ECCV 2024 Papers — Page 10
European Conference on Computer Vision · 2387 papers
GenRC: Generative 3D Room Completion from Sparse Image Collections
Ming-Feng Li (Carnegie Mellon University), Min Sun (National Yang Ming Chiao Tung University)
GenerationData SynthesisDepth EstimationPrompt EngineeringDiffusion modelImageMesh
🎯 What it does: This paper proposes a training-free automated pipeline that can convert sparse RGB-D images into complete, texture-preserving high-fidelity 3D meshes of indoor rooms.
GenView: Enhancing View Quality with Pretrained Generative Model for Self-Supervised Learning
Xiaojie Li (Harbin Institute of Technology), Min Zhang (Harbin Institute of Technology)
ClassificationObject DetectionSegmentationRepresentation LearningConvolutional Neural NetworkTransformerDiffusion modelContrastive LearningImage
🎯 What it does: Studied a self-supervised learning view generation framework called GenView based on pre-trained generative models to enhance the quality and diversity of positive sample views.
GeoCalib: Learning Single-image Calibration with Geometric Optimization
Alexander Veicht (ETH Zurich), Marc Pollefeys (ETH Zurich)
Pose EstimationOptimizationConvolutional Neural NetworkImage
🎯 What it does: Propose GeoCalib, a method that combines a deep network to predict the Perspective Field with differentiable geometric optimization (Levenberg–Marquardt) to achieve single-image camera calibration (focal length, gravity direction, distortion).
GeoGaussian: Geometry-aware Gaussian Splatting for Scene Rendering
Yanyan Li (National University of Singapore), Federico Tombari (Technical University of Munich)
GenerationGaussian SplattingImagePoint Cloud
🎯 What it does: This paper proposes GeoGaussian, a geometry-based Gaussian Splatting method aimed at enhancing geometric fidelity and image quality in scene rendering.
Geometry Fidelity for Spherical Images
Anders Christensen (Technical University of Denmark), Andrea Colaco (Google)
Convolutional Neural NetworkImage
🎯 What it does: This paper proposes two new metrics for evaluating the geometric fidelity of spherical images: OmniFID and Discontinuity Score (DS).
GeometrySticker: Enabling Ownership Claim of Recolorized Neural Radiance Fields
Xiufeng HUANG, Renjie Wan (NVIDIA)
Safty and PrivacyConvolutional Neural NetworkNeural Radiance FieldImage
🎯 What it does: Propose the GeometrySticker method, which can embed binary watermarks into the geometric surface points of NeRF, enabling the watermarks to remain recoverable even after recolorization (e.g., CLIP-based, palette-based), thereby achieving ownership claims for recolorized NeRF.
Geospecific View Generation - Geometry-Context Aware High-resolution Ground View Inference from Satellite Views
Ningli Xu (Ohio State University), Rongjun Qin (Ohio State University)
Image TranslationGenerationConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: This paper proposes a complete cross-perspective synthesis pipeline for generating high-resolution ground panoramic views from multi-view satellite images, focusing on leveraging satellite texture and geometric information to produce realistic ground images with high geographical consistency.
GeoWizard: Unleashing the Diffusion Priors for 3D Geometry Estimation from a Single Image
Xiao Fu (Chinese University of Hong Kong), Xiaoxiao Long (University of Hong Kong)
Depth EstimationVision Language ModelDiffusion modelImage
🎯 What it does: Proposed a generic single-image depth and normal estimation framework GeoWizard based on diffusion models.
Get Your Embedding Space in Order: Domain-Adaptive Regression for Forest Monitoring
Sizhuo Li (University of Copenhagen), Philippe Ciais (LSCE)
Domain AdaptationTransformerDiffusion modelImageAgriculture Related
🎯 What it does: This paper proposes a cross-border forest monitoring image-level regression framework: first, constructing and publicly releasing the DRIFT dataset, then training an ordered embedding space on the source domain using Geometric Order Learning (GOL), and finally achieving regression of canopy height, tree count, and crown coverage for high-resolution remote sensing images across domains through few-shot adaptation on the target domain and transductive post-processing via Manifold Diffusion (MDR).
Getting it Right: Improving Spatial Consistency in Text-to-Image Models
Agneet Chatterjee (Arizona State University), Yezhou Yang (Arizona State University)
GenerationLarge Language ModelSupervised Fine-TuningDiffusion modelImageText
🎯 What it does: This paper re-annotates existing visual-language datasets to construct a spatial relationship annotation dataset called SPRIGHT, containing over 60,000 images, and performs efficient fine-tuning of Stable Diffusion on this dataset, significantly improving the model's performance in spatial consistency.
GGRt: Towards Generalizable 3D Gaussians without Pose Priors in Real-Time
Hao Li (Northwestern Polytechnical University), Junwei Han (Baidu Inc.)
GenerationConvolutional Neural NetworkTransformerNeural Radiance FieldGaussian SplattingImage
🎯 What it does: Designed the first pose-free generalizable 3D Gaussian Splatting (3D-GS) framework, GGRt, which jointly learns an Iterative Pose Optimization Network (IPO-Net) and a general 3D-Gaussian model (G-3DG) to adaptively estimate relative poses from unlabelled multi-view images and generate high-resolution images renderable in real-time from arbitrary new viewpoints.
GiT: Towards Generalist Vision Transformer through Universal Language Interface
Haiyang Wang (Peking University), Liwei Wang (Peking University)
Representation LearningTransformerVision Language ModelImage
🎯 What it does: Developed a general-purpose visual Transformer named GiT that solely uses ViT, unifying various visual tasks.
