CVPR 2026 Papers — Page 32
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 4071 papers
See and Fix the Flaws: Enabling VLMs and Diffusion Models to Comprehend Visual Artifacts via Agentic Data Synthesis
Jaehyun Park (KAIST), Dongmin Park (KRAFTON)
Data SynthesisExplainability and InterpretabilityAgentic AIVision Language ModelDiffusion modelImageBenchmark
🎯 What it does: Proposed an automated three-agent framework named ArtiAgent for generating diverse and interpretable visual artifacts in real images, along with constructing corresponding annotated data;
See Further, Think Deeper: Advancing VLM's Reasoning Ability with Low-level Visual Cues and Reflection
Zhiheng Wu, Yumeng Zhang (Zhejiang University)
Explainability and InterpretabilityReinforcement LearningAgentic AIVision Language ModelImageMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Propose the ForeSight framework, integrating low-level visual tools and visual feedback into the VLM reasoning process, and learning tool calling strategies through reinforcement learning.
See It, Say It, Sorted: An Iterative Training-Free Framework for Visually-Grounded Multimodal Reasoning in LVLMs
Yongchang Zhang (Southeast University), Yang Chen (Southeast University)
Computational EfficiencyVision Language ModelMultimodalityBenchmark
🎯 What it does: Propose a no-training, plug-and-play visual evidence-driven decoding framework ECRD, which real-time supervises the multimodal reasoning process to avoid visual hallucinations.
See Less, See Right: Bi-directional Perceptual Shaping For Multimodal Reasoning
Shuoshuo Zhang (Tsinghua University), Rui Wang (Microsoft Research Asia)
Representation LearningReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningVision Language ModelTextMultimodalityGraphBenchmark
🎯 What it does: Propose the BiPS method, which enhances the perception capability of VLMs for fine-grained visual evidence by employing evidence-preserving and evidence-ablated views under problem conditions during training, through bidirectional KL constraints.
See Through the Noise: Improving Domain Generalization in Gaze Estimation
Yanming Peng (Beijing Jiaotong University), Yi Tian (Beijing Jiaotong University)
Pose EstimationDomain AdaptationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: Propose the SeeTN framework, which identifies and suppresses annotation noise by leveraging semantic manifolds and feature-label affinity consistency, thereby enhancing the domain generalization capability of pupil estimation.
See What I Mean: Aligning Vision and Language Representations for Video Fine-grained Object Understanding
Boyuan Sun (Nankai University), Qibin Hou (Nankai University)
Object DetectionSegmentationRepresentation LearningTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelVideoTextMultimodalityBenchmark
🎯 What it does: Constructed the NL-Refer dataset and performed cross-attention alignment training on multimodal large language models, enabling the model to precisely locate target objects in videos based solely on text prompts.
See What We Cannot See: A Geo-guided Reasoning Benchmark for Object Counting under Adverse Earth Observation Conditions
Jiayi Wang (Wuhan University), Zhenzhong Chen (Wuhan University)
Object DetectionTransformerLarge Language ModelImageMultimodalityPoint CloudBenchmark
🎯 What it does: Proposed a large-scale remote sensing object counting dataset called GROC, constructed a controllable degradation synthesis and interactive annotation engine, and further developed a Geo-guided Reasoning Agent based on a multi-modal large model;
See, Think, Act: Teaching Multimodal Agents to Effectively Interact with GUI by Identifying Toggles
Zongru Wu (Shanghai Jiao Tong University), Gongshen Liu (Shanghai Jiao Tong University)
Robotic IntelligenceTransformerLarge Language ModelReinforcement LearningAgentic AIPrompt EngineeringVision-Language-Action ModelMultimodalityBenchmark
🎯 What it does: Built a binary switch instruction state control benchmark and proposed the StaR multimodal reasoning method, enabling agents to first perceive the current state, infer the target state, and then determine whether a switch is needed when executing switch operations;
SeeGroup: Multi-Layer Depth Estimation of Transparent Surfaces via Self-Determined Grouping
Hongyu Wen (Princeton University), Jia Deng (Princeton University)
Depth EstimationTransformerImage
🎯 What it does: This paper proposes a multi-layer depth estimation framework called SeeGroup, specifically designed for multi-layer depth problems of transparent objects, which can predict a multi-layer depth sequence for each pixel from a single RGB image;
Seeing as Experts Do: A Knowledge-Augmented Agent for Open-Set Fine-Grained Visual Understanding
Junhan Chen (Beijing University of Posts and Telecommunications), Zhanyu Ma (Beijing University of Posts and Telecommunications)
RecognitionObject DetectionRetrievalExplainability and InterpretabilityTransformerLarge Language ModelVision-Language-Action ModelDiffusion modelImageMultimodalityBenchmarkRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Propose a knowledge-enhanced fine-grained reasoning agent (KFRA), achieving closed-loop inference for fine-grained visual understanding through open-vocabulary detection, web retrieval, text knowledge retrieval, region localization, and multimodal reasoning.
Seeing Beyond 8bits: Subjective and Objective Quality Assessment of HDR-UGC Videos
Shreshth Saini (University of Texas at Austin), Alan C. Bovik (University of Texas at Austin)
Explainability and InterpretabilityData-Centric LearningReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningVision Language ModelContrastive LearningVideoMultimodalityBenchmark
🎯 What it does: This paper constructs a large HDR-UGC video quality assessment dataset named Beyond8Bits and proposes HDR-Q, a model based on a multimodal large language model. HDR-Q achieves HDR video quality prediction and interpretation using an HDR-aware visual encoder and the HAPO reinforcement learning strategy.
