CVPR 2025 Papers — Page 13
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2871 papers
HOTFormerLoc: Hierarchical Octree Transformer for Versatile Lidar Place Recognition Across Ground and Aerial Views
Ethan Griffiths (CSIRO Robotics Data61), Milad Ramezani (CSIRO Robotics Data61)
RecognitionPose EstimationTransformerPoint Cloud
🎯 What it does: A hierarchical octree-based Transformer (HOTFormerLoc) is proposed for point cloud pose recognition across viewpoints (ground and aerial) and various environments (urban, forest).
HotSpot: Signed Distance Function Optimization with an Asymptotically Sufficient Condition
Zimo Wang, Tzu-Mao Li
OptimizationPoint Cloud
🎯 What it does: We propose HOTSPOT, an optimization method for neural signature distance functions based on the screened Poisson equation, which utilizes heat loss to achieve asymptotic sufficient conditions for the true distance function and significantly improves the accuracy of surface reconstruction and distance field approximation.
HoVLE: Unleashing the Power of Monolithic Vision-Language Models with Holistic Vision-Language Embedding
Chenxin Tao (Tsinghua University), Jifeng Dai (Tsinghua University)
Knowledge DistillationRepresentation LearningTransformerLarge Language ModelVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: A single-volume visual language model, HoVLE, has been developed, which uses a holistic embedding module to project images and text into a shared space, allowing the LLM to process images as it does text.
How Do I Do That? Synthesizing 3D Hand Motion and Contacts for Everyday Interactions
Aditya Prakash (University of Illinois Urbana-Champaign), Harpreet Sawhney (Microsoft)
Data SynthesisPose EstimationTransformerAuto EncoderImageVideo
🎯 What it does: Predict future 3D hand poses and contact maps (interaction trajectories) from a single RGB image, action text, and 3D contact points.
How to Merge Your Multimodal Models Over Time?
Sebastian Dziadzio (Tuebingen AI Center, University of Tuebingen), Matthias Bethge (Tuebingen AI Center, University of Tuebingen)
Vision Language ModelMultimodality
🎯 What it does: Proposed the TIME framework for time series merging of continuously appearing expert models and systematically evaluated different initialization, deployment, and merging techniques.
HRAvatar: High-Quality and Relightable Gaussian Head Avatar
Dongbin Zhang (Tsinghua University), Haoqian Wang (Tsinghua University)
GenerationPose EstimationGaussian SplattingVideo
🎯 What it does: Reconstructing high-quality, animatable, and re-lightable 3D head avatars from monocular videos
HSI-GPT: A General-Purpose Large Scene-Motion-Language Model for Human Scene Interaction
Yuan Wang (Tsinghua University), Shengjin Wang (Tsinghua University)
GenerationPose EstimationTransformerLarge Language ModelSupervised Fine-TuningMultimodalityPoint Cloud
🎯 What it does: A general-purpose large scene-action-language model named HSI-GPT is proposed for generating human actions that interact with 3D scenes under various control conditions and for completing understanding and prediction tasks.
HSI: A Holistic Style Injector for Arbitrary Style Transfer
Shuhao Zhang (Jilin University), Hongjuan Li (Jilin University)
Image TranslationConvolutional Neural NetworkAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: Proposes the Holistic Style Injector (HSI) module, achieving arbitrary style transfer with global style injection.
Human Motion Instruction Tuning
Lei Li (University of Copenhagen), Jenq-Neng Hwang (University of Washington)
RecognitionPose EstimationTransformerLarge Language ModelVideoMultimodality
🎯 What it does: The LLaMo framework is proposed, which implements the direct input of motion data as an independent modality into large language models for instruction tuning.
Human-centered Interactive Learning via MLLMs for Text-to-Image Person Re-identification
Yang Qin (Sichuan University), Peng Hu (Sichuan University)
RecognitionRetrievalTransformerLarge Language ModelVision Language ModelImageTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: This paper proposes an interactive text-image person re-identification (TIReID) framework called ICL, which utilizes a multimodal large language model (MLLM) to achieve multi-turn question answering for refining query text, and combines data augmentation RDA to enhance the model's ability to discriminate fine-grained descriptions.
HumanDreamer: Generating Controllable Human-Motion Videos via Decoupled Generation
Boyuan Wang (Institute of Automation, Chinese Academy of Sciences), Xingang Wang (Institute of Automation, Chinese Academy of Sciences)
GenerationData SynthesisPose EstimationTransformerDiffusion modelVideoText
🎯 What it does: This paper proposes the HumanDreamer framework, which separates text-to-pose generation and pose-to-video generation, and constructs the MotionVid dataset with 1.2 million pairs to train MotionDiT for high-quality text-controllable human motion video generation.
HumanMM: Global Human Motion Recovery from Multi-shot Videos
Yuhong Zhang (Tsinghua University), Lei Zhang (IDEA Research)
Pose EstimationTransformerSimultaneous Localization and MappingVideo
🎯 What it does: Proposed the HumanMM framework to recover long sequences of 3D human motion from multi-camera videos and align them to a world coordinate system.
HumanRig: Learning Automatic Rigging for Humanoid Character in a Large Scale Dataset
Zedong Chu (Alibaba Group), Mu Xu (Alibaba Group)
GenerationData SynthesisPose EstimationTransformerMesh
🎯 What it does: A large-scale AI-generated 3D humanoid character dataset called HumanRig is proposed, and an automatic rigging framework is built based on this dataset.
