CVPR 2023 Papers — Page 9
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2353 papers
Generalist: Decoupling Natural and Robust Generalization
Hongjun Wang (Peking University), Yisen Wang (Peking University)
ClassificationOptimizationAdversarial AttackConvolutional Neural NetworkMixture of ExpertsImage
🎯 What it does: A Generalist framework is proposed, which separately trains two base learners for natural classification and adversarial robustness, and then periodically aggregates their parameters to generate a global model, thereby improving adversarial robustness while maintaining high natural accuracy.
Generalizable Implicit Neural Representations via Instance Pattern Composers
Chiheon Kim (Kakao Brain), Wook-Shin Han (POSTECH)
CompressionRepresentation LearningMeta LearningImagePoint CloudAudio
🎯 What it does: This paper proposes a method to achieve generalizable implicit neural representations by modulating a set of weights (instance pattern synthesizer) only in the early layers of a coordinate-based MLP, while allowing the remaining weights to learn general pattern combination rules.
Generalizable Local Feature Pre-Training for Deformable Shape Analysis
Souhaib Attaiki (Ecole Polytechnique), Maks Ovsjanikov (Ecole Polytechnique)
ClassificationSegmentationDomain AdaptationConvolutional Neural NetworkContrastive LearningPoint CloudMesh
🎯 What it does: Pre-trained general local features for deformable 3D shapes and proposed a differentiable receptive field optimization method, allowing features to remain efficient in cross-category transfer tasks;
Generalization Matters: Loss Minima Flattening via Parameter Hybridization for Efficient Online Knowledge Distillation
Tianli Zhang (Zhejiang University), Mingli Song (Zhejiang University)
OptimizationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: In multi-student online knowledge distillation, a Hybrid-Weight Model is formed through parameter mixing, guiding student learning with its supervised loss.
Generalized Decoding for Pixel, Image, and Language
Xueyan Zou (University of Wisconsin Madison), Jianfeng Gao (Microsoft)
SegmentationRetrievalTransformerContrastive LearningImageTextMultimodality
🎯 What it does: A universal decoder, X-Decoder, is proposed, capable of performing various visual and vision-language tasks such as pixel-level segmentation (semantic/instance/panoptic segmentation), referential segmentation, image retrieval, image captioning, and VQA within the same model framework.
Generalized Deep 3D Shape Prior via Part-Discretized Diffusion Process
Yuhan Li (Shanghai Jiao Tong University), Fuzhen Wang (Shanghai Jiao Tong University)
GenerationData SynthesisTransformerDiffusion modelAuto EncoderMultimodalityPoint CloudMesh
🎯 What it does: A unified 3D shape prior model, 3DQD, is proposed, which combines part-based discrete encoding, a discrete diffusion generator, and a multi-frequency fusion module to achieve high-quality and diverse shape generation, completion, and cross-modal generation.
Generalized Relation Modeling for Transformer Tracking
Shenyuan Gao (Hong Kong University of Science and Technology), Jun Zhang (Hong Kong University of Science and Technology)
Object TrackingTransformerVideo
🎯 What it does: A general relation modeling method GRM is proposed, which achieves adaptive interaction between the template and search area in the Transformer tracker through learnable token partitioning.
Generalized UAV Object Detection via Frequency Domain Disentanglement
Kunyu Wang (University of Science and Technology of China), Zheng-Jun Zha (University of Science and Technology of China)
Object DetectionDomain AdaptationContrastive LearningImageVideo
🎯 What it does: The study enhances the generalization ability of drone target detection in unseen domains by introducing learnable filters in the frequency domain to separate domain-invariant and domain-specific spectra.
Generalizing Dataset Distillation via Deep Generative Prior
George Cazenavette (Massachusetts Institute of Technology), Jun-Yan Zhu (Carnegie Mellon University)
GenerationData SynthesisKnowledge DistillationConvolutional Neural NetworkTransformerGenerative Adversarial NetworkImage
🎯 What it does: A GLaD method is proposed for distilling datasets in the latent space of deep generative models, parameterizing synthetic images in the intermediate latent space of the generator rather than in pixel space, to enhance the cross-architecture generalization and high-resolution performance of the distilled data.
Generating Aligned Pseudo-Supervision From Non-Aligned Data for Image Restoration in Under-Display Camera
Ruicheng Feng (Nanyang Technological University), Chen Change Loy (Nanyang Technological University)
RestorationTransformerOptical FlowImage
🎯 What it does: This paper utilizes real UDC and reference image pairs collected from a dual-camera setup, employing AlignFormer to generate high-quality, well-aligned pseudo-label data, thereby achieving end-to-end training for UDC image restoration.
Generating Anomalies for Video Anomaly Detection With Prompt-Based Feature Mapping
Zuhao Liu (Sun Yat-sen University), Wei-Shi Zheng (Sun Yat-sen University)
Domain AdaptationAnomaly DetectionConvolutional Neural NetworkPrompt EngineeringAuto EncoderVideo
🎯 What it does: A prompt-based feature mapping framework is proposed, which generates an infinite variety of anomalies using a virtual anomaly dataset and detects anomalous events in real monitoring videos.
Generating Features With Increased Crop-Related Diversity for Few-Shot Object Detection
Jingyi Xu (Stony Brook University), Dimitris Samaras (Stony Brook University)
Object DetectionAuto EncoderImageAgriculture Related
🎯 What it does: By training a conditional variational autoencoder (Norm-VAE), the generated features have controllable diversity in terms of Crop-related difficulty, thereby enhancing the robustness of few-shot object detection.
