CVPR 2023 Papers — Page 20
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2353 papers
Semantic-Promoted Debiasing and Background Disambiguation for Zero-Shot Instance Segmentation
Shuting He (Zhejiang University), Wei Jiang (Zhejiang University)
Object DetectionSegmentationTransformerImage
🎯 What it does: The D Zero 2 method is proposed to address the bias and background confusion issues in zero-shot instance segmentation, constructing a semantically enhanced debiasing training and image adaptive background demixing model.
Semi-DETR: Semi-Supervised Object Detection With Detection Transformers
Jiacheng Zhang (Sun Yat-sen University), Guanbin Li (Sun Yat-sen University)
Object DetectionTransformerImage
🎯 What it does: A semi-supervised object detection framework called Semi-DETR based on DETR is proposed and implemented, addressing the issues of training inefficiency and query uncertainty caused by one-to-one matching in semi-supervised scenarios.
Semi-Supervised 2D Human Pose Estimation Driven by Position Inconsistency Pseudo Label Correction Module
Linzhi Huang (Beijing University of Posts and Telecommunications), Jieping Ye (Beike)
Pose EstimationImage
🎯 What it does: A semi-supervised 2D human pose estimation framework (SSPCM) based on a position inconsistency pseudo-label correction module is proposed, combined with pseudo keypoint-aware Cut-Occlude enhancement (SSCO).
Semi-Supervised Domain Adaptation With Source Label Adaptation
Yu-Chu Yu (National Taiwan University), Hsuan-Tien Lin (National Taiwan University)
Domain AdaptationImage
🎯 What it does: This paper proposes a Source Label Adaptation (SLA) framework for Semi-Supervised Domain Adaptation (SSDA), which dynamically corrects source domain labels to better match the target domain feature space, thereby improving classification performance.
Semi-Supervised Hand Appearance Recovery via Structure Disentanglement and Dual Adversarial Discrimination
Zimeng Zhao (Southeast University), Yangang Wang (Southeast University)
Image TranslationRestorationSegmentationGenerationTransformerGenerative Adversarial NetworkImage
🎯 What it does: By first using ViT-Sketcher to separate the bare hand structure from hand images with markers or occlusions, and then utilizing the Dual Adversarial Discriminator (DAD) framework to wrap the original appearance onto this structure under semi-supervised conditions, the bare hand appearance recovery of labeled MoCap data is achieved.
Semi-Supervised Learning Made Simple With Self-Supervised Clustering
Enrico Fini (University of Trento), Elisa Ricci (Fondazione Bruno Kessler)
ClassificationRepresentation LearningContrastive LearningImage
🎯 What it does: A simple and effective semi-supervised learning framework is proposed, which aligns self-supervised clustering methods (such as SwAV and DINO) with supervised objectives using labeled data through a single cross-entropy loss multi-task training, achieving alignment between self-supervised clustering centers and class prototypes.
Semi-Supervised Parametric Real-World Image Harmonization
Ke Wang (Adobe Inc.), Eli Shechtman (Adobe Inc.)
Image HarmonizationRestorationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: A semi-supervised dual-stream training parametric image harmonization method is proposed, utilizing global RGB curves and local shadow mapping to achieve high-resolution image harmonization.
Semi-Supervised Stereo-Based 3D Object Detection via Cross-View Consensus
Wenhao Wu (City University of Hong Kong), Si Wu (South China University of Technology)
Object DetectionDepth EstimationAutonomous DrivingKnowledge DistillationImage
🎯 What it does: A semi-supervised 3D object detection method based on a teacher-student framework is proposed on limited labeled stereo image data, utilizing cross-view consistency to filter and correct pseudo-labels, and guiding the student model to improve depth estimation accuracy through temporal aggregation of disparity consistency.
Semi-Supervised Video Inpainting With Cycle Consistency Constraints
Zhiliang Wu (Nanjing University of Science and Technology), Yan Yan (Illinois Institute of Technology)
RestorationSegmentationConvolutional Neural NetworkTransformerVideo
🎯 What it does: A semi-supervised video inpainting framework is proposed, which only requires labeling a single frame mask to complete the filling of missing areas in the entire video segment.
Semi-Weakly Supervised Object Kinematic Motion Prediction
Gengxin Liu (Shenzhen University), Ruizhen Hu (Shenzhen University)
Object DetectionPose EstimationGraph Neural NetworkPoint CloudMesh
🎯 What it does: A semi-weakly supervised framework is designed to transfer limited labeled motion information to large-scale unlabeled 3D models, achieving pseudo-label generation and improving motion prediction.
SemiCVT: Semi-Supervised Convolutional Vision Transformer for Semantic Segmentation
Huimin Huang (Zhejiang University), Yefeng Zheng (Tencent)
SegmentationConvolutional Neural NetworkTransformerImage
🎯 What it does: Proposes the SemiCVT scheme, which integrates a parallel structure of CNN and Transformer, utilizing Fourier domain interactions and cross-model category consistency for semi-supervised semantic segmentation.
Semidefinite Relaxations for Robust Multiview Triangulation
Linus Härenstam-Nielsen, Daniel Cremers (Karlsruhe University of Applied Sciences)
OptimizationPoint Cloud
🎯 What it does: This paper proposes two robust multi-view triangulation frameworks based on semidefinite relaxation, which can still achieve a global optimal solution in the presence of noise and outliers.