HuMoCon: Concept Discovery for Human Motion Understanding
Qihang Fang (University of Hong Kong), Yanchao Yang (University of Hong Kong)
RecognitionPose EstimationTransformerLarge Language ModelAuto EncoderVideoMultimodality
🎯 What it does: Developed the HuMoCon framework, which utilizes a multimodal encoder for joint learning of video and motion data, mining human behavior concepts and supporting question-and-answer-based understanding.
HUNet: Homotopy Unfolding Network for Image Compressive Sensing
Feiyang Shen (Northwestern Polytechnical University), Hongping Gan (Northwestern Polytechnical University)
RestorationCompressionTransformerImage
🎯 What it does: A deep unfolding network HUNet based on homotopy algorithms is proposed for image compressed sensing reconstruction.
HunyuanPortrait: Implicit Condition Control for Enhanced Portrait Animation
Zunnan Xu (Tsinghua University), Qinglin Lu (Hunyuan)
GenerationData SynthesisDiffusion modelImageVideo
🎯 What it does: This paper proposes an implicit conditional control framework called HunyuanPortrait based on stable video diffusion, which achieves high-fidelity and controllable portrait animation using a single portrait image and a driving video.
HuPerFlow: A Comprehensive Benchmark for Human vs. Machine Motion Estimation Comparison
Yung-Hao Yang (Kyoto University), Shin'ya Nishida (Nippon Telegraph and Telephone Corporation)
Autonomous DrivingOptical FlowVideoBenchmark
🎯 What it does: This paper constructs the HuPerFlow benchmark, collecting 38,400 human-perceived optical flow positions on ten commonly used optical flow datasets using an online psychological experiment method;
HUSH: Holistic Panoramic 3D Scene Understanding using Spherical Harmonics
Jongsung Lee (Ulsan National Institute of Science and Technology), Kyungdon Joo (Ulsan National Institute of Science and Technology)
SegmentationDepth EstimationTransformerImage
🎯 What it does: Proposes the HUSH framework, which utilizes spherical harmonics (SH) for unified 3D scene understanding of 360° panoramic images, supporting multi-tasks such as depth, normals, and indoor layout.
HVI: A New Color Space for Low-light Image Enhancement
Qingsen Yan (Northwestern Polytechnical University), Yanning Zhang (Northwestern Polytechnical University)
RestorationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A new HVI color space based on polarized HS plane and a learnable intensity collapse function is proposed, along with a corresponding dual-branch CIDNet network for low-light image enhancement.
Hybrid Concept Bottleneck Models
Yang Liu (University of Electronic Science and Technology of China), Shi Gu (University of Electronic Science and Technology of China)
ClassificationExplainability and InterpretabilityLarge Language ModelContrastive LearningImageMultimodality
🎯 What it does: Proposes HybridCBM, which combines static and dynamic concept buckets to achieve an interpretable concept bottleneck model.
Hybrid Global-Local Representation with Augmented Spatial Guidance for Zero-Shot Referring Image Segmentation
Ting Liu (Northwestern Polytechnical University), Siyuan Li (Northwestern Polytechnical University)
SegmentationContrastive LearningImage
🎯 What it does: A training-independent method for zero-shot referential image segmentation is proposed, which combines global-local feature extraction with multi-spatial guidance, utilizing SAM to generate masks and aligning features through the CLIP visual/text encoder.
Hybrid Reciprocal Transformer with Triplet Feature Alignment for Scene Graph Generation
Jiawei Fu (Chinese University of Hong Kong), Qi Dou (Chinese University of Hong Kong)
Object DetectionGenerationTransformerImage
🎯 What it does: This paper proposes a Hybrid Reciprocal Transformer for scene graph generation, which jointly learns triplet mask features with component features of subject, object, and predicate in an end-to-end single-stage framework, addressing the difficulty of triplet feature alignment caused by multi-role objects.
Hybrid-Level Instruction Injection for Video Token Compression in Multi-modal Large Language Models
Zhihang Liu (University of Science and Technology of China), Hongtao Xie (University of Science and Technology of China)
RecognitionCompressionTransformerLarge Language ModelPrompt EngineeringVideoMultimodality
🎯 What it does: A dual-layer instruction injection conditional compression method called HICom is proposed for the video understanding tasks of multimodal large language models, which significantly reduces visual tokens while retaining information relevant to user instructions.
HybridGS: Decoupling Transients and Statics with 2D and 3D Gaussian Splatting
Jingyu Lin (University of Science and Technology of China), Jieping Ye (Independent Researcher)
SegmentationGenerationGaussian SplattingImage
🎯 What it does: This paper proposes HybridGS, a hybrid Gaussian representation that uses 2D Gaussians for single-view modeling of transient objects and 3D Gaussians for multi-view consistent modeling of static scenes, thereby achieving separation and synthesis of transient and static scenes.
HybridMQA: Exploring Geometry-Texture Interactions for Colored Mesh Quality Assessment
Armin Shafiee Sarvestani (University of Waterloo), Zhou Wang (University of Waterloo)
Graph Neural NetworkTransformerMesh
🎯 What it does: This paper proposes a hybrid model-based and projection-based full-reference 3D mesh quality assessment framework called HybridMQA.
Hyperbolic Category Discovery
Yuanpei Liu (Visual AI Lab, University of Hong Kong), Kai Han (Visual AI Lab, University of Hong Kong)
ClassificationRepresentation LearningContrastive LearningImage
🎯 What it does: The HypCD framework is proposed, which migrates General Category Discovery (GCD) to hyperbolic space to better capture the hierarchical structure of samples.
