ECCV 2024 Papers — Page 6
European Conference on Computer Vision · 2387 papers
Detecting As Labeling: Rethinking LiDAR-camera Fusion in 3D Object Detection
Junjie Huang (PhiGent Robotics), Dalong Du (PhiGent Robotics)
Object DetectionAutonomous DrivingMultimodalityPoint CloudBenchmark
🎯 What it does: This paper proposes the Detecting As Labeling (DAL) framework, which re-examines the fundamental principles of LiDAR-Camera fusion in 3D object detection and constructs a concise network that performs regression using only point cloud features.
DeTra: A Unified Model for Object Detection and Trajectory Forecasting
Sergio Casas (University of Toronto), Raquel Urtasun (University of Toronto)
Object DetectionAutonomous DrivingRecurrent Neural NetworkTransformerMultimodalityPoint Cloud
🎯 What it does: The unified model DETRA simultaneously performs object detection and trajectory prediction through trajectory refinement.
DetToolChain: A New Prompting Paradigm to Unleash Detection Ability of MLLM
Yixuan Wu (Zhejiang University), Jian Wu (Zhejiang University)
Object DetectionTransformerPrompt EngineeringVision Language ModelImageMultimodalityChain-of-Thought
🎯 What it does: Propose DetToolChain, a novel prompting paradigm that enhances the detection capabilities of multi-modal large language models (MLLMs) in tasks such as zero-shot object detection, descriptive object detection, referential expression understanding, and directional object detection through visual processing tools and chain-of-thought (CoT) methods.
DEVIAS: Learning Disentangled Video Representations of Action and Scene
Kyungho Bae (Kyung Hee University), Jinwoo Choi (Kyung Hee University)
Representation LearningTransformerAuto EncoderVideo
🎯 What it does: This study proposes DEVIAS, an end-to-end framework based on a decomposition encoder-decoder (Slot Attention + Action Mask Decoder), to learn disentangled representations of actions and scenes in videos.
DG-PIC: Domain Generalized Point-In-Context Learning for Point Cloud Understanding
Jincen Jiang (Bournemouth University), Jian Jun Zhang (Bournemouth University)
RestorationDomain AdaptationRepresentation LearningTransformerPoint Cloud
🎯 What it does: Proposes DG-PIC, a multi-domain and multi-task point cloud understanding framework that achieves domain generalization through dual-layer feature shifting during testing without requiring model updates.
DGD: Dynamic 3D Gaussians Distillation
Isaac Labe (Hebrew University of Jerusalem), Sagie Benaim (Hebrew University of Jerusalem)
Object TrackingSegmentationVision Language ModelGaussian SplattingVideoTextPoint Cloud
🎯 What it does: This paper proposes a unified representation based on dynamic 3D Gaussian distribution (DGD), which can learn color, geometry, and semantic features from a single monocular video, achieving dense 3D semantic segmentation and tracking, and supporting text or 3D click interactions.
DGE: Direct Gaussian 3D Editing by Consistent Multi-view Editing
Minghao Chen (University of Oxford), Andrea Vedaldi (University of Oxford)
GenerationTransformerDiffusion modelGaussian SplattingPoint CloudMesh
🎯 What it does: Propose Direct Gaussian Editor (DGE), which achieves text-driven 3D scene editing by directly fitting a high-precision 3D Gaussian Splatting representation after multi-view consistency editing.
DGInStyle: Domain-Generalizable Semantic Segmentation with Image Diffusion Models and Stylized Semantic Control
Yuru Jia (ETH Zürich), Anton Obukhov (ETH Zürich)
SegmentationDomain AdaptationAutonomous DrivingSupervised Fine-TuningPrompt EngineeringDiffusion modelImageText
🎯 What it does: Developed the DGInStyle data generation pipeline, leveraging a pre-trained text-to-image diffusion model combined with semantic masks and style prompts to generate diverse street view images, thereby enhancing domain generalization performance in autonomous driving semantic segmentation.
DGR-MIL: Exploring Diverse Global Representation in Multiple Instance Learning for Whole Slide Image Classification
Wenhui Zhu (Arizona State University), Yalin Wang (Arizona State University)
ClassificationTransformerContrastive LearningBiomedical Data
🎯 What it does: This paper proposes a global diversity aggregation method called DGR-MIL based on multi-instance learning, which models the diversity between WSI instances using a learnable global vector and cross-attention mechanism;
DHR: Dual Features-Driven Hierarchical Rebalancing in Inter- and Intra-Class Regions for Weakly-Supervised Semantic Segmentation
Sanghyun Jo (OGQ), Kyungsu Kim (Massachusetts General Hospital and Harvard Medical School)
SegmentationImageBenchmark
🎯 What it does: Designed and implemented the DHR (Dual Features-Driven Hierarchical Rebalancing) method, which utilizes unsupervised features (USS) and weakly supervised features (WSS) for hierarchical rebalancing to recover the minority classes overlooked in weakly supervised semantic segmentation.
Diagnosing and Re-learning for Balanced Multimodal Learning
Yake Wei (Renmin University of China), Di Hu (Renmin University of China)
ClassificationRepresentation LearningConvolutional Neural NetworkTransformerVideoTextMultimodalityAudio
🎯 What it does: This paper proposes a "Diagnosing & Re-learning" strategy, which diagnoses the separability of single-modal representation spaces for each modality and subsequently performs soft re-initialization of the corresponding encoder to achieve balanced and enhanced modal learning.
