNeRF: Gaussian Directional Encoding for Specular Reflections

Li Ma (Hong Kong University of Science and Technology), Christian Richardt (Meta Reality Labs)

GenerationData SynthesisNeural Radiance FieldImage

🎯 What it does: This paper proposes a NeRF model based on three-dimensional Gaussian directional encoding, which can more accurately simulate view-dependent specular reflections under indoor near-field lighting.

Spectral and Polarization Vision: Spectro-polarimetric Real-world Dataset

Yujin Jeon (POSTECH), Seung-Hwan Baek (POSTECH)

RestorationCompressionAuto EncoderImage

🎯 What it does: This paper constructs two large-scale real-world full-wavelength full-Stokes (including linear and circular polarization) image datasets and systematically analyzes their statistical properties, compressed representations, and performance in tasks such as shape recovery and environmental perception.

Spectral Meets Spatial: Harmonising 3D Shape Matching and Interpolation

Dongliang Cao (University of Bonn), Florian Bernard (University of Bonn)

Diffusion modelPoint CloudMeshBiomedical Data

🎯 What it does: A unified unsupervised framework is proposed that can simultaneously learn the spectral mapping and spatial mapping of 3D shapes, achieving precise correspondence and realistic interpolation.

Spectrum AUC Difference (SAUCD): Human-aligned 3D Shape Evaluation

Tianyu Luan (State University of New York), Junsong Yuan (State University of New York)

Mesh

🎯 What it does: A new three-dimensional mesh shape difference metric SAUCD has been designed and evaluated.

Specularity Factorization for Low-Light Enhancement

Saurabh Saini (International Institute of Information Technology Hyderabad), P J Narayanan (International Institute of Information Technology Hyderabad)

RestorationConvolutional Neural NetworkSupervised Fine-TuningImage

🎯 What it does: Proposes a Recursive Brightness Factorization (RSFNet) and applies it to zero-reference low-light enhancement, image relighting, and multi-task dehazing/de-raining/de-blurring.

Spherical Mask: Coarse-to-Fine 3D Point Cloud Instance Segmentation with Spherical Representation

Sangyun Shin (University of Oxford), Niki Trigoni (University of Oxford)

Object DetectionSegmentationConvolutional Neural NetworkPoint Cloud

🎯 What it does: A coarse-to-fine 3D instance segmentation framework called Spherical Mask is proposed, which utilizes spherical polygons for coarse detection and achieves refinement through spherical point migration.

SpiderMatch: 3D Shape Matching with Global Optimality and Geometric Consistency

Paul Roetzer (University of Bonn), Florian Bernard (University of Bonn)

OptimizationMesh

🎯 What it does: A 3D shape matching method based on SpiderCurve is proposed, using integer linear programming to solve for globally optimal and geometrically consistent correspondences.

SPIDeRS: Structured Polarization for Invisible Depth and Reflectance Sensing

Tomoki Ichikawa (Kyoto University), Ko Nishino (Kyoto University)

Depth EstimationOptimizationImage

🎯 What it does: Using variable angle linear polarization projection to achieve invisible 3D depth and reflectance perception

Spike-guided Motion Deblurring with Unknown Modal Spatiotemporal Alignment

Jiyuan Zhang (Peking University), Tiejun Huang (Peking University)

RestorationConvolutional Neural NetworkSpiking Neural NetworkOptical FlowImageVideoMultimodality

🎯 What it does: Designed and implemented a three-stage spike camera guided motion deblurring network UaSDN, achieving high-quality deblurring under the condition of unaligned RGB and spike data.

SpikeNeRF: Learning Neural Radiance Fields from Continuous Spike Stream

Lin Zhu (Beijing Institute of Technology), Hua Huang (Beijing Normal University)

GenerationData SynthesisPose EstimationSpiking Neural NetworkNeural Radiance FieldImageVideo

🎯 What it does: A NeRF model based on continuous pulse streams from Spike cameras, called SpikeNeRF, is proposed, which can learn dense 3D scene representations and generate high-quality novel view images using only pulse data.