Hyperbolic Safety-Aware Vision-Language Models
Tobia Poppi (University of Modena and Reggio Emilia), Rita Cucchiara (Istituto Italiano di Tecnologia)
RetrievalSafty and PrivacyVision Language ModelContrastive LearningImageText
🎯 What it does: Construct a hierarchical relationship between safe and unsafe content in hyperbolic space to achieve safety-aware CLIP (HySAC), supporting safe retrieval and controllable traversal.
Hyperbolic Uncertainty-Aware Few-Shot Incremental Point Cloud Segmentation
Tanuj Sur (Chennai Mathematical Institute), Biplab Banerjee (Indian Institute of Technology Bombay)
RecognitionSegmentationPoint Cloud
🎯 What it does: A Hyperbolic Ideal Prototypes Optimization (HIPO) framework is proposed for few-shot incremental semantic segmentation of 3D point clouds.
Hyperdimensional Uncertainty Quantification for Multimodal Uncertainty Fusion in Autonomous Vehicles Perception
Luke Chen (University of California Irvine), Mohammad Abdullah Al Faruque (University of California Irvine)
Object DetectionSegmentationAutonomous DrivingComputational EfficiencyMultimodality
🎯 What it does: A deterministic uncertainty quantification method based on hyperdimensional computing, HyperDUM, is proposed, which can efficiently assess the feature uncertainty of each modality before multimodal feature fusion and weight the fused features.
HyperFree: A Channel-adaptive and Tuning-free Foundation Model for Hyperspectral Remote Sensing Imagery
Jingtao Li (Wuhan University), Yanfei Zhong (Wuhan University)
ClassificationObject DetectionSegmentationAnomaly DetectionPrompt EngineeringImage
🎯 What it does: A hyperspectral remote sensing foundation model called HyperFree is proposed, which can directly process hyperspectral images with different numbers of bands and achieve multi-task inference through single-point prompts.
HyperGLM: HyperGraph for Video Scene Graph Generation and Anticipation
Trong-Thuan Nguyen (University of Arkansas), Khoa Luu (University of Arkansas)
Object DetectionGenerationGraph Neural NetworkTransformerLarge Language ModelVideoGraph
🎯 What it does: A framework is proposed that unifies entity scene graphs and program graphs into a hypergraph and injects it into a large language model for video scene graph generation, prediction, and reasoning.
Hypergraph Vision Transformers: Images are More than Nodes, More than Edges
Joshua Fixelle (University of Virginia)
ClassificationRetrievalTransformerImage
🎯 What it does: This paper proposes the Hypergraph Vision Transformer (HgVT), which integrates a hierarchical bipartite hypergraph structure into the visual transformer to capture high-order semantic relationships without using clustering algorithms.
HyperGS: Hyperspectral 3D Gaussian Splatting
Christopher Thirgood (University of Surrey), Simon Hadfield (University of Surrey)
GenerationData SynthesisCompressionConvolutional Neural NetworkNeural Radiance FieldAuto EncoderGaussian SplattingImage
🎯 What it does: The HyperGS framework is proposed, which uses 3D Gaussian splatting to achieve hyperspectral novel view synthesis in latent space.
HyperLoRA: Parameter-Efficient Adaptive Generation for Portrait Synthesis
Mengtian Li (Intelligent Creation, ByteDance), Qian He (Intelligent Creation, ByteDance)
GenerationData SynthesisTransformerSupervised Fine-TuningImage
🎯 What it does: Proposes HyperLoRA, a method for zero-shot personalized portrait synthesis by generating LoRA weights;
HyperNet Fields: Efficiently Training Hypernetworks without Ground Truth by Learning Weight Trajectories
Eric Hedlin (University of British Columbia), Shweta Mahajan (Qualcomm AI Research)
GenerationOptimizationTransformerDiffusion modelImagePoint Cloud
🎯 What it does: Learn the hypernetwork field to estimate the complete optimization trajectory of task network weights without pre-computing converged weights.
HyperNVD: Accelerating Neural Video Decomposition via Hypernetworks
Maria Pilligua (Universitat Autonoma de Barcelona), Javier Vazquez-Corral (Universitat Autonoma de Barcelona)
Computational EfficiencyMeta LearningOptical FlowVideo
🎯 What it does: A meta-learning framework based on hypernetworks is designed, utilizing VideoMAE embeddings to generate parameters for an implicit neural video decomposition model, thereby achieving rapid adaptation and efficient hierarchical decomposition for new videos.
HyperPose: Hypernetwork-Infused Camera Pose Localization and an Extended Cambridge Landmarks Dataset
Ron Ferens (Bar Ilan University), Yosi Keller (Bar Ilan University)
Pose EstimationDomain AdaptationConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes HyperPose, which reduces the domain gap between training and testing and improves localization accuracy by introducing hypernetwork dynamically generated regression head weights in absolute camera pose regression (APR).
HyperSeg: Hybrid Segmentation Assistant with Fine-grained Visual Perceiver
Cong Wei (Tsinghua University), Yujiu Yang (Meituan Inc.)
Object DetectionSegmentationTransformerLarge Language ModelVision Language ModelImageVideo
🎯 What it does: We propose HyperSeg, a general pixel-level image and video segmentation framework based on Visual Large Language Models (VLLM), capable of handling a variety of tasks from conventional segmentation to complex reasoning segmentation.
