ECCV 2024 Papers — Page 5
European Conference on Computer Vision · 2387 papers
Contrastive Region Guidance: Improving Grounding in Vision-Language Models without Training
David Wan (UNC Chapel Hill), Mohit Bansal (UNC Chapel Hill)
Object DetectionSegmentationPrompt EngineeringVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Propose a training-free contrastive region guidance method, CRG, to enhance the localization and reasoning performance of vision-language models in visual prompt tasks.
Contribution-based Low-Rank Adaptation with Pre-training Model for Real Image Restoration
Dongwon Park (IPAI & INMC), Se Young Chun (Seoul National University)
RestorationConvolutional Neural NetworkTransformerSupervised Fine-TuningImage
🎯 What it does: Proposes contribution-based low-rank adapter CoLoRA and random sequential degradation pre-training method PROD for efficient fine-tuning of low-level image restoration tasks.
ControlCap: Controllable Region-level Captioning
Yuzhong Zhao (University of Chinese Academy of Sciences), Fang Wan (University of Chinese Academy of Sciences)
GenerationTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: Propose a controllable region-level image description method called ControlCap, which reduces description degradation and enables diverse, customizable region descriptions through control words.
Controllable Contextualized Image Captioning: Directing the Visual Narrative through User-Defined Highlights
Shunqi Mao (University of Sydney), Weidong Cai (University of Sydney)
GenerationTransformerLarge Language ModelPrompt EngineeringVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Propose the Ctrl-CIC task and design two controllers (Prompting-based Controller and Recalibration-based Controller) to generate contextualized image descriptions based on user-specified highlights.
Controllable Human-Object Interaction Synthesis
Jiaman Li, C. Karen Liu (Stanford University)
GenerationData SynthesisTransformerVision Language ModelDiffusion modelVideoTextMesh
🎯 What it does: Propose a system named CHOIS that utilizes conditional diffusion models to simultaneously generate object motion and human motion, with inputs including language descriptions, initial object/human states, and sparse object trajectories; generates synchronized, realistic dynamic interactions guided by object geometry loss and contact constraints during sampling;
Controllable Navigation Instruction Generation with Chain of Thought Prompting
Xianghao Kong, Si Liu
GenerationAutonomous DrivingTransformerLarge Language ModelPrompt EngineeringVision Language ModelMultimodalityChain-of-Thought
🎯 What it does: Proposes C-Instructor, which utilizes large language models combined with path and visual information to achieve controllable and interpretable navigation instruction generation.
Controlling the World by Sleight of Hand
Sruthi Sudhakar (Columbia University), Richard Zemel (Columbia University)
GenerationTransformerDiffusion modelVideo
🎯 What it does: Propose the CoSHAND model, which controls object interactions in images through hand masks to predict future states.
ControlLLM: Augment Language Models with Tools by Searching on Graphs
Zhaoyang Liu (Hong Kong University of Science and Technology), Wenhai Wang (Chinese University of Hong Kong)
Graph Neural NetworkTransformerLarge Language ModelPrompt EngineeringMultimodalityBenchmarkChain-of-Thought
🎯 What it does: This paper proposes the ControlLLM framework, which helps large language models (LLMs) accurately invoke multimodal tools to complete complex tasks through a three-stage process (task decomposition, graph-based thinking search, and execution engine).
ControlNet-XS: Rethinking the Control of Text-to-Image Diffusion Models as Feedback-Control Systems
Denis Zavadski (Heidelberg University), Carsten Rother (Heidelberg University)
GenerationDiffusion modelImageText
🎯 What it does: This paper redesigned the control network of text-to-image diffusion models, proposing ControlNet-XS, which achieves high-frequency, high-bandwidth bidirectional communication, significantly improving image quality and control precision under pixel-level guidance (depth, edge, semantic map).