SeqTrack: Sequence to Sequence Learning for Visual Object Tracking
Xin Chen (Dalian University of Technology), Han Hu (Microsoft Research)
Object TrackingTransformerVideo
🎯 What it does: A SeqTrack framework is proposed that transforms visual object tracking into a sequence generation task, utilizing a simple encoding-decoding Transformer to complete object bounding box localization.
Sequential Training of GANs Against GAN-Classifiers Reveals Correlated "Knowledge Gaps" Present Among Independently Trained GAN Instances
Arkanath Pathak (Google Research), Nicholas Dufour (Google Research)
GenerationData SynthesisAdversarial AttackConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper explores the knowledge gaps (artifacts) between different GAN-generated instances and their impact on training dynamics, generation quality, and detector generalization by alternately training GANs and classifiers targeting GANs.
SeSDF: Self-Evolved Signed Distance Field for Implicit 3D Clothed Human Reconstruction
Yukang Cao (Hong Kong University), Kwan-Yee K. Wong (Hong Kong University)
GenerationPose EstimationNeural Radiance FieldPoint CloudMesh
🎯 What it does: Proposes the SeSDF framework, which utilizes the SMPL-X prior to reconstruct 3D models of clothed humans from single-view or uncalibrated multi-view images.
SFD2: Semantic-Guided Feature Detection and Description
Fei Xue (University of Cambridge), Roberto Cipolla (University of Cambridge)
Object DetectionSegmentationAutonomous DrivingConvolutional Neural NetworkSimultaneous Localization and MappingImage
🎯 What it does: This paper studies how to utilize semantic information to enhance the robustness of feature detection and description in visual localization tasks, and proposes a feature network that implicitly embeds semantics through semantic guidance during the training phase.
SfM-TTR: Using Structure From Motion for Test-Time Refinement of Single-View Depth Networks
Sergio Izquierdo (University of Zaragoza), Javier Civera (University of Zaragoza)
Depth EstimationAutonomous DrivingSimultaneous Localization and MappingPoint Cloud
🎯 What it does: This paper proposes the SfM-TTR method, which utilizes sparse SfM reconstruction as a self-supervised signal during testing for fine-tuning the network in single-view depth estimation.
SGLoc: Scene Geometry Encoding for Outdoor LiDAR Localization
Wen Li (Xiamen University), Chenglu Wen (Xiamen University)
Pose EstimationAutonomous DrivingSimultaneous Localization and MappingPoint Cloud
🎯 What it does: The SGLoc framework is proposed, decoupling LiDAR localization into 3D-3D correspondence regression and RANSAC pose estimation, and providing the PQEE method to enhance data quality.
ShadowDiffusion: When Degradation Prior Meets Diffusion Model for Shadow Removal
Lanqing Guo (Nanyang Technological University), Bihan Wen (Harvard University)
RestorationDiffusion modelImage
🎯 What it does: A shadow removal framework called ShadowDiffusion based on diffusion models is proposed, which can recover shadow areas and remove shadows from a single image.
ShadowNeuS: Neural SDF Reconstruction by Shadow Ray Supervision
Jingwang Ling (Tsinghua University), Feng Xu (SenseTime Research)
RestorationData SynthesisDepth EstimationNeural Radiance FieldImage
🎯 What it does: This paper proposes a neural field method utilizing shadow ray supervision (ShadowNeuS), which can reconstruct a complete 3D scene from single-view multi-light source shadows or RGB images.
Shakes on a Plane: Unsupervised Depth Estimation From Unstabilized Photography
Ilya Chugunov (Princeton University), Felix Heide (Princeton University)
Depth EstimationOptimizationImage
🎯 What it does: By performing unsupervised optimization on long-duration handheld stabilized RAW burst sequences, we jointly estimate camera motion and high-precision depth maps.
Shape-Aware Text-Driven Layered Video Editing
Yao-Chih Lee (University of Maryland), Jia-Bin Huang (University of Maryland)
GenerationOptimizationDiffusion modelVideoText
🎯 What it does: This paper proposes a shape-aware video editing method based on text prompts, utilizing layered video representation to achieve appearance and shape consistency editing of target objects in videos.
Shape-Constraint Recurrent Flow for 6D Object Pose Estimation
Yang Hai (Xidian University), Yinlin Hu (MagicLeap)
Pose EstimationRecurrent Neural NetworkOptical FlowImage
🎯 What it does: A shape-constrained recursive matching framework is proposed, which refines 6D pose estimation using 3D shape information and optical flow.
Shape-Erased Feature Learning for Visible-Infrared Person Re-Identification
Jiawei Feng (Sun Yat-sen University), Wei-Shi Zheng (Sun Yat-sen University)
RecognitionRetrievalConvolutional Neural NetworkContrastive LearningImageVideo
🎯 What it does: A shape erasure feature learning paradigm is proposed, which uses orthogonal decomposition to divide visible-infrared re-identification features into shape-related and shape-erased parts, explicitly enhancing the feature diversity for cross-modal recognition.
Shape, Pose, and Appearance From a Single Image via Bootstrapped Radiance Field Inversion
Dario Pavllo, Federico Tombari
GenerationPose EstimationNeural Radiance FieldGenerative Adversarial NetworkImage
🎯 What it does: A hybrid GAN inversion framework is proposed, which utilizes a single image to recover 3D shape (SDF), pose, and appearance, balancing diversity and high quality;
ShapeClipper: Scalable 3D Shape Learning From Single-View Images via Geometric and CLIP-Based Consistency
Zixuan Huang (Georgia Institute of Technology), James M. Rehg (Georgia Institute of Technology)
SegmentationGenerationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper presents ShapeClipper, a framework that learns 3D object shapes from a single unannotated real-world RGB image. It guides the learning of implicit SDF through CLIP's semantic consistency constraints and offline surface normal consistency constraints, enabling the simultaneous reconstruction of global structure and local geometric details.