Hyperspectral Pansharpening via Diffusion Models with Iteratively Zero-Shot Guidance
Jin-Liang Xiao (University of Electronic Science and Technology of China), Qibin Zhao (University of Electronic Science and Technology of China)
RestorationSuper ResolutionDiffusion modelImage
🎯 What it does: This paper proposes a hyperspectral super-resolution fusion method based on a zero-shot guided diffusion model and neural spatial-spectral decomposition.
I2VGuard: Safeguarding Images against Misuse in Diffusion-based Image-to-Video Models
Dongnan Gui (Microsoft Research), Yan Lu (Microsoft Research)
GenerationAdversarial AttackDiffusion modelContrastive LearningImageVideo
🎯 What it does: This paper proposes an adversarial protection method named I2VGuard, which adds imperceptible perturbations to input images, causing a decline in the quality of animated videos generated by image-to-video (I2V) generation models based on diffusion models, thereby preventing malicious use of images.
IAAO: Interactive Affordance Learning for Articulated Objects in 3D Environments
Can Zhang (National University of Singapore), Gim Hee Lee (National University of Singapore)
Object DetectionRobotic IntelligenceContrastive LearningGaussian SplattingPoint Cloud
🎯 What it does: The IAAO framework is proposed, utilizing 3D Gaussian Splatting and large foundational models (SAM, MaskCLIP, DINOv2, CLIP) to construct an explicit 3D high-dimensional feature field, supporting operability (affordance) through interactive learning and reconstructing joint movements.
ICE: Intrinsic Concept Extraction from a Single Image via Diffusion Models
Fernando Julio Cendra (University of Hong Kong), Kai Han (University of Hong Kong)
SegmentationGenerationRetrievalDiffusion modelImage
🎯 What it does: This paper proposes a two-stage framework called ICE (Intrinsic Concept Extraction) that can automatically extract and decompose object-level concepts and their intrinsic attributes (such as color, material, etc.) using only a single image and a pre-trained text-image diffusion model.
IceDiff: High Resolution and High-Quality Arctic Sea Ice Forecasting with Generative Diffusion Prior
Jingyi Xu (Fudan University), Lei Bai (Fudan University)
GenerationData SynthesisTransformerDiffusion modelTime Series
🎯 What it does: The IceDiff framework is proposed, which first uses a Vision Transformer to generate sea ice concentration predictions at a 25 km level, and then employs a pre-trained unconditional diffusion model and zero-shot sampling strategy to generate high-resolution sea ice concentration maps at a 6.25 km level.
ICP: Immediate Compensation Pruning for Mid-to-high Sparsity
Xin Luo (University of Science and Technology of China), S. Kevin Zhou (University of Science and Technology of China)
TransformerLarge Language ModelImageTextMultimodality
🎯 What it does: A sparse method suitable for large language and vision models, called ICP, is designed to achieve performance improvement at medium to high sparsity through block-level error compensation and sparsity reallocation.
ICT: Image-Object Cross-Level Trusted Intervention for Mitigating Object Hallucination in Large Vision-Language Models
Junzhe Chen (Tsinghua University), Xuming Hu (Hong Kong University of Science and Technology)
RecognitionObject DetectionComputational EfficiencyTransformerVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes a training-free, plug-and-play intervention method called ICT, which enhances visual attention to overall scenes and fine-grained objects by adjusting the activation values of attention heads during the forward propagation phase of LVLM, reducing object misreporting caused by the model's language bias.
ID-Patch: Robust ID Association for Group Photo Personalization
Yimeng Zhang (ByteDance Inc.), Linjie Luo (ByteDance Inc.)
GenerationData SynthesisSafty and PrivacyDiffusion modelImage
🎯 What it does: This paper proposes the ID-Patch method for achieving precise identity localization and identity preservation in multi-identity group photo generation, avoiding identity leakage.
IDEA-Bench: How Far are Generative Models from Professional Designing?
Chen Liang (Institute of Automation, Chinese Academy of Sciences), Yu Liu (Alibaba Group)
GenerationTransformerLarge Language ModelPrompt EngineeringImageTextMultimodalityBenchmark
🎯 What it does: They created IDEA-Bench, a comprehensive benchmark that includes 100 professional image generation tasks, 275 cases, and 1,650 binary evaluation questions to assess the performance of generative models in professional design tasks.
IDEA: Inverted Text with Cooperative Deformable Aggregation for Multi-modal Object Re-Identification
Yuhao Wang (Dalian University of Technology), Huchuan Lu (Dalian University of Technology)
RecognitionRetrievalTransformerLarge Language ModelContrastive LearningTextMultimodality
🎯 What it does: Construct three sets of text-enhanced multimodal re-identification datasets and propose a text-guided multimodal feature learning framework called IDEA;
Identifying and Mitigating Position Bias of Multi-image Vision-Language Models
Xinyu Tian (Australian National University), Jing Zhang (Australian National University)
RecognitionOptimizationTransformerVision Language ModelImageMultimodalityBenchmark
🎯 What it does: Analyze and mitigate the position bias problem of large visual language models (LVLM) in multi-image reasoning.
Identifying and Mitigating Spurious Correlation in Multi-Task Learning
Junyi Chai (Purdue University), Xiaoqian Wang (Purdue University)
ClassificationRecognitionSegmentationDepth EstimationAdversarial AttackConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: To address the pseudo-correlation between tasks in multi-task learning, we first identify pseudo-related tasks by calculating the differences in correlation coefficients of task labels on the training set and the re-sampled training set by category. Then, we perform debiased adversarial training on each task, ensuring that the predictor for that task no longer utilizes information from other tasks identified as pseudo-related, thereby enhancing the model's generalization ability.
