CVPR 2023 Papers — Page 10
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2353 papers
High Fidelity 3D Hand Shape Reconstruction via Scalable Graph Frequency Decomposition
Tianyu Luan (State University of New York), Junsong Yuan (State University of New York)
GenerationPose EstimationGraph Neural NetworkMesh
🎯 What it does: A scalable frequency domain decomposition network is designed and implemented to achieve high-fidelity 3D hand model reconstruction from a single image.
High-Fidelity 3D Face Generation From Natural Language Descriptions
Menghua Wu (Nanjing University), Xun Cao (Nanjing University)
GenerationPrompt EngineeringGenerative Adversarial NetworkMesh
🎯 What it does: The research focuses on methods for generating high-quality 3D face models from natural language descriptions, proposing a two-stage generation and optimization process.
High-Fidelity 3D GAN Inversion by Pseudo-Multi-View Optimization
Jiaxin Xie (Hong Kong University of Science and Technology), Qifeng Chen (Hong Kong University of Science and Technology)
GenerationOptimizationGenerative Adversarial NetworkImage
🎯 What it does: A high-fidelity 3D GAN inversion framework based on pseudo-multi-view optimization is proposed, capable of generating high-quality, 3D-consistent view synthesis from a single image, and supports attribute editing and texture modification.
High-Fidelity 3D Human Digitization From Single 2K Resolution Images
Sang-Hun Han (Gwangju Institute of Science and Technology), Hae-Gon Jeon (Gwangju Institute of Science and Technology)
SegmentationGenerationPose EstimationDepth EstimationComputational EfficiencyConvolutional Neural NetworkImageMesh
🎯 What it does: This paper proposes a full-body 3D human digitization framework (2K2K) based on a single 2K (2048×2048) high-resolution color image. It generates a high-fidelity 3D mesh through segmentation (body parts) extraction, surface normal prediction, low-resolution full-body depth prediction, and high-resolution depth fusion.
High-Fidelity and Freely Controllable Talking Head Video Generation
Yue Gao (Microsoft Research), Yan Lu (Microsoft Research)
GenerationData SynthesisPose EstimationGenerative Adversarial NetworkVideo
🎯 What it does: This paper studies a high-fidelity speaker video generation model PECHead that allows for free control of head pose and facial expressions.
High-Fidelity Clothed Avatar Reconstruction From a Single Image
Tingting Liao (University of Chinese Academy of Sciences), Zhen Lei (University of Chinese Academy of Sciences)
GenerationPose EstimationImage
🎯 What it does: This paper proposes a coarse-to-fine two-stage framework for reconstructing human avatars from a single image (CAR), enabling the rapid generation of high-fidelity clothing pose avatars.
High-Fidelity Event-Radiance Recovery via Transient Event Frequency
Jin Han (University of Tokyo), Imari Sato (University of Tokyo)
RestorationDepth EstimationImage
🎯 What it does: Directly reconstruct the scene radiance values using the transient event frequency (TEF) of event cameras under active illumination.
High-Fidelity Facial Avatar Reconstruction From Monocular Video With Generative Priors
Yunpeng Bai (Tsinghua University), Ying Shan (Tencent AI Lab)
GenerationData SynthesisGenerative Adversarial NetworkVideoMultimodalityAudio
🎯 What it does: Constructing high-fidelity 3D facial avatars from monocular videos, supporting facial reenactment and free-viewpoint rendering.
High-Fidelity Generalized Emotional Talking Face Generation With Multi-Modal Emotion Space Learning
Chao Xu (Zhejiang University), Yong Liu (Zhejiang University)
GenerationTransformerGenerative Adversarial NetworkImageTextMultimodalityAudio
🎯 What it does: This paper proposes a one-shot emotion-driven facial animation framework for speakers, supporting multimodal emotion control (text, image, audio), and generating high-resolution, naturally expressive, and audio-synchronized animated faces.
High-Fidelity Guided Image Synthesis With Latent Diffusion Models
Jaskirat Singh (Australian National University), Liang Zheng (Australian National University)
GenerationOptimizationDiffusion modelImage
🎯 What it does: A framework for image generation guided by user doodles and text prompts is proposed, addressing the domain transfer distortion problem of traditional methods.
High-Frequency Stereo Matching Network
Haoliang Zhao (Guizhou University), Yong Zhao (Peking University)
Depth EstimationDomain AdaptationConvolutional Neural NetworkRecurrent Neural NetworkTransformerImageBenchmark
🎯 What it does: A new iterative high-frequency information retention stereo matching network, DLNR, is proposed, which can obtain more refined disparity maps on details, edges, and thin objects.
High-Res Facial Appearance Capture From Polarized Smartphone Images
Dejan Azinović (Technical University of Munich), Justus Thies (Max Planck Institute for Intelligent Systems)
RestorationOptimizationImageVideo
🎯 What it does: Using a smartphone equipped with a polarizing filter, short video sequences of cross-polarized and parallel-polarized light were captured in a dark room. By combining structured light illumination + multi-view stereo reconstruction, FLAME model fitting, and differentiable rendering for photometric optimization, high-resolution facial geometry, diffuse reflection, specular reflection, and normal maps were recovered, which can be directly used in rendering software such as Blender.
High-Resolution Image Reconstruction With Latent Diffusion Models From Human Brain Activity
Yu Takagi (Osaka University), Shinji Nishimoto (Osaka University)
RestorationGenerationDiffusion modelImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A high-resolution image reconstruction method based on the latent diffusion model (LDM) is proposed, which reconstructs images from human brain activity (fMRI signals).