ControlNet++: Improving Conditional Controls with Efficient Consistency Feedback
Ming Li (University of Central Florida), Chen Chen (University of Central Florida)
SegmentationGenerationReinforcement LearningDiffusion modelImageText
🎯 What it does: Propose ControlNet++, explicitly optimizing the image conditional controllability of text-to-image diffusion models through pixel-level cycle consistency rewards.
Convex Relaxations for Manifold-Valued Markov Random Fields with Approximation Guarantees
Robin Kenis (Vrije Universiteit Brussel), Panagiotis Patrinos (Vrije Universiteit Brussel)
OptimizationGraph
🎯 What it does: This paper proposes a graph cut optimization framework based on optimal transport (OT) theory, unifying the potential functions of nodes and edges, and obtaining a solvable continuous optimal problem through dual transformation;
CoPT: Unsupervised Domain Adaptive Segmentation using Domain-Agnostic Text Embeddings
Cristina Mata (Stony Brook University), Michael S Ryoo (Stony Brook University)
SegmentationDomain AdaptationLarge Language ModelVision Language ModelImageText
🎯 What it does: Propose CoPT, achieving unsupervised domain adaptation for semantic segmentation by leveraging covariance consistency loss on domain-agnostic text embeddings.
CoR-GS: Sparse-View 3D Gaussian Splatting via Co-Regularization
Jiawei Zhang, Xiao Bai (Beihang University)
GenerationNeural Radiance FieldGaussian SplattingImage
🎯 What it does: This paper proposes the CoR-GS method, which simultaneously trains two 3D Gaussian radiance fields under sparse views. It uses point difference and rendering difference as two inconsistency metrics to unsupervisedly evaluate and suppress inaccurate reconstructions, thereby improving the geometric accuracy and rendering quality of sparse-view 3D Gaussian Splatting (3DGS).
CoReS: Orchestrating the Dance of Reasoning and Segmentation
Xiaoyi Bao (University of Chinese Academy of Sciences), Xingang Wang (Alibaba Group)
SegmentationTransformerSupervised Fine-TuningVision Language ModelMultimodalityChain-of-Thought
🎯 What it does: Propose the CoReS framework, leveraging multimodal chain reasoning and a dual-chain structure for segmentation to address complex reasoning and segmentation tasks.
Correspondence-Free SE(3) Point Cloud Registration in RKHS via Unsupervised Equivariant Learning
Ray Zhang (University of Michigan), Arnie Sen (Amazon Lab126)
Pose EstimationPoint Cloud
🎯 What it does: Proposes an SE(3) point cloud registration framework called EquivAlign that does not require correspondence, based on point cloud functions in RKHS and unsupervised equivariant feature learning, achieving direct feature space registration.
Correspondences of the Third Kind: Camera Pose Estimation from Object Reflection
Kohei Yamashita (Kyoto University), Ko Nishino (Kyoto University)
Pose EstimationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: Propose and utilize reflective correspondences to jointly estimate camera pose and object shape by combining pixel, 3D, and reflective correspondences from two images of a textureless, non-smooth object.
CoSIGN: Few-Step Guidance of ConSIstency Model to Solve General INverse Problems
Jiankun Zhao (University of Michigan), Liyue Shen (University of Michigan)
RestorationSuper ResolutionDiffusion modelScore-based ModelImageBiomedical DataComputed Tomography
🎯 What it does: Propose the CoSIGN framework, which combines a pre-trained Consistency Model with ControlNet to solve various inverse problems (super-resolution, inpainting, deblurring, sparse-view CT reconstruction) within 1-2 sampling steps (or with multi-step refinement).
COSMU: Complete 3D human shape from monocular unconstrained images
Marco Pesavento (University of Surrey), Adrian Hilton (University of Surrey)
GenerationPose EstimationConvolutional Neural NetworkNeural Radiance FieldImage
🎯 What it does: Given a target RGB image and a set of uncalibrated, unaligned monocular images, the proposed COSMU framework utilizes body-based reference selection and registration to simulate a multi-view scene, then generates a complete 3D human shape through a multi-view attention neural implicit model.