DIAL: Dense Image-text ALignment for Weakly Supervised Semantic Segmentation
Soojin Jang (Chung-Ang University), YoungBin Kim
SegmentationTransformerContrastive LearningImageTextMultimodality
🎯 What it does: Proposes DALNet, a single-stage weakly supervised semantic segmentation framework that leverages dual-layer image-text alignment;
Diff-Reg: Diffusion Model in Doubly Stochastic Matrix Space for Registration Problem
Qianliang Wu (Nanjing University of Science and Technology), Jian Yang (Nanjing University of Science and Technology)
Pose EstimationTransformerDiffusion modelPoint Cloud
🎯 What it does: Propose a diffusion model-based matching matrix iterative optimization framework, Diff-Reg, which generates high-quality correspondences by utilizing forward noise diffusion and backward denoising iteration in the doubly stochastic matrix space, further applied to 3D and 2D-3D registration;
Diff-Tracker: Text-to-Image Diffusion Models are Unsupervised Trackers
Zhengbo Zhang (Singapore University of Technology and Design), Jun Liu (Singapore University of Technology and Design)
Object TrackingTransformerDiffusion modelVideo
🎯 What it does: Develop Diff-Tracker, achieving unsupervised visual tracking by learning initial and online prompts while leveraging a pre-trained text-to-image diffusion model.
Diff3DETR: Agent-based Diffusion Model for Semi-supervised 3D Object Detection
Jiacheng Deng (University of Science and Technology of China), Tianzhu Zhang (University of Science and Technology of China)
Object DetectionTransformerAgentic AIDiffusion modelPoint Cloud
🎯 What it does: This paper proposes a proxy-based diffusion model called Diff3DETR for semi-supervised 3D object detection.
DiffBIR: Toward Blind Image Restoration with Generative Diffusion Prior
Xinqi Lin (Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences), Chao Dong (Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences)
RestorationDiffusion modelAuto EncoderImage
🎯 What it does: Propose a two-stage blind image restoration framework called DiffBIR, which first removes degradation and then reconstructs missing information using a generative diffusion prior.
DiffCD: A Symmetric Differentiable Chamfer Distance for Neural Implicit Surface Fitting
Linus Härenstam-Nielsen (Technical University of Munich), Daniel Cremers (Munich Center for Machine Learning)
RestorationPoint CloudOrdinary Differential Equation
🎯 What it does: Propose a new symmetric differentiable Chamfer distance loss, DiffCD, for fitting neural implicit surfaces from sparse noisy point clouds.
DiffClass: Diffusion-Based Class Incremental Learning
Zichong Meng (Northeastern University), Yanzhi Wang (Northeastern University)
ClassificationDomain AdaptationSupervised Fine-TuningDiffusion modelImage
🎯 What it does: Proposes a DiffClass method that achieves sample-free class incremental learning by leveraging multi-distribution matching diffusion models and selective synthetic image augmentation.
DIFFender: Diffusion-Based Adversarial Defense against Patch Attacks
Caixin Kang (Beihang University), Xingxing Wei (Beihang University)
RestorationAdversarial AttackPrompt EngineeringDiffusion modelImage
🎯 What it does: Propose DIFFender, which utilizes a pre-trained text-guided diffusion model to locate and recover adversarial patches, forming a unified defense framework.
Differentiable Convex Polyhedra Optimization from Multi-view Images
Daxuan Ren (Nanyang Technological University), Lei Yang (Sensetime Research)
OptimizationImageMesh
🎯 What it does: A method for directly optimizing convex polyhedron shapes from multi-view images is proposed by combining non-differentiable dual transformations of half-spaces with a differentiable linear solver.
Differentiable Product Quantization for Memory Efficient Camera Relocalization
Zakaria Laskar (Czech Technical University in Prague), Juho Kannala (Aalto University)
Pose EstimationCompressionComputational EfficiencyAuto EncoderContrastive LearningImagePoint CloudBenchmark
🎯 What it does: To address the 3D map storage problem in camera relocalization, this paper proposes a differentiable product quantization (DPQ) autoencoder (D-PQED) for efficiently compressing local descriptors, and uses a decoder to recover the descriptors after compression while maintaining matching accuracy.
DiffFAS: Face Anti-Spoofing via Generative Diffusion Models
Xinxu Ge (Tianjin University), Heikki Kälviäinen (Lappeenranta-Lahti University of Technology LUT)
Anomaly DetectionSafty and PrivacyConvolutional Neural NetworkDiffusion modelImageBenchmark
🎯 What it does: Achieve high-fidelity generation from live faces to spoof faces using diffusion models, and enhance cross-domain and cross-attack facial anti-spoofing performance by incorporating image quality priors.
DiffiT: Diffusion Vision Transformers for Image Generation
Ali Hatamizadeh (NVIDIA), Arash Vahdat (NVIDIA)
GenerationTransformerDiffusion modelAuto EncoderImage
🎯 What it does: This paper proposes a new diffusion model based on Vision Transformer called DiffiT, achieving high-quality image generation in both latent and image spaces.
DiffPMAE: Diffusion Masked Autoencoders for Point Cloud Reconstruction
Yanlong LI, Kanchana Thilakarathna (University of Sydney)
RestorationCompressionTransformerDiffusion modelAuto EncoderPoint Cloud
🎯 What it does: Propose DiffPMAE, a self-supervised point cloud reconstruction architecture that combines Masked Autoencoder with Diffusion Model, applicable for compression, up-sampling, and completion tasks.
DiffSurf: A Transformer-based Diffusion Model for Generating and Reconstructing 3D Surfaces in Pose
Yusuke Yoshiyasu (National Institute of Advanced Industrial Science and Technology), Leyuan Sun (National Institute of Advanced Industrial Science and Technology)
GenerationPose EstimationTransformerDiffusion modelScore-based ModelPoint CloudMesh
🎯 What it does: This paper proposes DiffSurf, a Transformer-based denoising diffusion model for generating, editing, and reconstructing 3D surfaces, capable of unconditional generation, shape deformation, pose control, and single-image 3D human mesh recovery across multiple poses, shapes, and object categories.