Highly Confident Local Structure Based Consensus Graph Learning for Incomplete Multi-View Clustering
Jie Wen (Harbin Institute of Technology), Yong Xu (Pengcheng Laboratory)
OptimizationGraph Neural NetworkContrastive LearningGraph
🎯 What it does: A consensus graph learning method based on high-confidence local structures, HCLS CGL, is proposed to address the problem of incomplete multi-view clustering.
Hint-Aug: Drawing Hints From Foundation Vision Transformers Towards Boosted Few-Shot Parameter-Efficient Tuning
Zhongzhi Yu (Georgia Institute of Technology), Yingyan (Celine) Lin (Georgia Institute of Technology)
ClassificationAdversarial AttackTransformerSupervised Fine-TuningImage
🎯 What it does: Proposes the Hint-Aug framework, which utilizes the attention information of pre-trained visual Transformers to detect overfitting and adaptively enhance few-shot data through adversarially confused feature injection, thereby improving the few-shot fine-tuning performance of FViT.
Histopathology Whole Slide Image Analysis With Heterogeneous Graph Representation Learning
Tsai Hor Chan (University of Hong Kong), Lequan Yu (University of Hong Kong)
ClassificationExplainability and InterpretabilityRepresentation LearningGraph Neural NetworkTransformerImageBiomedical Data
🎯 What it does: This paper constructs a heterogeneous graph that includes cell types and continuous similarity, utilizing the HEAT layer and pseudo-label pooling for whole slide image (WSI) analysis, and provides causal explanations.
HNeRV: A Hybrid Neural Representation for Videos
Hao Chen (University of Maryland), Abhinav Shrivastava (University of Maryland)
RestorationCompressionConvolutional Neural NetworkVideo
🎯 What it does: A hybrid neural representation method HNeRV is proposed, which uses learnable content-adaptive embeddings and a specially designed decoder to store and reconstruct videos.
HOICLIP: Efficient Knowledge Transfer for HOI Detection With Vision-Language Models
Shan Ning (ShanghaiTech University), Xuming He (ShanghaiTech University)
Object DetectionKnowledge DistillationTransformerVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: Proposes the HOICLIP framework, which efficiently transfers CLIP's visual-language knowledge to the human-object interaction (HOI) detection task.
HOLODIFFUSION: Training a 3D Diffusion Model Using 2D Images
Animesh Karnewar (University College London), Niloy J. Mitra (University College London)
GenerationData SynthesisDiffusion modelImageVideo
🎯 What it does: A 3D diffusion generative model called HOLODIFFUSION has been developed, which is trained solely on posed 2D images and can generate perspective-consistent 3D scenes.
HOOD: Hierarchical Graphs for Generalized Modelling of Clothing Dynamics
Artur Grigorev (ETH Zurich), Otmar Hilliges (Max Planck Institute for Intelligent Systems)
GenerationGraph Neural NetworkMesh
🎯 What it does: A clothing dynamics prediction framework based on graph neural networks (HOOD) is proposed, capable of generating real-time dynamic effects of clothing of different types, sizes, materials, and even topologically variable without labeled self-supervised training.
HOTNAS: Hierarchical Optimal Transport for Neural Architecture Search
Jiechao Yang, Hongteng Xu
GenerationNeural Architecture SearchRecurrent Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A multi-task learning text generation model based on the Transformer architecture is proposed.
HouseDiffusion: Vector Floorplan Generation via a Diffusion Model With Discrete and Continuous Denoising
Mohammad Amin Shabani (Simon Fraser University), Yasutaka Furukawa (Simon Fraser University)
GenerationData SynthesisTransformerDiffusion modelGraph
🎯 What it does: Directly generate vectorized floor plans using diffusion models, achieving geometric consistency through continuous and discrete denoising.
How Can Objects Help Action Recognition?
Xingyi Zhou (Google Research), Cordelia Schmid (Google Research)
RecognitionObject DetectionTransformerVideo
🎯 What it does: Improving video action recognition using external object detection information, proposing Object-Guided Token Sampling (OGS) and Object-Aware Attention Module (OAM) to achieve the dual goals of reducing token inefficiency and enhancing accuracy.
How to Backdoor Diffusion Models?
Sheng-Yen Chou (National Tsing Hua University), Tsung-Yi Ho (Chinese University of Hong Kong)
GenerationAdversarial AttackDiffusion modelImage
🎯 What it does: This paper proposes a backdoor attack framework for diffusion models called BadDiffusion, and demonstrates the feasibility of this attack in image generation tasks.
How To Prevent the Continuous Damage of Noises To Model Training?
Xiaotian Yu (Zhejiang University), Li Sun (Ningbo Innovation Center Zhejiang University)
ClassificationData-Centric LearningConvolutional Neural NetworkImage
🎯 What it does: To address the issue of gradient misguidance caused by label noise during training, this paper proposes a Gradient Switching Strategy (GSS) that suppresses misguidance by dynamically switching gradient directions through a pool of gradient directions, thereby enhancing the model's robustness on noisy data.
How To Prevent the Poor Performance Clients for Personalized Federated Learning?