Generating Holistic 3D Human Motion From Speech
Hongwei Yi (Max Planck Institute for Intelligent Systems), Michael J. Black (Max Planck Institute for Intelligent Systems)
GenerationData SynthesisPose EstimationAuto EncoderVideoMeshAudio
🎯 What it does: This paper studies a model called TalkSHOW that generates 3D full-body motion (facial expressions, gestures, and body postures) from speech.
Generating Human Motion From Textual Descriptions With Discrete Representations
Jianrong Zhang (Jilin University), Ying Shan (Tencent AI Lab)
GenerationData SynthesisPose EstimationConvolutional Neural NetworkTransformerLarge Language ModelAuto EncoderVideoText
🎯 What it does: Text-driven human motion generation framework based on VQ-VAE and GPT
Generating Part-Aware Editable 3D Shapes Without 3D Supervision
Konstantinos Tertikas (National and Kapodistrian University of Athens), Leonidas Guibas (Stanford University)
GenerationData SynthesisTransformerNeural Radiance FieldAuto EncoderPoint CloudMesh
🎯 What it does: Proposes PartNeRF, which generates 3D shapes using local NeRF and supports editable part-level control.
Generative Bias for Robust Visual Question Answering
Jae Won Cho (Korea Advanced Institute of Science and Technology), In So Kweon (Korea Advanced Institute of Science and Technology)
GenerationKnowledge DistillationGenerative Adversarial NetworkMultimodality
🎯 What it does: A generative bias model called GenB is proposed to directly learn biases from the target VQA model and to perform debiasing training on the target model using this model.
Generative Diffusion Prior for Unified Image Restoration and Enhancement
Ben Fei, Bo Dai
RestorationDiffusion modelImage
🎯 What it does: A general image restoration prior GDP based on a pre-trained diffusion model is proposed, capable of addressing linear, nonlinear, and blind inverse problems within a single network.
Generative Semantic Segmentation
Jiaqi Chen (Fudan University), Li Zhang (Fudan University)
SegmentationGenerationTransformerAuto EncoderImage
🎯 What it does: Redefines the semantic segmentation problem as a mask generation task conditioned on images, using discrete latent variable learning to model the posterior distribution of the masks, and then learning the latent prior through an image encoder, allowing for the generation of segmentation masks given an input image.
Generic-to-Specific Distillation of Masked Autoencoders
Wei Huang (University of Chinese Academy of Sciences), Qixiang Ye (University of Chinese Academy of Sciences)
Object DetectionSegmentationKnowledge DistillationTransformerAuto EncoderImage
🎯 What it does: A two-stage knowledge distillation framework G2SD is proposed, which first performs general knowledge distillation on a pre-trained masked autoencoder, and then conducts task-specific knowledge distillation on downstream tasks to enhance the performance of lightweight Vision Transformers.
Genie: Show Me the Data for Quantization
Yongkweon Jeon (Samsung Research), Ho-young Kim (Samsung Research)
Data SynthesisOptimizationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: A zero-shot quantization framework called GENIE is proposed, which utilizes the batch normalization statistics of pre-trained models to optimize the generation of synthetic data suitable for quantization through a generator and latent vectors. The PTQ submodule GENIE-M jointly optimizes the step size and rounding to reduce information loss and enhance quantization performance.
GeoLayoutLM: Geometric Pre-Training for Visual Information Extraction
Chuwei Luo (Alibaba Group), Cong Yao (Alibaba Group)
RecognitionTransformerVision Language ModelImageTextMultimodality
🎯 What it does: A multimodal pre-training framework named GeoLayoutLM is proposed, specifically designed to learn geometric layout representations for semantic entity recognition (SER) and relation extraction (RE) tasks in visual information extraction (VIE).
GeoMAE: Masked Geometric Target Prediction for Self-Supervised Point Cloud Pre-Training
Xiaoyu Tian (Tsinghua University), Hang Zhao (Tsinghua University)
Object DetectionSegmentationAutonomous DrivingRepresentation LearningTransformerPoint Cloud
🎯 What it does: A geometry-aware self-supervised pre-training framework called GeoMAE has been designed and implemented for point cloud representation learning.
Geometric Visual Similarity Learning in 3D Medical Image Self-Supervised Pre-Training
Yuting He (Southeast University), Shuo Li (Southeast University)
SegmentationDomain AdaptationRepresentation LearningConvolutional Neural NetworkContrastive LearningImageBiomedical DataComputed Tomography
🎯 What it does: A self-supervised pre-training framework for 3D medical imaging based on geometric visual similarity learning is proposed, utilizing geometric matching to achieve cross-image semantic similarity learning and train consistent feature representations.
Geometry and Uncertainty-Aware 3D Point Cloud Class-Incremental Semantic Segmentation
Yuwei Yang (Sichuan University), Yinjie Lei (Sichuan University)
SegmentationKnowledge DistillationGraph Neural NetworkPoint Cloud
🎯 What it does: A category incremental learning framework for 3D point cloud semantic segmentation is proposed, which can gradually learn new categories and maintain the performance of old categories without storing old data.
GeoMVSNet: Learning Multi-View Stereo With Geometry Perception
Zhe Zhang (Peking University), Ronggang Wang (Peking University)
Depth EstimationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes GeoMVSNet, which explicitly injects coarse geometric information into the fine stage through a two-branch geometric fusion network and probabilistic volume embedding, enhancing the accuracy and completeness of multi-view stereo reconstruction.