Seeing Beyond: Extrapolative Domain Adaptive Panoramic Segmentation
Yuanfan Zheng (Hunan University), Kailun Yang (Hunan University)
SegmentationDomain AdaptationTransformerImage
🎯 What it does: Propose an open-set domain adaptation based panoptic semantic segmentation framework EDA-PSeg, which trains on local perspectives and infers on 360° panoramas.
Seeing Both Sides: Towards Bidirectional Semantic Alignment for Open-Vocabulary Camouflaged Object Segmentation
Guohui Zhang (Dalian Minzu University), Fasheng Wang (Dalian Minzu University)
SegmentationConvolutional Neural NetworkTransformerPrompt EngineeringVision Language ModelContrastive LearningImage
🎯 What it does: Proposed the BaCLIP framework, which achieves open-vocabulary camouflage object segmentation through bidirectional semantic alignment;
Seeing Clearly, Reasoning Confidently: Plug-and-Play Remedies for Vision Language Model Blindness
Xin Hu (Tulane University), Zhengming Ding (Tulane University)
Explainability and InterpretabilityComputational EfficiencyRepresentation LearningPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: Designed a pluggable lightweight module that refines visual tokens and injects object prompts into text prompts, enhancing the Vision-Language Model's (VLM) ability to recognize and reason about rare objects.
Seeing Conversations: Communication Context Identification in Egocentric Video
Tobias Dorszewski (Technical University of Denmark), Jens Hjortkjær (Technical University of Denmark)
ClassificationObject DetectionPose EstimationRecurrent Neural NetworkTransformerVideo
🎯 What it does: Proposes the task of identifying dialogue partners in first-person videos (Communication Context Identification, CCI), and develops the CoCoNet model to achieve this task through temporal and mutual relationship reasoning.
Seeing Depth Through Frequency and Motion: A Progressive Training Paradigm for Monocular Depth Estimation
Ke Li (Dalian Maritime University), Hongbo Liu (Dalian Maritime University)
Pose EstimationDepth EstimationTransformerImageVideo
🎯 What it does: This paper proposes a frequency-guided self-supervised monocular depth estimation framework that considers the complementarity between depth and pose networks;
Seeing is Improving: Visual Feedback for Iterative Text Layout Refinement
Junrong Guo (University of Science and Technology of China), Hongtao Xie (University of Science and Technology of China)
GenerationTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningVision Language ModelImageTextMultimodality
🎯 What it does: Propose a self-improving text layout generation framework VFLM, combining a generate-render-reflect-modify cycle, leveraging visual feedback to achieve iterative optimization.
Seeing Motion Through Polarity for Event-based Action Recognition
Meiqi Cao (Nanjing University of Science and Technology), Xiangbo Shu (Nanjing University of Science and Technology)
RecognitionLarge Language ModelPrompt EngineeringVision-Language-Action ModelContrastive LearningMultimodality
🎯 What it does: Proposes an event camera action recognition framework called POKER, which explicitly enhances event stream representations in visual and text modalities by leveraging polarity-driven motion knowledge.
Seeing the Scene Matters: Revealing Forgetting in Video Understanding Models with a Scene-Aware Long-Video Benchmark
Seng Nam Chen (Chinese University of Hong Kong (Shenzhen)), Chao Li (University of Cambridge)
Large Language ModelVideoTextMultimodalityBenchmarkRetrieval-Augmented GenerationAudio
🎯 What it does: Propose the SceneBench long-video understanding benchmark to evaluate models' reasoning and memory capabilities at the scene level.
Seeing Through Blur: Tackling Defocus in Spike-Based Imaging
Xiantao Ma (Beijing Institute of Technology), Hua Huang (Beijing Normal University)
RestorationSpiking Neural NetworkTransformerImage
🎯 What it does: To address defocus blur in event cameras caused by depth of field, this paper proposes an end-to-end DeSpike framework that can recover sharp images from captured pulse sequences.
Seeing through boxes: Non-Line-of-Sight 3D Reconstruction from Radar Signals
Jiachen Lu, Haitham Al Hassanieh (École Polytechnique Fédérale de Lausanne)
GenerationNeural Radiance FieldImageMultimodalityPoint Cloud
🎯 What it does: Propose GeRaF 2.0, a unified 3D shape reconstruction framework that integrates visible light Line-of-Sight (LoS) information with radar Non-Line-of-Sight (NLoS) signals.
Seeing through Light and Darkness: Sensor-Physics Grounded Deblurring HDR NeRF from Single-Exposure Images and Events
Yunshan Qi (Beihang University), Jia Li (Beihang University)
RestorationGenerationNeural Radiance FieldImageMultimodalityPhysics Related
🎯 What it does: A unified perceptual physics-driven NeRF framework called See-NeRF is constructed, achieving HDR scene deblurring and novel view synthesis by leveraging single-exposure blurred LDR images and corresponding event data.
Seeing Through the Noise: Improving Infrared Small Target Detection and Segmentation from Noise Suppression Perspective
Maoxun Yuan (Beihang University), Xingxing Wei (Beihang University)
RestorationObject DetectionSegmentationImage
🎯 What it does: Proposed a noise suppression feature pyramid network (NS-FPN) that enhances the accuracy of infrared small target detection and segmentation by significantly reducing false alarms through two modules: low-frequency guided feature purification (LFP) and spiral sensing feature sampling (SFS);
Seeing Through the Shift: Causality-Inspired Robust Generalized Category Discovery
Wei Feng (Monash University), Zongyuan Ge (Monash University)
ClassificationRecognitionDomain AdaptationTransformerImage
🎯 What it does: Proposes a causal-inspired cross-domain generalization category discovery framework, CausalGCD, which can automatically discover new categories and maintain identification of known categories in scenarios with domain shift and unknown class confounding.