Identity-Clothing Similarity Modeling for Unsupervised Clothing Change Person Re-Identification
Zhiqi Pang (Harbin Institute of Technology), Chunyu Wang (Harbin Institute of Technology)
RecognitionRetrievalConvolutional Neural NetworkVision Language ModelContrastive LearningImage
🎯 What it does: This paper proposes an Identity-Clothing Similarity Modeling (ICSM) framework for unsupervised clothing change person re-identification, addressing the issue of misclustering samples of the same identity with different outfits.
Identity-preserving Distillation Sampling by Fixed-Point Iterator
SeonHwa Kim (Korea University), Eunju Cha (Sookmyung Women's University)
GenerationData SynthesisKnowledge DistillationDiffusion modelScore-based ModelNeural Radiance FieldImage
🎯 What it does: A method for identity-preserving distillation sampling (IDS) is proposed, which corrects the text-conditioned score function using fixed-point iterative regularization (FPR) to generate images and NeRF edits that maintain the identity of the source image (such as pose, structure, etc.) while completing text-guided tasks.
Identity-Preserving Text-to-Video Generation by Frequency Decomposition
Shenghai Yuan (Peking University), Li Yuan (Peking University)
GenerationTransformerDiffusion modelVideoText
🎯 What it does: This paper proposes ConsisID, a text-to-video generation model based on Diffusion Transformer that does not require fine-tuning. It utilizes frequency decomposition to inject low-frequency global facial features into shallow layers and high-frequency identity features into attention blocks, thereby ensuring consistent character identity in the video.
IDOL: Instant Photorealistic 3D Human Creation from a Single Image
Yiyu Zhuang (Nanjing University), Wei Liu (Tencent)
GenerationData SynthesisTransformerDiffusion modelGaussian SplattingImage
🎯 What it does: Designed the IDOL model and the HuGe100K dataset, achieving fast generation of animatable high-fidelity 3D human figures from a single image;
IDProtector: An Adversarial Noise Encoder to Protect Against ID-Preserving Image Generation
Yiren Song (National University of Singapore), Mike Zheng Shou (National University of Singapore)
RecognitionGenerationSafty and PrivacyTransformerImage
🎯 What it does: A noise encoder based on Vision Transformer has been designed and trained to add small, invisible adversarial perturbations to portrait photos, thereby preventing Encoder-based identity-preserving image generation models (such as InstantID, IP-Adapter, PhotoMaker, etc.) from accurately reproducing the identity of the portraits.
iG-6DoF: Model-free 6DoF Pose Estimation for Unseen Object via Iterative 3D Gaussian Splatting
Tuo Cao (Wuhan University), Chunxia Xiao (Wuhan University)
Pose EstimationGaussian SplattingImage
🎯 What it does: A model-free 6DoF pose estimation framework called iG-6DoF is proposed, which can infer the pose of unknown objects using only reference images.
ILIAS: Instance-Level Image retrieval At Scale
Giorgos Kordopatis-Zilos (Czech Technical University in Prague), Giorgos Tolias (Czech Technical University in Prague)
RetrievalVision Language ModelImageBenchmark
🎯 What it does: This work proposes the instance-level image retrieval dataset ILIAS, which consists of 1,000 objects and 100M background images, and benchmarks various foundational models and retrieval methods.
Illumination Spectrum Estimation for Multispectral Images via Surface Reflectance Modeling and Spatial-Spectral Feature Generation
Hyejin Oh (Ewha W. University), Je-Won Kang (Ewha W. University)
RestorationConvolutional Neural NetworkRecurrent Neural NetworkGaussian SplattingImage
🎯 What it does: This paper proposes a method for estimating illumination spectra based on multispectral images, achieving the separation and reconstruction of illumination spectra and reflectance through surface reflectance modeling and spatial-spectral feature generation.
IM-Portrait: Learning 3D-aware Video Diffusion for Photorealistic Talking Heads from Monocular VideosC
Yuan Li (Zhejiang University), Yinda Zhang (Google)
GenerationData SynthesisDiffusion modelVideo
🎯 What it does: A video diffusion model based on 3D perception is proposed, which can directly generate photorealistic talking head videos in the form of multi-plane images (MPI) from a single identity image and expression control signals.
IM-Zero: Instance-level Motion Controllable Video Generation in a Zero-shot Manner
Yuyang Huang (Shanghai Jiao Tong University), Qi Tian (Huawei Inc.)
GenerationData SynthesisDiffusion modelVideoText
🎯 What it does: IM-Zero is proposed, a zero-shot instance-level action controllable video generation framework that can synthesize high-quality, coherent videos under the conditions of given instance locations, motion trajectories, masks, or reference videos.
Image Generation Diversity Issues and How to Tame Them
Mischa Dombrowski (Friedrich-Alexander-Universität Erlangen-Nürnberg), Bernhard Kainz (Imperial College London)
GenerationData SynthesisRetrievalDiffusion modelImage
🎯 What it does: A diversity evaluation metric IRS based on image retrieval is proposed, revealing the shortcomings of existing generative models in terms of diversity and introducing a diversity-aware diffusion model DiADM.
Image is All You Need to Empower Large-scale Diffusion Models for In-Domain Generation
Pu Cao (Beijing University of Posts and Telecommunications), Qing Song (Beijing University of Posts and Telecommunications)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: By using only image data, personalized fine-tuning of large-scale diffusion models for in-domain generation is performed, maintaining the original open-world control capabilities and generating high-fidelity, highly diverse results.