GIVT: Generative Infinite-Vocabulary Transformers
Michael Tschannen (Google DeepMind), Fabian Mentzer (Google DeepMind)
SegmentationGenerationDepth EstimationTransformerMixture of ExpertsAuto EncoderImage
🎯 What it does: Propose Generative Infinite-Vocabulary Transformer (GIVT), which directly generates continuous vector sequences instead of discrete tokens, using the continuous latent space of β-VAE as input; two modifications are made to the Transformer: input uses linear projection instead of vocabulary embeddings, and output uses a multivariate Gaussian mixture model instead of classification distribution.
GKGNet: Group K-Nearest Neighbor based Graph Convolutional Network for Multi-Label Image Recognition
Ruijie Yao (Tsinghua University), Ji Wu (Tsinghua University)
RecognitionGraph Neural NetworkImage
🎯 What it does: Proposes a fully graph convolution-based multi-label image recognition model called GKGNet, which utilizes a unified graph structure to combine image patches and label embeddings for interaction, enabling more accurate label predictions on targets of different scales and shapes.
GLAD: Towards Better Reconstruction with Global and Local Adaptive Diffusion Models for Unsupervised Anomaly Detection
Hang Yao (Harbin Institute of Technology), Wangmeng Zuo (Harbin Institute of Technology)
Anomaly DetectionDiffusion modelContrastive LearningImage
🎯 What it does: Proposed a global and local adaptive diffusion model GLAD for unsupervised anomaly detection, achieving more accurate reconstruction and anomaly localization by dynamically determining denoising steps per image and fusing local features.
GLARE: Low Light Image Enhancement via Generative Latent Feature based Codebook Retrieval
Han Zhou (McMaster University), Jun Chen (McMaster University)
RestorationFlow-based ModelAuto EncoderImage
🎯 What it does: Proposes the GLARE method, which utilizes three modules: a VQ codebook learned from normally illuminated images, invertible latent normalization flows, and adaptive feature transformations to achieve natural enhancement of low-light images to normal illumination.
Global Counterfactual Directions
Bartłomiej Sobieski (Warsaw University of Technology), Przemyslaw Biecek (Warsaw University of Technology)
Explainability and InterpretabilityDiffusion modelAuto EncoderImageBiomedical Data
🎯 What it does: Proposes a black-box visual counterfactual explanation method that generates cross-image set counterfactual images by leveraging global directions (GCDs) in the semantic latent space of a diffusion autoencoder (DiffAE), and implements black-box latent integral gradient (BB-LIG) explanations using finite differences.
Global Structure-from-Motion Revisited
Linfei Pan (ETH Zurich), Johannes L Schönberger
Pose EstimationRetrievalOptimizationSimultaneous Localization and MappingImageBenchmark
🎯 What it does: Proposed a Global Structure Optical Flow (GLOMAP) system that achieves joint localization of camera pose and 3D points, bypassing traditional translation averaging plus global triangulation steps;
Global-Local Collaborative Inference with LLM for Lidar-Based Open-Vocabulary Detection
Xingyu Peng (Beihang University), Si Liu (Meituan)
Object DetectionTransformerLarge Language ModelPoint CloudChain-of-Thought
🎯 What it does: This paper proposes a global-local collaborative reasoning framework called GLIS, which combines global scene information and local object information using LiDAR point clouds, and employs a Large Language Model (LLM) for chain reasoning to achieve open-vocabulary detection in point clouds.
Global-to-Pixel Regression for Human Mesh Recovery
Yabo Xiao (Beijing University of Posts and Telecommunications), Dongdong Yu (AISphere Tech)
Pose EstimationConvolutional Neural NetworkImageMesh
🎯 What it does: Propose a global-to-pixel-level prediction framework (GLNet), achieving fine recovery of human meshes through a 2D keypoint-guided local encoding module and dynamic matching strategy, while maintaining visual mesh alignment.
GlobalPointer: Large-Scale Plane Adjustment with Bi-Convex Relaxation
Bangyan Liao (Zhejiang University), Peidong Liu (Westlake University)
Pose EstimationAutonomous DrivingOptimizationPoint Cloud
🎯 What it does: For the problem of planar adjustment in large-scale multi-frame LiDAR point clouds, two algorithms based on Bi-Convex Relaxation, GlobalPointer and GlobalPointer++, are proposed, which can simultaneously estimate LiDAR pose and 3D planes.
Glyph-ByT5: A Customized Text Encoder for Accurate Visual Text Rendering
Zeyu Liu (Microsoft Research Asia), Yuhui Yuan (Microsoft Research Asia)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelDiffusion modelContrastive LearningImageTextMultimodality
🎯 What it does: Propose a character-aware, cursor-aligned text encoder Glyph-ByT5 based on ByT5, integrating it into SDXL through region-level multi-head cross-attention, significantly enhancing the visual text rendering accuracy of design images and scene text.
GMM-IKRS: Gaussian Mixture Models for Interpretable Keypoint Refinement and Scoring
Emanuele Santellani (Graz University of Technology), Friedrich Fraundorfer (Graz University of Technology)
Pose EstimationExplainability and InterpretabilityImage
🎯 What it does: This paper proposes the GMM-IKRS framework, which can refine keypoints generated by any keypoint detector and assign interpretable robustness and localization accuracy scores to them.
GMT: Enhancing Generalizable Neural Rendering via Geometry-Driven Multi-Reference Texture Transfer
Youngho Yoon (KAIST), Kuk-Jin Yoon (KAIST)
GenerationConvolutional Neural NetworkTransformerNeural Radiance FieldImagePoint Cloud
🎯 What it does: Proposed a Geometry-Driven Multi-Reference Texture Transfer (GMT) module to enhance the quality of novel view synthesis images based on Generalizable NeRF (G-NeRF) through geometry-driven feature alignment and multi-reference texture transfer.