Zhe Qu (Central South University), Lixing Chen (Shanghai Jiaotong University)
Federated LearningConvolutional Neural NetworkImage
🎯 What it does: The PLGU strategy has been designed and validated, achieving a unification of local personalization and global generalization in personalized federated learning through hierarchical freezing and LWSAM in Scheme I and Scheme II, preventing poor performance from individual clients.
How You Feelin'? Learning Emotions and Mental States in Movie Scenes
Dhruv Srivastava (International Institute of Information Technology Hyderabad), Makarand Tapaswi (International Institute of Information Technology Hyderabad)
ClassificationRecognitionTransformerVideoMultimodality
🎯 What it does: A multi-modal Transformer architecture called EmoTx is proposed, capable of multi-label emotion and psychological state prediction at both the movie scene and character levels.
HRDFuse: Monocular 360deg Depth Estimation by Collaboratively Learning Holistic-With-Regional Depth Distributions
Hao Ai (Hong Kong University of Science and Technology), Lin Wang (Hong Kong University of Science and Technology)
Depth EstimationConvolutional Neural NetworkTransformerPoint Cloud
🎯 What it does: A framework called HRDFuse is proposed, which utilizes ERP and TP projection for collaborative learning of panoramic depth distribution, directly outputting smooth and accurate depth maps in the ERP space.
HS-Pose: Hybrid Scope Feature Extraction for Category-Level Object Pose Estimation
Linfang Zheng (Southern University of Science and Technology), Hyung Jin Chang (University of Birmingham)
Object DetectionPose EstimationGraph Neural NetworkPoint Cloud
🎯 What it does: This paper proposes a new hybrid scale feature extraction layer (HS-layer) and applies it to the category-level object pose estimation framework HS-Pose, which can simultaneously capture local and global geometric information, encode scale and translation information, and is robust to outliers.
Hubs and Hyperspheres: Reducing Hubness and Improving Transductive Few-Shot Learning With Hyperspherical Embeddings
Daniel J. Trosten (UiT Arctic University of Norway), Michael C. Kampffmeyer (UiT Arctic University of Norway)
ClassificationRepresentation LearningImage
🎯 What it does: This paper proposes a method to eliminate the hubness problem in transductive few-shot learning by achieving uniform embedding on the sphere, enhancing classification performance while maintaining local similarity (LSP);
Human Body Shape Completion With Implicit Shape and Flow Learning
Boyao Zhou (University of Grenoble Alpes), Edmond Boyer (University of Grenoble Alpes)
RestorationSegmentationDepth EstimationConvolutional Neural NetworkOptical FlowPoint CloudMesh
🎯 What it does: Learn to jointly complete human shape completion based on implicit function dual-frame depth shape and flow estimation.
Human Guided Ground-Truth Generation for Realistic Image Super-Resolution
Du Chen (Hong Kong Polytechnic University), Lei Zhang (Hong Kong Polytechnic University)
RestorationSuper ResolutionConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: A multi-version high-quality GT generation process based on human judgment has been developed, and a specialized HGGT dataset for super-resolution in real scenarios has been constructed based on this.
Human Pose As Compositional Tokens
Zigang Geng (University of Science and Technology of China), Han Hu (University of Science and Technology of China)
Pose EstimationTransformerImage
🎯 What it does: Proposes a structured representation called Pose as Compositional Tokens (PCT), which maps human poses to discrete sub-structure tokens and completes pose estimation through classification tasks.
Human Pose Estimation in Extremely Low-Light Conditions
Sohyun Lee (POSTECH), Suha Kwak (POSTECH)
Pose EstimationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: To address human pose estimation in extremely low light environments, the ExLPose dataset was constructed with corresponding well-lit images, and an end-to-end training strategy utilizing 'privileged information' (LUPI) and light-specific batch normalization (LSBN) was proposed.
Human-Art: A Versatile Human-Centric Dataset Bridging Natural and Artificial Scenes
Xuan Ju (International Digital Economy Academy), Lei Zhang (International Digital Economy Academy)
Object DetectionPose EstimationConvolutional Neural NetworkTransformerDiffusion modelImage
🎯 What it does: This paper presents the Human-Art dataset, aimed at bridging the gap between natural scenes and artificial scenes (such as sculptures, paintings, cartoons, digital art, etc.) in the tasks of human detection and pose estimation, and provides rich annotations (bounding boxes, 21 key points, self-contact points, text descriptions) to support various downstream tasks.
HumanBench: Towards General Human-Centric Perception With Projector Assisted Pretraining
Shixiang Tang (University of Sydney), Wanli Ouyang (Shanghai AI Laboratory)
Object DetectionSegmentationPose EstimationRepresentation LearningTransformerImageBenchmark
🎯 What it does: This paper proposes HumanBench (which includes 19 datasets covering 6 categories of human perception tasks) and PATH (a projector-based hierarchical weight sharing pre-training method) for learning general human visual representations.
HumanGen: Generating Human Radiance Fields With Explicit Priors
Suyi Jiang (ShanghaiTech University), Lan Xu (ShanghaiTech University)
GenerationData SynthesisNeural Radiance FieldGenerative Adversarial NetworkImage
🎯 What it does: A 3D human generation framework called HumanGen has been constructed, capable of synthesizing light radiation fields with high-detail geometric structures and realistic textures from all viewpoints.
HuManiFlow: Ancestor-Conditioned Normalising Flows on SO(3) Manifolds for Human Pose and Shape Distribution Estimation
Akash Sengupta (University of Cambridge), Roberto Cipolla (University of Cambridge)
Pose EstimationFlow-based ModelImage
🎯 What it does: A probabilistic distribution model based on regularized flows, HuManiFlow, is proposed for predicting the distribution of 3D human body pose and shape from a monocular image.