CoTracker: It is Better to Track Together
Nikita Karaev (Meta AI), Christian Rupprecht (Meta AI)
Object TrackingTransformerVideoPoint Cloud
🎯 What it does: CoTracker is a Transformer-based point tracker that can simultaneously track up to 70k points on a single GPU, achieving long-term online tracking through sliding windows and recursive training.
CountFormer: Multi-View Crowd Counting Transformer
Hong Mo (Hubei University Of Arts & Science), Wenqi Ren (Sun Yat Sen University)
Object DetectionTransformerImage
🎯 What it does: Propose a Transformer-based multi-view crowd counting framework called CountFormer, which elevates multi-view image features to a 3D voxel representation and directly regresses a scene-level density map in the voxel space.
CPM: Class-conditional Prompting Machine for Audio-visual Segmentation
Yuanhong Chen (University of Adelaide), Gustavo Carneiro (University of Adelaide)
SegmentationTransformerPrompt EngineeringContrastive LearningMultimodality
🎯 What it does: This paper designs and implements a Transformer-based audio-visual segmentation training framework called CPM, integrating class-agnostic queries and sampled class-conditional queries to enhance the stability of cross-modal attention and bilateral matching.
CPT-VR: Improving Surface Rendering via Closest Point Transform with View-Reflection Appearance
Zhipeng Hu (NetEase Fuxi AI Lab), Xin Yu (University of Queensland)
GenerationNeural Radiance FieldImage
🎯 What it does: Proposes a differentiable surface rendering method called CPT-V R based on the closest point transform (CPT) and view-reflection vectors, which can more accurately project sample points onto the zero isosurface and improve the appearance reconstruction of highlight regions.
CriSp: Leveraging Tread Depth Maps for Enhanced Crime-Scene Shoeprint Matching
Samia Shafique (University of California, Irvine), Charless Fowlkes (University of California, Irvine)
Depth EstimationRetrievalConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: Propose the CriSp method, achieving more accurate shoe print retrieval by matching crime scene shoe prints with sole depth maps.
CRM: Single Image to 3D Textured Mesh with Convolutional Reconstruction Model
Zhengyi Wang (Tsinghua University), Jun Zhu (Tsinghua University)
Image TranslationGenerationConvolutional Neural NetworkDiffusion modelImageMesh
🎯 What it does: Achieved a fast generation model CRM that creates high-quality 3D textured meshes from a single image
CroMo-Mixup: Augmenting Cross-Model Representations for Continual Self-Supervised Learning
Erum Mushtaq (University of Southern California), Salman Avestimehr (University of Southern California)
ClassificationKnowledge DistillationRepresentation LearningContrastive LearningImage
🎯 What it does: In unlabelled continual self-supervised learning, this paper proposes the CroMo-Mixup framework, which enhances the model's ability to retain knowledge of previous tasks and distinguish task IDs through cross-task data mixing and cross-model feature mixing.
Cross-Domain Few-Shot Object Detection via Enhanced Open-Set Object Detector
Yuqian Fu (Fudan University), Xingqun Jiang (BOE Technology)
Object DetectionDomain AdaptationTransformerSupervised Fine-TuningContrastive LearningImageBenchmark
🎯 What it does: This work studies cross-domain few-shot object detection (CD-FSOD). First, a benchmark containing three metrics—style, inter-class variance, and undefined boundary—is constructed. Subsequently, the open-set detector DE-ViT is improved on this benchmark by introducing modules such as learnable instance features, instance re-weighting, and domain trigger, combined with fine-tuning to form CD-ViTO.
Cross-Domain Learning for Video Anomaly Detection with Limited Supervision
Yashika Jain (University of Delhi), Min Xu (Carnegie Mellon University)
Domain AdaptationAnomaly DetectionTransformerContrastive LearningVideo
🎯 What it does: Proposes a weakly supervised cross-domain video anomaly detection framework (CDL), which enhances cross-domain generalization by leveraging external unlabeled data through self-supervised learning, even when only video-level labels are available in the source domain.