Image Over Text: Transforming Formula Recognition Evaluation with Character Detection Matching
Bin Wang (Shanghai Artificial Intelligence Laboratory), Conghui He (Shanghai Artificial Intelligence Laboratory)
RecognitionObject DetectionImage
🎯 What it does: This paper proposes a mathematical formula recognition evaluation metric based on Character Detection Matching (CDM), replacing traditional text-level BLEU and Edit Distance metrics with image-level character matching.
Image Quality Assessment: From Human to Machine Preference
Chunyi Li (Shanghai Jiao Tong University), Guangtao Zhai (Shanghai Jiao Tong University)
SegmentationRetrievalLarge Language ModelContrastive LearningImageMultimodality
🎯 What it does: This paper proposes an image quality assessment (IQA) framework that transitions from the human visual system (HVS) to the machine visual system (MVS), and constructs a large-scale database focused on machine preferences, referred to as MPD, for the first time.
Image Quality Assessment: Investigating Causal Perceptual Effects with Abductive Counterfactual Inference
Wenhao Shen (Chongqing University), Weijia Jia (Beijing Normal University)
Representation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes a full-reference image quality assessment (FR-IQA) method based on inductive counterfactual reasoning, which predicts quality by exploring the causal relationship between deep network features and human perception of distortion.
Image Reconstruction from Readout-Multiplexed Single-Photon Detector Arrays
Shashwath Bharadwaj (Boston University), Vivek K Goyal (Boston University)
RestorationImage
🎯 What it does: This paper proposes a multiphoton resonance analytical method for a row-column multiplexed single-photon detection array, achieving probability estimation and image reconstruction for up to four simultaneously excited photons.
Image Referenced Sketch Colorization Based on Animation Creation Workflow
Dingkun Yan (Institute of Science Tokyo), Jiaxian Guo (University of Tokyo)
Image TranslationGenerationTransformerDiffusion modelImage
🎯 What it does: This paper proposes a coloring framework for image reference line art based on an animation production workflow, which can automatically map the color distribution of reference images to line art while avoiding spatial aliasing.
Imagine and Seek: Improving Composed Image Retrieval with an Imagined Proxy
You Li (Zhejiang University), Yi Yang (Zhejiang University)
RetrievalTransformerLarge Language ModelContrastive LearningImageText
🎯 What it does: A training-independent and pluggable IP-CIR method is proposed to enhance the performance of composite image retrieval by generating proxy images aligned with query images and relative text.
ImagineFSL: Self-Supervised Pretraining Matters on Imagined Base Set for VLM-based Few-shot Learning
Haoyuan Yang (Dalian University of Technology), Peihua Li (Dalian University of Technology)
ClassificationGenerationData SynthesisTransformerVision Language ModelContrastive LearningImageTextChain-of-Thought
🎯 What it does: This paper proposes a two-stage few-shot learning framework based on CLIP called ImagineFSL, which first conducts self-supervised pre-training on an independent image set iBase synthesized by a text generation model, and then fine-tunes the model using real and synthetic images for downstream tasks.
IMFine: 3D Inpainting via Geometry-guided Multi-view Refinement
Zhihao Shi (Huawei Canada Research Institute), Xinxin Zuo (Concordia University)
RestorationDepth EstimationDiffusion modelGaussian SplattingImage
🎯 What it does: A geometry-guided multi-view refinement 3D inpainting framework is proposed, achieving seamless filling of 3D scenes after object removal.
Img-Diff: Contrastive Data Synthesis for Multimodal Large Language Models
Qirui Jiao (Sun Yat-Sen University), Ying Shen (Alibaba Group)
RecognitionGenerationData SynthesisTransformerLarge Language ModelDiffusion modelContrastive LearningImageMultimodality
🎯 What it does: An automated 'object replacement' image pair dataset, Img-Diff, has been designed and constructed, utilizing contrastive learning and difference annotation techniques to provide fine-grained visual difference recognition training data for multimodal large language models (MLLMs).
Immune: Improving Safety Against Jailbreaks in Multi-modal LLMs via Inference-Time Alignment
Soumya Suvra Ghosal (University of Maryland), Amrit Singh Bedi (University of Central Florida)
Safty and PrivacyAdversarial AttackTransformerLarge Language ModelReinforcement LearningTextMultimodality
🎯 What it does: A reasoning-time safe alignment framework named Immune has been developed to defend against jailbreak attacks on multimodal large language models.
Implicit Bias Injection Attacks against Text-to-Image Diffusion Models
Huayang Huang (Wuhan University), Yu Wu (Wuhan University)
GenerationAdversarial AttackTransformerLarge Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: An implicit bias injection attack framework IBI-Attacks is proposed in the text-to-image diffusion model. It first uses LLM to generate neutral and biased text pairs, calculates the average difference vector in the text embedding space as the bias direction, and then dynamically adjusts this direction through an adaptive feature selection module during inference, injecting bias into the user's prompt embedding, achieving covert attacks without model retraining or prompt modification.
Implicit Correspondence Learning for Image-to-Point Cloud Registration
Xinjun Li (University of Science and Technology of China), Tianzhu Zhang (University of Science and Technology of China)
Pose EstimationAutonomous DrivingConvolutional Neural NetworkImagePoint Cloud
🎯 What it does: A method for image-to-point cloud registration based on implicit correspondence learning is proposed.