Humans As Light Bulbs: 3D Human Reconstruction From Thermal Reflection
Ruoshi Liu (Columbia University), Carl Vondrick (Columbia University)
GenerationPose EstimationGenerative Adversarial NetworkImage
🎯 What it does: A method is proposed to recover the 3D position and posture of the human body on the surfaces of everyday objects using thermal reflection.
Hunting Sparsity: Density-Guided Contrastive Learning for Semi-Supervised Semantic Segmentation
Xiaoyang Wang (University of Liverpool), Jimin Xiao (XJTLU)
SegmentationContrastive LearningImage
🎯 What it does: This paper proposes a density-guided contrastive learning framework based on feature space (DGCL), which enhances semi-supervised semantic segmentation by locating sparse features and guiding them to cluster around high-density centers.
Hybrid Active Learning via Deep Clustering for Video Action Detection
Aayush J. Rana, Yogesh S. Rawat
RecognitionObject DetectionConvolutional Neural NetworkVideo
🎯 What it does: A hybrid active learning framework that combines video-level and frame-level selection is proposed to reduce the labeling cost of video action detection.
Hybrid Neural Rendering for Large-Scale Scenes With Motion Blur
Peng Dai (University of Hong Kong), Xiaojuan Qi (Google)
RestorationGenerationConvolutional Neural NetworkNeural Radiance FieldImagePoint Cloud
🎯 What it does: A hybrid neural rendering model is proposed, combining image-based rendering and 3D NeRF to generate high-fidelity, view-consistent new views in large scenes.
Hyperbolic Contrastive Learning for Visual Representations Beyond Objects
Songwei Ge (University of Maryland), David Jacobs (University of Maryland)
Object DetectionSegmentationRepresentation LearningContrastive LearningImage
🎯 What it does: This paper proposes a hyperplane contrastive learning framework (HCL) that learns object representations in Euclidean space and scene representations in Poincaré ball hyperplane space, thereby constructing a hierarchical structure of objects and scenes within the same feature space.
HyperCUT: Video Sequence From a Single Blurry Image Using Unsupervised Ordering
Bang-Dang Pham (VinAI Research), Minh Hoai (VinAI Research)
Image TranslationRestorationContrastive LearningImageVideo
🎯 What it does: This paper proposes an unsupervised HyperCUT method to address the frame sequence ambiguity problem in the task of deblurring single-frame blurry images to videos, and trains a high-quality image-to-video deblurring model based on this.
HyperMatch: Noise-Tolerant Semi-Supervised Learning via Relaxed Contrastive Constraint
Beitong Zhou (Hikvision Research Institute), Yi Niu (Hikvision Research Institute)
ClassificationRecognitionContrastive LearningImage
🎯 What it does: A semi-supervised learning method called HyperMatch is proposed, which separates clean and noisy pseudo-labels after calibration and uses superclasses (top-K approximate categories) for relaxed contrastive learning on noisy samples, effectively utilizing a large amount of unlabeled data.
HyperReel: High-Fidelity 6-DoF Video With Ray-Conditioned Sampling
Benjamin Attal (Carnegie Mellon University), Changil Kim (Meta)
Data SynthesisCompressionComputational EfficiencyNeural Radiance FieldVideo
🎯 What it does: A 6-degree-of-freedom video representation method named HyperReel is proposed, which achieves high-quality, real-time viewpoint synthesis while maintaining low memory usage.
Hyperspherical Embedding for Point Cloud Completion
Junming Zhang (University of Michigan), Matthew Johnson-Roberson (Carnegie Mellon University)
GenerationData SynthesisPoint Cloud
🎯 What it does: This paper proposes a hyperspherical embedding module that maps the encoder output to the unit hypersphere within the point cloud completion encoder-decoder framework, making the embedding directionality more concentrated, stabilizing training, and improving completion performance.
HypLiLoc: Towards Effective LiDAR Pose Regression With Hyperbolic Fusion
Sijie Wang (Nanyang Technological University), Wee Peng Tay (Nanyang Technological University)
Pose EstimationAutonomous DrivingSimultaneous Localization and MappingMultimodalityPoint Cloud
🎯 What it does: This paper proposes a new LiDAR pose regression network called HypLiLoc, which utilizes multimodal features from 3D point clouds and spherical projections, and integrates them in Euclidean and hyperbolic spaces.
I2-SDF: Intrinsic Indoor Scene Reconstruction and Editing via Raytracing in Neural SDFs
Jingsen Zhu (Zhejiang University), Rui Wang (Zhejiang University)
RestorationGenerationData SynthesisNeural Radiance FieldImagePoint Cloud
🎯 What it does: This paper proposes the I2-SDF framework, which utilizes differentiable Monte Carlo ray tracing to achieve joint recovery of shape, radiance, and material in indoor scenes on neural SDF, and supports lighting and material editing.