Seeing Through Touch: Tactile-Driven Visual Localization of Material Regions
Seongyu Kim (Korea Advanced Institute of Science and Technology), Arda Senocak (Ulsan National Institute of Science and Technology)
SegmentationTransformerVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: Proposed the 'Seeing Through Touch' framework to achieve visual region localization (material segmentation) based on tactile input, generating tactile saliency maps through dense cross-modal feature interactions;
Seeing What Matters: A Training-Free Self-Guided Framework for Multimodal Detail Perception and Reasoning
Mingjie Ma (Huazhong University of Science and Technology), Guohui Li (Huazhong University of Science and Technology)
Super ResolutionExplainability and InterpretabilityComputational EfficiencyTransformerVision Language ModelContrastive LearningMultimodality
🎯 What it does: This paper proposes a training-free, self-guided framework called SLoFo to enhance the detail perception and reasoning capabilities of multimodal large language models in high-resolution visual tasks.
Seeing What Matters: Visual Preference Policy Optimization for Visual Generation
Ziqi Ni (Southeast University), Xuelong Li (Institute of Artificial Intelligence (TeleAI), China Telecom)
GenerationOptimizationReinforcement Learning from Human FeedbackConvolutional Neural NetworkTransformerReinforcement LearningDiffusion modelFlow-based ModelImageVideo
🎯 What it does: Proposes Visual Preference Policy Optimization (ViPO), an improved GRPO method that converts global scalar rewards into pixel-level advantages in visual generation tasks.
Seeing without Pixels: Perception from Camera Trajectories
Zihui Xue (Google DeepMind), Tengda Han (Google DeepMind)
ClassificationRecognitionRetrievalRepresentation LearningTransformerVision Language ModelContrastive LearningMultimodalitySequential
🎯 What it does: Leverage camera trajectory (camera pose sequence) for semantic awareness of video content and propose the CamFormer encoder for contrastive learning.
Seele: A Unified Acceleration Framework for Real-Time Gaussian Splatting on Mobile Devices
He Zhu (Shanghai Jiao Tong University), Yu Feng (Shanghai Jiao Tong University)
Computational EfficiencyGaussian Splatting
🎯 What it does: Propose the SEELE framework, accelerating mobile 3D Gaussian splatting rendering through hybrid preprocessing and contribution-aware rasterization.
SeeThrough3D: Occlusion Aware 3D Control in Text-to-Image Generation
Vaibhav Agrawal (IIIT Hyderabad), Venkatesh Babu Radhakrishnan (IISc Bengaluru)
GenerationData SynthesisSupervised Fine-TuningVision Language ModelDiffusion modelAuto EncoderImageTextBenchmark
🎯 What it does: Propose the SeeThrough3D method, which introduces transparent colorful 3D boxes (OSCR) to achieve occlusion awareness and viewpoint control during text generation, enabling precise layout of multi-object scenes.
SeeU: Seeing the Unseen World via 4D Dynamics-aware Generation
Yu Yuan (Purdue University), Stanley H. Chan (Purdue University)
GenerationData SynthesisGaussian SplattingVideo
🎯 What it does: Proposes the SeeU framework, which learns continuous 4D dynamics from sparse monocular 2D frames through a 2D → 4D → 2D process, and generates visual content for unobserved times and perspectives.
SegCompass: Exploring Interpretable Alignment with Sparse Autoencoders for Enhanced Reasoning Segmentation
Zhenyu Lu (Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences), Yaowei Wang (Peng Cheng Laboratory)
SegmentationExplainability and InterpretabilityTransformerLarge Language ModelReinforcement LearningAuto EncoderMultimodality
🎯 What it does: This paper proposes the SegCompass model, which employs sparse autoencoders to construct interpretable links, aligning language reasoning with visual segmentation.
SegEarth-R2: Towards Comprehensive Language-guided Segmentation for Remote Sensing Images
Zepeng Xin (Xi'an Jiaotong University), Xiangyong Cao (Xi'an Jiaotong University)
SegmentationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: Proposed a new large language-driven remote sensing image segmentation framework called SegEarth-R2, and constructed the LaSeRS dataset covering four dimensions: hierarchical fine-grainedness, target diversity, inference requirements, and linguistic variation
SegGBC: Justifiable Coarse-to-Fine Granular-Ball Computing for Enhancing Clustering Image Segmentation
Qianpeng Chong (Beijing Normal University), Xin Zheng (Beijing Normal University)
SegmentationImageAgriculture Related
🎯 What it does: Proposed an unsupervised image segmentation framework named SegGBC based on granular ball computing;
SegMo: Co-Designing Content-Aware Sparsity and Locally-Cohesive Segment Parallelism for Efficient VLM Inference
Haojuan Li (Shanghai Jiao Tong University), Jiaqi Wang (Shanghai Innovation Institute)
Computational EfficiencyVision Language ModelVideo
🎯 What it does: Proposes the SegMo framework, which jointly optimizes long video VLM inference through content-aware sparsification and localized coherent segment parallelization.
SegMoTE: Token-Level Mixture of Experts for Medical Image Segmentation
Yujie Lu (Sichuan University), Junlong Cheng (Sichuan University)
SegmentationTransformerSupervised Fine-TuningPrompt EngineeringMixture of ExpertsBiomedical Data
🎯 What it does: Propose the SegMoTE framework, introducing token-level expert routing and Progressive Prompt Tokenization based on SAM to achieve efficient adaptation for medical image segmentation;
SegQuant: A Semantics-Aware and Generalizable Quantization Framework for Diffusion Models
Jiaji Zhang (Zhejiang University), Shuiguang Deng (Zhejiang University)
GenerationComputational EfficiencyDiffusion modelImage
🎯 What it does: Propose the SegQuant framework for post-training quantization of diffusion models, including semantic segmentation quantization SegLinear and hardware-friendly dual-scale quantization DualScale;
Select Less, Reason More: Prioritizing Evidence Purity for Video Reasoning
Xuchen Li (Institute of Automation, Chinese Academy of Sciences), Kaiqi Huang (Institute of Automation, Chinese Academy of Sciences)
Computational EfficiencyTransformerSupervised Fine-TuningReinforcement LearningVision Language ModelVideoBenchmark
🎯 What it does: Proposed an evidence priority-based adaptive pixel space video inference framework (EARL), achieving efficient inference for long videos through dynamic frame selection and local resampling.