ShapeTalk: A Language Dataset and Framework for 3D Shape Edits and Deformations
Panos Achlioptas (Stanford University), Leonidas Guibas (Stanford University)
SegmentationGenerationData SynthesisTransformerAuto EncoderTextPoint CloudMesh
🎯 What it does: This paper presents a large-scale shape comparison language dataset called ShapeTalk, and based on this, develops a language-driven shape editing framework called ChangeIt3D that can work with any 3D generative model. Additionally, it provides quantifiable evaluation metrics for editing quality.
Sharpness-Aware Gradient Matching for Domain Generalization
Pengfei Wang (Hong Kong Polytechnic University), Lei Zhang (Hong Kong Polytechnic University)
Domain AdaptationImage
🎯 What it does: The Sharpness-Aware Gradient Matching (SAGM) method is proposed, which improves domain generalization ability by simultaneously minimizing empirical risk, perturbation loss, and the gap between the two, allowing the model to converge to low-loss and flat regions.
Shepherding Slots to Objects: Towards Stable and Robust Object-Centric Learning
Jinwoo Kim (Yonsei University), Seon Joo Kim (Yonsei University)
Object DetectionSegmentationImage
🎯 What it does: A single-view image object-centric learning framework called SLASH is proposed, which addresses the bleeding problem caused by background noise by incorporating two modules, ARK (Attention Refinement Kernel) and IPPE (Intermediate Point Prediction and Encoding), based on Slot Attention, achieving more stable and robust object decomposition.
Shifted Diffusion for Text-to-Image Generation
Yufan Zhou (State University of New York at Buffalo), Jinhui Xu (State University of New York at Buffalo)
GenerationData SynthesisTransformerDiffusion modelImageTextMultimodality
🎯 What it does: The Corgi model is proposed, which utilizes Shifted Diffusion to generate image embeddings in the CLIP multimodal space, enabling text-to-image generation and supporting fully supervised, semi-supervised, and no-language (image-only) training.
Shortcomings of Top-Down Randomization-Based Sanity Checks for Evaluations of Deep Neural Network Explanations
Alexander Binder (SUTD Singapore), Wojciech Samek (Fraunhofer Heinrich Hertz Institute)
Explainability and InterpretabilityConvolutional Neural NetworkImage
🎯 What it does: This paper reveals the shortcomings of using top-down randomization sanity checks to evaluate deep network explanation methods through theoretical analysis and experimental validation, and suggests that such checks should be used with caution in isolation.
SHS-Net: Learning Signed Hyper Surfaces for Oriented Normal Estimation of Point Clouds
Qing Li (Tsinghua University), Zhizhong Han (Wayne State University)
OptimizationPoint Cloud
🎯 What it does: Designed and implemented SHS-Net, which achieves one-shot directed normal vector estimation by learning signature hypersurfaces.
Siamese DETR
Zeren Chen (Beihang University), Lu Sheng (Beihang University)
Object DetectionTransformerContrastive LearningImage
🎯 What it does: A Siamese self-supervised pre-training framework for the Transformer detector DETR is proposed, utilizing two-view contrast to achieve pre-training tasks for region detection and semantic distinction.
Siamese Image Modeling for Self-Supervised Vision Representation Learning
Chenxin Tao (Tsinghua University), Jifeng Dai (Tsinghua University)
Object DetectionSegmentationRepresentation LearningTransformerContrastive LearningImage
🎯 What it does: The SiameseIM framework is proposed, which predicts dense representations of different augmented views in self-supervised learning through a dual-branch network, integrating the advantages of Instance Discrimination and Masked Image Modeling;
Sibling-Attack: Rethinking Transferable Adversarial Attacks Against Face Recognition
Zexin Li (University of California), Cong Liu (University of California)
RecognitionAdversarial AttackMeta LearningImage
🎯 What it does: A transferable adversarial attack method based on multi-task learning, called Sibling-Attack, is proposed, which utilizes facial attribute recognition tasks to enhance the black-box attack success rate against face recognition models.
Side Adapter Network for Open-Vocabulary Semantic Segmentation
Mengde Xu (Huazhong University of Science and Technology), Xiang Bai (Huazhong University of Science and Technology)
SegmentationTransformerContrastive LearningImage
🎯 What it does: This paper proposes the Side Adapter Network (SAN), which attaches a lightweight side network to a frozen CLIP model to achieve end-to-end open vocabulary semantic segmentation.
SIEDOB: Semantic Image Editing by Disentangling Object and Background
Wuyang Luo (Fudan University), Weishan Zhang (China University of Petroleum)
Image TranslationGenerationGenerative Adversarial NetworkImage
🎯 What it does: For the task of semantic image editing, a framework called SIEDOB is proposed, which separates the background and foreground objects for individual generation and fusion, significantly improving editing quality in complex scenes.