I2MVFormer: Large Language Model Generated Multi-View Document Supervision for Zero-Shot Image Classification
Muhammad Ferjad Naeem (ETH Zurich), Federico Tombari
ClassificationTransformerLarge Language ModelPrompt EngineeringImageTextMultimodality
🎯 What it does: Proposes the I2MVFormer model, which utilizes large language models to generate multi-view text supervision for unsupervised zero-shot image classification.
iCLIP: Bridging Image Classification and Contrastive Language-Image Pre-Training for Visual Recognition
Yixuan Wei (Tsinghua University), Baining Guo (Microsoft Research Asia)
ClassificationRecognitionRetrievalTransformerContrastive LearningImageText
🎯 What it does: By deeply integrating the image classification task with CLIP's contrastive learning, utilizing shared visual and text encoders, the traditional linear classifier is rewritten as a cosine classifier generated by the text encoder, and external dictionary enhancement is applied to category names, further unifying the input space of the two tasks.
Identity-Preserving Talking Face Generation With Landmark and Appearance Priors
Weizhi Zhong (Sun Yat-sen University), Guanbin Li (Sun Yat-sen University)
GenerationData SynthesisTransformerGenerative Adversarial NetworkVideoAudio
🎯 What it does: A two-stage framework for generating talking facial videos driven by audio is proposed, which first maps audio to mouth and jaw landmarks, and then generates a complete facial video that synchronizes lip movements while preserving identity using the landmarks and multiple reference images.
IDGI: A Framework To Eliminate Explanation Noise From Integrated Gradients
Ruo Yang (Illinois Institute of Technology), Mustafa Bilgic (Illinois Institute of Technology)
Explainability and InterpretabilityConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: Proposes the Important Direction Gradient Integration (IDGI) framework, which removes the explanation noise in the Integrated Gradients (IG) method and enhances interpretability.
iDisc: Internal Discretization for Monocular Depth Estimation
Luigi Piccinelli (Computer Vision Lab ETH Zurich), Fisher Yu (Computer Vision Lab ETH Zurich)
Depth EstimationTransformerImage
🎯 What it does: This paper proposes an internal discretization module named iDisc for monocular depth estimation, which achieves an adaptive discrete representation of the scene by embedding a continuous-discrete-continuous bottleneck in the network, ultimately generating high-quality depth maps.
IFSeg: Image-Free Semantic Segmentation via Vision-Language Model
Sukmin Yun (Mohamed bin Zayed University of Artificial Intelligence), Jinwoo Shin (Korea Advanced Institute of Science and Technology)
SegmentationTransformerSupervised Fine-TuningVision Language ModelImageText
🎯 What it does: Using a pre-trained visual language encoder-decoder model, artificial images are generated from semantic category words to complete the semantic segmentation task without using any task-specific images or annotations.
Im2Hands: Learning Attentive Implicit Representation of Interacting Two-Hand Shapes
Jihyun Lee (KAIST), Tae-Kyun Kim (Imperial College London)
GenerationPose EstimationGraph Neural NetworkImagePoint Cloud
🎯 What it does: Predicts the high-resolution 3D shapes of two interacting hands from a single RGB image and hand keypoints using a neural implicit occupancy function.
Image as a Foreign Language: BEiT Pretraining for Vision and Vision-Language Tasks
Wenhui Wang (Microsoft Corporation), Furu Wei (Microsoft Corporation)
ClassificationObject DetectionSegmentationRetrievalTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: This paper presents BEIT-3, a general-purpose multimodal foundation model capable of processing images, text, and image-text pairs simultaneously.
Image Cropping With Spatial-Aware Feature and Rank Consistency
Chao Wang (Shanghai Jiao Tong University), Liqing Zhang (Shanghai Jiao Tong University)
Image TranslationSegmentationOptimizationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes a spatially aware feature for encoding the spatial relationship between candidate cropping boxes and aesthetic elements (such as salient objects and semantic edges), and improves image cropping by transferring ranking knowledge and enforcing ranking consistency between labeled and unlabeled images.
Image Quality-Aware Diagnosis via Meta-Knowledge Co-Embedding
Haoxuan Che (Hong Kong University of Science and Technology), Hao Chen (Hong Kong University of Science and Technology)
ClassificationMeta LearningConvolutional Neural NetworkImageBiomedical DataComputed Tomography
🎯 What it does: This paper proposes the Image Quality Aware Diagnosis (IQAD) problem and constructs the Meta-knowledge Co-embedding Network (MKCNet), which utilizes low-quality images and their quality labels to achieve more robust medical image diagnosis.
Image Super-Resolution Using T-Tetromino Pixels
Simon Grosche (Friedrich Alexander University Erlangen Nuremberg), André Kaup (Friedrich Alexander University Erlangen Nuremberg)
RestorationSuper ResolutionConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a novel image sensor layout based on four T-shaped tetromino pixels (T-tetromino) and conducts compressed sensing recovery under this layout, evaluating its performance in low light and high-speed imaging.
ImageBind: One Embedding Space To Bind Them All
Rohit Girdhar (FAIR), Ishan Misra (FAIR)
RecognitionGenerationRetrievalTransformerContrastive LearningImageTextMultimodalityAudio
🎯 What it does: We propose IMAGEBIND, a multimodal joint embedding framework that utilizes image alignment to bind six modalities—image, text, audio, depth, thermal imaging, and IMU—into the same space without the need for cross-modal samples, supporting novel functionalities such as cross-modal retrieval, combination, and generation.
Imagen Editor and EditBench: Advancing and Evaluating Text-Guided Image Inpainting
Su Wang (Google Research), William Chan (Google Research)
Image TranslationRestorationObject DetectionGenerationConvolutional Neural NetworkDiffusion modelImageTextBenchmark
🎯 What it does: Developed Imagen Editor, which can perform high-quality and controllable local editing on specified masked areas of images based on text prompts.