DiffuMatting: Synthesizing Arbitrary Objects with Matting-level Annotation
Xiaobin Hu (Tencent Youtu Lab), Rongrong Ji (Tencent Youtu Lab)
GenerationData SynthesisDiffusion modelAuto EncoderImage
🎯 What it does: Propose DiffuMatting, a method that utilizes diffusion models to generate arbitrary objects on a fixed green background and directly outputs high-precision matting annotations.
Diffusion Bridges for 3D Point Cloud Denoising
Mathias Vogel Hüni (ETH Zurich), Francis Engelmann (ETH Zurich)
RestorationConvolutional Neural NetworkDiffusion modelPoint CloudStochastic Differential Equation
🎯 What it does: Propose a point cloud denoising method based on the Schrödinger bridge (P2P-Bridge), treating denoising as a reversible data-to-data diffusion process from noisy point clouds to clean point clouds.
Diffusion for Natural Image Matting
Yihan Hu (Beijing Jiaotong University), Humphrey Shi (Georgia Institute of Technology)
SegmentationConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: Developed a multi-step iterative matting framework called DiffMatte based on a pixel-level denoising diffusion model, which can further refine the alpha matte on top of existing matting encoders.
Diffusion for Out-of-Distribution Detection on Road Scenes and Beyond
Silvio Galesso (University of Freiburg), Thomas Brox (University of Freiburg)
Anomaly DetectionAutonomous DrivingTransformerDiffusion modelScore-based ModelImageBenchmark
🎯 What it does: Propose a pixel-level anomaly detection method based on diffusion models called DOoD, and construct a diverse ADE-OoD benchmark dataset.
Diffusion Model for Robust Multi-Sensor Fusion in 3D Object Detection and BEV Segmentation
Duy Tho Le, Hamid Rezatofighi
Object DetectionSegmentationAutonomous DrivingDiffusion modelMultimodality
🎯 What it does: Designed and implemented a multi-modal fusion framework called DifFUSER based on diffusion models for 3D object detection and BEV semantic segmentation.
Diffusion Model is a Good Pose Estimator from 3D RF-Vision
Junqiao Fan (Nanyang Technological University), Lihua Xie (Nanyang Technological University)
Pose EstimationGraph Neural NetworkTransformerDiffusion modelPoint Cloud
🎯 What it does: Proposes a mmDiff method for 3D human pose estimation from millimeter-wave radar point clouds based on diffusion models, utilizing radar information as a condition to guide pose generation.
Diffusion Models are Geometry Critics: Single Image 3D Editing Using Pre-Trained Diffusion Priors
Ruicheng Wang (University of Science and Technology of China), Xin Tong (Microsoft Research Asia)
GenerationDepth EstimationSupervised Fine-TuningDiffusion modelImageStochastic Differential Equation
🎯 What it does: Use a pre-trained diffusion model to perform 3D editing on a single image (e.g., rotation, translation), and achieve consistency between texture and geometry through an iterative closed-loop.
Diffusion Models as Data Mining Tools
Ioannis Siglidis (University of Gustave Eiffel), Shiry Ginosar (University of California)
Data-Centric LearningSupervised Fine-TuningVision Language ModelDiffusion modelImage
🎯 What it does: Utilize fine-tuned diffusion models for conditional sampling, define a pixel-level typicality measure, and aggregate and cluster image patches to achieve large-scale visual data mining and visualization.
Diffusion Models as Optimizers for Efficient Planning in Offline RL
Renming Huang (University of Electronic Science and Technology of China), Heng Tao Shen
OptimizationTransformerReinforcement LearningDiffusion modelBenchmark
🎯 What it does: This paper proposes an offline reinforcement learning method called Trajectory Diffuser, which significantly improves sampling efficiency while maintaining or enhancing sampling quality by splitting the diffusion model's sampling process into two steps: 'generating feasible trajectories' and 'trajectory optimization'.
Diffusion Models for Monocular Depth Estimation: Overcoming Challenging Conditions
Fabio Tosi (University of Bologna), Matteo Poggi (University of Bologna)
Depth EstimationAutonomous DrivingKnowledge DistillationSupervised Fine-TuningPrompt EngineeringDiffusion modelImage
🎯 What it does: Utilize a text-to-image diffusion model to synthesize images with challenging conditions such as rainy days, night scenes, and transparent/mirrored surfaces while preserving the original depth consistency; subsequently, adopt a self-distillation approach to fine-tune existing monocular depth networks, enhancing their robustness on discretely distributed samples.
Diffusion Models for Open-Vocabulary Segmentation
Laurynas Karazija (University of Oxford), Christian Rupprecht (University of Oxford)
SegmentationLarge Language ModelVision Language ModelDiffusion modelImageText
🎯 What it does: Developed OVDiff, a method for annotation-free open-vocabulary semantic segmentation that generates support images using pre-trained text-to-image diffusion models and extracts foreground/background prototypes through unsupervised instance segmentation.
Diffusion Prior-Based Amortized Variational Inference for Noisy Inverse Problems
Sojin Lee (Korea University), Hyunwoo J. Kim (Korea University)
RestorationSuper ResolutionDiffusion modelScore-based ModelImageStochastic Differential Equation
🎯 What it does: Propose a bootstrap variational inference method (DAVI) based on diffusion model priors, which can directly map measurements to implicit posterior distributions under single-step inference to solve noisy inverse problems.
Diffusion Reward: Learning Rewards via Conditional Video Diffusion
Tao Huang (Shanghai Qi Zhi Institute), Huazhe Xu (Shanghai Qi Zhi Institute)
Robotic IntelligenceReinforcement LearningDiffusion modelAuto EncoderVideo
🎯 What it does: Proposed the Diffusion Reward framework, which trains conditional video diffusion models in the latent space to learn reward signals from expert videos, guiding visual reinforcement learning for complex robotic manipulation tasks.