Cross-Domain Semantic Segmentation on Inconsistent Taxonomy using VLMs
Jeongkee Lim (Sungkyunkwan University), Yusung Kim (Sungkyunkwan University)
SegmentationDomain AdaptationVision Language ModelImage
🎯 What it does: Proposes the CSI method, achieving semantic segmentation for cross-domain inconsistent classification vocabularies in unsupervised domain adaptation through vision-language models.
Cross-Input Certified Training for Universal Perturbations
Changming Xu (University of Illinois Urbana Champaign), Gagandeep Singh (University of Illinois Urbana Champaign)
ClassificationAdversarial AttackImage
🎯 What it does: Proposed and implemented CITRUS, a certified training method for general adversarial perturbations.
Cross-Platform Video Person ReID: A New Benchmark Dataset and Adaptation Approach
Shizhou Zhang (Northwestern Polytechnical University), Yanning Zhang (Northwestern Polytechnical University)
RecognitionDomain AdaptationTransformerPrompt EngineeringVision Language ModelImageVideoBenchmark
🎯 What it does: Constructed the cross-platform video person re-identification benchmark dataset G2AVReID, and proposed the cross-platform video re-identification method VSLA-CLIP based on CLIP.
Cross-view image geo-localization with Panorama-BEV Co-Retrieval Network
Junyan Ye (Sun Yat Sen University), Conghui He (SenseTime Research)
RetrievalConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: Designed the Panorama-BEV Co-Retrieval Network, combining panoramic street views and explicit bird's-eye-view dual branches to achieve cross-view retrieval.
CrossGLG: LLM Guides One-shot Skeleton-based 3D Action Recognition in a Cross-level Manner
Tingbing Yan (Huazhong University of Science and Technology), Joey Tianyi Zhou (Agency for Science, Technology and Research)
RecognitionGraph Neural NetworkTransformerLarge Language ModelVision-Language-Action ModelTextGraph
🎯 What it does: Proposed the CrossGLG model, which leverages global and local action text descriptions generated by large language models to guide skeleton encoders in learning features through a global→local→global process, achieving one-shot skeleton action recognition;
CrossScore: A Multi-View Approach to Image Evaluation and Scoring
Zirui Wang (University of Oxford), Victor Adrian Prisacariu (University of Oxford)
TransformerImage
🎯 What it does: Propose CrossScore, a cross-referencing image quality assessment method based on multi-view unaligned reference images, which can evaluate the quality of newly synthesized images from novel views without requiring real ground truth.
Crowd-SAM:SAM as a smart annotator for object detection in crowded scenes
Zhi Cai (Beihang University), Di Huang (Beihang University)
Object DetectionTransformerPrompt EngineeringContrastive LearningImage
🎯 What it does: This paper proposes Crowd-SAM, a framework that utilizes SAM as an intelligent annotator for one-class object detection with few samples in crowded scenes;
Cs2K: Class-specific and Class-shared Knowledge Guidance for Incremental Semantic Segmentation
Wei Cong (Shenyang Institute of Automation, Chinese Academy of Sciences), Gan Sun (South China University of Technology)
SegmentationConvolutional Neural NetworkImage
🎯 What it does: Propose a Cs K model for incremental semantic segmentation, which utilizes class-specific prototypes and class-shared model weights to simultaneously suppress old class forgetting and new class bias.
CSOT: Cross-Scan Object Transfer for Semi-Supervised LiDAR Object Detection
Jinglin Zhan (IEIT Systems), Yuntao Chen (Centre for Artificial Intelligence and Robotics)
Object DetectionAutonomous DrivingTransformerPoint Cloud
🎯 What it does: Proposes the Cross-scan Object Transfer (CSOT) method, which uses HotspotNet to predict suitable placement positions, copying and pasting annotated objects into unannotated LiDAR scans to generate sparsely annotated training data, thereby achieving semi-supervised LiDAR object detection.