SIM: Semantic-Aware Instance Mask Generation for Box-Supervised Instance Segmentation
Ruihuang Li (Hong Kong Polytechnic University), Lei Zhang (Hong Kong Polytechnic University)
Object DetectionSegmentationImage
🎯 What it does: A method for instance mask generation based on semantic prototypes is proposed, which achieves instance segmentation relying solely on box supervision.
Similarity Maps for Self-Training Weakly-Supervised Phrase Grounding
Tal Shaharabany (Tel Aviv University), Lior Wolf (Tel Aviv University)
Object DetectionSegmentationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: By extracting and aggregating the self-similarity maps of the image encoder as pseudo-labels, the weakly supervised phrase localization model is improved using a self-training approach.
Similarity Metric Learning for RGB-Infrared Group Re-Identification
Jianghao Xiong (Sun Yat-Sen University), Jianhuang Lai (Sun Yat-Sen University)
RecognitionRetrievalConvolutional Neural NetworkGraph Neural NetworkImageMultimodalityBenchmark
🎯 What it does: This paper proposes a metric learning framework for RGB-IR cross-modal group re-identification (G-ReID) called Closest Permutation Matching (CPM), and based on this, designs a weakly supervised Relation-Aware Module (RAM).
Simple Cues Lead to a Strong Multi-Object Tracker
Jenny Seidenschwarz (Technical University of Munich), Laura Leal-Taixé (Technical University of Munich)
Object TrackingDomain AdaptationConvolutional Neural NetworkSupervised Fine-TuningImageVideo
🎯 What it does: This paper proposes an online multi-object tracker GHOST based on the Hungarian algorithm, which combines improved re-identification features and a simple linear motion model to achieve robust association performance.
SimpleNet: A Simple Network for Image Anomaly Detection and Localization
Zhikang Liu (University of Science and Technology of China), Zilei Wang (University of Science and Technology of China)
Anomaly DetectionConvolutional Neural NetworkGaussian SplattingImage
🎯 What it does: A simple network called SimpleNet is proposed for unsupervised image defect detection and localization.
SimpSON: Simplifying Photo Cleanup With Single-Click Distracting Object Segmentation Network
Chuong Huynh (University of Maryland), Abhinav Shrivastava (University of Maryland)
RestorationObject DetectionSegmentationConvolutional Neural NetworkTransformerImage
🎯 What it does: An interactive framework based on single-click (SimpSON) is proposed, which can automatically segment and remove visual distractions (such as small, rare, or frequently appearing objects) from photos, and supports batch selection of similar distractions with a single click.
Simulated Annealing in Early Layers Leads to Better Generalization
Amir M. Sarfi (Concordia University), Eugene Belilovsky (Concordia University)
Domain AdaptationOptimizationConvolutional Neural NetworkImage
🎯 What it does: The iterative training method of gradient ascent followed by descent in the early layers of the network (SEAL) aims to enhance the model's generalization and transfer performance.
Simultaneously Short- and Long-Term Temporal Modeling for Semi-Supervised Video Semantic Segmentation
Jiangwei Lao (Ant Group), Wei Chu (Ant Group)
SegmentationTransformerVideo
🎯 What it does: A SSLTM network is proposed, designed for video semantic segmentation with only one frame labeled per video segment, utilizing both short-term and long-term temporal information to enhance feature representation.
SINE: Semantic-Driven Image-Based NeRF Editing With Prior-Guided Editing Field
Chong Bao (Zhejiang University), Zhaopeng Cui (Zhejiang University)
GenerationData SynthesisTransformerNeural Radiance FieldImage
🎯 What it does: Utilize a single edited image or text prompt to perform semantic-driven geometric and texture editing on a pre-trained NeRF, producing multi-view consistent and realistic edited results.
SINE: SINgle Image Editing With Text-to-Image Diffusion Models
Zhixing Zhang (Rutgers University), Jian Ren (Snap Inc.)
Image TranslationGenerationDiffusion modelImage
🎯 What it does: This paper proposes a single image editing framework called SINE, based on a pre-trained text-to-image diffusion model, which can perform various edits such as style transfer, content addition, and object modification using only one image and a text prompt, and supports output at any resolution.
Single Domain Generalization for LiDAR Semantic Segmentation
Hyeonseong Kim (Korea Advanced Institute of Science and Technology), Kuk-Jin Yoon (Korea Advanced Institute of Science and Technology)
SegmentationDomain AdaptationAutonomous DrivingConvolutional Neural NetworkPoint Cloud
🎯 What it does: This paper proposes a LiDAR semantic segmentation method for single-source domain generalization, DGLSS, which learns domain-invariant representations using sparse consistency and semantic-related consistency constraints, maintaining good performance in both the training source domain and unseen target domain.
Single Image Backdoor Inversion via Robust Smoothed Classifiers
Mingjie Sun (Carnegie Mellon University), Zico Kolter (Bosch Center for AI)
OptimizationAdversarial AttackDiffusion modelImageBenchmark
🎯 What it does: This paper proposes a single-image reverse inference backdoor trigger method (SmoothInv) that can recover high-success-rate triggers that are highly similar to the original backdoor from a single clean image.
Single Image Depth Prediction Made Better: A Multivariate Gaussian Take
Ce Liu (ETH Zurich), Luc Van Gool (ETH Zurich)
Depth EstimationAutonomous DrivingTransformerImage
🎯 What it does: This paper studies the modeling of multivariate Gaussian distribution for monocular image depth prediction and proposes a negative log-likelihood loss with low-rank covariance approximation.