Diffusion Soup: Model Merging for Text-to-Image Diffusion Models
Benjamin J Biggs, Stefano Soatto (AWS AI Labs)
GenerationSupervised Fine-TuningMixture of ExpertsDiffusion modelImageTextMultimodalityStochastic Differential Equation
🎯 What it does: Propose the Diffusion Soup method, which averages the weights of diffusion models trained on different data shards to obtain a single model, achieving incremental learning, no-learning, and zero-shot mixing without additional inference costs;
Diffusion-Based Image-to-Image Translation by Noise Correction via Prompt Interpolation
Junsung Lee (Seoul National University), Bohyung Han (Seoul National University)
Image TranslationPrompt EngineeringDiffusion modelImage
🎯 What it does: Propose a training-agnostic prompt interpolation noise correction method to correct inverse diffusion starting errors in image-to-image translation with diffusion models, preserving the background structure while precisely editing the target region.
Diffusion-Driven Data Replay: A Novel Approach to Combat Forgetting in Federated Class Continual Learning
Jinglin Liang (South China University of Technology), Qiang Yang (Hong Kong University of Science and Technology)
ClassificationData SynthesisFederated LearningSafty and PrivacyKnowledge DistillationConvolutional Neural NetworkDiffusion modelContrastive LearningImage
🎯 What it does: This paper proposes a federated continual learning framework named DDDR based on diffusion models, aimed at mitigating the catastrophic forgetting problem in federated classification continual learning.
Diffusion-Generated Pseudo-Observations for High-Quality Sparse-View Reconstruction
Xinhang Liu (Hong Kong University of Science and Technology), Chi-Keung Tang (Hong Kong University of Science and Technology)
GenerationData SynthesisSuper ResolutionDiffusion modelNeural Radiance FieldGaussian SplattingImage
🎯 What it does: Generate pseudo observations using a pre-trained 2D diffusion model to densify sparse input images, and train NeRF / 3D Gaussian Splatting with these pseudo observations together with real views to achieve high-quality 3D reconstruction.
Diffusion-Guided Weakly Supervised Semantic Segmentation
Sung-Hoon Yoon (Korea Advanced Institute of Science and Technology), Kuk-Jin Yoon (Korea Advanced Institute of Science and Technology)
SegmentationTransformerDiffusion modelImage
🎯 What it does: This paper proposes a weakly supervised semantic segmentation framework that combines diffusion models with vision transformers, enhancing CAM quality through Local Fusion Cross-Attention (LFCA) and Patch Affinity Consistency (PAC).
Diffusion-Refined VQA Annotations for Semi-Supervised Gaze Following
Qiaomu Miao (Stony Brook University), Dimitris Samaras (University of Adelaide)
TransformerVision Language ModelDiffusion modelImageVideoBenchmark
🎯 What it does: Propose a semi-supervised gaze tracking method that generates Grad-CAM heatmaps using a pre-trained vision-language (VQA) model, denoises and refines them with a diffusion model, and finally produces high-quality pseudo-labels for training the gaze tracking network.
DiffusionDepth: Diffusion Denoising Approach for Monocular Depth Estimation
Yiqun Duan, Zheng Zhu (GigaAI)
Depth EstimationTransformerDiffusion modelImage
🎯 What it does: Propose the DiffusionDepth model, reformulating monocular depth estimation as an iterative denoising process in the latent space.
DiffusionPen: Towards Controlling the Style of Handwritten Text Generation
Konstantina Nikolaidou (Luleå University of Technology), Marcus Liwicki (Luleå University of Technology)
GenerationData SynthesisConvolutional Neural NetworkDiffusion modelAuto EncoderImage
🎯 What it does: Propose a few-shot handwriting generation method called DiffusionPen based on latent diffusion models, which can reproduce the writing styles of known and unknown writers using only 5 samples and generate highly readable word images.
DiffuX2CT: Diffusion Learning to Reconstruct CT Images from Biplanar X-Rays
Xuhui Liu (Beihang University), Baochang Zhang (Central Research Institute, United Imaging Healthcare)
Image TranslationRestorationData SynthesisDiffusion modelBiomedical DataComputed Tomography
🎯 What it does: Propose a CT reconstruction framework called DiffuX2CT based on a 3D conditional diffusion model, which can generate high-quality 3D CT images from biplane orthogonal X-ray projections.
DIM: Dyadic Interaction Modeling for Social Behavior Generation
Minh Tran (University of Southern California), Mohammad Soleymani (University of Southern California)
GenerationTransformerAuto EncoderContrastive LearningVideoMultimodalityAudio
🎯 What it does: Propose a framework based on Dyadic Interaction Modeling (DIM) for generating three-dimensional facial and head movements of listeners in two-person dialogues.
DINO-Tracker: Taming DINO for Self-Supervised Point Tracking in a Single Video
Narek Tumanyan (Weizmann Institute of Science), Tali Dekel (Weizmann Institute of Science)
Object TrackingTransformerContrastive LearningOptical FlowVideo
🎯 What it does: Proposed a DINO-Tracker framework that achieves self-supervised long-term dense point tracking by combining test-time training with pre-trained DINO-ViT features.
Direct Distillation between Different Domains
Jialiang Tang (Nanjing University of Science and Technology), Masashi Sugiyama (RIKEN)
ClassificationDomain AdaptationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: Propose a one-stage direct knowledge distillation method called 4Ds, which utilizes a teacher network pre-trained on the source domain to directly train a small student network on the target domain without requiring source data.