Select, Hypothesize and Verify: Towards Verified Neuron Concept Interpretation
ZeBin Ji (Chongqing University of Posts and Telecommunications), Bin Xiao (Jinan Inspur Data Technology Co., Ltd.)
Explainability and InterpretabilityConvolutional Neural NetworkTransformerVision Language ModelDiffusion modelContrastive LearningImageText
🎯 What it does: Proposes the SIEVE (Select-Hypothesize-Verify) framework, which first selects high-discriminative activation samples, then generates conceptual hypotheses based on these samples, and verifies the concepts using generated images, ultimately achieving more accurate neuronal concept explanations.
Selection-as-Nonlinearity: Bridging Attention and Activation via a Joint Game-Decision Lens for Interpretable, Discriminative Visual Representations
Sudong Cai (Hong Kong Polytechnic University), Bing Wang (Hong Kong Polytechnic University)
ClassificationObject DetectionExplainability and InterpretabilityTransformerImage
🎯 What it does: Proposes the 'Select-as-Nonlinearity' (SaN) explanation framework to reveal why self-attention performs poorly without FFN, and based on this designs the CSaN mechanism, which enhances attention expressiveness through hierarchical budget calibration and public-private readout while maintaining standard attention regularization.
Selective Amnesia using Contrastive Subnet Erasure for Class Level Unlearning in Vision Models
Vishal Pramanik (University of Florida), Sumit Kumar Jha (University of Florida)
ClassificationRecognitionComputational EfficiencyConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: Proposes a training-free encoder editing method based on Contrastive Subnet Erasure (CSE), which removes specified class concepts in visual models without affecting other model capabilities.
Selective, Regularized, and Calibrated: Harnessing Vision Foundation Models for Cross-Domain Few-Shot Semantic Segmentation
Junyuan Ma (Nanjing University), Yang Gao (Nanjing University)
SegmentationDomain AdaptationRepresentation LearningTransformerSupervised Fine-TuningContrastive LearningImageBiomedical Data
🎯 What it does: A three-stage framework named HERA is designed for cross-domain few-shot semantic segmentation tasks, which utilizes a frozen visual foundation model (VFM) to adapt during testing without accessing source data, requiring only fine-tuning on a few labeled samples.
Selectively Extracting and Injecting Visual Attributes into Text-to-Image Models
Seunghwan Choi (Korea Advanced Institute of Science and Technology), Jaegul Choo (Korea Advanced Institute of Science and Technology)
GenerationTransformerVision Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: Extract attribute-level concepts from a single reference image and inject them into a text-to-image model to achieve controllable generation
SelecTKD: Selective Token-Weighted Knowledge Distillation for LLMs
Haiduo Huang (Xi'an Jiaotong University), Pengju Ren (Xi'an Jiaotong University)
Knowledge DistillationTransformerLarge Language ModelVision Language ModelText
🎯 What it does: Proposed a Selective Token-Weighted Knowledge Distillation (SelecTKD) framework that applies complete distillation loss only to tokens approved by the teacher through a proposal-verification mechanism, thereby enhancing the learning quality of small models.
Self-Attention Driven Tensor Representation for High-Order Data Recovery
Zhi-Wei Shi (Southwest Jiaotong University), Heng-Chao Li (Southwest Jiaotong University)
RestorationTransformerImageVideoMultimodality
🎯 What it does: Developed a self-attention driven tensor representation framework (SADTR) for high-order data recovery tasks
Self-Consistency for LLM-Based Motion Trajectory Generation and Verification
Jiaju Ma (Stanford University), Maneesh Agrawala (Stanford University)
GenerationData SynthesisLarge Language ModelTextBenchmark
🎯 What it does: Proposes extending self-consistency to trajectory generation and verification in motion graphics animation, generating multiple trajectory samples with LLM and selecting the most self-consistent trajectory through geometric transformation clustering.
Self-Corrected Image Generation with Explainable Latent Rewards
Yinyi Luo (Carnegie Mellon University), Shengfeng He (Carnegie Mellon University)
GenerationTransformerLarge Language ModelReinforcement LearningVision Language ModelDiffusion modelAuto EncoderMultimodality
🎯 What it does: Propose xLARD, a lightweight framework that uses interpretable latent rewards for self-correction during text-to-image generation; it inserts residual correctors into the latent space and guides generation using evaluations from multimodal large language models.
Self-Critical Distillation Network for Video-based Commonsense Captioning
Mengqi Yuan (Nanjing University of Posts and Telecommunications), Bing-Kun Bao (Hefei University of Technology)
GenerationKnowledge DistillationTransformerVideoMultimodality
🎯 What it does: Proposed a self-critical distillation network (SCD-Net) for generating commonsense descriptions of videos, addressing the weaknesses of traditional models in visual-semantic alignment and the lack of mutual reasoning among different commonsense categories.
Self-Diffusion Driven Blind Imaging
Yanlong Yang, Guanxiong Luo
RestorationConvolutional Neural NetworkDiffusion modelScore-based ModelImage
🎯 What it does: Proposed a self-supervised blind image restoration framework called DeblurSDI, which requires zero training and zero hyperparameters, capable of simultaneously restoring clear images and unknown blur kernels (PSF) while achieving joint removal of optical distortions and motion blur.