GeoNet: Benchmarking Unsupervised Adaptation Across Geographies
Tarun Kalluri (University of California San Diego), Manmohan Chandraker (University of California San Diego)
ClassificationDomain AdaptationConvolutional Neural NetworkTransformerContrastive LearningImageBenchmark
🎯 What it does: This paper presents the GeoNet dataset for studying geographic adaptation and systematically evaluates the performance of existing unsupervised domain adaptation algorithms and large pre-trained models in cross-regional scene and object classification tasks.
GeoVLN: Learning Geometry-Enhanced Visual Representation With Slot Attention for Vision-and-Language Navigation
Jingyang Huo (Fudan University), Yanwei Fu (Fudan University)
Representation LearningRecurrent Neural NetworkTransformerReinforcement LearningVision Language ModelImageMultimodality
🎯 What it does: This paper proposes an indoor visual and language navigation framework named GeoVLN, which utilizes a slot attention two-stage mechanism to perform geometric enhanced representation on RGB, depth, and normal three-modal images, and combines BERT's cross-modal encoder and multiway attention module for decision-making.
GFIE: A Dataset and Baseline for Gaze-Following From 2D to 3D in Indoor Environments
Zhengxi Hu (Nankai University), Jingtai Liu (Nankai University)
Object TrackingDepth EstimationConvolutional Neural NetworkImagePoint CloudBenchmark
🎯 What it does: This paper presents the GFIE dataset and baseline methods, achieving semi-automated gaze point collection using a laser rangefinder and Azure Kinect, and provides a 2D/3D gaze tracking benchmark.
GFPose: Learning 3D Human Pose Prior With Gradient Fields
Hai Ci (National Key Laboratory of General Artificial Intelligence Beijing Institute for General Artificial Intelligence), Yizhou Wang (Peking University)
GenerationPose EstimationDiffusion modelScore-based ModelPoint CloudStochastic Differential Equation
🎯 What it does: A gradient field-based 3D human pose prior model, GFPose, is proposed, which can generate diverse poses that meet task constraints through conditional gradient reverse diffusion.
GINA-3D: Learning To Generate Implicit Neural Assets in the Wild
Bokui Shen (Stanford University), Dragomir Anguelov (Waymo LLC)
GenerationData SynthesisAutonomous DrivingNeural Radiance FieldImage
🎯 What it does: By training on sensor data from field driving, a framework called GINA-3D is proposed to generate implicit neural assets, capable of producing various 3D assets for vehicles and pedestrians.
GIVL: Improving Geographical Inclusivity of Vision-Language Models With Pre-Training Methods
Da Yin (University of California, Los Angeles), Kai-Wei Chang (University of California, Los Angeles)
Object DetectionRetrievalTransformerVision Language ModelImageTextMultimodality
🎯 What it does: A geography-inclusive visual-language pre-training model GIVL is proposed, which learns geographically diverse knowledge through a novel pre-training objective.
GKEAL: Gaussian Kernel Embedded Analytic Learning for Few-Shot Class Incremental Task
Huiping Zhuang (South China University of Technology), Ziqian Zeng (South China University of Technology)
ClassificationGaussian SplattingImage
🎯 What it does: This paper proposes a Gaussian kernel embedded analytic learning (GKEAL) method to address the problem of catastrophic forgetting in few-shot class incremental learning.
GlassesGAN: Eyewear Personalization Using Synthetic Appearance Discovery and Targeted Subspace Modeling
Richard Plesh (Clarkson University), Vitomir Struc (University of Ljubljana)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: This paper presents GlassesGAN, a personalized virtual try-on framework for optical lenses based on StyleGAN2, which can add and continuously adjust the appearance of glasses on high-resolution facial images.
GLeaD: Improving GANs With a Generator-Leading Task
Qingyan Bai (Tsinghua University), Yujun Shen (Ant Group)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a generator-led task (GLeaD) that allows the discriminator to extract sufficient information for a frozen generator to reconstruct the input image while making discriminations, thereby enhancing the representational capability of the discriminator and achieving fairer GAN training.
GLIGEN: Open-Set Grounded Text-to-Image Generation
Yuheng Li (University of Wisconsin-Madison), Yong Jae Lee (University of Wisconsin-Madison)
GenerationData SynthesisDiffusion modelImageTextMultimodality
🎯 What it does: This paper proposes the GLIGEN model, which can control generation under various grounding conditions (such as boxes, images, key points, etc.) by adding new learnable gated self-attention layers to a frozen text-to-image diffusion model.
Global and Local Mixture Consistency Cumulative Learning for Long-Tailed Visual Recognitions
Fei Du (Yunnan University), Yun Yang (Yunnan University)
ClassificationRecognitionContrastive LearningImage
🎯 What it does: A single-stage training framework GLMC is proposed, which enhances long-tail visual recognition using global and local mixed consistency loss and cumulative class-balanced reweighting loss.
Global Vision Transformer Pruning With Hessian-Aware Saliency
Huanrui Yang (University of California), Jan Kautz (NVIDIA)
ClassificationSegmentationOptimizationComputational EfficiencyKnowledge DistillationTransformerImage
🎯 What it does: Global structural pruning is performed on the Vision Transformer, reallocating dimensions of QKV, MLP, etc. within each layer to achieve efficient parameter utilization, resulting in the proposed NViT series models.
Global-to-Local Modeling for Video-Based 3D Human Pose and Shape Estimation
Xiaolong Shen (Zhejiang University), Yi Yang (Zhejiang University)
Pose EstimationTransformerVideo
🎯 What it does: A video-based 3D human pose and shape estimation framework named GLoT is proposed, which utilizes global and local Transformers to model long-term and short-term information respectively, and achieves synergy between the two through cross-attention.