CTRLorALTer: Conditional LoRAdapter for Efficient 0-Shot Control & Altering of T2I Models
Nick Stracke (CompVis @ LMU Munich), Bjorn Ommer
GenerationPrompt EngineeringDiffusion modelImageText
🎯 What it does: Propose LoRAdapter, a lightweight adapter based on conditional LoRA, which can achieve bidirectional unified control of structure (local) and style (global) in text-to-image diffusion models, supporting zero-shot fine-grained generation.
Curved Diffusion: A Generative Model With Optical Geometry Control
Andrey Voynov (Google), Daniel Cohen-Or (Tel Aviv University)
GenerationData SynthesisSupervised Fine-TuningDiffusion modelImageText
🎯 What it does: Introduce pixel-level coordinate conditions in diffusion models to achieve control over camera geometry, generating images with different lens distortion effects (fisheye, panoramic, spherical textures).
Customize-A-Video: One-Shot Motion Customization of Text-to-Video Diffusion Models
Yixuan Ren (University of Maryland), Abhinav Shrivastava (University of Maryland)
GenerationData SynthesisConvolutional Neural NetworkTransformerSupervised Fine-TuningDiffusion modelVideoText
🎯 What it does: This paper proposes a one-time video motion customization method (Customize-A-Video) based on a pre-trained text-to-video diffusion model, which can learn motion patterns from a single reference video and transfer them to new subjects and scenes.
Customized Generation Reimagined: Fidelity and Editability Harmonized
Jian Jin (Nanjing University of Science and Technology), Jian Yang (Nanjing University of Science and Technology)
GenerationPrompt EngineeringDiffusion modelImageText
🎯 What it does: Propose the DCI framework and ICO refinement strategy to address the fidelity-editability trade-off in text-to-image model customization
Cut out the Middleman: Revisiting Pose-based Gait Recognition
Yang Fu (Beijing Normal University), Yongzhen Huang (Beijing Normal University)
RecognitionPose EstimationConvolutional Neural NetworkTransformerContrastive LearningVideo
🎯 What it does: This paper systematically reconstructs pose-based gait recognition, proposing to directly use heatmaps rather than skeletons as input, improving preprocessing and heatmap alignment processes, and constructing a global-local network with multi-stage fusion branches, significantly enhancing recognition performance and cross-dataset generalization capabilities.
CVT-Occ: Cost Volume Temporal Fusion for 3D Occupancy Prediction
Zhangchen Ye (Tsinghua University), Hang Zhao (Tsinghua University)
Autonomous DrivingConvolutional Neural NetworkTransformerImageTime Series
🎯 What it does: Propose a CVT-Occ module that constructs a 3D cost volume based on gaze sampling and historical frame projection for 3D occupancy prediction in multi-view time series.
D-SCo: Dual-Stream Conditional Diffusion for Monocular Hand-Held Object Reconstruction
Bowen Fu (Tsinghua University), Federico Tombari (Technical University of Munich)
GenerationPose EstimationDiffusion modelImagePoint Cloud
🎯 What it does: This paper proposes a single-view handheld object 3D reconstruction method called D-SCo based on dual-stream conditional diffusion.
D4-VTON: Dynamic Semantics Disentangling for Differential Diffusion based Virtual Try-On
Zhaotong Yang (Beijing Institute Of Technology), Yong Du (Beijing Institute Of Technology)
Image TranslationGenerationDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: A multi-task generative adversarial network (GAN) was constructed that can achieve multi-style transfer and precise pixel-level reconstruction while maintaining image content accuracy.
DA-BEV: Unsupervised Domain Adaptation for Bird's Eye View Perception
Kai Jiang (Xidian University), Shijian Lu (Nanyang Technological University)
Object DetectionSegmentationDomain AdaptationTransformerImage
🎯 What it does: A query-driven unsupervised domain adaptation framework for camera-only Bird’s Eye View (BEV) perception, named DA-BEV, is proposed to enhance the performance of cross-domain 3D object detection and scene segmentation.