DISCO: Embodied Navigation and Interaction via Differentiable Scene Semantics and Dual-level Control
Xinyu Xu (Shanghai Jiao Tong University), Cewu Lu (Shanghai Jiao Tong University)
Robotic IntelligenceConvolutional Neural NetworkVision-Language-Action ModelImageTextMultimodalityPoint Cloud
🎯 What it does: Built a mobile manipulation framework DISCO based on differentiable scene semantic representation and a two-layer coarse-to-fine control, which can complete navigation and interaction according to verb-noun pairs in commands.
DiscoMatch: Fast Discrete Optimisation for Geometrically Consistent 3D Shape Matching
Paul Roetzer (University of Bonn), Paul Swoboda (Heinrich-Heine University Düsseldorf)
OptimizationMesh
🎯 What it does: Proposed a fast discrete optimization framework called DiscoMatch, achieving geometrically consistent 3D shape matching through integer linear programming and pre-trained features.
Discover-then-Name: Task-Agnostic Concept Bottlenecks via Automated Concept Discovery
Sukrut Rao (Max Planck Institute for Informatics), Bernt Schiele (Max Planck Institute for Informatics)
Explainability and InterpretabilityRepresentation LearningVision Language ModelAuto EncoderImageTextMultimodality
🎯 What it does: Propose a reverse concept bottleneck model (DN-CBM), which first automatically discovers interpretable concepts from CLIP features using a sparse autoencoder, then matches text embeddings to name these concepts, and finally constructs an interpretable classifier by using these named concepts as a bottleneck layer.
Discovering Novel Actions from Open World Egocentric Videos with Object-Grounded Visual Commonsense Reasoning
Sanjoy Kundu (Auburn University), Sathyanarayanan N Aakur
RecognitionObject DetectionTransformerVision Language ModelVideo
🎯 What it does: Propose the ALGO two-step neuro-symbolic framework, which first grounds objects in self-perspective videos using evidence reasoning based on knowledge graphs and CLIP, and then discovers and identifies new actions in open worlds through energy-based symbolic pattern reasoning and visual-semantic action grounding.
Discovering Unwritten Visual Classifiers with Large Language Models
Mia Chiquier (Columbia University), Carl Vondrick (Columbia University)
ClassificationExplainability and InterpretabilityTransformerLarge Language ModelImage
🎯 What it does: Automatically discover interpretable visual attribute sets for fine-grained classification using evolutionary search combined with large language models.
Disentangled Clothed Avatar Generation from Text Descriptions
Jionghao Wang, Wenping Wang
GenerationVision Language ModelDiffusion modelScore-based ModelTextMultimodalityMesh
🎯 What it does: This paper proposes a method to generate animatable and separable 3D human avatars with clothing from textual descriptions. By training in two stages, it first generates a human body model without clothing, then generates an aligned clothing model based on it, ultimately producing human and clothing meshes that can be separately driven in a physical simulator.
Disentangled Generation and Aggregation for Robust Radiance Fields
Shihe Shen (Peking University), Ronggang Wang (University of Birmingham)
GenerationPose EstimationOptimizationNeural Radiance FieldImage
🎯 What it does: Proposes a decoupled three-plane representation method that separates three-plane generation and aggregation for robust light field reconstruction and joint camera pose optimization.
Disentangling Masked Autoencoders for Unsupervised Domain Generalization
An Zhang (National University Of Singapore), Tat-Seng Chua (National University Of Singapore)
Domain AdaptationRepresentation LearningTransformerAuto EncoderContrastive LearningImageBenchmark
🎯 What it does: This paper proposes a Disentangled Masked AutoEncoder (DisMAE), which learns domain-invariant semantic features and domain-specific variation features through a dual-branch architecture to achieve unsupervised domain generalization.
Dissecting Dissonance: Benchmarking Large Multimodal Models Against Self-Contradictory Instructions
Jin Gao (Shanghai Jiao Tong University), Dequan Wang (Shanghai Jiao Tong University)
TransformerLarge Language ModelPrompt EngineeringImageTextMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Proposed the Self-Contradictory Instructions (SCI) benchmark, constructed a dataset of 20K self-contradictory instructions, and evaluated the conflict detection capabilities of various large multimodal models.
Dissolving Is Amplifying: Towards Fine-Grained Anomaly Detection
Jian Shi (King Abdullah University of Science and Technology), Peter Wonka (NEC Laboratories China)
Anomaly DetectionConvolutional Neural NetworkDiffusion modelContrastive LearningBiomedical Data
🎯 What it does: Propose the DIA (Dissolving Is Amplifying) framework to achieve detection of fine-grained anomalies in medical images.
Distill Gold from Massive Ores: Bi-level Data Pruning towards Efficient Dataset Distillation
Yue Xu (Shanghai Jiao Tong University), Chi-Keung Tang (Hong Kong University Of Science And Technology)
Computational EfficiencyKnowledge DistillationImage
🎯 What it does: Study data redundancy and propose a two-layer data pruning strategy to improve dataset distillation efficiency.
Distilling Diffusion Models into Conditional GANs
MinGuk Kang, Taesung Park (Adobe Research)
GenerationData SynthesisKnowledge DistillationDiffusion modelGenerative Adversarial NetworkImageTextMultimodalityBenchmark
🎯 What it does: Distill multi-step diffusion models into single-step conditional GANs to significantly enhance generation speed while maintaining image quality.
Distilling Knowledge from Large-Scale Image Models for Object Detection
Gang Li (Nanjing University of Science and Technology), Shanshan Zhang (OpenGVLab, Shanghai AI Laboratory)
Object DetectionKnowledge DistillationTransformerImage
🎯 What it does: This paper proposes a knowledge distillation framework called DLIM-Det for large-scale visual models, leveraging a frozen pre-trained trunk teacher model and a query distillation method specifically designed for DETR, significantly enhancing the performance of small detectors.