Self-Evaluation Unlocks Any-Step Text-to-Image Generation
Xin Yu (University of Hong Kong), Yotam Nitzan (Adobe Research)
GenerationTransformerScore-based ModelFlow-based ModelImageTextMultimodality
🎯 What it does: Propose a self-assessment text-to-image model named Self-E, which can be trained from scratch and supports arbitrary-step inference.
Self-guided Semantic Inspection for Zero-Shot Composed Image Retrieval
Jingjing Zhang (University of Science and Technology of China), Zhendong Mao (University of Science and Technology of China)
RetrievalVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: Propose the DiffComp framework, which adopts a difference-driven self-supervised approach to address the semantic inconsistency between vision and text in zero-shot compositional image retrieval.
Self-Paced and Self-Corrective Masked Prediction for Movie Trailer Generation
Sidan Zhu, Dixin Luo (Beijing Institute Of Technology)
GenerationTransformerVideo
🎯 What it does: This work proposes a self-paced self-correcting mask prediction framework named SSMP for automated movie trailer generation.
Self-supervised Dynamic Heterogeneous Degradation Modeling for Unified Zero-Shot Image Restoration
XiaoWan Hu (Beihang University), Mai Xu (Beihang University)
RestorationDiffusion modelContrastive LearningImage
🎯 What it does: Proposed a unified physics-informed zero-shot image restoration framework, UP-ZeroIR, which maps various heterogeneous degradations to a low-dimensional unified distribution and performs physically consistent posterior sampling in the latent diffusion space;
SelfHVD: Self-Supervised Handheld Video Deblurring
Honglei Xu (Harbin Institute of Technology), Wangmeng Zuo (Harbin Institute of Technology)
RestorationOptical FlowVideo
🎯 What it does: Proposed a self-supervised handheld video deblurring method called SelfHVD, which utilizes clear frames present in the video as supervision signals;
Selfi: Self-improving Reconstruction Engine via 3D Geometric Feature Alignment
Youming Deng (Cornell University), Lucy Chai (Google)
GenerationData SynthesisPose EstimationDepth EstimationConvolutional Neural NetworkContrastive LearningGaussian SplattingImage
🎯 What it does: Propose Selfi, a self-enhancing pipeline for 3D reconstruction and novel view synthesis from pose-free images;
Semantic Alignment for Pose-Invariant Identity Preserving Diffusion
Jiwon Kim (Korea University), Kyong Hwan Jin (Korea University)
GenerationPose EstimationConvolutional Neural NetworkDiffusion modelImageTextMultimodality
🎯 What it does: This paper proposes SeAl, a zero-shot text-to-image diffusion framework that can precisely align the target identity and structure with text prompts while maintaining the pose unchanged.
Semantic Audio-Visual Navigation in Continuous Environments
Yichen Zeng (Wuhan University), Gongping Huang (Wuhan University)
Robotic IntelligenceTransformerReinforcement LearningVision-Language-Action ModelMultimodality
🎯 What it does: Proposed the SAVN-CE task, i.e., semantic audio-visual navigation in continuous 3D environments, and implemented continuous reasoning and navigation toward targets through MAGNet;
Semantic Context Matters: Improving Conditioning for Autoregressive Models
Dongyang Jin (Alibaba Group), Xiangxiang Chu (Alibaba Group)
GenerationTransformerVision Language ModelImage
🎯 What it does: Propose the SCAR method to achieve efficient semantic context conditional autoregressive image generation and editing.
Semantic Derivative Flow: Graph-Guided Diffusion for Controllable Instance Interactions
Shibin Mei (Huawei), Bingbing Ni (Shanghai Jiao Tong University)
Object DetectionGenerationGraph Neural NetworkDiffusion modelImageText
🎯 What it does: Construct a semantic derivative flow model based on an interaction graph, achieving precise control over the semantics and spatial relationships of multi-instance interactions.
Semantic Foam: Unifying Spatial and Semantic Scene Decomposition
Amr Sharafeldin (Simon Fraser University), Andrea Tagliasacchi (Simon Fraser University)
SegmentationConvolutional Neural NetworkNeural Radiance FieldImage
🎯 What it does: Propose Semantic Foam, combining the voxel segmentation of Radiant Foam with semantic segmentation to construct a directly editable 3D scene representation.
Semantic Noise Reduction via Teacher-Guided Dual-Path Audio-Visual Representation Learning
Linge Wang (Foundation Model Research Center, Institute of Automation, Chinese Academy of Sciences), Jinqiao Wang (Foundation Model Research Center, Institute of Automation, Chinese Academy of Sciences)
RestorationRetrievalRepresentation LearningTransformerAuto EncoderContrastive LearningVideoMultimodalityAudio
🎯 What it does: Propose a teacher-guided dual-path framework TG-DP, which splits audio-visual reconstruction and alignment into two independent forward channels. It uses full-view attention generated by the teacher to guide the mask in the alignment branch and incorporates teacher-student distillation, significantly reducing semantic noise and optimizing interference.
Semantic Scale Space: A Framework for Controllable Image Abstraction
Kazu Mishiba (Tottori University)
GenerationDiffusion modelImage
🎯 What it does: Proposes the Semantic Scale Space framework and its controllable abstraction method AGSS for non-photorealistic rendering, aiming to preserve important structures while suppressing detailed textures.
Semantic-Adaptive Diffusion for Dynamic Spatiotemporal Fusion
Jinsong Zhang (Beijing Normal University), Zhenzhou Shao (Capital Normal University)
RestorationDiffusion modelImageAgriculture Related
🎯 What it does: Propose a dynamic spatiotemporal fusion framework based on semantic adaptive diffusion (SA-STF) for restoring fine-grained images in satellite images with different resolutions and temporal sequences.