DailyDVS-200: A Comprehensive Benchmark Dataset for Event-Based Action Recognition
Qi Wang (Xidian University), Liang Zhang (Xidian University)
RecognitionConvolutional Neural NetworkSpiking Neural NetworkTransformerVideoMultimodalityBenchmark
🎯 What it does: Proposed and made publicly available a large-scale event camera action recognition benchmark dataset named DailyDVS-200, and conducted systematic evaluations of multiple models on it.
DAMSDet: Dynamic Adaptive Multispectral Detection Transformer with Competitive Query Selection and Adaptive Feature Fusion
Junjie Guo, Xinbo Gao (Chongqing University Of Posts And Telecommunications)
Object DetectionConvolutional Neural NetworkTransformerImageMultimodality
🎯 What it does: Proposed DAMSDet, a Transformer framework for infrared-visible light fusion object detection, addressing the problems of multi-modal information competition and alignment.
Data Augmentation via Latent Diffusion for Saliency Prediction
Bahar Aydemir (EPFL), Sabine Süsstrunk (EPFL)
Data SynthesisPrompt EngineeringDiffusion modelImage
🎯 What it does: Generate new images with predictable saliency changes by performing controllable photometric and semantic editing on images through saliency-guided cross-attention in the latent space of Stable Diffusion, and use these images as data augmentation to enhance the training set of deep saliency prediction models.
Data Collection-free Masked Video Modeling
Yuchi Ishikawa (LY Corporation), Yoshimitsu Aoki (Keio University)
ClassificationRecognitionData SynthesisRepresentation LearningTransformerImageVideo
🎯 What it does: Propose a method that recursively transforms a single image to generate pseudo-motion videos and performs unsupervised pre-training on video transformers through masked video modeling (VideoMAE), completely eliminating the need for real video data.
Data Overfitting for On-Device Super-Resolution with Dynamic Algorithm and Compiler Co-Design
Gen Li (Clemson University), Xiaolong Ma (Clemson University)
Super ResolutionVideo
🎯 What it does: This paper proposes a "Dy-DCA" framework that transfers the video block super-resolution task into a single dynamic deep neural network, combined with a content-aware data preprocessing pipeline, thereby significantly reducing model switching overhead.
Data Poisoning Quantization Backdoor Attack
Tran Huynh (VinAI Research), Tung Pham (VinAI Research)
Adversarial AttackConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a quantization backdoor attack method that leverages data poisoning, allowing attackers to induce the model to activate the backdoor after quantization without interfering with the target model's training process. The attack does not require any prior information about the target network structure or training details.
Data-to-Model Distillation: Data-Efficient Learning Framework
Ahmad Sajedi (University of Toronto), Konstantinos N. Plataniotis (University of Toronto)
GenerationData SynthesisKnowledge DistillationGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a data-to-model distillation framework, D2M, which compresses knowledge from large real-world datasets into the learnable parameters of a pre-trained generative model, thereby generating small yet information-rich synthetic images for efficient training.
DataDream: Few-shot Guided Dataset Generation
Jae Myung Kim (University of Tübingen), Zeynep Akata (Helmholtz Munich)
ClassificationData SynthesisTransformerSupervised Fine-TuningDiffusion modelContrastive LearningImage
🎯 What it does: This paper proposes DataDream, a few-shot data generation framework based on LoRA fine-tuning of Stable Diffusion, and further fine-tunes CLIP with LoRA, using a small number of real images to generate a high-quality synthetic training set, significantly enhancing classification performance.
Dataset Distillation by Automatic Training Trajectories
Dai Liu (Technical University of Munich), Martin Schulz (Technical University of Munich)
Data SynthesisKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: Propose an adaptive training trajectory (ATT) method that dynamically selects the optimal trajectory length for long-range dataset distillation, generating a more representative synthetic dataset.