SpikingResformer: Bridging ResNet and Vision Transformer in Spiking Neural Networks

Xinyu Shi (Peking University), Zhaofei Yu (Peking University)

Spiking Neural NetworkTransformerImage

🎯 What it does: A dual-pulse self-attention mechanism (DSSA) is proposed, and based on it, a multi-stage ResNet-Transformer structure called SpikingResformer is designed, constructing a complete and directly trainable spiking neural network.

Spin-UP: Spin Light for Natural Light Uncalibrated Photometric Stereo

Zongrui Li (Nanyang Technological University), Xudong Jiang (Nanyang Technological University)

RestorationData SynthesisNeural Radiance FieldImage

🎯 What it does: An unsupervised natural light non-calibrated photogrammetry method called Spin-UP is proposed, which utilizes a rotating platform to achieve uniform ambient light and recovers surface normals, ambient light, and isotropic reflectance through inverse rendering.

SPIN: Simultaneous Perception Interaction and Navigation

Shagun Uppal (Carnegie Mellon University), Deepak Pathak (Carnegie Mellon University)

Robotic IntelligenceReinforcement LearningPoint Cloud

🎯 What it does: This paper proposes an end-to-end mobile manipulation framework called SPIN, which trains robots to collaboratively complete perception, manipulation, and navigation tasks in an unstructured environment without maps, relying solely on active vision.

SplaTAM: Splat Track & Map 3D Gaussians for Dense RGB-D SLAM

Nikhil Keetha (Carnegie Mellon University), Jonathon Luiten (Carnegie Mellon University)

Pose EstimationOptimizationGaussian SplattingSimultaneous Localization and MappingPoint Cloud

🎯 What it does: Using 3D Gaussian point clouds (Gaussian Splatting) as an explicit voxel representation, we optimize the camera pose of RGB-D frames in real-time and construct high-fidelity scene maps.

Splatter Image: Ultra-Fast Single-View 3D Reconstruction

Stanislaw Szymanowicz (Visual Geometry Group University of Oxford), Andrea Vedaldi (Visual Geometry Group University of Oxford)

GenerationData SynthesisComputational EfficiencyGaussian SplattingImage

🎯 What it does: Proposes the Splatter Image method, which maps single-view images to a 3D Gaussian mixture representation, achieving efficient single-view/multi-view 3D reconstruction.

SplattingAvatar: Realistic Real-Time Human Avatars with Mesh-Embedded Gaussian Splatting

Zhijing Shao (Hong Kong University of Science and Technology), Zeyu Wang (Hong Kong University of Science and Technology)

GenerationOptimizationComputational EfficiencyGaussian SplattingVideoMesh

🎯 What it does: Using monocular video and a parametric mesh (SMPL/FLAME) for training, a 3D Gaussian distribution is embedded onto the mesh to construct digital models of human heads and full bodies that can capture high-frequency details and render efficiently.

Split to Merge: Unifying Separated Modalities for Unsupervised Domain Adaptation

Xinyao Li (University of Electronic Science and Technology of China), Jingjing Li (University of Electronic Science and Technology of China)

Domain AdaptationKnowledge DistillationContrastive LearningImage

🎯 What it does: This paper proposes the UniMoS framework, which separates the CLIP visual features into language-associated (LAC) and vision-associated (VAC) parts through a modality separation network, and achieves unsupervised domain adaptation using modality fusion training and a modality discriminator.

SPOC: Imitating Shortest Paths in Simulation Enables Effective Navigation and Manipulation in the Real World

Kiana Ehsani (Allen Institute for AI), Aniruddha Kembhavi (University of Washington)

Object DetectionRobotic IntelligenceTransformerReinforcement LearningAgentic AIImageText

🎯 What it does: Using a large-scale procedurally generated simulation environment, trajectories generated by a shortest path expert were used to train a behavior cloning model (SPOC), achieving navigation and manipulation under RGB-only perception, and enabling direct transfer to real houses without adaptation.

SportsHHI: A Dataset for Human-Human Interaction Detection in Sports Videos

Tao Wu (Shanghai AI Lab), Limin Wang (Nanjing University)

RecognitionObject DetectionTransformerVideo

🎯 What it does: The SportsHHI dataset is proposed, and based on this, a human-human interaction detection task is conducted.