Glocal Energy-Based Learning for Few-Shot Open-Set Recognition
Haoyu Wang (Northwestern Polytechnical University), Yanning Zhang (Northwestern Polytechnical University)
ClassificationRecognitionContrastive LearningImage
🎯 What it does: A Glocal Energy-based Learning (GEL) framework for few-shot open set recognition is proposed.
Gloss Attention for Gloss-Free Sign Language Translation
Aoxiong Yin (Zhejiang University), Zhou Zhao (Zhejiang University)
TransformerVideo
🎯 What it does: This paper proposes a Gloss Attention mechanism for sign language translation without subtitles, utilizing semantic temporal locality and knowledge transfer to compensate for the lack of Gloss supervision.
GM-NeRF: Learning Generalizable Model-Based Neural Radiance Fields From Multi-View Images
Jianchuan Chen (Dalian University of Technology), Huchuan Lu (Dalian University of Technology)
GenerationData SynthesisPose EstimationNeural Radiance FieldImage
🎯 What it does: This paper proposes a general geometric model-based neural radiance field (GM-NeRF) framework that generates high-fidelity free-viewpoint images of arbitrary human shapes from sparse multi-view images.
Good Is Bad: Causality Inspired Cloth-Debiasing for Cloth-Changing Person Re-Identification
Zhengwei Yang (Wuhan University), Zheng Wang (Wuhan University)
RecognitionRetrievalKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: A dual-branch model based on causal self-intervention (AIM) is proposed to automatically eliminate clothing bias and enhance the performance of clothing variation person re-identification.
GP-VTON: Towards General Purpose Virtual Try-On via Collaborative Local-Flow Global-Parsing Learning
Zhenyu Xie (Shenzhen Campus of Sun Yat-Sen University), Xiaodan Liang (Shenzhen Campus of Sun Yat-Sen University)
Image TranslationSegmentationGenerationPose EstimationConvolutional Neural NetworkGenerative Adversarial NetworkOptical FlowImage
🎯 What it does: A general-purpose virtual try-on framework GP-VTON is proposed, which achieves high-resolution virtual try-on for different categories of clothing and can handle complex poses and garment shapes.
Grad-PU: Arbitrary-Scale Point Cloud Upsampling via Gradient Descent With Learned Distance Functions
Yun He (Fudan University), Yanwei Fu (Fudan University)
GenerationOptimizationPoint Cloud
🎯 What it does: This paper proposes a point cloud upsampling method that supports arbitrary magnification: first, a coarse-grained upsampling result is obtained through midpoint interpolation in Euclidean space, and then iterative optimization is performed using a learned point-to-point distance function to achieve high-precision high-resolution point clouds.
GradICON: Approximate Diffeomorphisms via Gradient Inverse Consistency
Lin Tian (University of North Carolina at Chapel Hill), Marc Niethammer (University of North Carolina at Chapel Hill)
Image TranslationOptimizationConvolutional Neural NetworkBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: Proposes GradICON regularization, utilizing gradient inverse consistency to train medical image registration networks, achieving high-quality, seamless deformations.
Gradient Norm Aware Minimization Seeks First-Order Flatness and Improves Generalization
Xingxuan Zhang (Tsinghua University), Peng Cui (Tsinghua University)
OptimizationConvolutional Neural NetworkTransformerImage
🎯 What it does: The concept of first-order flatness is proposed, and based on this, the Gradient Norm Aware Minimization (GAM) algorithm is designed to explicitly optimize the gradient norm of model weights during training, thereby seeking flatter minima and enhancing the model's generalization performance.
Gradient-Based Uncertainty Attribution for Explainable Bayesian Deep Learning
Hanjing Wang (Rensselaer Polytechnic Institute), Qiang Ji (Rensselaer Polytechnic Institute)
Explainability and InterpretabilityConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A gradient-based Bayesian deep learning uncertainty attribution method (UA-Backprop) is proposed, and the attribution results are used as an attention mechanism for uncertainty mitigation.
GradMA: A Gradient-Memory-Based Accelerated Federated Learning With Alleviated Catastrophic Forgetting
Kangyang Luo (East China Normal University), Ming Gao (East China Normal University)
Federated LearningConvolutional Neural NetworkImage
🎯 What it does: This paper proposes GradMA, an accelerated federated learning framework that simultaneously utilizes gradient memory on both the server and client sides, aimed at alleviating catastrophic forgetting caused by data heterogeneity and partial participation.
Graph Representation for Order-Aware Visual Transformation
Yue Qiu (National Institute of Advanced Industrial Science and Technology), Hirokatsu Kataoka (National Institute of Advanced Industrial Science and Technology)
RecognitionData SynthesisGraph Neural NetworkTransformerImage
🎯 What it does: This paper proposes a new visual reasoning task: identifying changes from two images and inferring their temporal order.
Graph Transformer GANs for Graph-Constrained House Generation
Hao Tang (ETH Zurich), Luc Van Gool (ETH Zurich)
GenerationData SynthesisGraph Neural NetworkTransformerGenerative Adversarial NetworkGraph
🎯 What it does: This paper proposes a Graph Transformer-based Generative Adversarial Network (GTGAN) for generating realistic house layouts and roof structures from graph constraints (room connectivity relationships).