GOEmbed: Gradient Origin Embeddings for Representation Agnostic 3D Feature Learning
Animesh Karnewar, David Novotny
SegmentationComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: Proposed a semantic segmentation network based on an attention mechanism.
Goldfish: Vision-Language Understanding of Arbitrarily Long Videos
Kirolos Ataallah (King Abdullah University of Science and Technology), Mohamed Elhoseiny (King Abdullah University of Science and Technology)
TransformerLarge Language ModelVision Language ModelVideoTextMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: Propose the Goldfish framework, which first selects the most relevant short video clips using a retrieval mechanism and then generates answers with an LLM; simultaneously introduce MiniGPT4-Video for generating short video summaries and design the TVQA-long long video evaluation benchmark based on TVQA.
Good Teachers Explain: Explanation-Enhanced Knowledge Distillation
Amin Parchami-Araghi (Max Planck Institute for Informatics), Bernt Schiele (Max Planck Institute for Informatics)
Explainability and InterpretabilityKnowledge DistillationContrastive LearningImage
🎯 What it does: Incorporating explanations from the teacher model (GradCAM, B-cos) into knowledge distillation to train the student model
GPSFormer: A Global Perception and Local Structure Fitting-based Transformer for Point Cloud Understanding
Changshuo Wang (Nanyang Technological University), Thambipillai Srikanthan (Nanyang Technological University)
ClassificationSegmentationRepresentation LearningGraph Neural NetworkTransformerPoint Cloud
🎯 What it does: Proposed a Transformer-based point cloud understanding model called GPSFormer, with core components including the Global Perception Module (GPM) (using ADGConv, residual cross-attention, and multi-head attention) and the Local Structure Fitting Convolution (LSFConv) (leveraging low-order and high-order convolutions derived from Taylor series decomposition).
GRA: Detecting Oriented Objects through Group-wise Rotating and Attention
Jiangshan Wang (Tsinghua University), Gao Huang (Tsinghua University)
Object DetectionConvolutional Neural NetworkImage
🎯 What it does: Designed and implemented a lightweight GRA module to replace traditional convolutions, enhancing orientation perception capability for tilted object detection.
GRACE: Graph-Based Contextual Debiasing for Fair Visual Question Answering
Yifeng Zhang (University of Minnesota), Qi Zhao (University of Minnesota)
RetrievalExplainability and InterpretabilityGraph Neural NetworkLarge Language ModelImageTextGraphRetrieval-Augmented Generation
🎯 What it does: Propose a graph-based unsupervised context graph learning and graph similarity retrieval framework called GRACE, aimed at eliminating bias in visual question answering during context learning in large language models.
Gradient-Aware for Class-Imbalanced Semi-supervised Medical Image Segmentation
Wenbo Qi (University of Hong Kong), S. C. Chan (University of Hong Kong)
SegmentationConvolutional Neural NetworkBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: This paper addresses the class imbalance problem in semi-supervised medical image segmentation by proposing a gradient-aware (GA) loss, improving the gradient behavior of Dice and cross-entropy to enhance training stability and performance on small classes.
Gradient-based Out-of-Distribution Detection
Taha Entesari (Johns Hopkins University), Mahyar Fazlyab (Johns Hopkins University)
Anomaly DetectionImage
🎯 What it does: Propose a gradient-regularized OOD detection method GReg (along with its accompanying energy sampling algorithm GReg+), which allows the network to learn both OOD scores and their local gradients during training, thereby achieving a smoother score manifold;
GRAPE: Generalizable and Robust Multi-view Facial Capture
Jing Li (Harbin Institute of Technology), Zhenyu He (Tencent AI Lab)
GenerationConvolutional Neural NetworkImageMesh
🎯 What it does: Propose a general and robust multi-view facial capture system called GRAPE, which can efficiently directly predict topologically consistent 3D face meshes under different camera arrays.
Graph Neural Network Causal Explanation via Neural Causal Models
Arman Behnam (Illinois Institute of Technology), Binghui Wang (Illinois Institute of Technology)
Explainability and InterpretabilityDrug DiscoveryGraph Neural NetworkGraphBiomedical Data
🎯 What it does: This paper proposes a graph neural network interpreter CXGNN based on causal inference, which can automatically identify the causal subgraph leading to model predictions.
GraphBEV: Towards Robust BEV Feature Alignment for Multi-Modal 3D Object Detection
Ziying Song (Beijing Jiaotong University), Li Wang (Beijing Institute of Technology)
Object DetectionAutonomous DrivingConvolutional Neural NetworkTransformerImageMultimodalityPoint Cloud
🎯 What it does: This paper proposes the GraphBEV framework, addressing the issue of local and global misalignment in LiDAR and camera Bird’s-Eye-View (BEV) feature fusion caused by projection errors, and constructs a robust multi-modal 3D object detection model.
GraspXL: Generating Grasping Motions for Diverse Objects at Scale
Hui Zhang (ETH Zürich), Jie Song (ETH Zürich)
Robotic IntelligenceReinforcement LearningMesh
🎯 What it does: Propose GraspXL, an RL framework capable of generating diverse grasping actions for multiple grasping targets without requiring hand-object interaction data.
Gravity-aligned Rotation Averaging with Circular Regression
Linfei Pan (ETH Zurich), Daniel Barath
Pose EstimationAutonomous DrivingOptimizationSimultaneous Localization and MappingImageVideo
🎯 What it does: Studied a rotational averaging method that utilizes gravity direction information to reduce degrees of freedom and improve the accuracy of camera poses in global SfM.