SportsSloMo: A New Benchmark and Baselines for Human-centric Video Frame Interpolation

Jiaben Chen (University of California San Diego), Huaizu Jiang (Northeastern University)

Image TranslationRestorationSegmentationPose EstimationTransformerOptical FlowVideoBenchmark

🎯 What it does: This study constructed a high-resolution slow-motion sports video benchmark dataset named SportsSloMo and retrained and evaluated various video frame interpolation methods on it.

SPOT: Self-Training with Patch-Order Permutation for Object-Centric Learning with Autoregressive Transformers

Ioannis Kakogeorgiou (National Technical University of Athens), Nikos Komodakis (University of Crete)

Object DetectionSegmentationTransformerAuto EncoderImage

🎯 What it does: This paper enhances the segmentation performance of unsupervised object center learning by incorporating self-training and sequence permutation techniques into the slot-based autoencoder.

SPU-PMD: Self-Supervised Point Cloud Upsampling via Progressive Mesh Deformation

Yanzhe Liu (Dalian Maritime University), Xuehou Tan (Tokai University)

RestorationGenerationData SynthesisSuper ResolutionTransformerPoint CloudMesh

🎯 What it does: A point cloud upsampling method based on self-supervised learning utilizes a mesh interpolation and recursive feature aggregation deformation module to achieve the transformation from sparse point clouds to high-resolution uniformly distributed point clouds.

SRTube: Video-Language Pre-Training with Action-Centric Video Tube Features and Semantic Role Labeling

Ju-Hee Lee (Ewha Womans University), Je-Won Kang (Ewha Womans University)

RetrievalTransformerVision-Language-Action ModelVideoTextMultimodality

🎯 What it does: This paper proposes a novel video-language pre-training framework called SRTube, which significantly enhances cross-modal alignment by utilizing action-centered video tube features based on object trajectories and language features based on Semantic Role Labeling (SRL).

SSR-Encoder: Encoding Selective Subject Representation for Subject-Driven Generation

Yuxuan Zhang (Shanghai Jiao Tong University), Zhongliang Jing (Shanghai Jiao Tong University)

GenerationData SynthesisTransformerDiffusion modelContrastive LearningImageText

🎯 What it does: This paper proposes the SSR-Encoder, a selective subject representation encoder for zero-training-time fine-tuning, capable of extracting and aligning target subjects from single or multiple reference images, thereby driving any custom diffusion model for zero-shot, single/multi-subject generation.

Stable Neighbor Denoising for Source-free Domain Adaptive Segmentation

Dong Zhao (Xidian University), Zhun Zhong (University of Nottingham)

SegmentationDomain AdaptationKnowledge DistillationImage

🎯 What it does: Proposes the Stable Neighbor Denoising method for source-agnostic domain adaptive semantic segmentation by denoising unstable samples.

StableVITON: Learning Semantic Correspondence with Latent Diffusion Model for Virtual Try-On

Jeongho Kim (KAIST), Jaegul Choo (KAIST)

Image TranslationGenerationDiffusion modelImage

🎯 What it does: This paper proposes StableVITON, which utilizes the pre-trained Stable Diffusion to learn the semantic correspondence between clothing and the human body in the latent space, achieving high-quality virtual try-on.

State Space Models for Event Cameras

Nikola Zubic (University of Zurich), Davide Scaramuzza (University of Zurich)

Object DetectionAutonomous DrivingComputational EfficiencyRecurrent Neural NetworkTransformerSupervised Fine-TuningImageVideo

🎯 What it does: A ViT network based on the state space model (SSM) is designed for the target detection task of event cameras, allowing the model to maintain performance at different sampling frequencies and accelerate training.

Stationary Representations: Optimally Approximating Compatibility and Implications for Improved Model Replacements

Niccolò Biondi (University of Florence), Alberto Del Bimbo (University of Florence)

Representation LearningConvolutional Neural NetworkContrastive LearningImage

🎯 What it does: This study investigates the use of a fixed d-Simplex classifier to learn static feature representations for achieving model compatibility, and proposes a High-Order Compatibility (HOC) loss.