ImageNet-E: Benchmarking Neural Network Robustness via Attribute Editing
Xiaodan Li (Alibaba Group), Hui Xue (Alibaba Group)
ClassificationObject DetectionData SynthesisConvolutional Neural NetworkTransformerDiffusion modelImageBenchmark
🎯 What it does: A controllable editing tool for object attributes (background, size, position, direction) has been developed, and based on this, the ImageNet-E dataset has been generated to evaluate the robustness of models under these attribute variations;
Images Speak in Images: A Generalist Painter for In-Context Visual Learning
Xinlong Wang (Beijing Academy of Artificial Intelligence), Tiejun Huang (Peking University)
SegmentationDepth EstimationTransformerPrompt EngineeringImage
🎯 What it does: This paper presents Painter, a general visual model that uses image pairs as task prompts to support reasoning for various visual tasks in context.
Imagic: Text-Based Real Image Editing With Diffusion Models
Bahjat Kawar (Google Research), Michal Irani (Weizmann Institute of Science)
Image TranslationGenerationDiffusion modelImageBenchmark
🎯 What it does: Imagic has been developed, a method that allows for complex non-rigid semantic editing with just a single real image and target text input, capable of altering pose, geometry, or composition while maintaining the overall structure and details of the original image.
Imitation Learning As State Matching via Differentiable Physics
Siwei Chen (National University of Singapore), Zhongwen Xu (Sea AI Lab)
OptimizationRobotic IntelligenceReinforcement Learning from Human FeedbackReinforcement LearningSequentialPhysics Related
🎯 What it does: By utilizing a differentiable physics simulator to directly embed the state matching target into gradient descent, single-loop imitation learning is achieved.
IMP: Iterative Matching and Pose Estimation With Adaptive Pooling
Fei Xue (University of Cambridge), Roberto Cipolla (University of Cambridge)
Pose EstimationTransformerPoint Cloud
🎯 What it does: This paper proposes an Iterative Matching and Pose Estimation framework (IMP), which simultaneously updates feature matching and relative pose at each step, achieving mutual promotion between matching and pose.
Implicit 3D Human Mesh Recovery Using Consistency With Pose and Shape From Unseen-View
Hanbyel Cho (Korea Advanced Institute of Science and Technology), Junmo Kim (Korea Advanced Institute of Science and Technology)
Pose EstimationNeural Radiance FieldMesh
🎯 What it does: Implicitly construct a 3D human mesh through neural feature fields, and improve the accuracy of SMPL parameter regression from a single image using arbitrary viewpoint consistency and geometric guidance self-supervision.
Implicit Diffusion Models for Continuous Super-Resolution
Sicheng Gao (Beihang University), Baochang Zhang (Beihang University)
RestorationSuper ResolutionDiffusion modelImage
🎯 What it does: Combining implicit neural representations with diffusion models, IDM is proposed to achieve high-fidelity image super-resolution at arbitrary scales.
Implicit Identity Driven Deepfake Face Swapping Detection
Baojin Huang (Wuhan University), Dengpan Ye (Wuhan University)
ClassificationRecognitionConvolutional Neural NetworkContrastive LearningImageVideo
🎯 What it does: A framework for detecting facial swap forgery based on implicit identity-driven methods is proposed, utilizing the differences between explicit and implicit identities in facial recognition to distinguish between real and forged images.
Implicit Identity Leakage: The Stumbling Block to Improving Deepfake Detection Generalization
Shichao Dong (MEGVII Technology), Zheng Ge (MEGVII Technology)
ClassificationObject DetectionConvolutional Neural NetworkImageVideo
🎯 What it does: This paper analyzes the fundamental reasons for the poor performance of deepfake detection models in cross-dataset evaluations and proposes the theory of 'implicit identity leakage'. It then designs an ID-unaware deepfake detection model based on local forgery trace detection.
Implicit Neural Head Synthesis via Controllable Local Deformation Fields
Chuhan Chen (Carnegie Mellon University), Pablo Garrido (Flawless AI)
GenerationData SynthesisNeural Radiance FieldVideo
🎯 What it does: The research proposes an implicit neural head synthesis model based on local deformation fields, achieving high-quality and controllable 3D head animations through local control.
Implicit Occupancy Flow Fields for Perception and Prediction in Self-Driving
Ben Agro (Waabi), Raquel Urtasun (Waabi)
Autonomous DrivingConvolutional Neural NetworkOptical FlowPoint Cloud
🎯 What it does: A unified implicit occupancy flow field prediction framework named IMPLICITO is proposed for the perception and future motion prediction of autonomous vehicles.
Implicit Surface Contrastive Clustering for LiDAR Point Clouds
Zaiwei Zhang (Nuro Inc), Erran Li
Object DetectionSegmentationAutonomous DrivingContrastive LearningPoint Cloud
🎯 What it does: Pre-training on large-scale LiDAR point clouds under unsupervised conditions, two self-supervised tasks are proposed: global semantic clustering and local implicit surface occupancy prediction, and the pre-trained model enhances the performance of semantic segmentation and object detection.