Single View Scene Scale Estimation Using Scale Field
Byeong-Uk Lee (KAIST), In So Kweon (KAIST)
Depth EstimationTransformerImage
🎯 What it does: A scale field representation based on a single image is proposed, training a network to achieve the mapping from image to scale field, and annotating the network through geometric reasoning before network training.
SinGRAF: Learning a 3D Generative Radiance Field for a Single Scene
Minjung Son (Samsung Advanced Institute of Technology), Gordon Wetzstein (Stanford University)
GenerationData SynthesisNeural Radiance FieldGenerative Adversarial NetworkImage
🎯 What it does: Using a small number of uncalibrated single-scene images to train the generative model SinGRAF, it can produce a variety of visually realistic and diverse 3D scene realizations.
Sketch2Saliency: Learning To Detect Salient Objects From Human Drawings
Ayan Kumar Bhunia (University of Surrey), Yi-Zhe Song (University of Surrey)
Object DetectionSegmentationConvolutional Neural NetworkRecurrent Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: Utilizing hand-drawn sketches as weak labels, we learn image saliency detection through a photo-to-sketch generation network.
SketchXAI: A First Look at Explainability for Human Sketches
Zhiyu Qu (University of Surrey), Yi-Zhe Song (University of Surrey)
ClassificationRecognitionExplainability and InterpretabilityConvolutional Neural NetworkRecurrent Neural NetworkTransformerImage
🎯 What it does: An interpretable framework for human hand-drawn sketches is proposed, and dynamic visual explanations of the hand-drawn classifier are achieved through the 'Stroke Location Inversion (SLI)' method.
Skinned Motion Retargeting With Residual Perception of Motion Semantics & Geometry
Jiaxu Zhang (Wuhan University), Zhigang Tu (Wuhan University)
Pose EstimationTransformerVideo
🎯 What it does: A residual structure motion remapping network R2ET is proposed, which can maintain the source action semantics while avoiding the issues of target character crossover and self-collision in a single-step inference.
SkyEye: Self-Supervised Bird's-Eye-View Semantic Mapping Using Monocular Frontal View Images
Nikhil Gosala (University of Freiburg), Abhinav Valada (University of Freiburg)
SegmentationDepth EstimationAutonomous DrivingConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: The SkyEye framework is proposed, which generates bird's eye view (BEV) semantic maps using single front-view images and self-supervised methods, eliminating the need for BEV annotations.
SLACK: Stable Learning of Augmentations With Cold-Start and KL Regularization
Juliette Marrie (NAVER LABS Europe), Julien Mairal (University of Grenoble Alpes)
OptimizationData-Centric LearningConvolutional Neural NetworkReinforcement LearningImage
🎯 What it does: Learn data augmentation strategies directly from a complete transformation space without prior knowledge, skipping manual pre-selection of transformations;
Sliced Optimal Partial Transport
Yikun Bai (Vanderbilt University), Soheil Kolouri (Vanderbilt University)
OptimizationPoint Cloud
🎯 What it does: An efficient primal-dual algorithm is proposed to solve optimal partial transport (OPT) on one-dimensional discrete measures, and it is extended to multi-dimensional space to obtain the sliced-optimal partial transport (SOPT) distance through projection, which is used for point cloud registration and color matching.
SliceMatch: Geometry-Guided Aggregation for Cross-View Pose Estimation
Ted Lentsch (Delft University of Technology), Julian F. P. Kooij (Delft University of Technology)
Pose EstimationContrastive LearningImage
🎯 What it does: This paper proposes SliceMatch, a cross-view 3-DoF camera pose estimation framework based on slice aggregation and geometric guidance.
Slide-Transformer: Hierarchical Vision Transformer With Local Self-Attention
Xuran Pan (Tsinghua University), Gao Huang (Tsinghua University)
ClassificationObject DetectionSegmentationTransformerImage
🎯 What it does: An efficient and flexible local attention module called Slide Attention is proposed, which balances convolutional local attribution and self-attention dynamic feature selection, allowing for easy replacement in various Vision Transformer architectures.
Slimmable Dataset Condensation
Songhua Liu (National University of Singapore), Xinchao Wang (National University of Singapore)
Data SynthesisCompressionImage
🎯 What it does: This study investigates methods for further compressing existing synthetic data (slimmable dataset condensation).
SLOPER4D: A Scene-Aware Dataset for Global 4D Human Pose Estimation in Urban Environments
Yudi Dai (Xiamen University), Cheng Wang (Xiamen University)
Pose EstimationSimultaneous Localization and MappingMultimodalityPoint CloudBenchmark
🎯 What it does: We constructed SLOPER4D—a large-scale global 4D human pose dataset in urban environments that includes LiDAR, camera, and IMU data. Based on this, we conducted benchmark experiments on 3D human pose estimation (HPE) and global human pose estimation (GHPE).
SlowLiDAR: Increasing the Latency of LiDAR-Based Detection Using Adversarial Examples
Han Liu (Washington University in St. Louis), Ning Zhang (Washington University in St. Louis)
Object DetectionAutonomous DrivingComputational EfficiencyAdversarial AttackPoint Cloud
🎯 What it does: A time delay attack on the LiDAR detection pipeline named SlowLiDAR is proposed, which can significantly increase the running time of detection models by perturbing the point cloud or adding points, while maintaining the imperceptibility of the adversarial samples.