Distractor-Free Novel View Synthesis via Exploiting Memorization Effect in Optimization
Yukun Wang (Sun Yat-sen University), Yulan Guo (Sun Yat-sen University)
GenerationOptimizationNeural Radiance FieldGaussian SplattingImage
🎯 What it does: Propose an unsupervised plugin module called MemE, which can automatically filter distractors in images during NeRF and 3D Gaussian Splatting training, achieving noise-free novel view synthesis.
Distractors-Immune Representation Learning with Cross-modal Contrastive Regularization for Change Captioning
Yunbin Tu (University of Chinese Academy of Sciences), Qingming Huang (Hangzhou Dianzi University)
GenerationRepresentation LearningTransformerVision Language ModelContrastive LearningImageText
🎯 What it does: This paper studies the visual change description task, proposing an interference-immune representation learning network (DIRL) and cross-modal contrast regularization (CCR). It enhances the robustness of image representations against interferences such as viewpoint and illumination through self-supervised channel correlation/disassociation mechanisms, and generates accurate change descriptions using Transformers.
Distributed Active Client Selection With Noisy Clients Using Model Association Scores
Kwang In Kim (Pohang University of Science and Technology)
Federated LearningData-Centric LearningImage
🎯 What it does: This paper proposes an active client selection algorithm based on model association scores, aiming to identify and suppress the impact of noisy label clients in federated learning to accelerate model convergence.
Distributed Semantic Segmentation with Efficient Joint Source and Task Decoding
Danish Nazir (Technische Universität Braunschweig), Tim Fingscheidt (Technische Universität Braunschweig)
SegmentationCompressionOptimizationConvolutional Neural NetworkAuto EncoderImage
🎯 What it does: This paper proposes a joint decoder (JD) that simultaneously performs source encoding and task decoding in distributed semantic segmentation, thereby minimizing the scale of the cloud network;
Distribution Alignment for Fully Test-Time Adaptation with Dynamic Online Data Streams
Ziqiang Wang (Concordia University), Yang Wang (Concordia University)
Domain AdaptationConvolutional Neural NetworkImage
🎯 What it does: In the fully test-time adaptation task, a distribution alignment (DA) loss function is proposed, which pulls the feature distribution of each test batch back to the precomputed source domain feature distribution, maintaining model performance under non-i.i.d. dynamic data streams. A domain shift detection mechanism is designed to support continuous domain adaptation.
Distribution-Aware Robust Learning from Long-Tailed Data with Noisy Labels
Jae Soon Baik (Hanyang University), Jun Won Choi (Seoul National University)
ClassificationData-Centric LearningContrastive LearningImage
🎯 What it does: Proposed a new distribution-aware robust learning framework called DaSC, which can simultaneously address long-tailed distributions and noisy label problems.
Distributionally Robust Loss for Long-Tailed Multi-Label Image Classification
Dekun Lin (Chengdu Institute of Computer Applications, Chinese Academy of Sciences), Xiaolin Qin (Chengdu Institute of Computer Applications, Chinese Academy of Sciences)
ClassificationOptimizationConvolutional Neural NetworkImage
🎯 What it does: Proposed a distribution-robust loss (DR Loss) for long-tailed multi-label image classification, enhancing model robustness on long-tailed data through class-level computation of LSEP loss (C-LSEP) and incorporating a negative gradient constraint (NGC).
Diverse Text-to-3D Synthesis with Augmented Text Embedding
Uy Dieu Tran (VinAI Research), Binh-Son Hua (VinAI Research)
GenerationData SynthesisPrompt EngineeringDiffusion modelScore-based ModelNeural Radiance FieldImageMeshBenchmark
🎯 What it does: Under a single text prompt, by introducing HiPer text inversion tokens based on reference images and shared learning tokens on each particle, the variational score distillation (VSD) optimization process of 3D neural radiance fields (NeRF) is improved, achieving diverse text-to-3D synthesis.
Divide and Fuse: Body Part Mesh Recovery from Partially Visible Human Images
Tianyu Luan (State University Of New York At Buffalo), Ziyan Wu (United Imaging Intelligence)
Pose EstimationTransformerImageMeshBenchmark
🎯 What it does: Propose a bottom-up partitioning and fusion method named Divide and Fuse to recover complete human meshes from partially visible human images.
DMiT: Deformable Mipmapped Tri-Plane Representation for Dynamic Scenes
Jing-Wen Yang (Beijing Key Laboratory of Mobile Computing and Pervasive Device, Institute of Computing Technology, Chinese Academy of Sciences), Lin Gao (Beijing Key Laboratory of Mobile Computing and Pervasive Device, Institute of Computing Technology, Chinese Academy of Sciences)
GenerationNeural Radiance FieldImageVideo
🎯 What it does: Proposed the DMIT framework, utilizing deformable multi-level Mip tri-plane representations to achieve high-fidelity rendering and efficient training/inference for dynamic scenes.
DNI: Dilutional Noise Initialization for Diffusion Video Editing
Sunjae Yoon (Korea Advanced Institute of Science and Technology), Chang D. Yoo (Korea Advanced Institute of Science and Technology)
GenerationDiffusion modelAuto EncoderVideo
🎯 What it does: Proposed a framework called Dilutional Noise Initialization (DNI), which introduces noise dilution operations in the initial latent noise of video diffusion editing, enabling the model to achieve non-rigid editing (such as motion changes) as well as traditional rigid editing;
Do Generalised Classifiers really work on Human Drawn Sketches?
Hmrishav Bandyopadhyay (University of Surrey), Yi-Zhe Song (University of Surrey)
ClassificationSupervised Fine-TuningPrompt EngineeringContrastive LearningImageTextMultimodality
🎯 What it does: This paper combines the large-scale pre-trained CLIP model with image-text alignment techniques, leveraging visual and text prompt learning along with sparse vectorization auxiliary tasks to construct a general classifier capable of achieving zero/few-shot classification on hand-drawn sketches with unknown categories and varying levels of abstraction.