Steerers: A Framework for Rotation Equivariant Keypoint Descriptors

Georg Bökman (Chalmers University of Technology), Fredrik Kahl (Linkoping University)

RecognitionObject DetectionRetrievalOptimizationContrastive LearningImage

🎯 What it does: This study investigates the implementation of rotation invariance for image keypoint descriptors through learning linear transformations (steerers) in the descriptor space, enabling matching under large-angle rotations while maintaining discriminability.

Steganographic Passport: An Owner and User Verifiable Credential for Deep Model IP Protection Without Retraining

Qi Cui (Nanyang Technological University), Chip-Hong Chang (Nanyang Technological University)

Safty and PrivacyGenerative Adversarial NetworkImage

🎯 What it does: A new passport mechanism called steganographic passport is proposed, aimed at protecting the intellectual property of deep models, allowing ownership and usage rights to be verified without retraining the model.

StegoGAN: Leveraging Steganography for Non-Bijective Image-to-Image Translation

Sidi Wu (ETH Zurich), Loic Landrieu (Univ Gustave Eiffel)

Image TranslationGenerationGenerative Adversarial NetworkImageBiomedical DataMagnetic Resonance Imaging

🎯 What it does: A StegoGAN model based on CycleGAN is proposed, which utilizes steganography techniques to explicitly separate matching and non-matching information in the feature space, thereby suppressing the generation of pseudo-features in non-injective image translation.

Step Differences in Instructional Video

Tushar Nagarajan (Meta), Lorenzo Torresani (Meta)

TransformerLarge Language ModelSupervised Fine-TuningVideoTextBenchmark

🎯 What it does: The StepDiff model is proposed, which can compare two teaching videos with the same key steps and automatically describe fine-grained differences.

StraightPCF: Straight Point Cloud Filtering

Dasith de Silva Edirimuni (Deakin University), Hongdong Li (Australian National University)

RestorationOptimizationGraph Neural NetworkPoint CloudOrdinary Differential Equation

🎯 What it does: A lightweight point cloud denoising method called StraightPCF is proposed, which directly transfers noise points to the clean surface along straight trajectories using a constant velocity vector field and distance scalar;

Stratified Avatar Generation from Sparse Observations

Han Feng (Wuhan University), Huijuan Xu (Pennsylvania State University)

GenerationPose EstimationTransformerDiffusion modelAuto EncoderMesh

🎯 What it does: A hierarchical full-body avatar generation (SAGE) framework is proposed to address sparse observations with only the head and hands, first reconstructing the upper body and then predicting the lower body based on the upper body, thus obtaining a complete 3D full-body pose.

Streaming Dense Video Captioning

Xingyi Zhou (Google), Cordelia Schmid (Google)

GenerationTransformerVision Language ModelVideoTextMultimodality

🎯 What it does: A streaming dense video subtitle model is proposed, capable of generating time-aligned text descriptions in real-time for long videos.

StreamingFlow: Streaming Occupancy Forecasting with Asynchronous Multi-modal Data Streams via Neural Ordinary Differential Equation

Yining Shi (Tsinghua University), Diange Yang (Tsinghua University)

Autonomous DrivingRecurrent Neural NetworkMultimodalityPoint CloudTime SeriesOrdinary Differential Equation

🎯 What it does: Proposes the StreamingFlow framework, which utilizes asynchronous multi-sensor data streams for real-time BEV occupancy prediction and supports generating occupancy maps at any future time.

StrokeFaceNeRF: Stroke-based Facial Appearance Editing in Neural Radiance Field

Xiao-Juan Li (Chinese Academy of Sciences), Feng-Lin Liu (Chinese Academy of Sciences)

GenerationData SynthesisTransformerNeural Radiance FieldGenerative Adversarial NetworkImage

🎯 What it does: A 3D face NeRF editing framework based on color strokes, StrokeFaceNeRF, has been developed, allowing users to draw strokes from any perspective and achieve local appearance modifications.

Strong Transferable Adversarial Attacks via Ensembled Asymptotically Normal Distribution Learning

Zhengwei Fang (Beijing Jiaotong University), Liping Jing (Beijing Jiaotong University)

Adversarial AttackConvolutional Neural NetworkImage

🎯 What it does: Two transferable adversarial attack methods based on learning perturbation distribution, ANDA and MultiANDA, are proposed.