Grid-Attention: Enhancing Computational Efficiency of Large Vision Models without Fine-Tuning
Pengyu Li (Terminus Labs), Xian-Sheng Hua (Terminus Labs)
Computational EfficiencyTransformerDiffusion modelImageMultimodality
🎯 What it does: Propose the Grid-Attention (GridAttn) module, which can directly replace the multi-head attention layers in large visual Transformers (e.g., SAM, Stable Diffusion, SegFormer) without training or fine-tuning, to improve inference efficiency while maintaining or slightly enhancing performance.
GRIDS: Grouped Multiple-Degradation Restoration with Image Degradation Similarity
Shuo Cao (University of Science and Technology of China), Chao Dong (Shanghai Artificial Intelligence Laboratory)
RestorationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes the GRIDS (Grouped Multiple-Degradation Restoration with Image Degradation Similarity) method, which first quantifies the similarity between different degradations through deep degradation feature statistics. It then groups multiple degradation tasks into several groups based on similarity and trains a single model for each group. During inference, it estimates the degradation distribution of the input image using self-incremental cropping technology, automatically matches the most similar group, selects the corresponding model, and can also predict the generalization performance of the model without executing network inference.
Griffon: Spelling out All Object Locations at Any Granularity with Large Language Models
Yufei Zhan (Chinese Academy of Sciences), Jinqiao Wang (Chinese Academy of Sciences)
Object DetectionTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes a pure LVLM model named Griffon, which can locate all objects in an image based on any given text.
GRiT: A Generative Region-to-text Transformer for Object Understanding
Jialian Wu (State University of New York at Buffalo), Lijuan Wang (State University of New York at Buffalo)
RecognitionObject DetectionTransformerLarge Language ModelVision Language ModelImageText
🎯 What it does: A generative region-to-text Transformer (GRiT) was studied, which can simultaneously locate objects in images and freely describe their attributes with natural language, achieving a unified framework for object detection and dense description.
GRM: Large Gaussian Reconstruction Model for Efficient 3D Reconstruction and Generation
Yinghao Xu (Stanford University), Gordon Wetzstein (Shanghai AI Laboratory)
GenerationData SynthesisComputational EfficiencyTransformerDiffusion modelNeural Radiance FieldGaussian SplattingImageTextPoint CloudMesh
🎯 What it does: Propose GRM (Gaussian Reconstruction Model), a sparse multi-view reconstruction model based on Transformer, which can recover high-quality 3D scenes from four perspective images within approximately 0.1 seconds, and can be combined with multi-view diffusion models to achieve text-to-3D and single-image-to-3D generation.
GroCo: Ground Constraint for Metric Self-Supervised Monocular Depth
Aurélien Cecille (Visual Behavior), Rémi Agier (Visual Behavior)
Depth EstimationConvolutional Neural NetworkImage
🎯 What it does: Propose the GroCo framework, which leverages ground prior in self-supervised monocular depth estimation to recover scale and enhance model generalization.
Groma: Localized Visual Tokenization for Grounding Multimodal Large Language Models
Chuofan Ma (University of Hong Kong), Xiaojuan Qi (University of Hong Kong)
Object DetectionSegmentationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: Proposes Groma, a multimodal large language model that achieves fine-grained region identification and visual localization by leveraging local visual segmentation;
Grounding DINO: Marrying DINO with Grounded Pre-Training for Open-Set Object Detection
Shilong Liu (Tsinghua University), Lei Zhang (International Digital Economy Academy)
Object DetectionTransformerLarge Language ModelVision Language ModelContrastive LearningMultimodality
🎯 What it does: Developed an open-set object detection model called Grounding DINO based on DINO, which can detect any object through text or pointing expressions.
Grounding Image Matching in 3D with MASt3R
Vincent Leroy (Naver Labs Europe), Jerome Revaud (Naver Labs Europe)
Pose EstimationRetrievalTransformerContrastive LearningImage
🎯 What it does: Propose MASt3R, a 3D-aware dense image matching framework based on Transformer, which can simultaneously perform 3D reconstruction and pixel-level matching.
Grounding Language Models for Visual Entity Recognition
Zilin Xiao (Rice University), Vicente Ordonez (Microsoft STCA)
RecognitionTransformerLarge Language ModelVision Language ModelContrastive LearningMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: Propose AutoVER, a retrieval-enhanced constraint generation-based autoregressive multimodal large language model for visual entity recognition.
GroundUp: Rapid Sketch-Based 3D City Massing
Gizem Esra Unlu, Gabriel Brostow (University College London)
GenerationData SynthesisConvolutional Neural NetworkDiffusion modelImageMesh
🎯 What it does: Propose the GroundUp system to enable rapid 3D city modeling based on top views and perspective sketches.
Group Testing for Accurate and Efficient Range-Based Near Neighbor Search for Plagiarism Detection
Harsh Shah (Indian Institute of Technology Bombay), Ajit Rajwade (Indian Institute of Technology Bombay)
RetrievalConvolutional Neural NetworkImage
🎯 What it does: Proposes a high-dimensional nearest neighbor search framework based on adaptive binary grouping testing, utilizing vector accumulation summing/element-wise max pooling to rapidly construct sub-pools, achieving error-free range queries.
GroupDiff: Diffusion-based Group Portrait Editing
Yuming Jiang (Nanyang Technological University), Ziwei Liu (Nanyang Technological University)
GenerationPose EstimationDiffusion modelImage
🎯 What it does: Propose GroupDiff, a unified framework based on diffusion models for group portrait editing, enabling insertion, deletion, and interactive editing of individuals.
GS-LRM: Large Reconstruction Model for 3D Gaussian Splatting
Kai Zhang (Adobe Research), Zexiang Xu (Adobe Research)
GenerationTransformerGaussian SplattingImagePoint Cloud
🎯 What it does: Propose a Transformer-based GS-LRM model that can rapidly predict high-quality 3D Gaussian primitives from 2-4 sparse images with pose information, achieving sparse view reconstruction for objects and large-scale scenes.