Semantic-Guided Global-Local Collaborative Prompt Learning for Few-Shot Class Incremental Learning
Yongxin Yan (Hikvision), Nong Sang (Hikvision)
ClassificationKnowledge DistillationRepresentation LearningMeta LearningTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: Propose a Semantic-Guided Global-Local Collaborative Prompt Learning (SGLC) framework based on Vision-Language Models (VLM) for few-shot class-incremental learning (FSCIL), achieving rapid adaptation to new classes while maintaining memory of old classes.
Semantics Lead the Way: Harmonizing Semantic and Texture Modeling with Asynchronous Latent Diffusion
Yueming Pan (Xi'an Jiaotong University), Nanning Zheng (Xi'an Jiaotong University)
GenerationTransformerDiffusion modelAuto EncoderImageOrdinary Differential Equation
🎯 What it does: Proposes Semantic-First Diffusion (SFD), which in a three-stage asynchronous denoising process first generates semantic latent representations and then guides texture generation, significantly accelerating training and improving image quality.
SemanticVLA: Towards Semantic Reasoning over Action Memorization via Synergistic Explicit Trace and Latent Action Planning
Fei Ni (Imperial College London), Stefanos Zafeiriou (Imperial College London)
Robotic IntelligenceTransformerVision-Language-Action ModelFlow-based ModelAuto EncoderMultimodality
🎯 What it does: Propose the SemanticVLA framework, combining explicit trajectory reasoning and implicit latent action planning, leveraging VLM's spatial reasoning to achieve semantic understanding and execution of task instructions;
Semi-Supervised Conformal Prediction With Unlabeled Nonconformity Score
Xuanning Zhou (Southern University of Science and Technology), Hongxin Wei (Southern University of Science and Technology)
ClassificationImage
🎯 What it does: Propose SemiCP, a semi-supervised consistency prediction framework that leverages labeled and unlabeled data for calibration;
Semi-supervised Echocardiography Video Segmentation via Anchor Semantic Awareness and Continuous Pseudo-label Reforging
Yunpeng Fang (Shenzhen University), Jing Qin (Shenzhen University)
SegmentationConvolutional Neural NetworkTransformerVideoBiomedical DataUltrasound
🎯 What it does: This paper proposes EchoForge, a semi-supervised cardiac ultrasound video segmentation framework, which achieves accurate real-time segmentation of ventricular structures with minimal annotations through two modules: Anchor Semantic Awareness (ASA) and Continuous Pseudo-label Reforging (CPR).
SemiGDA: Generative Dual-distribution Alignment for Semi-Supervised Medical Image Segmentation
Kaiwen Huang (Nanjing University of Science and Technology), Tao Zhou (Nanjing University of Science and Technology)
SegmentationDiffusion modelAuto EncoderGenerative Adversarial NetworkBiomedical DataUltrasound
🎯 What it does: Propose a semi-supervised medical image segmentation framework called SemiGDA based on generative adversarial distribution alignment. It uses dual encoders to align the latent distributions of images and segmentation masks, and enhances multi-scale semantic consistency through a jump adapter, significantly improving segmentation performance in low-annotation scenarios.
SemLayer: Semantic-aware Generative Segmentation and Layer Construction for Abstract Icons
Haiyang Xu (University Of San Diego), Zhaowen Wang (Adobe Research)
RestorationSegmentationGenerationOptimizationVision Language ModelDiffusion modelImageText
🎯 What it does: To address the semantic layer reconstruction problem in flat vector icons, SemLayer restores editable hierarchical structures through three steps: generative segmentation, semi-modal completion, and hierarchical sorting.
SemLT3D: Semantic-Guided Expert Distillation for Camera-only Long-Tailed 3D Object Detection
Hao Vo (University of Arkansas), Ngan Le (University of Arkansas)
Object DetectionAutonomous DrivingKnowledge DistillationTransformerMixture of ExpertsVision Language ModelContrastive LearningImage
🎯 What it does: Propose the SemLT3D framework to address the long-tail imbalance, class confusion, and intra-class diversity issues in single-camera 3D object detection.
SemVideo: Reconstructs What You Watch from Brain Activity via Hierarchical Semantic Guidance
Minghan Yang, Yi-Zhe Song (Beijing University Of Posts And Telecommunications)
TransformerLarge Language ModelVision Language ModelDiffusion modelVideoTextBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This study proposes SemVideo, a hierarchical semantic guidance-based fMRI-to-video reconstruction framework.
SenCache: Accelerating Diffusion Model Inference via Sensitivity-Aware Caching
Yasaman Haghighi (Ecole Polytechnique Federale De Lausanne), Alexandre Alahi (Ecole Polytechnique Federale De Lausanne)
Computational EfficiencyDiffusion modelVideoOrdinary Differential Equation
🎯 What it does: Propose a dynamic caching framework called SenCache, which determines whether to cache based on the local sensitivity of the network to input perturbations (noise latent variables and time steps), to accelerate diffusion model inference.
SenseSearch: Empowering Vision-Language Models with High-Resolution Agentic Search-Reasoning via Reinforcement Learning
Yong Xien Chng (SenseTime Research), Lewei Lu (SenseTime Research)
RetrievalSupervised Fine-TuningReinforcement LearningAgentic AIVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: Built SenseSearch, an agentic visual-language model capable of adaptively invoking image search, text search, and image cropping tools during multi-round reasoning.
Sensor2Sensor: Cross-Embodiment Sensor Conversion for Autonomous Driving
Jiahao Wang (Waymo), Chiyu Max Jiang (Waymo)
GenerationData SynthesisAutonomous DrivingDiffusion modelAuto EncoderGaussian SplattingImageVideoPoint Cloud
🎯 What it does: Convert monocular dashcam videos from the natural world into high-fidelity multi-modal sensor logs for specific vehicle models (multi-view cameras + LiDAR)
SFR-Net: Steering-Fusion-Refining Network in Multi-label Zero-Shot Sewer Defect Detection
Zhao-Min Chen (Wenzhou University), Yu Li (Zhejiang College of Security Technology)
Domain AdaptationAnomaly DetectionConvolutional Neural NetworkGraph Neural NetworkLarge Language ModelSupervised Fine-TuningVision Language ModelImage
🎯 What it does: Propose a three-stage Steering-Fusion-Refining network (SFR-Net) for multi-label zero-shot sewage pipeline defect detection.