Improve Representation for Imbalanced Regression through Geometric Constraints
Zijian Dong (National University of Singapore), Juan Helen Zhou (National University of Singapore)
Representation LearningConvolutional Neural NetworkContrastive LearningTabular
🎯 What it does: This study investigates representation learning in deep imbalance regression, proposing geometric constraints to achieve a uniform distribution in the feature space.
Improved Monocular Depth Prediction Using Distance Transform Over Pre-semantic Contours with Self-supervised Neural Networks
Marwane Hariat (Institut Polytechnique de Paris), David Filliat (Institut Polytechnique de Paris)
Depth EstimationAutonomous DrivingConvolutional Neural NetworkImage
🎯 What it does: A self-supervised monocular depth estimation framework is proposed, which enhances the image variance in low-texture areas by utilizing distance transformation of pre-semantic contours, thereby reducing the uncertainty of pixel matching.
Improved Video VAE for Latent Video Diffusion Model
Pingyu Wu (University of Science and Technology of China), Zheng-Jun Zha (University of Science and Technology of China)
GenerationCompressionAuto EncoderGenerative Adversarial NetworkVideo
🎯 What it does: An improved Video VAE (IV-VAE) is proposed for more efficient video compression, reconstruction, and implicit video diffusion models.
Improving Accuracy and Calibration via Differentiated Deep Mutual Learning
Han Liu (Tsinghua University), Xiaolin Hu (Tsinghua University)
ClassificationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: Proposes the Diff-DML method, which improves the deep mutual learning framework, allowing a single model to enhance both accuracy and significantly improve probability calibration.
Improving Adversarial Transferability on Vision Transformers via Forward Propagation Refinement
Yuchen Ren (Xi'an Jiaotong University), Chao Shen
Adversarial AttackTransformerImage
🎯 What it does: By diversifying the attention maps (AMD) during the forward propagation of the Vision Transformer and applying momentum smoothing to token embeddings (MTE), a Forward Propagation Refinement (FPR) method is proposed to enhance the transfer attack capability of adversarial samples.
Improving Autoregressive Visual Generation with Cluster-Oriented Token Prediction
Teng Hu (Shanghai Jiao Tong University), Lizhuang Ma (Shanghai Jiao Tong University)
GenerationTransformerLarge Language ModelImage
🎯 What it does: An improved autoregressive visual generation method IAR is proposed, which enhances the generation quality of LLM by utilizing code cluster rearrangement and clustering-guided cross-entropy loss.
Improving Diffusion Inverse Problem Solving with Decoupled Noise Annealing
Bingliang Zhang (California Institute of Technology), Yang Song (OpenAI)
Image TranslationRestorationSuper ResolutionDiffusion modelImageMagnetic Resonance Imaging
🎯 What it does: The Decoupled Annealing Posterior Sampling (DAPS) method is proposed to achieve better posterior sampling in inverse problems by decoupling the noise annealing process.
Improving Editability in Image Generation with Layer-wise Memory
Daneul Kim (Seoul National University), Jaesik Park (Seoul National University)
GenerationLarge Language ModelDiffusion modelImageBenchmark
🎯 What it does: An interactive multi-step image editing framework based on coarse masks and prompts has been constructed, supporting the natural addition and deletion of multiple objects while maintaining background consistency.
Improving Gaussian Splatting with Localized Points Management
Haosen Yang (University of Surrey), Xiatian Zhu (Fudan University)
RestorationOptimizationGaussian SplattingPoint Cloud
🎯 What it does: A Localized Point Management (LPM) framework based on error contribution is proposed to improve the point distribution and geometric correction of 3D Gaussian Splatting models.
Improving Personalized Search with Regularized Low-Rank Parameter Updates
Fiona Ryan (Georgia Tech), Bryan Russell (Georgia Tech)
RetrievalTransformerContrastive LearningImageText
🎯 What it does: This paper proposes a low-rank parameter update-based fine-tuning method for the CLIP text encoder, aimed at few-shot personalized visual-language retrieval.
Improving Semi-Supervised Semantic Segmentation with Sliced-Wasserstein Feature Alignment and Uniformity
Chen-Yi Lu (Purdue University), Somali Chaterji (Purdue University)
SegmentationConvolutional Neural NetworkImage
🎯 What it does: A semi-supervised semantic segmentation method SWSEG is proposed, which enhances the model's utilization of unlabeled data by directly optimizing alignment and uniformity in the feature space of the backbone encoder.
Improving Sound Source Localization with Joint Slot Attention on Image and Audio
Inho Kim (POSTECH), Suha Kwak (POSTECH)
RecognitionObject DetectionRetrievalRecurrent Neural NetworkContrastive LearningImageMultimodalityAudio
🎯 What it does: A self-supervised audio-visual sound source localization method based on joint slot attention is proposed, which can separate target and non-target features from unlabeled data and achieve precise localization.
Improving the Training of Data-Efficient GANs via Quality Aware Dynamic Discriminator Rejection Sampling
Zhaoyu Zhang (Queen's University Belfast), Seán McLoone (Queen's University Belfast)
GenerationData SynthesisDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: A quality-aware dynamic discriminator rejection sampling (QADDRS) method is proposed, which dynamically rejects high-scoring real samples and low-scoring fake samples during the data-efficient GAN training process to alleviate discriminator overfitting and improve generation quality.