GS-Pose: Category-Level Object Pose Estimation via Geometric and Semantic Correspondence
Pengyuan Wang (Technical University of Munich), Koichi Nishiwaki (Woven by Toyota)
Pose EstimationTransformerImagePoint Cloud
🎯 What it does: Proposes a category-level object pose estimation method GS-Pose based on geometric and semantic correspondence, by projecting 2D semantic features extracted from the pre-trained DINOv2 model into 3D point clouds to form semantic point clouds, and learning point cloud matching in a self-supervised manner under the condition of only synthetic CAD models, thus achieving cross-domain pose reasoning without real annotated data.
GS2Mesh: Surface Reconstruction from Gaussian Splatting via Novel Stereo Views
Yaniv Wolf (Technion - Israel Institute of Technology), Ron Kimmel (Technion - Israel Institute of Technology)
GenerationPose EstimationDepth EstimationNeural Radiance FieldGaussian SplattingPoint CloudMeshBenchmark
🎯 What it does: Proposed the GS2Mesh method, which first uses 3D Gaussian Splatting to render stereo-calibrated images that match the original viewpoint, then feeds these images into a pre-trained stereo matching network to obtain depth maps, followed by fusing all depths using TSDF and generating smooth 3D meshes via Marching-Cubes.
GSD: View-Guided Gaussian Splatting Diffusion for 3D Reconstruction
Yuxuan Mu (University of Alberta), Li Cheng (Huawei Noah's Ark Lab)
GenerationTransformerDiffusion modelGaussian SplattingImagePoint CloudMesh
🎯 What it does: This work proposes a diffusion model based on Gaussian Splatting (GS) representation, capable of performing high-quality 3D reconstruction from a single-view image, and achieves detail enhancement through view-guided sampling and an auxiliary 2D diffusion model.
GTMS: A Gradient-driven Tree-guided Mask-free Referring Image Segmentation Method
Haoxin Lv (Beijing Institute of Technology), Sanyuan Zhao (Beijing Institute of Technology)
SegmentationTransformerVision Language ModelImageTextMultimodality
🎯 What it does: Proposed a gradient-driven, tree-guided, mask-free reference image segmentation method called GTMS, which uses only bounding box supervision. It combines structural trees with GradCAM to generate high-quality pseudo-labels, thereby improving the performance of weakly supervised RIS.
GTP-4o: Modality-prompted Heterogeneous Graph Learning for Omni-modal Biomedical Representation
Chenxin Li, Yixuan Yuan (Shanghai Ai Laboratory)
Representation LearningGraph Neural NetworkTransformerPrompt EngineeringContrastive LearningImageTextMultimodalityGraphBiomedical Data
🎯 What it does: Proposed the GTP-4o framework, which leverages heterogeneous graph learning to obtain unified representations of multiple clinical multimodal data, including genomics, pathological images, cell graphs, and diagnostic texts.
GTPT: Group-based Token Pruning Transformer for Efficient Human Pose Estimation
Haonan Wang (Nanjing University), Yong Wang (Cainiao Network)
Pose EstimationTransformerImage
🎯 What it does: Proposed a Transformer-based efficient human pose estimation framework called GTPT, which significantly reduces computational cost while maintaining high accuracy, especially suitable for full-body pose estimation with a large number of keypoints.
Guide-and-Rescale: Self-Guidance Mechanism for Effective Tuning-Free Real Image Editing
Vadim Titov (AIRI), Aibek Alanov (Constructor University)
Image TranslationDiffusion modelImage
🎯 What it does: Propose a self-guided 'Guide-and-Rescale' framework that enables various edits (e.g., local modifications, style transfer) on real images without fine-tuning or reconstructing diffusion models
GVGEN: Text-to-3D Generation with Volumetric Representation
Xianglong He (Shanghai AI Laboratory), Tong He (Shanghai AI Laboratory)
GenerationData SynthesisConvolutional Neural NetworkDiffusion modelGaussian SplattingTextPoint CloudMesh
🎯 What it does: Propose the GVGEN framework to directly generate high-quality 3D Gaussian Volumes from text.
H-V2X: A Large Scale Highway Dataset for BEV Perception
Chang Liu (ADLab, Tencent), Cong Ma (ADLab, Tencent)
Object DetectionObject TrackingAutonomous DrivingConvolutional Neural NetworkImageVideoPoint CloudGraphBenchmark
🎯 What it does: This paper constructs the first large-scale highway roadside infrastructure BEV perception dataset, H-V2X, and proposes three tasks (BEV detection, tracking, and trajectory prediction) along with their benchmark methods based on this dataset.
HAC: Hash-grid Assisted Context for 3D Gaussian Splatting Compression
Yihang Chen (Shanghai Jiao Tong University), Jianfei Cai (Monash University)
CompressionGaussian Splatting
🎯 What it does: This paper proposes a 3D Gaussian rendering compression method based on Hash Grid Assisted Context (HAC), which utilizes hash grids to provide context for anchor attributes, achieving efficient entropy coding;
HaloQuest: A Visual Hallucination Dataset for Advancing Multimodal Reasoning
Zhecan Wang (Columbia University), Golnaz Ghiasi (Google DeepMind)
Data SynthesisLarge Language ModelSupervised Fine-TuningDiffusion modelImageTextMultimodalityBenchmark
🎯 What it does: Propose the HaloQuest visual question answering dataset, combining real and synthetic images to specifically evaluate multimodal hallucinations and provide fine-tuning resources.
HandDAGT: A Denoising Adaptive Graph Transformer for 3D Hand Pose Estimation
WENCAN CHENG (Sungkyunkwan University), Jong Hwan Ko (Sungkyunkwan University)
Pose EstimationGraph Neural NetworkTransformerImagePoint Cloud
🎯 What it does: Designed a denoising adaptive graph transformer, HandDAGT, which accurately estimates 3D hand pose using multi-modal inputs of depth maps and point clouds.