SMAE: Few-Shot Learning for HDR Deghosting With Saturation-Aware Masked Autoencoders
Qingsen Yan (Northwestern Polytechnical University), Yanning Zhang (Northwestern Polytechnical University)
RestorationTransformerAuto EncoderImage
🎯 What it does: A two-stage semi-supervised HDR ghost removal method called SSHDR is proposed: first, a self-supervised Saturated Mask AutoEncoder (SMAE) is used to fill in saturated areas, and then adaptive pseudo-label iterative learning is applied to remove ghosting.
SmallCap: Lightweight Image Captioning Prompted With Retrieval Augmentation
Rita Ramos (Instituto Superior Técnico), Yova Kementchedjhieva
GenerationRetrievalDomain AdaptationComputational EfficiencyTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageVideoTextRetrieval-Augmented Generation
🎯 What it does: A lightweight image description model called SMALLCAP is designed, which utilizes retrieval-enhanced prompts to generate descriptions and can achieve cross-domain transfer without additional training.
SmartAssign: Learning a Smart Knowledge Assignment Strategy for Deraining and Desnowing
Yinglong Wang (Huawei Noah's Ark Lab), Jianzhuang Liu (Huawei Noah's Ark Lab)
Image TranslationRestorationTransformerContrastive LearningImage
🎯 What it does: A multi-task learning framework called SmartAssign is proposed, which can simultaneously remove raindrops and snowflakes.
SmartBrush: Text and Shape Guided Object Inpainting With Diffusion Model
Shaoan Xie (Carnegie Mellon University), Kun Zhang (Mohamed bin Zayed University of Artificial Intelligence)
RestorationGenerationDiffusion modelImageTextMultimodality
🎯 What it does: A multimodal object filling method based on diffusion models is designed, which can simultaneously utilize text descriptions and shape masks to complete the insertion of missing areas, and supports precision control from exact to rough masks.
SMOC-Net: Leveraging Camera Pose for Self-Supervised Monocular Object Pose Estimation
Tao Tan (University of Chinese Academy of Sciences), Qiulei Dong (University of Chinese Academy of Sciences)
Pose EstimationKnowledge DistillationSimultaneous Localization and MappingImage
🎯 What it does: Proposes SMOC-Net, a self-supervised monocular 6D object pose estimation network that utilizes camera poses from unlabeled real images.
SMPConv: Self-Moving Point Representations for Continuous Convolution
Sanghyeon Kim (Sungkyunkwan University), Eunbyung Park (Sungkyunkwan University)
ClassificationSegmentationOptimizationConvolutional Neural NetworkImageTime SeriesSequential
🎯 What it does: A continuous convolution based on self-moving point representation (SMPConv) is proposed, where the convolution kernel parameters are represented by learnable point coordinates, weights, and radii, allowing for the construction of convolution kernels of arbitrary resolution without the need for neural networks.
Soft Augmentation for Image Classification
Yang Liu (Argo AI), Deva Ramanan (Argo AI)
ClassificationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a Soft Augmentation technique that dynamically softens target labels or sample weights based on the degree of transformation, allowing the network to train under more aggressive augmentations without losing performance.
Soft-Landing Strategy for Alleviating the Task Discrepancy Problem in Temporal Action Localization Tasks
Hyolim Kang (Yonsei University), Seon Joo Kim (Yonsei University)
RecognitionObject DetectionConvolutional Neural NetworkContrastive LearningVideo
🎯 What it does: A lightweight Soft-Landing (SoLa) module is proposed, placed between the frozen snippet encoder and the TAL head, and trained using self-supervised Similarity Matching to enhance the temporal sensitivity of segment features and alleviate the task gap issue.
Solving 3D Inverse Problems Using Pre-Trained 2D Diffusion Models
Hyungjin Chung (Korea Advanced Institute of Science and Technology), Jong Chul Ye (Korea Advanced Institute of Science and Technology)
RestorationGenerationDiffusion modelScore-based ModelImageBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: This paper proposes a method that combines a pre-trained two-dimensional diffusion model with traditional TV-based models to address the problem of three-dimensional medical image reconstruction.
Solving Oscillation Problem in Post-Training Quantization Through a Theoretical Perspective
Yuexiao Ma (Xiamen University), Rongrong Ji (Xiamen University)
CompressionOptimizationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a theoretical perspective to explain and solve the oscillation problem in post-training quantization (PTQ), further improving the accuracy of PTQ.
Solving Relaxations of MAP-MRF Problems: Combinatorial In-Face Frank-Wolfe Directions
Vladimir Kolmogorov (Institute of Science and Technology Austria)
OptimizationGraphBenchmark
🎯 What it does: An efficient implementation of LP relaxation for the MAP-MRF problem is proposed using the Frank-Wolfe algorithm combined with in-face directions.
SOOD: Towards Semi-Supervised Oriented Object Detection
Wei Hua (Huazhong University of Science and Technology), Xiang Bai (Huazhong University of Science and Technology)
Object DetectionImageBenchmark
🎯 What it does: A semi-supervised directional object detection framework SOOD for aerial images is proposed.
Sound to Visual Scene Generation by Audio-to-Visual Latent Alignment
Kim Sung-Bin (POSTECH), Tae-Hyun Oh (Yonsei University)
GenerationData SynthesisConvolutional Neural NetworkGenerative Adversarial NetworkContrastive LearningImageVideoMultimodalityAudio
🎯 What it does: Train an audio-to-visual alignment model to generate high-quality scene images corresponding to input audio without labeled or language supervision.