Do text-free diffusion models learn discriminative visual representations?
Soumik Mukhopadhyay (University of Maryland), Abhinav Shrivastava (University of Maryland)
ClassificationObject DetectionSegmentationRepresentation LearningConvolutional Neural NetworkTransformerDiffusion modelImage
🎯 What it does: Investigate the intermediate features of unconditional diffusion models, exploring whether they can learn discriminative visual representations under self-supervised learning without labels, and evaluate their performance on various downstream tasks.
DOCCI: Descriptions of Connected and Contrasting Images
Yasumasa Onoe (Google), Jason M Baldridge (Google)
Supervised Fine-TuningVision Language ModelMultimodalityBenchmark
🎯 What it does: This paper constructs the DOCCI dataset, providing 15k carefully selected images along with detailed human-generated descriptions of approximately 136 words, and utilizes this dataset to evaluate the capabilities of text-to-image and image-to-text models.
Dolfin: Diffusion Layout Transformers without Autoencoder
Yilin Wang (Tsinghua University), Zhuowen Tu (University of California, San Diego)
GenerationTransformerDiffusion modelImage
🎯 What it does: Propose a diffusion layout Transformer (Dolfin) without an autoencoder for unconditional layout and line segment generation.
Dolphins: Multimodal Language Model for Driving
Yingzi Ma (University of Wisconsin-Madison), Chaowei Xiao (University of Wisconsin-Madison)
Autonomous DrivingTransformerSupervised Fine-TuningVision Language ModelVideoTextMultimodalityRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: Proposed and implemented Dolphin, a multimodal vision-language model for autonomous driving, which can receive videos, textual instructions, and historical control signals, and output corresponding driving decisions and explanations.
Domain Generalization of 3D Object Detection by Density-Resampling
Shuangzhi Li (University of Alberta), Xingyu Li (University of Alberta)
Object DetectionDomain AdaptationAutonomous DrivingConvolutional Neural NetworkPoint Cloud
🎯 What it does: This paper proposes a generalization method for 3D object detection from a single source domain. The core idea is to enhance the model's robustness to varying point cloud densities through physical constraint-based point cloud density resampling (PDDA), and to incorporate a self-supervised point cloud densification task into the detection framework, achieving multi-task learning during training and lightweight model adaptation during inference.
Domain Reduction Strategy for Non-Line-of-Sight Imaging
Hyunbo Shim (Yonsei University), Seon Joo Kim (Yonsei University)
OptimizationComputational EfficiencyImagePhysics Related
🎯 What it does: Proposes an optimization method based on domain reduction that can rapidly reconstruct albedo and surface normals of non-line-of-sight (NLOS) scenes under sparse scanning.
Domain Shifting: A Generalized Solution for Heterogeneous Cross-Modality Person Re-Identification
Yan Jiang (Nanjing University of Information Science and Technology), Guoying Zhao (University of Oulu)
RecognitionDomain AdaptationKnowledge DistillationConvolutional Neural NetworkMultimodality
🎯 What it does: Proposes a general domain shifting method for cross-modal person re-identification tasks.
Domain-Adaptive 2D Human Pose Estimation via Dual Teachers in Extremely Low-Light Conditions
Yihao Ai (National University of Singapore), Robby T. Tan (National University of Singapore)
Pose EstimationDomain AdaptationConvolutional Neural NetworkImage
🎯 What it does: Propose a domain adaptation method based on dual teachers (center detection + keypoint detection) to achieve 2D human pose estimation under extremely low illumination conditions without requiring real annotations for low-light images.
Domain-adaptive Video Deblurring via Test-time Blurring
Jin-Ting He (National Yang Ming Chiao Tung University), Yen-Yu Lin (Qualcomm Technologies, Inc.)
RestorationDomain AdaptationSupervised Fine-TuningDiffusion modelOptical FlowVideo
🎯 What it does: Propose a domain adaptive video deblurring framework based on test-time fuzziness, which utilizes generated pseudo-sharp points and synthetic blurred images to self-optimize the deblurring model, thereby enhancing the restoration performance on target domain videos.
DomainFusion: Generalizing To Unseen Domains with Latent Diffusion Models
Yuyang Huang (Shanghai Jiao Tong University), Qi Tian (Shanghai Jiao Tong University)
Domain AdaptationKnowledge DistillationDiffusion modelScore-based ModelImageBenchmark
🎯 What it does: Propose a framework named DomainFusion that jointly leverages latent knowledge distillation from LDM and online lightweight image enhancement to improve domain generalization performance.
Domesticating SAM for Breast Ultrasound Image Segmentation via Spatial-frequency Fusion and Uncertainty Correction
Wanting Zhang (Shenzhen University), Jing Qin (Hong Kong Polytechnic University)
SegmentationConvolutional Neural NetworkTransformerSupervised Fine-TuningImageBiomedical DataUltrasound
🎯 What it does: For breast ultrasound image segmentation, the SFRecSAM model based on SAM is proposed, achieving more accurate segmentation through the introduction of spatial-frequency domain feature fusion and dual error correction.
DoubleTake: Geometry Guided Depth Estimation
Mohamed Sayed (Niantic), Michael Firman (Niantic)
Depth EstimationSimultaneous Localization and MappingImage
🎯 What it does: Proposes a network that utilizes historical geometric hints (TSDF rendered depth and confidence) in multi-view stereo depth estimation, capable of providing more accurate instant depth maps under online real-time conditions.