SG-LoRA: Semantic-guided LoRA Parameters Generation
Miaoge Li (Hong Kong Polytechnic University), Jingcai Guo (Hong Kong Polytechnic University)
RetrievalVision Language ModelAuto EncoderImageText
🎯 What it does: Automatically generate LoRA weights for any new task without requiring any target task data by leveraging task descriptions through a semantic-guided conditional generative model;
SGAD-SLAM: Splatting Gaussians at Adjusted Depth for Better Radiance Fields in RGBD SLAM
Pengchong Hu (Wayne State University), Zhizhong Han (Wayne State University)
Neural Radiance FieldGaussian SplattingSimultaneous Localization and MappingImage
🎯 What it does: Propose SGAD-SLAM, which uses pixel-aligned Gaussian models in RGB-D SLAM, allowing each Gaussian to adjust depth along the line of sight to improve rendering quality and scalability; simultaneously, efficient and robust camera pose estimation is achieved through geometric similarity and GICP.
SGDE: Self-supervised Geometry Degradation Estimation Framework for Coded Aperture Compressive Spectral Imaging
Yuqiao He (Hunan University), Wenbin He (Hunan University)
RestorationOptimizationComputational EfficiencyConvolutional Neural NetworkImagePhysics Related
🎯 What it does: A self-supervised geometric degradation estimation framework named SGDE is studied for correcting reconstruction distortions caused by mask mismatch in CASSI systems.
SGDrive: Scene-to-Goal Hierarchical World Cognition for Autonomous Driving
Jingyu Li (Fudan University), Li Zhang (Fudan University)
Autonomous DrivingTransformerVision Language ModelDiffusion modelImage
🎯 What it does: Propose the SGDrive framework, which leverages pre-trained vision-language models (VLM) to explicitly learn and predict driving-related world knowledge through scene-agent-goal hierarchical queries, then generates safe trajectories using Diffusion Transformer.
SGI: Structured 2D Gaussians for Efficient and Compact Large Image Representation
Zixuan Pan (Tohoku University), Yiyu Shi (University of Notre Dame)
CompressionComputational EfficiencyRepresentation LearningGaussian SplattingImage
🎯 What it does: Propose Structured 2D Gaussians (SGI), generating local 2D Gaussian primitives through multi-scale seed regions and lightweight MLP to achieve efficient and compact representation of high-resolution images.
SGS-Intrinsic: Semantic-Invariant Gaussian Splatting for Sparse-View Indoor Inverse Rendering
Jiahao Niu (Sun Yat-sen University), Qing Zhang (Sun Yat-sen University)
GenerationDiffusion modelGaussian SplattingImage
🎯 What it does: Propose a two-stage framework named SGS-Intrinsic for sparse perspective indoor inverse rendering.
SGSoft: Learning Fused Semantic-Geometric Features for 3D Shape Correspondence via Template-Guided Soft Signals
Soyeon Yoon (KAIST AI), Hyunjung Shim (KAIST AI)
Pose EstimationRetrievalTransformerContrastive LearningMesh
🎯 What it does: This paper proposes SGSoft, a template-guided soft signal framework for efficient and general-purpose dense 3D shape correspondence.
ShadowDraw: From Any Object to Shadow-Drawing Compositional Art
Rundong Luo (Cornell University), Wei-Chiu Ma (Cornell University)
GenerationTransformerPrompt EngineeringVision Language ModelDiffusion modelFlow-based ModelImageTextPoint CloudMesh
🎯 What it does: This paper proposes a method for generating shadow painting art from arbitrary 3D objects. The system simultaneously predicts scene parameters (lighting direction, object pose) and partial line drawings, enabling projected shadows to complete the entire image.
SHands: A Multi-View Dataset and Benchmark for Surgical Hand-Gesture and Error Recognition Toward Medical Training
Le Ma (MIRALab), Katarzyna Wac (Quality Of Life Technologies Lab, University Of Geneva)
RecognitionTransformerVideoBiomedical DataBenchmark
🎯 What it does: This paper proposes the SHANDS dataset and constructs a benchmark experiment for multi-view gesture and error recognition on this dataset.
Shape-of-You: Fused Gromov-Wasserstein Optimal Transport for Semantic Correspondence in-the-Wild
Jiin Im (Hanyang University), Je Hyeong Hong (Hanyang University)
RecognitionOptimizationImagePoint Cloud
🎯 What it does: This paper proposes a semantic correspondence framework called Shape-of-You based on fused Gromov-Wasserstein (FGW) optimal transport, achieving pixel-level semantic correspondence in wild images without explicit geometric annotations.
SHAPE: Structure-aware Hierarchical Unsupervised Domain Adaptation with Plausibility Evaluation for Medical Image Segmentation
Linkuan Zhou (Northwestern Polytechnical University), Qiangguo Jin (Macao Polytechnic University)
SegmentationDomain AdaptationTransformerMultimodalityBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: This paper proposes the SHAPE framework, achieving unsupervised domain adaptation for medical image segmentation. By generating high-quality pseudo-labels through structure-aware hierarchical feature modulation (HFM) and hierarchical feasibility evaluation (HPE+SAP), it significantly enhances cross-modal segmentation performance.
ShapeAR: Generating Editable Shape Layers via Autoregressive Diffusion
Souymodip Chakraborty (Adobe Inc.), Ankit Phogat (Adobe Inc.)