Source-Free Adaptive Gaze Estimation by Uncertainty Reduction
Xin Cai (Institute of Computing Technology Chinese Academy of Sciences), Xilin Chen (Institute of Computing Technology Chinese Academy of Sciences)
RecognitionDomain AdaptationConvolutional Neural NetworkImage
🎯 What it does: A passive data cross-domain gaze estimation method called UnReGA is proposed, which achieves adaptation by reducing sample and model uncertainty.
Source-Free Video Domain Adaptation With Spatial-Temporal-Historical Consistency Learning
Kai Li (NEC Labs America), Martin Renqiang Min (NEC Labs America)
Domain AdaptationConvolutional Neural NetworkVideo
🎯 What it does: A source-agnostic video domain adaptation method based on spatio-temporal-historical consistency learning (STHC) is proposed, achieving model adaptation by randomly applying spatial and temporal augmentations on the target video and enforcing three types of consistency losses.
SPARF: Neural Radiance Fields From Sparse and Noisy Poses
Prune Truong (ETH Zurich), Federico Tombari (Technical University of Munich)
GenerationPose EstimationNeural Radiance FieldImage
🎯 What it does: This paper proposes SPARF, a joint optimization method for training NeRF scene representation and camera pose under conditions of sparse views and significant camera pose noise.
Sparse Multi-Modal Graph Transformer With Shared-Context Processing for Representation Learning of Giga-Pixel Images
Ramin Nakhli (University of British Columbia), Ali Bashashati (University of British Columbia)
Representation LearningGraph Neural NetworkTransformerImageMultimodalityBiomedical Data
🎯 What it does: This paper proposes a multimodal sparse graph transformer (AMIGO) that constructs cell-level cell graphs and performs shared context processing and cross-modal attention aggregation between different staining markers to generate patient-level representations for tumor survival prediction.
SparseFusion: Distilling View-Conditioned Diffusion for 3D Reconstruction
Zhizhuo Zhou (Carnegie Mellon University), Shubham Tulsiani (Carnegie Mellon University)
GenerationData SynthesisDepth EstimationKnowledge DistillationTransformerDiffusion modelNeural Radiance FieldImage
🎯 What it does: SparseFusion proposes a method to reconstruct a complete and geometrically consistent 3D neural field from very few views (such as 2-3 images) and can render from any new viewpoint.
Sparsely Annotated Semantic Segmentation With Adaptive Gaussian Mixtures
Linshan Wu (Hunan University), Hao Chen (Hong Kong University of Science and Technology)
SegmentationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes a sparse annotation semantic segmentation framework based on an Adaptive Gaussian Mixture Model (AGMM), using labeled pixels as the centers of the Gaussian mixtures to generate soft GMM predictions online for self-supervision.
SparsePose: Sparse-View Camera Pose Regression and Refinement
Samarth Sinha (University of Toronto), David B. Lindell (Vector Institute)
Pose EstimationRecurrent Neural NetworkTransformerVideo
🎯 What it does: This paper proposes a method called SparsePose for camera pose regression and iterative refinement based on sparse views, which can accurately estimate camera poses and achieve high-fidelity novel view synthesis with fewer than 10 wide baseline images.
SparseViT: Revisiting Activation Sparsity for Efficient High-Resolution Vision Transformer
Xuanyao Chen (Fudan University), Song Han (Shanghai Qi Zhi Institute)
Object DetectionSegmentationAutonomous DrivingComputational EfficiencyTransformerImage
🎯 What it does: Introduce window-level activation pruning on high-resolution Swin Transformer, significantly reducing computation by skipping low-information windows while maintaining accuracy.
Sparsifiner: Learning Sparse Instance-Dependent Attention for Efficient Vision Transformers
Cong Wei (University of Toronto), Florian Shkurti (University of Toronto)
Computational EfficiencyKnowledge DistillationTransformerImage
🎯 What it does: This paper proposes Sparsifiner, which learns instance-dependent sparse attention patterns to reduce the computational cost of MHSA in Vision Transformers.
SpaText: Spatio-Textual Representation for Controllable Image Generation
Omri Avrahami (Meta AI), Xi Yin (Meta AI)
GenerationDiffusion modelImageText
🎯 What it does: This paper proposes an image generation method called SpaText based on sparse free-text scene control, which can generate images that conform to layout and semantics given global text and local free text prompts.
Spatial-Frequency Mutual Learning for Face Super-Resolution
Chenyang Wang (Harbin Institute of Technology), Xianming Liu (Harbin Institute of Technology)
RecognitionSuper ResolutionGenerative Adversarial NetworkImage
🎯 What it does: A dual-branch spatial frequency mutual learning network (SFMNet) is proposed for the task of face super-resolution.
Spatial-Temporal Concept Based Explanation of 3D ConvNets
Ying Ji (Nagoya University), Jien Kato (Ritsumeikan University)
Explainability and InterpretabilityConvolutional Neural NetworkVideo
🎯 What it does: A spatiotemporal concept-based explanation framework (STCE) is proposed for global interpretability analysis of 3D ConvNet.
Spatial-Then-Temporal Self-Supervised Learning for Video Correspondence
Rui Li (University of Science and Technology of China), Dong Liu (University of Science and Technology of China)
SegmentationRepresentation LearningContrastive LearningImageVideo
🎯 What it does: This paper proposes a two-stage self-supervised learning framework, which first trains spatial representations using image contrastive learning, and then enhances temporal representations on videos through multi-scale frame reconstruction and correlation distillation, while introducing global correlation distillation to retain spatial discriminative ability.