Graphics Capsule: Learning Hierarchical 3D Face Representations From 2D Images
Chang Yu (Chinese Academy of Sciences), Zhen Lei (Chinese Academy of Sciences)
SegmentationRepresentation LearningAuto EncoderContrastive LearningImage
🎯 What it does: Using the unsupervised image learning framework IGC-Net, hierarchical 3D representations of faces are learned and decomposed into semantically consistent parts.
GraVoS: Voxel Selection for 3D Point-Cloud Detection
Oren Shrout (Technion), Ayellet Tal (Technion)
Object DetectionAutonomous DrivingConvolutional Neural NetworkPoint Cloud
🎯 What it does: A gradient-based voxel selection method (GraVoS) is proposed to select important voxels for the network during the training phase of 3D point cloud detection, aiming to balance foreground/background and class imbalance.
GRES: Generalized Referring Expression Segmentation
Chang Liu (Nanyang Technological University), Xudong Jiang (Nanyang Technological University)
Object DetectionSegmentationTransformerVision Language ModelImageText
🎯 What it does: A general framework for referring expression segmentation (GRES) is proposed, supporting an arbitrary number of target and non-target expressions. A large-scale gRefCOCO dataset is constructed, and a ReLA baseline model based on Region-Language interaction is designed.
Grid-Guided Neural Radiance Fields for Large Urban Scenes
Linning Xu (Chinese University of Hong Kong), Dahua Lin (Chinese University of Hong Kong)
GenerationData SynthesisNeural Radiance FieldImage
🎯 What it does: To achieve high-quality view synthesis in large-scale urban scenarios, a dual-branch grid-guided NeRF model is proposed.
Ground-Truth Free Meta-Learning for Deep Compressive Sampling
Xinran Qin (South China University of Technology), Hui Ji (National University of Singapore)
RestorationMeta LearningConvolutional Neural NetworkImageBiomedical DataMagnetic Resonance ImagingAlzheimer's Disease
🎯 What it does: A target-free (GT) meta-learning method is proposed for compressed sensing (CS) image reconstruction, which utilizes external measurement data for unsupervised training and captures internal sample features through model fine-tuning (adaptation) during testing.
Grounding Counterfactual Explanation of Image Classifiers to Textual Concept Space
Siwon Kim (Seoul National University), Tara Taghavi (Amazon)
Explainability and InterpretabilityPrompt EngineeringContrastive LearningImageTextMultimodality
🎯 What it does: This paper proposes a concept counterfactual explanation framework called CounTEX, which does not require manual annotation of images. It directly generates concept directions using text prompts in the joint embedding space of CLIP and maps them to the target classifier through projection/inverse projection to obtain concept importance scores.
GrowSP: Unsupervised Semantic Segmentation of 3D Point Clouds
Zihui Zhang (Hong Kong Polytechnic University), Bo Li (Hong Kong Polytechnic University)
SegmentationConvolutional Neural NetworkPoint Cloud
🎯 What it does: A completely unsupervised 3D point cloud semantic segmentation method called GrowSP is proposed, which achieves point-level semantic segmentation using progressively expanded superpoints and semantic primitive clustering.
gSDF: Geometry-Driven Signed Distance Functions for 3D Hand-Object Reconstruction
Zerui Chen (Inria), Ivan Laptev (Inria)
Object DetectionPose EstimationTransformerImageVideoMesh
🎯 What it does: This paper proposes a geometry-driven Signed Distance Field (gSDF) model for reconstructing the 3D mesh of hands and handheld objects from a monocular RGB image.
Guided Depth Super-Resolution by Deep Anisotropic Diffusion
Nando Metzger (ETH Zurich), Konrad Schindler (ETH Zurich)
RestorationDepth EstimationSuper ResolutionConvolutional Neural NetworkDiffusion modelImagePoint Cloud
🎯 What it does: A guided deep super-resolution method that integrates deep learning with anisotropic diffusion is proposed, utilizing RGB guidance images to achieve high-quality depth image magnification.
Guided Recommendation for Model Fine-Tuning
Hao Li (Amazon Web Services Artificial Intelligence Labs), Stefano Soatto (Amazon Web Services Artificial Intelligence Labs)
Recommendation SystemImage
🎯 What it does: This paper studies a model selection recommendation system based on historical training records to predict the performance of pre-trained models after fine-tuning on downstream tasks.
Guiding Pseudo-Labels With Uncertainty Estimation for Source-Free Unsupervised Domain Adaptation
Mattia Litrico (Istituto Italiano di Tecnologia), Pietro Morerio (Istituto Italiano di Tecnologia)
Domain AdaptationContrastive LearningImage
🎯 What it does: A source-free unsupervised domain adaptation method is proposed, which enhances target domain performance by re-weighting pseudo-labels through neighbor knowledge aggregation and entropy estimation.
H2ONet: Hand-Occlusion-and-Orientation-Aware Network for Real-Time 3D Hand Mesh Reconstruction
Hao Xu (Chinese University of Hong Kong), Chi-Wing Fu (Chinese University of Hong Kong)
Pose EstimationConvolutional Neural NetworkMesh
🎯 What it does: A real-time 3D hand mesh reconstruction network called H2ONet is proposed, which utilizes multi-frame information and finger-level and hand-level occlusion awareness to enhance reconstruction quality in hand-object interaction scenarios.
HAAV: Hierarchical Aggregation of Augmented Views for Image Captioning
Chia-Wen Kuo (Georgia Tech), Zsolt Kira (Georgia Tech)
Object DetectionGenerationTransformerContrastive LearningImageText
🎯 What it does: This paper proposes a method for generating image captions using multiple heterogeneous image views (object detection, grid features, text descriptions). The method treats each view as an enhanced version of the image, independently encoding them using a shared Transformer encoder, and improving encoding quality through contrastive loss; subsequently, a hierarchical decoder aggregates information at both the token level and view level, dynamically weighing the contribution of each view to caption generation.