DoughNet: A Visual Predictive Model for Topological Manipulation of Deformable Objects
Dominik Bauer (Columbia University), Shuran Song (Columbia University)
Robotic IntelligenceTransformerWorld ModelImagePoint Cloud
🎯 What it does: This paper proposes DoughNet, a Transformer-based visual prediction model that can predict the geometric deformation and topological changes (separation, merging, changes in the number of loops) of flexible objects under a series of manipulation actions from a single RGB-D observation.
DPA-Net: Structured 3D Abstraction from Sparse Views via Differentiable Primitive Assembly
Fenggen Yu (Amazon Simon Fraser University), Hao Zhang (Amazon Simon Fraser University)
GenerationConvolutional Neural NetworkNeural Radiance FieldImageMesh
🎯 What it does: Learn unsupervised 3D structured abstraction from sparse RGB images with large viewpoint variations, outputting an editable primitive set composed of convex tetrahedra.
DQ-DETR: DETR with Dynamic Query for Tiny Object Detection
Yi-Xin Huang (National Yang Ming Chiao Tung University), Wen-Huang Cheng (National Yang Ming Chiao Tung University)
Object DetectionTransformerImage
🎯 What it does: Propose a dynamic query DETR model (DQ-DETR) for small object detection.
Drag Anything: Motion Control for Anything using Entity Representation
Weijia Wu (Kuaishou Technology), Zhang Di
Object TrackingSegmentationGenerationConvolutional Neural NetworkDiffusion modelVideo
🎯 What it does: Propose DragAnything, a trajectory-driven video generation method based on entity representations, achieving precise motion control of arbitrary objects.
DragAPart: Learning a Part-Level Motion Prior for Articulated Objects
Ruining Li (University of Oxford), Andrea Vedaldi (University of Oxford)
GenerationData SynthesisTransformerDiffusion modelImageMesh
🎯 What it does: Trained an interactive image generation model called DragAPart based on drag input, which can achieve physically plausible deformations at the object part level on a single image according to drag operations.
DragVideo: Interactive Drag-style Video Editing
Yufan Deng (Hong Kong University Of Science And Technology), Chi-Keung Tang (Hong Kong University Of Science And Technology)
GenerationTransformerSupervised Fine-TuningDiffusion modelVideo
🎯 What it does: Implemented an interactive video editing framework based on drag instructions, supporting point/mask dragging while ensuring video spatiotemporal consistency.
DreamDiffusion: High-Quality EEG-to-Image Generation with Temporal Masked Signal Modeling and CLIP Alignment
Yunpeng Bai (Tsinghua Shenzhen International Graduate School), Ying Shan (Tencent PCG)
GenerationData SynthesisTransformerDiffusion modelAuto EncoderContrastive LearningImageMultimodalityTime SeriesBiomedical Data
🎯 What it does: Propose the DreamDiffusion method to generate high-quality images directly from EEG signals.
DreamDissector: Learning Disentangled Text-to-3D Generation from 2D Diffusion Priors
Zizheng Yan (Shenzhen Future Network of Intelligence Institute), Xiaoguang Han (Chinese University of Hong Kong Shenzhen)
GenerationPrompt EngineeringDiffusion modelScore-based ModelNeural Radiance FieldTextMesh
🎯 What it does: Propose a method called DreamDissector, which can decompose a generated multi-object text-to-3D NeRF into multiple independent, interactive object Meshes through Neural Category Field (NeCF) and Deep Concept Mining (DCM), and further refine geometry and texture based on this.
DreamDrone: Text-to-Image Diffusion Models are Zero-shot Perpetual View Generators
Hanyang Kong (National University of Singapore), Xinchao Wang (National University of Singapore)
GenerationData SynthesisDepth EstimationTransformerDiffusion modelImageVideoText
🎯 What it does: Developed a zero-training, zero-fine-tuning pipeline named DreamDrone, which can generate infinitely long flight perspective sequences along any user-defined camera trajectory from a single RGBD image and text prompts, without constructing 3D point clouds.
DreamMesh: Jointly Manipulating and Texturing Triangle Meshes for Text-to-3D Generation
Haibo Yang (Fudan University), Tao Mei (HiDream.ai Inc.)
GenerationData SynthesisDiffusion modelScore-based ModelNeural Radiance FieldTextMeshBenchmark
🎯 What it does: Propose DreamMesh, a text-to-3D generation framework based on fully explicit triangular meshes, employing a two-stage coarse-to-fine deformation and texturing process.
DreamMotion: Space-Time Self-Similar Score Distillation for Zero-Shot Video Editing
Hyeonho Jeong (KAIST), Jong Chul Ye (KAIST)
GenerationData SynthesisDiffusion modelScore-based ModelVideoText
🎯 What it does: Proposed a zero-shot video editing framework called DreamMotion, which can modify video content according to target text while preserving the original video's motion and structure.
DreamMover: Leveraging the Prior of Diffusion Models for Image Interpolation with Large Motion
Liao Shen (Huazhong University of Science and Technology), Zhiguo Cao (Huazhong University of Science and Technology)
GenerationSupervised Fine-TuningDiffusion modelOptical FlowImageBenchmark
🎯 What it does: Propose a framework called DreamMover based on pre-trained text-to-image diffusion models, which can generate semantically consistent intermediate frames between two images with large motion and form a short video.
DreamReward: Aligning Human Preference in Text-to-3D Generation
Junliang Ye (Tsinghua University), Jun Zhu (Tsinghua University)
GenerationData SynthesisReinforcement Learning from Human FeedbackReinforcement LearningVision Language ModelDiffusion modelScore-based ModelTextPoint CloudMesh
🎯 What it does: Built a 3D generation framework based on human preferences called DreamReward. First, collected 25k expert comparison data and trained the Reward3D reward model, then embedded this model into a multi-view diffusion model (MVDream). By integrating Reward3D-Feedback Learning (DreamFL), model-free fine-tuning was achieved to improve text-3D alignment and multi-view consistency.