GenerationTransformerDiffusion modelAuto EncoderImageText
🎯 What it does: Propose ShapeAR, an autoregressive latent diffusion framework that decomposes raster images into editable RGBA shape layers;
ShapeR: Robust Conditional 3D Shape Generation from Casual Captures
Yawar Siddiqui (Meta Reality Labs Research), Jakob Engel (Meta Reality Labs Research)
Object DetectionSegmentationGenerationData SynthesisPose EstimationTransformerVision Language ModelFlow-based ModelRectified FlowAuto EncoderSimultaneous Localization and MappingImageTextMultimodalityPoint CloudMeshBenchmark
🎯 What it does: Leverages multimodal inputs (sparse SLAM point clouds, calibrated multi-view images, and text descriptions generated by VLM) along with a flow-matching based Transformer to automatically generate high-quality, metric-consistent 3D object shapes and reconstruct complete scenes by stitching individual objects.
SHARP: Short-Window Streaming for Accurate and Robust Prediction in Motion Forecasting
Alexander Prutsch (Graz University of Technology), Horst Possegger (Graz University of Technology)
Autonomous DrivingComputational EfficiencyTransformerTime SeriesSequential
🎯 What it does: Proposed the SHARP framework to achieve short-window streaming motion prediction, capable of continuously updating and accurately predicting future trajectories in dynamic scenes.
SharpTimeGS: Sharp and Stable Dynamic Gaussian Splatting via Lifespan Modulation
Zhanfeng Liao (Tsinghua University), Yebin Liu (Beijing Normal University)
GenerationGaussian SplattingOptical FlowBenchmark
🎯 What it does: This paper proposes SharpTimeGS, a 4D Gaussian representation based on a learnable lifetime, capable of uniformly handling visibility and motion in both static and dynamic scenes.
Shedding Light on VLN Robustness: A Black-box Framework for Indoor Lighting-based Adversarial Attack
Chenyang Li (Nanyang Technological University), Yang Liu (Nanyang Technological University)
Adversarial AttackMultimodality
🎯 What it does: Propose a black-box adversarial attack framework based on indoor lighting (ILA), which adjusts global illumination intensity or switches lights on/off at critical moments to evaluate and undermine the robustness of vision-and-language navigation (VLN) models.
ShelfOcc: Native 3D Supervision beyond LiDAR for Vision-Based Occupancy Estimation
Simon Boeder (Bosch Research), Benjamin Risse (University of Münster)
Autonomous DrivingVision Language ModelImage
🎯 What it does: Propose the ShelfOcc framework, which generates high-quality 3D pseudo labels using 3D geometry and semantic foundation models, directly supervising the occupancy network in native 3D voxel space without requiring LiDAR
ShiftLUT: Spatial Shift Enhanced Look-Up Tables for Efficient Image Restoration
Xiaolong Zeng (Tsinghua University), Bin Wang (Tsinghua University)
RestorationImage
🎯 What it does: Propose the ShiftLUT framework, integrating Learnable Spatial Shift, asymmetric dual-branch architecture, and Error-bounded Adaptive Sampling (EAS) LUT compression to achieve efficient image restoration.
Shoe Style-Invariant and Ground-Aware Learning for Dense Foot Contact Estimation
Daniel Sungho Jung, Kyoung Mu Lee (Seoul National University)
Pose EstimationConvolutional Neural NetworkTransformerImageMultimodality
🎯 What it does: Propose the FECO framework, which can predict fine-grained foot contact points from a single RGB image, and overcome the challenges of shoe diversity and insufficient ground information through shoe style content randomization and ground perception learning.
ShotDirector: Directorially Controllable Multi-Shot Video Generation with Cinematographic Transitions
Xiaoxue Wu (Fudan University), Yu Qiao (Shanghai Artificial Intelligence Laboratory)
GenerationTransformerDiffusion modelVideo
🎯 What it does: Proposes the ShotDirector framework, which can generate multi-shot videos based on director instructions and achieve controllable cinematic transitions.
SHOW3D: Capturing Scenes of 3D Hands and Objects in the Wild
Patrick Rim, Kun He (Yale University)
SegmentationPose EstimationTransformerVision Language ModelVideoTextMeshBenchmark
🎯 What it does: This work proposes the SHOW3D dataset and designs a lightweight wearable multi-camera system capable of capturing hand-object interaction actions in real-world environments. It achieves unmarked 3D annotation (hand pose, object pose, segmentation, contact points, and text descriptions) through a self-developed ego-exo fusion pipeline.
ShowTable: Unlocking Creative Table Visualization with Collaborative Reflection and Refinement
Zhihang Liu (University of Science and Technology of China), Hongtao Xie (Tongyi Lab)
GenerationLarge Language ModelReinforcement LearningDiffusion modelMultimodalityTabularBenchmark
🎯 What it does: Proposed a novel creative table visualization task, developing the ShowTable multimodal pipeline to generate high-fidelity infographics from tables
ShowUI-p: Flow-based Generative Models as GUI Dexterous Hands
Siyuan Hu (National University of Singapore), Mike Zheng Shou (National University of Singapore)
GenerationData SynthesisTransformerVision Language ModelFlow-based ModelTime SeriesSequentialBenchmark
🎯 What it does: Propose ShowUIπ, a lightweight model based on flow-based generation, for real-time generation of continuous dragging trajectories in GUI interactions.
ShreddingNet: Coarse-to-Fine Restoration for Multi-Source Shredded Manuscripts
Haoyang Cui (Peking University), Yadong Mu (Peking University)
RestorationData SynthesisConvolutional Neural NetworkGraph Neural NetworkImage
🎯 What it does: Proposed ShreddingNet, a two-stage coarse-to-fine multi-source fragment stitching network, which can automatically reassemble mixed-source fragments without requiring prior conditions.