Spatially Adaptive Self-Supervised Learning for Real-World Image Denoising
Junyi Li (Harbin Institute of Technology), Wangmeng Zuo (Harbin Institute of Technology)
RestorationConvolutional Neural NetworkImage
🎯 What it does: A spatially adaptive supervised self-supervised image denoising framework is proposed, utilizing a blind neighborhood network to extract supervision from flat regions and a local perception network to extract supervision from textured regions, ultimately training a U-Net for denoising.
Spatio-Focal Bidirectional Disparity Estimation From a Dual-Pixel Image
Donggun Kim (Korea Advanced Institute of Science and Technology), Min H. Kim (Korea Advanced Institute of Science and Technology)
Depth EstimationSupervised Fine-TuningImage
🎯 What it does: This paper proposes a self-supervised learning method that utilizes the reflective symmetry blur kernel characteristics of dual-pixel cameras to directly estimate bidirectional disparity (i.e., complete depth information containing both positive and negative disparities) from a single dual-pixel image.
Spatio-Temporal Pixel-Level Contrastive Learning-Based Source-Free Domain Adaptation for Video Semantic Segmentation
Shao-Yuan Lo (Johns Hopkins University), Vishal M. Patel (Johns Hopkins University)
SegmentationDomain AdaptationContrastive LearningOptical FlowVideo
🎯 What it does: This paper studies the application of source-free domain adaptation (SFDA) in video semantic segmentation (VSS) and proposes a spatiotemporal fusion method STPL based on pixel-level contrastive learning.
Spatiotemporal Self-Supervised Learning for Point Clouds in the Wild
Yanhao Wu (Xi'an Jiaotong University), Mathieu Salzmann (EPFL)
SegmentationAutonomous DrivingConvolutional Neural NetworkContrastive LearningPoint Cloud
🎯 What it does: This paper proposes a LiDAR point cloud semantic segmentation method based on spatiotemporal self-supervised learning, utilizing unsupervised clustering and tracking to construct positive samples for end-to-end self-supervised pre-training.
Specialist Diffusion: Plug-and-Play Sample-Efficient Fine-Tuning of Text-to-Image Diffusion Models To Learn Any Unseen Style
Haoming Lu (Picsart AI Research), Humphrey Shi (U of Oregon)
GenerationData SynthesisComputational EfficiencySupervised Fine-TuningDiffusion modelImageText
🎯 What it does: A fine-tuning framework named Specialist Diffusion is proposed, which learns unknown styles using a very small number (≤10 images) of images, enabling a pre-trained text-to-image diffusion model to generate images of any object in that style.
Spectral Bayesian Uncertainty for Image Super-Resolution
Tao Liu (Huazhong University of Science and Technology), Shan Tan (Huazhong University of Science and Technology)
RestorationSuper ResolutionConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: A dual-domain learning (DDL) framework is proposed, integrating frequency domain learning into the SR network, and combining MC-dropout to achieve spectral uncertainty estimation. Additionally, a Spectral Uncertainty Decoupled Frequency (SUDF) training method is introduced to enhance perceptual quality.
Spectral Enhanced Rectangle Transformer for Hyperspectral Image Denoising
Miaoyu Li (Beijing Institute of Technology), Dejing Dou (BCG X)
RestorationTransformerImage
🎯 What it does: This paper proposes a Spectral Enhanced Rectangle Transformer (SERT) for hyperspectral image (HSI) denoising.
Sphere-Guided Training of Neural Implicit Surfaces
Andreea Dogaru (Skolkovo Institute of Science and Technology), Evgeny Burnaev (Artificial Intelligence Research Institute)
GenerationOptimizationNeural Radiance FieldImagePoint Cloud
🎯 What it does: This paper proposes a training framework guided by sphere clouds to learn neural implicit surfaces from multi-view images, achieving more efficient and accurate 3D reconstruction.
Spherical Transformer for LiDAR-Based 3D Recognition
Xin Lai (Chinese University of Hong Kong), Jiaya Jia (Chinese University of Hong Kong)
RecognitionObject DetectionSegmentationAutonomous DrivingTransformerPoint Cloud
🎯 What it does: SphereFormer is proposed to directly aggregate long-range information of dense points through radial window self-attention, enhancing the performance of semantic segmentation and detection for sparse long-distance LiDAR point clouds.
Spider GAN: Leveraging Friendly Neighbors To Accelerate GAN Training
Siddarth Asokan (Indian Institute of Science), Chandra Sekhar Seelamantula (Indian Institute of Science)
GenerationData SynthesisComputational EfficiencyGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes Spider GAN, which achieves fast and higher quality GAN training without paired constraints by replacing random noise with friendly neighbor image data that is close to the target distribution as input to the generator.
SPIn-NeRF: Multiview Segmentation and Perceptual Inpainting With Neural Radiance Fields
Ashkan Mirzaei (Samsung AI Centre), Alex Levinshtein (Samsung AI Centre)
RestorationSegmentationNeural Radiance FieldImageVideoBenchmark
🎯 What it does: A method is proposed that starts with a small number of single-view point annotations, utilizing interactive segmentation, semantic NeRF, and a 2D inpainting technique to generate consistent 3D masks, achieving viewpoint-consistent 3D object removal and filling in NeRF.