Implicit View-Time Interpolation of Stereo Videos Using Multi-Plane Disparities and Non-Uniform Coordinates
Avinash Paliwal (Texas A&M University), Nima Khademi Kalantari (Texas A&M University)
GenerationData SynthesisConvolutional Neural NetworkOptical FlowVideo
🎯 What it does: A perspective-time interpolation method for stereoscopic video is proposed, improving the X-Fields network to achieve low storage and near real-time perspective-time synthesis.
Improved Distribution Matching for Dataset Condensation
Ganlong Zhao (Sun Yat-sen University), Yizhou Yu (University of Hong Kong)
OptimizationData-Centric LearningImage
🎯 What it does: An improved distribution matching method is proposed for dataset condensation, allowing small-scale synthetic datasets to maintain effectiveness for model training.
Improved Test-Time Adaptation for Domain Generalization
Liang Chen (University of Adelaide), Lingqiao Liu (University of Adelaide)
Domain AdaptationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: To address the domain generalization problem, this paper proposes an improved test-time adaptation method called ITTA, which dynamically adjusts the model during the testing phase to adapt to the target domain.
Improving Commonsense in Vision-Language Models via Knowledge Graph Riddles
Shuquan Ye (Microsoft), Jing Liao (City University of Hong Kong)
RetrievalTransformerVision Language ModelImageText
🎯 What it does: This paper proposes the DANCE technique, which generates puzzle-like text and image pairs by linearizing conceptual graph knowledge and hiding entities, thereby injecting common sense knowledge into visual-language models during the training phase.
Improving Cross-Modal Retrieval With Set of Diverse Embeddings
Dongwon Kim (POSTECH), Suha Kwak (POSTECH)
RetrievalConvolutional Neural NetworkRecurrent Neural NetworkContrastive LearningImageText
🎯 What it does: This paper addresses the ambiguity issue between images and multiple descriptions in image-text cross-modal retrieval, proposing a retrieval framework based on set-based embedding.
Improving Fairness in Facial Albedo Estimation via Visual-Textual Cues
Xingyu Ren (Shanghai Jiao Tong University), Xiaokang Yang (Shanghai Jiao Tong University)
RecognitionGenerationData SynthesisPrompt EngineeringGenerative Adversarial NetworkImage
🎯 What it does: A method named ID2Albedo for illumination-independent facial albedo estimation is proposed, which directly generates high-quality, unbiased surface albedo maps using identity features.
Improving Generalization of Meta-Learning With Inverted Regularization at Inner-Level
Lianzhe Wang (Tsinghua University), Wenwu Zhu (Tsinghua University)
Meta LearningImageTabular
🎯 What it does: This paper studies the generalization problem of meta-learning at the adaptation layer and meta layer, proposing Minimax-Meta Regularization: using inverted regularization in the inner layer and standard regularization in the outer layer, and theoretically analyzing and experimentally validating its ability to enhance the generalization performance of meta-learning models on new tasks.
Improving Generalization With Domain Convex Game
Fangrui Lv (Beijing Institute of Technology), Di Liu (Beijing Institute of Technology)
Domain AdaptationMeta LearningConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: Proposes modeling domain generalization as a convex game between domains, enhancing the generalization ability of multi-source domains through hypermodel regularization and sample filtering.
Improving Graph Representation for Point Cloud Segmentation via Attentive Filtering
Nan Zhang (Peking University), Ge Li (Peking University)
SegmentationGraph Neural NetworkPoint Cloud
🎯 What it does: A hybrid graph convolutional network AF-GCN is proposed, which combines graph convolution with self-attention mechanisms for point cloud semantic segmentation.
Improving Image Recognition by Retrieving From Web-Scale Image-Text Data
Ahmet Iscen (Google Research), Cordelia Schmid (Google Research)
ClassificationRecognitionRetrievalTransformerImageTextRetrieval-Augmented Generation
🎯 What it does: A retrieval-enhanced visual classification model is proposed, which retrieves relevant samples from a large-scale image-text memory bank during prediction and learns their importance through an attention module, improving classification performance.
Improving Robust Generalization by Direct PAC-Bayesian Bound Minimization
Zifan Wang (Carnegie Mellon University), Radu Soricut (Google Research)
OptimizationAdversarial AttackTransformerImage
🎯 What it does: Proposed and implemented Trace of Hessian regularization based on minimizing the direct PAC-Bayes upper bound to enhance the generalization performance of adversarial training.
Improving Robustness of Semantic Segmentation to Motion-Blur Using Class-Centric Augmentation
Aakanksha (Indian Institute of Technology Madras), A. N. Rajagopalan (Indian Institute of Technology Madras)
SegmentationConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes a category center motion blur data augmentation method based on semantic segmentation masks (CCMBA), which helps semantic segmentation models maintain robustness on motion blurred images.
Improving Robustness of Vision Transformers by Reducing Sensitivity To Patch Corruptions
Yong Guo (Max Planck Institute for Informatics), Bernt Schiele (Max Planck Institute for Informatics)
ClassificationRecognitionTransformerImage
🎯 What it does: A new training method called RSPC is proposed, specifically designed to enhance the model's robustness against common image perturbations by reducing the sensitivity of the Transformer to patch corruption.
Improving Selective Visual Question Answering by Learning From Your Peers
Corentin Dancette (Meta AI), Marcus Rohrbach (Meta AI)
ClassificationRecognitionDomain AdaptationTransformerContrastive LearningImageMultimodality
🎯 What it does: This paper studies the selective prediction problem in Visual Question Answering (VQA) and proposes a learning method using peer models (Learning from Your Peers, LYP) to train a selector, enabling the model to self-reject when facing uncertainty in correctness.