HandDGP: Camera-Space Hand Mesh Prediction with Differentiable Global Positioning
Eugene Valassakis (Niantic), Guillermo Garcia-Hernando (Niantic)
Pose EstimationConvolutional Neural NetworkImageMesh
🎯 What it does: Propose the HandDGP framework, which unifies the learning of root-relative hand meshes and global translation in camera space, forming an end-to-end differentiable hand mesh prediction system;
Handling The Non-Smooth Challenge in Tensor SVD: A Multi-Objective Tensor Recovery Framework
Jingjing Zheng (University of British Columbia), Xianta Jiang (Memorial University of Newfoundland)
RestorationOptimizationImageVideo
🎯 What it does: This paper proposes a multi-objective tensor recovery framework based on a learnable tensor nuclear norm to address the non-smooth challenges in tensor SVD and achieve high-order tensor completion.
HARIVO: Harnessing Text-to-Image Models for Video Generation
Mingi Kwon (Yonsei University), Youngjung Uh (Yonsei University)
GenerationConvolutional Neural NetworkTransformerDiffusion modelScore-based ModelContrastive LearningVideoText
🎯 What it does: Built a high-quality video generation model called HARIVO by training only the temporal layers on the frozen text-to-image diffusion model (Stable Diffusion v1.5).
Harmonizing knowledge Transfer in Neural Network with Unified Distillation
yaomin huang, Guixu Zhang (East China Normal University)
ClassificationObject DetectionKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: Proposed a unified knowledge distillation framework called UniKD, which first aggregates multi-scale intermediate features through Adaptive Features Fusion (AFF), then maps these features into a multivariate Gaussian distribution using Feature Distribution Prediction (FDP), and finally unifies the distillation of intermediate layer distributions and final logits via KL divergence.
Harnessing Text-to-Image Diffusion Models for Category-Agnostic Pose Estimation
Duo Peng (Singapore University of Technology and Design), Jun Liu (Singapore University of Technology and Design)
Pose EstimationPrompt EngineeringDiffusion modelImageText
🎯 What it does: Propose a method for category-agnostic pose estimation that leverages cross-attention maps from text-to-image diffusion models. It first learns pseudo prompts from a few samples via Prompt Pose Matching (PPM), then uses these pseudo prompts to locate keypoints in test images.
HAT: History-Augmented Anchor Transformer for Online Temporal Action Localization
Sakib Reza (Northeastern University), Octavia Camps (Northeastern University)
RecognitionTransformerVideoBenchmark
🎯 What it does: Propose the History-Augmented Anchor Transformer (HAT), which introduces long-term historical information into online temporal action localization to enhance anchor features.
Head360: Learning a Parametric 3D Full-Head for Free-View Synthesis in 360°
Yuxiao He (Nanjing University), Hao Zhu (Tencent)
GenerationData SynthesisNeural Radiance FieldGenerative Adversarial NetworkImageMesh
🎯 What it does: Proposed a parameterizable 3D head model capable of achieving 360° panoramic free-viewpoint synthesis, supporting single-image fitting, animation, and text editing
HeadGaS: Real-Time Animatable Head Avatars via 3D Gaussian Splatting
Helisa Dhamo, Eduardo Pérez-Pellitero (Huawei Noah's Ark Lab)
GenerationComputational EfficiencyGaussian SplattingVideoPoint Cloud
🎯 What it does: This paper proposes a HeadGaS model based on 3D Gaussian Splatting, capable of real-time rendering and animating 3D head avatars driven by expression parameters.
HeadStudio: Text to Animatable Head Avatars with 3D Gaussian Splatting
Zhenglin Zhou (Zhejiang University), Yi Yang (Zhejiang University)
GenerationDiffusion modelScore-based ModelGaussian SplattingVideoTextMesh
🎯 What it does: Propose an animation-driven head avatar generation framework called HeadStudio, which combines 3D Gaussian Splatting with FLAME head prior to generate high-quality avatars that can be rendered in real-time.
HENet: Hybrid Encoding for End-to-end Multi-task 3D Perception from Multi-view Cameras
Zhongyu Xia (Wangxuan Institute of Computer Technology, Peking University), Ming-Hsuan Yang (University of California, Merced)
Object DetectionSegmentationAutonomous DrivingConvolutional Neural NetworkImage
🎯 What it does: Propose HENet, an end-to-end multi-task 3D perception framework that combines hybrid image encoding, temporal feature fusion, and task-specific BEV encoding to achieve simultaneous 3D object detection and bird's-eye-view semantic segmentation.
HERGen: Elevating Radiology Report Generation with Longitudinal Data
Fuying Wang (University of Hong Kong), Lequan Yu (University of Hong Kong)
GenerationTransformerVision Language ModelContrastive LearningImageTime SeriesBiomedical Data
🎯 What it does: This study proposes the HERGen framework for generating radiology reports using longitudinal imaging data.
Hetecooper: Feature Collaboration Graph for Heterogeneous Collaborative Perception
Congzhang Shao (Beijing University of Posts and Telecommunications), Jinglin Li (Beijing University of Posts and Telecommunications)
Autonomous DrivingGraph Neural NetworkTransformerImageMultimodalityPoint Cloud
🎯 What it does: Proposed the Hetecooper framework to achieve collaborative perception for heterogeneous perception models by constructing feature collaboration graphs and using graph Transformers for lossless feature fusion.