Stronger Fewer & Superior: Harnessing Vision Foundation Models for Domain Generalized Semantic Segmentation

Zhixiang Wei (University of Science and Technology of China), Jinjin Zheng (University of Science and Technology of China)

SegmentationDomain AdaptationConvolutional Neural NetworkSupervised Fine-TuningImage

🎯 What it does: This paper explores how to apply Visual Foundation Models (VFMs) to the Domain Generalization Semantic Segmentation (DGSS) task and proposes a new parameter-efficient fine-tuning method called Rein.

Structure Matters: Tackling the Semantic Discrepancy in Diffusion Models for Image Inpainting

Haipeng Liu (Hefei University of Technology), Yong Rui (Lenovo Research)

RestorationDiffusion modelScore-based ModelImage

🎯 What it does: A structure-guided diffusion model (StrDiffusion) is proposed for image inpainting.

Structure-Aware Sparse-View X-ray 3D Reconstruction

Yuanhao Cai (Johns Hopkins University), Angtian Wang (Johns Hopkins University)

RestorationTransformerNeural Radiance FieldImageComputed Tomography

🎯 What it does: A sparse perspective X-ray 3D reconstruction framework based on Transformer, SAX-NeRF, is proposed, utilizing Lineformer to capture the internal structure of X-rays.

Structure-Guided Adversarial Training of Diffusion Models

Ling Yang (Peking University), Bin Cui (Peking University)

GenerationDomain AdaptationDiffusion modelGenerative Adversarial NetworkImage

🎯 What it does: A structure-guided adversarial training framework SADM is proposed to enhance the training effectiveness of diffusion models; by learning the manifold structure among samples within each mini-batch and introducing a structural discriminator for adversarial optimization, it further approximates the real data distribution.

Structured Gradient-based Interpretations via Norm-Regularized Adversarial Training

Shizhan Gong (Chinese University of Hong Kong), Farzan Farnia (Chinese University of Hong Kong)

Explainability and InterpretabilityAdversarial AttackGenerative Adversarial NetworkImage

🎯 What it does: A norm-regularization-based adversarial training framework is proposed, which makes the model's simple gradient explanation maps sparser, more structured, and maintains the fidelity of the original explanations.

Structured Model Probing: Empowering Efficient Transfer Learning by Structured Regularization

Zhi-Fan Wu (Alibaba Group), Rong Jin (Alibaba Group)

ClassificationDomain AdaptationComputational EfficiencyTransformerSupervised Fine-TuningImage

🎯 What it does: A Structured Model Probing (SMP) method is proposed, which performs efficient transfer learning on fixed pre-trained models using structured regularization. It maintains the simplicity of linear probing in easy domain tasks while automatically increasing non-linear complexity in difficult domain tasks.

Style Aligned Image Generation via Shared Attention

Amir Hertz (Google Research), Daniel Cohen-Or (Google Research)

GenerationData SynthesisTransformerDiffusion modelImageText

🎯 What it does: Proposes the StyleAligned method, which achieves multi-image style-consistent generation in text-to-image diffusion models through shared attention.

Style Blind Domain Generalized Semantic Segmentation via Covariance Alignment and Semantic Consistence Contrastive Learning

Woo-Jin Ahn (Korea University), Myo-Taeg Lim (Korea University)

SegmentationDomain AdaptationContrastive LearningImage

🎯 What it does: This paper proposes BlindNet, a method for domain generalization semantic segmentation achieved through covariance alignment and semantic consistency contrastive learning.

Style Injection in Diffusion: A Training-free Approach for Adapting Large-scale Diffusion Models for Style Transfer

Jiwoo Chung (Sungkyunkwan University), Jae-Pil Heo (Sungkyunkwan University)

Image TranslationGenerationDiffusion modelImage

🎯 What it does: Without the need for any training or fine-tuning, high-quality artistic style transfer is achieved using the pre-trained Stable Diffusion large model by replacing the keys and values in the self-attention layer and incorporating query retention, attention temperature scaling, and initial latent variable AdaIN.