Improving Table Structure Recognition With Visual-Alignment Sequential Coordinate Modeling
Yongshuai Huang (Huawei Technologies Ltd), Wei Peng (Huawei Technologies Ltd)
RecognitionObject DetectionTransformerContrastive LearningTabular
🎯 What it does: An end-to-end table structure recognition framework called VAST is proposed, which first generates an HTML sequence and then uses a coordinate sequence decoder to predict the bounding boxes of non-empty cells, balancing logical and physical structures.
Improving the Transferability of Adversarial Samples by Path-Augmented Method
Jianping Zhang (Chinese University of Hong Kong), Michael R. Lyu (Chinese University of Hong Kong)
Adversarial AttackGenerative Adversarial NetworkImage
🎯 What it does: This paper studies a method for generating adversarial examples based on multi-path data augmentation (Path-Augmented Method, PAM) to enhance the transferability of black-box attacks.
Improving Vision-and-Language Navigation by Generating Future-View Image Semantics
Jialu Li (University of North Carolina Chapel Hill), Mohit Bansal (University of North Carolina Chapel Hill)
GenerationRobotic IntelligenceTransformerAuto EncoderMultimodality
🎯 What it does: The VLN-SIG method is proposed, allowing the navigation agent to assist decision-making by generating the semantics of future views;
Improving Visual Grounding by Encouraging Consistent Gradient-Based Explanations
Ziyan Yang (Rice University), Vicente Ordonez (Rice University)
Object DetectionExplainability and InterpretabilityTransformerVision Language ModelImage
🎯 What it does: This paper proposes a gradient-based interpretable consistency (AMC) loss to directly optimize GradCAM heatmaps in visual language models, aligning them with human-annotated regions to enhance visual localization capabilities.
Improving Visual Representation Learning Through Perceptual Understanding
Samyakh Tukra (Tractable AI), Ken Chatfield (Tractable AI)
Object DetectionSegmentationRepresentation LearningTransformerGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: Incorporating perceptual similarity loss and adversarial training based on MAE to enhance the effectiveness of visual representation learning.
Improving Weakly Supervised Temporal Action Localization by Bridging Train-Test Gap in Pseudo Labels
Jingqiu Zhou (Chinese University of Hong Kong), Hongsheng Li (Chinese University of Hong Kong)
RecognitionObject DetectionOptimizationConvolutional Neural NetworkGaussian SplattingVideo
🎯 What it does: This paper proposes a weakly supervised temporal action localization framework that utilizes pseudo-labels generated from predicted action boundaries to bridge the gap between training and testing.
Improving Zero-Shot Generalization and Robustness of Multi-Modal Models
Yunhao Ge (Google Research), Jiaping Zhao (Google Research)
ClassificationPrompt EngineeringVision Language ModelImageMultimodality
🎯 What it does: This paper addresses zero-shot classification in multimodal image-text models by proposing a confidence estimation method based on prompt and image transformation self-consistency, and enhances labels using the WordNet hierarchy on low-confidence samples to improve prediction accuracy.
In-Hand 3D Object Scanning From an RGB Sequence
Shreyas Hampali (Reality Labs at Meta), Vincent Lepetit (LIGM, Ecole des Ponts, Univ Gustave Eiffel, CNRS)
Depth EstimationOptimizationOptical FlowImageVideo
🎯 What it does: A method is proposed for 3D shape and texture reconstruction of handheld objects using only RGB sequences, under unknown camera-object poses.
Incremental 3D Semantic Scene Graph Prediction From RGB Sequences
Shun-Cheng Wu (Technische Universität München), Federico Tombari (Google)
RecognitionObject DetectionSegmentationComputational EfficiencyGraph Neural NetworkSimultaneous Localization and MappingImageVideoPoint Cloud
🎯 What it does: A real-time incremental 3D semantic scene graph inference framework based on RGB sequences is proposed, which can continuously construct a globally consistent 3D scene graph without relying on depth information.
Incrementer: Transformer for Class-Incremental Semantic Segmentation With Knowledge Distillation Focusing on Old Class
Chao Shang (University of Electronic Science and Technology of China), Lanxiao Wang (University of Electronic Science and Technology of China)
SegmentationKnowledge DistillationTransformerImage
🎯 What it does: Proposes an incremental semantic segmentation framework called Incrementer based on Transformer, and addresses the issues of old class forgetting and new class overfitting by combining Old Class Focused Distillation (FOD) and Class Disambiguation (CDS) schemes.
Independent Component Alignment for Multi-Task Learning
Dmitry Senushkin (Samsung Research), Anton Konushin (Samsung Research)
SegmentationDepth EstimationOptimizationConvolutional Neural NetworkReinforcement LearningImage
🎯 What it does: This paper proposes a new multi-task learning optimization method called Aligned-MTL, which eliminates gradient conflicts and dominance issues by aligning the gradient matrices, thereby enhancing the stability and performance of multi-task training.
Indescribable Multi-Modal Spatial Evaluator
Lingke Kong (Manteia Tech), Qichao Zhou (Manteia Tech)
Image TranslationOptimizationConvolutional Neural NetworkGenerative Adversarial NetworkMultimodalityBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: This paper proposes a self-supervised multimodal spatial evaluator (IMSE) to address the distribution discrepancy problem in multimodal image registration, and directly drives the optimization of the registration network through the evaluator's error.