Heterogeneous Graph Learning for Scene Graph Prediction in 3D Point Clouds
Yanni Ma, Yulan Guo
RecognitionConvolutional Neural NetworkRecurrent Neural NetworkGraph Neural NetworkPoint Cloud
🎯 What it does: Designed a 3D heterogeneous scene graph prediction framework called 3D-HetSGP, which first learns the heterogeneous graph structure through edge type prediction, then performs type-specific message passing on this structure, and finally outputs predictions for objects and relationships.
HGL: Hierarchical Geometry Learning for Test-time Adaptation in 3D Point Cloud Segmentation
Tianpei Zou (Tongji University), Changjun Jiang (Tongji University)
SegmentationDomain AdaptationConvolutional Neural NetworkContrastive LearningPoint Cloud
🎯 What it does: Propose a hierarchical geometric learning framework (HGL) that extracts geometric information from three levels—point-level, object-level, and frame-level—during test-time adaptation for point cloud segmentation to generate pseudo labels and update the model.
HiDiffusion: Unlocking Higher-Resolution Creativity and Efficiency in Pretrained Diffusion Models
Shen Zhang (MEGVII Technology), Jiajun Liang (MEGVII Technology)
GenerationConvolutional Neural NetworkTransformerDiffusion modelImageText
🎯 What it does: Propose a no-finetuning framework called HiDiffusion to enhance the effectiveness and efficiency of high-resolution image synthesis in pre-trained diffusion models.
Hiding Imperceptible Noise in Curvature-Aware Patches for 3D Point Cloud Attack
Mingyu Yang (Huazhong University of Science and Technology), Junyang Chen (Shenzhen University)
ClassificationAdversarial AttackPoint Cloud
🎯 What it does: Propose the Wavelet Patches Attack (WPA) method, which utilizes spectral graph wavelet transform to locate curvature-aware patches in point clouds, hiding imperceptible noise in smooth or sharp regions to attack 3D classification models.
HiEI: A Universal Framework for Generating High-quality Emerging Images from Natural Images
Jingmeng Li (Fudan University), Hui Wei (Fudan University)
GenerationSafty and PrivacyConvolutional Neural NetworkImage
🎯 What it does: Propose the HiEI framework, capable of generating high-quality black-and-white discrete point images (Emerging Image) from natural images, and verify its effectiveness as a CAPTCHA that can resist attacks from deep visual models.
Hierarchical Conditioning of Diffusion Models Using Tree-of-Life for Studying Species Evolution
Mridul Khurana (Virginia Tech), Anuj Karpatne (Virginia Tech)
GenerationData SynthesisDiffusion modelAuto EncoderImage
🎯 What it does: This study proposes the Phylo-Diffusion framework, which utilizes hierarchical embedding (HIER-Embed) to inject phylogenetic tree knowledge into diffusion models, enabling the automatic generation and controllable mutation of species images to reveal evolutionary features.
Hierarchical Gaussian Mixture Normalizing Flow Modeling for Unified Anomaly Detection
Xincheng Yao (Shanghai Jiao Tong University), Chongyang Zhang (Shanghai Jiao Tong University)
Anomaly DetectionFlow-based ModelImageBenchmark
🎯 What it does: This paper proposes a unified anomaly detection framework named HGAD, which models multi-class normal data using a hierarchical Gaussian mixture regularized normalizing flow, and further enhances inter-class discriminability through mutual information maximization and multi-center learning, thereby achieving detection and localization of multi-class anomalies within a single model.
Hierarchical Separable Video Transformer for Snapshot Compressive Imaging
Ping Wang (Zhejiang University), Xin Yuan (Westlake University)
RestorationCompressionTransformerVideo
🎯 What it does: Proposes a HiSViT architecture based on a hierarchical separable video Transformer for video reconstruction in snapshot compressive imaging (SCI).
Hierarchical Temporal Context Learning for Camera-based Semantic Scene Completion
Bohan Li (Shanghai Jiao Tong University), Wenjun Zeng (Shanghai Jiao Tong University)
SegmentationPose EstimationDepth EstimationAutonomous DrivingComputational EfficiencyRepresentation LearningConvolutional Neural NetworkContrastive LearningImageVideoPoint Cloud
🎯 What it does: Proposes a Hierarchical Temporal Context Learning (HTCL) framework based on cameras for completing 3D semantic scene completion from sparse RGB frames.
Hierarchical Unsupervised Relation Distillation for Source Free Domain Adaptation
Bowei Xing (Peking University), Wenzhen Yue (Peking University)
Domain AdaptationKnowledge DistillationImageBenchmark
🎯 What it does: Propose a Hierarchical Relation Distillation (HRD) framework that learns target domain features through a teacher-student model in source-agnostic domain adaptation.
Hierarchically Structured Neural Bones for Reconstructing Animatable Objects from Casual Videos
Subin Jeon (Yonsei University), Seon Joo Kim (Yonsei University)
GenerationPose EstimationNeural Radiance FieldVideo
🎯 What it does: Propose a hierarchical neural skeleton-based framework to generate animatable 3D models from daily videos and enable easy control.
HiFi-123: Towards High-fidelity One Image to 3D Content Generation
Wangbo Yu (Peking University), Yonghong Tian (Peking University)
GenerationData SynthesisDepth EstimationDiffusion modelScore-based ModelNeural Radiance FieldImageMesh
🎯 What it does: This paper proposes the HiFi-123 method, which utilizes depth-based DDIM inversion and attention injection to generate high-fidelity, multi-view consistent 3D content from a single image.
HiFi-Score: Fine-grained Image Description Evaluation with Hierarchical Parsing Graphs
Ziwei Yao (Chinese Academy of Sciences), Xilin Chen (Chinese Academy of Sciences)
SegmentationExplainability and InterpretabilityGraph Neural NetworkTransformerLarge Language ModelVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Proposed a HiFi-Score image caption evaluation metric based on hierarchical parsed graphs (HPG), which evaluates the consistency and completeness between text and images through fine-grained matching, and assesses linguistic fluency using a Large Language Model (LLM).