Habitat-Matterport 3D Semantics Dataset
Karmesh Yadav (Meta AI), Devendra Singh Chaplot (Meta AI)
SegmentationRobotic IntelligenceReinforcement LearningPoint Cloud
🎯 What it does: The HM3DSEM dataset has been released, constructing the largest and most finely annotated collection of real-world 3D indoor spaces, and its effectiveness has been validated in Embodied AI tasks.
HairStep: Transfer Synthetic to Real Using Strand and Depth Maps for Single-View 3D Hair Modeling
Yujian Zheng (Chinese University of Hong Kong Shenzhen), Xiaoguang Han (Chinese University of Hong Kong Shenzhen)
GenerationDepth EstimationDomain AdaptationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A new intermediate representation called HairStep is proposed, which transforms a single portrait into a dual-channel feature containing directional hair strand mapping and a depth map, and based on this, reconstructs a high-fidelity 3D hair model.
HaLP: Hallucinating Latent Positives for Skeleton-Based Self-Supervised Learning of Actions
Anshul Shah (Johns Hopkins University), Rama Chellappa (Johns Hopkins University)
RecognitionOptimizationRepresentation LearningContrastive LearningVideo
🎯 What it does: This paper proposes enhancing contrastive learning in self-supervised learning for skeleton action recognition by synthesizing hard positive samples (HaLP) in the latent space.
Ham2Pose: Animating Sign Language Notation Into Pose Sequences
Rotem Shalev Arkushin (Reichman University), Ohad Fried (Reichman University)
GenerationPose EstimationTransformerVideoText
🎯 What it does: A method for directly converting HamNoSys characters into a sequence of gesture poses is proposed, constructing the first general SLP (Sign Language Production) system aimed at multiple sign languages.
Hand Avatar: Free-Pose Hand Animation and Rendering From Monocular Video
Xingyu Chen, Heung-Yeung Shum
GenerationPose EstimationNeural Radiance FieldVideoMesh
🎯 What it does: We propose HandAvatar, a neural hand avatar framework that achieves free-pose hand animation and high-quality rendering from monocular videos.
HandNeRF: Neural Radiance Fields for Animatable Interacting Hands
Zhiyang Guo (University of Science and Technology of China), Houqiang Li (University of Science and Technology of China)
GenerationPose EstimationKnowledge DistillationNeural Radiance FieldMesh
🎯 What it does: Proposes the HandNeRF framework, which uses neural radiance fields to reconstruct the high-quality appearance and geometry of hands (single or dual hand interactions) in arbitrary poses and viewpoints.
HandsOff: Labeled Dataset Generation With No Additional Human Annotations
Austin Xu (Georgia Institute of Technology), Arjun Seshadri
SegmentationData SynthesisDepth EstimationGenerative Adversarial NetworkImage
🎯 What it does: Using GAN inversion technology and a label generator, a small number of existing labeled images (≤50) are mapped to the W+ space of StyleGAN2, thereby generating an unlimited number of high-quality image-label pairs that support tasks such as semantic segmentation, keypoint detection, and depth estimation.
Handwritten Text Generation From Visual Archetypes
Vittorio Pippi (University of Modena and Reggio Emilia), Rita Cucchiara (University of Modena and Reggio Emilia)
GenerationTransformerGenerative Adversarial NetworkImageText
🎯 What it does: A few-shot handwritten text generation model based on Transformer, VATr, is proposed, which utilizes visual prototypes to encode and generate images of a specific author's writing style.
Handy: Towards a High Fidelity 3D Hand Shape and Appearance Model
Rolandos Alexandros Potamias (Imperial College London), Stefanos Zafeiriou (Imperial College London)
GenerationData SynthesisPose EstimationGenerative Adversarial NetworkPoint CloudMesh
🎯 What it does: This work presents Handy, a high-fidelity hand shape and texture parameter model based on over 1200 hand scans, capable of reconstructing 3D hand shapes and high-frequency textures from a single photo.
Hard Patches Mining for Masked Image Modeling
Haochen Wang (Chinese Academy of Sciences), Zhaoxiang Zhang (Chinese Academy of Sciences)
ClassificationObject DetectionSegmentationTransformerContrastive LearningImage
🎯 What it does: This paper proposes a Hard Patches Mining (HPM) framework, where the model acts as both student and teacher during MIM pre-training, generating more challenging occlusion tasks by predicting the reconstruction loss of each patch, and jointly training the reconstruction network and loss predictor.
Hard Sample Matters a Lot in Zero-Shot Quantization
Huantong Li (South China University of Technology), Mingkui Tan (South China University of Technology)
CompressionOptimizationAdversarial AttackImage
🎯 What it does: This paper addresses the issue of performance degradation caused by the easy fitting of synthetic samples in zero-shot quantization, proposing the Hard Sample Synthesis and Training (HAST) method.
Harmonious Feature Learning for Interactive Hand-Object Pose Estimation
Zhifeng Lin (South China University of Technology), Shaoli Huang (Tencent AI Lab)
Pose EstimationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a new Harmonious Feature Learning Network (HFL-Net) for simultaneously estimating the 3D poses of hands and objects from a single RGB image.