Improving the Transferability of Adversarial Attacks on Face Recognition with Diverse Parameters Augmentation
Fengfan Zhou (Huazhong University of Science and Technology), Wenxuan Wang (Northwestern Polytechnical University)
RecognitionAdversarial AttackImage
🎯 What it does: A multi-parameter enhanced adversarial attack method DPA is proposed, which improves the transferability of attacks on facial recognition models by utilizing various parameter initializations and an intermediate training model set.
Improving Transferable Targeted Attacks with Feature Tuning Mixup
Kaisheng Liang (Hong Kong Polytechnic University), Bin Xiao (Hong Kong Polytechnic University)
Adversarial AttackConvolutional Neural NetworkTransformerMixture of ExpertsImage
🎯 What it does: This paper proposes a new feature-level attack method called Feature Tuning Mixup (FTM), which enhances the transferability of targeted attacks by adding learnable perturbations in the intermediate layers and mixing them with randomly clean features.
Improving Visual and Downstream Performance of Low-Light Enhancer with Vision Foundation Models Collaboration
Yuxuan Gu (University of Science and Technology of China), Enhong Chen (University of Science and Technology of China)
RestorationObject DetectionSegmentationGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: The FoCo framework is proposed, utilizing various foundational visual models (CLIP, SAM, ResNet, etc.) for self-supervised low-light image enhancement.
Imputation-free and Alignment-free: Incomplete Multi-view Clustering Driven by Consensus Semantic Learning
Yuzhuo Dai (National University of Defense Technology), Yu Feng (National University of Defense Technology)
OptimizationGraph Neural NetworkAuto EncoderContrastive LearningImage
🎯 What it does: A complete semantic learning framework called FreeCSL is proposed to eliminate the need for missing value imputation and alignment for incomplete multi-view clustering (IMVC).
ImViD: Immersive Volumetric Videos for Enhanced VR Engagement
Zhengxian Yang (Tsinghua University), Tao Yu (Tsinghua University)
GenerationData SynthesisGaussian SplattingVideoBenchmarkAudio
🎯 What it does: This work proposes the ImViD dataset and implements a baseline, providing 5K 60FPS dynamic recording of multi-view synchronized video and audio, as well as a complete light field and spatial audio reconstruction process, supporting 6-DoF immersive VR experiences.
IncEventGS: Pose-Free Gaussian Splatting from a Single Event Camera
Jian Huang (Zhejiang University), Peidong Liu (Westlake University)
Object DetectionPose EstimationDepth EstimationGaussian SplattingSimultaneous Localization and MappingImagePoint Cloud
🎯 What it does: An incremental 3D Gaussian spraying (IncEventGS) method for monocular event cameras is proposed, capable of achieving 3D scene reconstruction and camera motion estimation without prior camera pose information.
Incomplete Multi-modal Brain Tumor Segmentation via Learnable Sorting State Space Model
Zheyu Zhang (University of Science and Technology of China), Xiaoyan Sun (University of Science and Technology of China)
SegmentationMultimodalityBiomedical DataMagnetic Resonance Imaging
🎯 What it does: Proposed the Learnable Sorting State Space Model (LS3M), a framework specifically designed for brain tumor segmentation under missing modalities;
Incomplete Multi-View Multi-label Learning via Disentangled Representation and Label Semantic Embedding
Xu Yan (Shanghai Maritime University), Jie Wen (Harbin Institute of Technology)
ClassificationRepresentation LearningGraph Neural NetworkAuto EncoderMultimodality
🎯 What it does: A framework for dual missing multi-view multi-label learning, DRLS, is proposed, which can simultaneously address the issues of missing views and missing labels.
Incorporating Dense Knowledge Alignment into Unified Multimodal Representation Models
Yuhao Cui (Alibaba Cloud Computing), Jinyang Gao (Tongyi Lab)
RetrievalTransformerLarge Language ModelContrastive LearningImageVideoTextMultimodality
🎯 What it does: A unified multimodal retrieval model DeKR is constructed, and the alignment effect of multimodal embeddings is improved through the dense knowledge alignment dataset DeKon5M.
Incremental Object Keypoint Learning
Mingfu Liang (Northwestern University), Ying Wu (Northwestern University)
Object DetectionPose EstimationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes an Incremental Keypoint Learning (IKL) framework that allows for the gradual learning of newly defined keypoints without retaining old data, while maintaining the ability to recognize old keypoints.
IndoorGS: Geometric Cues Guided Gaussian Splatting for Indoor Scene Reconstruction
Cong Ruan (Huazhong University of Science and Technology), Lili Ju (University of South Carolina)
RestorationDepth EstimationGaussian SplattingSimultaneous Localization and MappingPoint Cloud
🎯 What it does: This paper proposes IndoorGS, a method for indoor scene reconstruction based on 3D Gaussian scattering.
Inference-Scale Complexity in ANN-SNN Conversion for High-Performance and Low-Power Applications
Tong Bu (Peking University), Zhaofei Yu (Peking University)
ClassificationObject DetectionSegmentationConvolutional Neural NetworkSpiking Neural NetworkImageVideo
🎯 What it does: This paper proposes an ANN-SNN conversion framework that can be completed solely during the inference stage, allowing for the direct conversion of pre-trained ANN models into high-performance, low-power SNNs.
Infighting in the Dark: Multi-Label Backdoor Attack in Federated Learning
Ye Li (Nanjing University of Aeronautics and Astronautics), Jiale Zhang (Yangzhou University)
Federated LearningAdversarial AttackImage
🎯 What it does: Proposes the Mirage method, studying the feasibility of non-cooperative multi-label backdoor attacks in federated learning and achieving simultaneous implantation by multiple attackers;