High-Fidelity 3D Textured Shapes Generation by Sparse Encoding and Adversarial Decoding
Qi Zuo (Institute of Intelligent Computing, Alibaba Group), Zilong Dong (SSE, CUHKSZ)
GenerationConvolutional Neural NetworkTransformerDiffusion modelGenerative Adversarial NetworkPoint CloudMeshBenchmark
🎯 What it does: Proposed a 3D texture shape generation framework (Sparse3D) based on sparse coding and adversarial decoding, which can achieve high-fidelity, multi-class, and open-vocabulary conditional generation while preserving high-frequency textures.
High-Fidelity and Transferable NeRF Editing by Frequency Decomposition
Yisheng He (Alibaba Group), Qixing Huang (University of Texas at Austin)
GenerationConvolutional Neural NetworkNeural Radiance FieldOptical FlowImageText
🎯 What it does: Propose a frequency-decomposition-based NeRF editing framework called Freditor, enabling high-fidelity and transferable 3D scene editing under text instructions.
High-Fidelity Modeling of Generalizable Wrinkle Deformation
Jingfan Guo, Hyun Soo Park (University of Minnesota)
GenerationData SynthesisDiffusion modelAuto EncoderMesh
🎯 What it does: This paper proposes a high-fidelity garment wrinkle generation model based on real capture data, decomposing the garment surface into a smooth base surface and fine wrinkles, and using a conditional diffusion model based on the Green-Lagrange strain field to generate high-frequency wrinkles.
High-Precision Self-Supervised Monocular Depth Estimation with Rich-Resource Prior
Jianbing Shen (University of Macau), Wencheng Han (University of Macau)
Depth EstimationConvolutional Neural NetworkTransformerImage
🎯 What it does: Leverage rich resources (such as high-resolution and multi-frame data) available during training to extract prior features, and perform depth estimation during inference using only low-resolution single-frame images. Improve the accuracy of single-frame models through prior feature retrieval, the Prior Depth Fusion module, and the Rich-resource Guided Loss.
High-Quality Mesh Blendshape Generation from Face Videos via Neural Inverse Rendering
Xin Ming (Tsinghua University), Feng Xu (Tsinghua University)
GenerationVideoMesh
🎯 What it does: Automatically generate personalized mesh blendshape models from single or sparse multi-view facial videos using neural inverse rendering.
High-Resolution and Few-shot View Synthesis from Asymmetric Dual-lens Inputs
Ruikang Xu (University of Science and Technology of China), Zhiwei Xiong (University of Science and Technology of China)
RestorationGenerationSuper ResolutionNeural Radiance FieldGaussian SplattingOptical FlowImage
🎯 What it does: Propose DL-GS, which leverages the different focal length information from the asynchronous dual-camera system of a smartphone to achieve few-shot and high-resolution view synthesis.
HIMO: A New Benchmark for Full-Body Human Interacting with Multiple Objects
Xintao Lv (Shanghai Jiao Tong University), Xiaokang Yang
GenerationData SynthesisTransformerVision-Language-Action ModelDiffusion modelVideoTextPoint CloudMeshBenchmark
🎯 What it does: Construct a large-scale 4D dataset HIMO for multi-object full-body interactions and design a text-driven HOI synthesis task with time period control.
HiT-SR: Hierarchical Transformer for Efficient Image Super-Resolution
XIANG ZHANG (ETH ZURICH), Fisher Yu
Super ResolutionTransformerImageBenchmark
🎯 What it does: Propose HiT-SR, a general strategy to convert conventional SR Transformers into hierarchical Transformers, efficiently aggregating multi-scale features for image super-resolution through hierarchical windows and spatial-channel correlation methods.
HO-Gaussian: Hybrid Optimization of 3D Gaussian Splatting for Urban Scenes
Zhuopeng Li (Zhejiang University), Liangjun Zhang (Baidu Research)
Autonomous DrivingOptimizationGaussian SplattingVideo
🎯 What it does: We propose a HO-Gaussian pipeline that combines grid-based volumes with 3D Gaussian splats to achieve real-time, SfM-initialization-free novel view synthesis for urban scenes.
HoloADMM: High-Quality Holographic Complex Field Recovery
Mazen Mel (University of Padova), Alexander Gatto (Sony Semiconductor Solutions Europe)
RestorationData SynthesisSuper ResolutionOptimizationExplainability and InterpretabilityConvolutional Neural NetworkDiffusion modelImagePhysics Related
🎯 What it does: This paper proposes HoloADMM, an interpretable deep network based on ADMM deconvolution, for recovering high-resolution complex amplitude fields from multi-frame low-resolution adaptive holograms, while simultaneously achieving optical focusing and spatial super-resolution capabilities.
Holodepth: Programmable Depth-Varying Projection via Computer-Generated Holography
Dorian Chan (Carnegie Mellon University), Jian Wang (Snap Inc)
Depth EstimationImagePhysics Related
🎯 What it does: Built a programmable depth-varying projection system (Holodepth) by integrating an optical lens array into a holographic projector, enabling the capability of simultaneous multi-plane projection.
How Far Can a 1-Pixel Camera Go? Solving Vision Tasks using Photoreceptors and Computationally Designed Visual Morphology
Andrei Atanov (Swiss Federal Institute of Technology Lausanne), Amir Zamir (Swiss Federal Institute of Technology Lausanne)
Autonomous DrivingOptimizationComputational EfficiencyTransformerReinforcement LearningImage
🎯 What it does: The study uses single-pixel photoreceptors with extremely low resolution to address visual tasks such as navigation and control.