Harmonious Teacher for Cross-Domain Object Detection
Jinhong Deng (University of Electronic Science and Technology of China), Lixin Duan (Shenzhen Institute for Advanced Study, University of Electronic Science and Technology of China)
Object DetectionDomain AdaptationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: Proposes the Harmonious Teacher framework, which enhances the classification and localization consistency of detection models in the source and target domains through self-supervised and unsupervised harmonious loss, and improves the self-training method for cross-domain object detection by using harmonious metrics for threshold-free weighting of pseudo-labels.
HARP: Personalized Hand Reconstruction From a Monocular RGB Video
Korrawe Karunratanakul (ETH Zurich), Siyu Tang (ETH Zurich)
GenerationPose EstimationImageVideo
🎯 What it does: This paper proposes a personalized hand avatar generation method called HARP, based on an explicit hand model and differentiable rendering, which can recover high-quality hand geometry, texture, and lighting information from short monocular RGB videos.
HDR Imaging With Spatially Varying Signal-to-Noise Ratios
Yiheng Chi (Purdue University), Stanley H. Chan (Purdue University)
RestorationTransformerImage
🎯 What it does: A network called SV-HDR is proposed for simultaneous denoising and HDR fusion, aimed at processing images with spatially varying signal-to-noise ratios in photon-limited scenes.
Heat Diffusion Based Multi-Scale and Geometric Structure-Aware Transformer for Mesh Segmentation
Chi-Chong Wong (University of Macau)
SegmentationTransformerMesh
🎯 What it does: This paper proposes MeshFormer, which combines Heat Diffusion Multi-Scale Self-Attention (HDMSA) and Heat Kernel Signature Structure Encoding (HKSSE) to achieve semantic segmentation of triangular meshes.
HelixSurf: A Robust and Efficient Neural Implicit Surface Learning of Indoor Scenes With Iterative Intertwined Regularization
Zhihao Liang (South China University of Technology), Kui Jia (South China University of Technology)
Depth EstimationComputational EfficiencyNeural Radiance FieldPoint CloudMesh
🎯 What it does: For surface reconstruction of multi-view indoor scenes, the HelixSurf method is proposed, which combines traditional PatchMatch MVS with neural implicit surface learning, and enhances geometric details and robustness through mutually iterative regularization.
Heterogeneous Continual Learning
Divyam Madaan (New York University), Pavlo Molchanov (NVIDIA)
Computational EfficiencyKnowledge DistillationImage
🎯 What it does: A heterogeneous continual learning framework HCL is proposed, allowing for dynamic switching to more powerful network architectures between tasks, and achieving knowledge transfer without data through knowledge distillation and quick deep inversion.
HexPlane: A Fast Representation for Dynamic Scenes
Ang Cao (University of Michigan), Justin Johnson (University of Michigan)
Data SynthesisComputational EfficiencyNeural Radiance FieldVideo
🎯 What it does: Proposes HexPlane, which explicitly represents 4D spatiotemporal voxels through six feature planes, enabling rapid reconstruction and viewpoint synthesis of dynamic scenes.
HGFormer: Hierarchical Grouping Transformer for Domain Generalized Semantic Segmentation
Jian Ding (Wuhan University), Dengxin Dai (Max Planck Institute for Informatics)
SegmentationDomain AdaptationTransformerImage
🎯 What it does: This paper proposes a Hierarchical Grouping Transformer (HGFormer), which first clusters pixels into part-level masks, then aggregates them into a whole-level mask, and performs classification on both levels of masks to obtain the final semantic segmentation results, enhancing cross-domain generalization robustness.
HGNet: Learning Hierarchical Geometry From Points, Edges, and Surfaces
Ting Yao (HiDream.ai Inc), Tao Mei (HiDream.ai Inc)
ClassificationSegmentationGraph Neural NetworkTransformerPoint Cloud
🎯 What it does: Proposes the Hierarchical Geometry Network (HGNet), which constructs a four-level hierarchical topology of points, edges, faces, and hyperfaces to aggregate features from top to bottom, and then uses a Transformer to unify multi-layer geometric features, achieving tasks such as point cloud classification and segmentation.
Hi-LASSIE: High-Fidelity Articulated Shape and Skeleton Discovery From Sparse Image Ensemble
Chun-Han Yao (University of California Merced), Varun Jampani (Google Research)
SegmentationGenerationPose EstimationTransformerContrastive LearningImage
🎯 What it does: Hi-LASSIE automatically learns the 3D skeleton, shape, camera perspective, and part deformations of animal categories from a small number (20-30) of field animal images to generate high-fidelity 3D models.
Hi4D: 4D Instance Segmentation of Close Human Interaction
Yifei Yin (ETH Zurich), Otmar Hilliges (ETH Zurich)
SegmentationPose EstimationVideo
🎯 What it does: A method for interactive human 4D instance segmentation and pose shape estimation using personalized neural implicit portraits (SNARF) is proposed, and based on this, the Hi4D dataset is constructed.
Hidden Gems: 4D Radar Scene Flow Learning Using Cross-Modal Supervision
Fangqiang Ding (University of Edinburgh), Chris Xiaoxuan Lu (Delft University of Technology)
Autonomous DrivingSimultaneous Localization and MappingOptical FlowMultimodalityPoint Cloud
🎯 What it does: This paper proposes a 4D radar scene flow estimation method based on multi-modal collaborative supervision, using co-located radar, LiDAR, camera, and odometry data for training without manual labeling.
HIER: Metric Learning Beyond Class Labels via Hierarchical Regularization
Sungyeon Kim (POSTECH), Suha Kwak (POSTECH)
RetrievalRepresentation LearningContrastive LearningImage
🎯 What it does: Proposes the HIER regularization method, which utilizes hierarchical proxies to self-supervise the learning of hidden semantic hierarchies in hyperbolic space and incorporates it as additional supervision into deep metric learning;
Hierarchical B-Frame Video Coding Using Two-Layer CANF Without Motion Coding
David Alexandre (National Yang Ming Chiao Tung University), Wen-Hsiao Peng (National Yang Ming Chiao Tung University)
CompressionVideo
🎯 What it does: A B-frame video coding framework based on a two-layer Conditional Augmented Normalizing Flow (CANF) is proposed, which completely avoids motion coding.
Hierarchical Dense Correlation Distillation for Few-Shot Segmentation
Bohao Peng (Chinese University of Hong Kong), Jiaya Jia (Chinese University of Hong Kong)
SegmentationKnowledge DistillationTransformerImage
🎯 What it does: A Hierarchically Decoupled Matching Network (HDMNet) is proposed for few-shot semantic segmentation.
Hierarchical Discriminative Learning Improves Visual Representations of Biomedical Microscopy
Cheng Jiang (University of Michigan), Todd C. Hollon (University of Michigan)
Representation LearningConvolutional Neural NetworkContrastive LearningImageBiomedical Data
🎯 What it does: A self-supervised contrastive learning framework called HiDisc is proposed, based on a patient-slice-patch hierarchy, to enhance the visual representation learning of medical microscopy images.
Hierarchical Fine-Grained Image Forgery Detection and Localization
Xiao Guo (Michigan State University), Xiaoming Liu (Michigan State University)
Anomaly DetectionConvolutional Neural NetworkImage
🎯 What it does: This paper studies a unified method for image forgery detection and localization in the fields of CNN synthesis and image editing.
Hierarchical Neural Memory Network for Low Latency Event Processing
Ryuhei Hamaguchi (National Institute of Advanced Industrial Science and Technology), Ken Sakurada (National Institute of Advanced Industrial Science and Technology)
Object DetectionSegmentationDepth EstimationComputational EfficiencyConvolutional Neural NetworkTransformerImageMultimodality
🎯 What it does: A low-latency event processing framework called HMNet is proposed, which utilizes a multi-level asynchronous memory network to achieve real-time dense predictions of events and images (semantic segmentation, object detection, monocular depth estimation, etc.).
Hierarchical Prompt Learning for Multi-Task Learning
Yajing Liu (JD Logistics), Chenguang Gui (University of Science and Technology of China)
ClassificationDomain AdaptationTransformerPrompt EngineeringVision Language ModelImageMultimodality
🎯 What it does: This paper proposes a Hierarchical Prompt Learning (HiPro) method in the context of multi-task learning to jointly adapt visual-language pre-trained models, enhancing the performance of multiple visual tasks.
Hierarchical Semantic Contrast for Scene-Aware Video Anomaly Detection
Shengyang Sun (Zhejiang University), Xiaojin Gong (Zhejiang University)
Anomaly DetectionAuto EncoderContrastive LearningVideo
🎯 What it does: A hierarchical semantic contrast (HSC) method is proposed for learning scene-aware video anomaly detection models.
Hierarchical Semantic Correspondence Networks for Video Paragraph Grounding
Chaolei Tan (Sun Yat-sen University), Jianhuang Lai (Sun Yat-sen University)
RecognitionObject DetectionConvolutional Neural NetworkRecurrent Neural NetworkVision Language ModelVideoText
🎯 What it does: This paper proposes a Hierarchical Semantic Correspondence Network (HSCNet) for video paragraph localization, achieving more accurate video event localization through multi-level visual-text correspondence.
Hierarchical Supervision and Shuffle Data Augmentation for 3D Semi-Supervised Object Detection
Chuandong Liu (Chongqing University of Posts and Telecommunications), Xinbo Gao (Chongqing University of Posts and Telecommunications)
Object DetectionAutonomous DrivingKnowledge DistillationPoint Cloud
🎯 What it does: Using a teacher-student framework, the pseudo-labels output by the teacher network are dynamically divided into three groups: high confidence, ambiguous, and low confidence. Shuffle data augmentation is introduced in the student network to enhance feature representation capabilities.
Hierarchical Temporal Transformer for 3D Hand Pose Estimation and Action Recognition From Egocentric RGB Videos
Yilin Wen (University of Hong Kong), Wenping Wang (Texas A&M University)
RecognitionPose EstimationTransformerVideo
🎯 What it does: This paper proposes a hierarchical temporal Transformer framework that can simultaneously estimate 3D hand poses and recognize hand actions in first-person RGB videos.
Hierarchical Video-Moment Retrieval and Step-Captioning
Abhay Zala (University of North Carolina at Chapel Hill), Mohit Bansal (University of North Carolina at Chapel Hill)
GenerationRetrievalTransformerSupervised Fine-TuningVision Language ModelVideoTextMultimodality
🎯 What it does: The HIREST dataset and a four-level video retrieval and step generation task are proposed, constructing an end-to-end retrieval and decomposition framework.
HierVL: Learning Hierarchical Video-Language Embeddings
Kumar Ashutosh (University of Texas at Austin), Kristen Grauman (Meta AI)
RecognitionRepresentation LearningTransformerVision Language ModelContrastive LearningVideoText
🎯 What it does: A hierarchical video-language embedding model, HierVL, is proposed, which can simultaneously learn short-term (segment-level) and long-term (whole video-level) alignment between video and text.