Dataset Enhancement with Instance-Level Augmentations
Orest Kupyn (University of Oxford), Christian Rupprecht (University of Oxford)
Object DetectionSegmentationData SynthesisSafty and PrivacyPrompt EngineeringVision Language ModelDiffusion modelImageMultimodality
🎯 What it does: Proposes an instance-level data augmentation method based on conditional latent diffusion models, generating diverse training samples by redrawing individual objects in images while retaining the original labels.
Dataset Growth
Ziheng Qin (National University of Singapore), Yang You (National University of Singapore)
RetrievalComputational EfficiencyData-Centric LearningVision Language ModelMultimodality
🎯 What it does: Propose InfoGrowth, an online data cleaning and incremental sampling framework that evaluates the noise and redundancy of new samples in a multi-modal embedding space, automatically filtering and incrementally building a high-quality, diverse dataset.
Dataset Quantization with Active Learning based Adaptive Sampling
Zhenghao Zhao (Illinois Institute of Technology), Yan Yan (Illinois Institute of Technology)
CompressionData-Centric LearningAuto EncoderImage
🎯 What it does: Propose an adaptive sampling dataset quantization method (DQAS) based on active learning, improving the traditional DQ process and achieving more precise category sample allocation.
DatasetNeRF: Efficient 3D-aware Data Factory with Generative Radiance Fields
Yu Chi, Adam Kortylewski
SegmentationGenerationData SynthesisNeural Radiance FieldGenerative Adversarial NetworkImageVideoPoint Cloud
🎯 What it does: This paper proposes DatasetNeRF, a 3D-aware data factory based on 3D generative models, capable of generating an infinite number of high-quality, multi-view consistent 2D segmentation maps and corresponding 3D point cloud segmentation results from only a few 2D annotations.
DATENeRF: Depth-Aware Text-based Editing of NeRFs
Sara Rojas Martinez (KAUST), Kalyan Sunkavalli (KAUST)
Image TranslationGenerationDepth EstimationDiffusion modelNeural Radiance FieldImageTextMultimodality
🎯 What it does: This paper proposes a depth-aware text editing method called DATENeRF based on NeRF geometry, achieving high-quality, view-consistent text editing in specified regions of NeRF scenes through ControlNet and depth-conditioned control;
DC-Solver: Improving Predictor-Corrector Diffusion Sampler via Dynamic Compensation
Wenliang Zhao (Tsinghua University), Jiwen Lu (Tsinghua University)
GenerationComputational EfficiencyDiffusion modelImageTextStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: Proposed a new fast sampling method, DC-Solver, which utilizes dynamic compensation (DC) to mitigate alignment errors in predictor-corrector sampling, and can rapidly predict compensation ratios through cascading polynomial regression, achieving efficient DPM sampling.
DCDM: Diffusion-Conditioned-Diffusion Model for Scene Text Image Super-Resolution
Shrey Singh (Indian Institute of Technology), Partha Pratim Roy (Indian Institute of Technology)
Super ResolutionVision Language ModelDiffusion modelImage
🎯 What it does: To address the super-resolution problem of scene text images, a dual diffusion model named DCDM is proposed, which uses low-resolution images and character-level text embeddings as conditions for super-resolution.
De-confounded Gaze Estimation
Ziyang Liang (Beihang University), Feng Lu (Beihang University)
Pose EstimationDomain AdaptationConvolutional Neural NetworkImage
🎯 What it does: Propose a causal intervention framework based on feature separation (FSCI) to achieve cross-domain generalization for gaze estimation without target domain data.
De-Confusing Pseudo-Labels in Source-Free Domain Adaptation
Idit Diamant (Sony Semiconductor Israel), Arnon Netzer (Sony Semiconductor Israel)
Domain AdaptationTransformerContrastive LearningImageBenchmark
🎯 What it does: Propose the DCPL method for source-free domain adaptation tasks, which learns a noise transition matrix to correct pseudo-label noise, enabling model transfer without source data;
DEAL: Disentangle and Localize Concept-level Explanations for VLMs
Tang Li (University of Delaware), Xi Peng (University of Delaware)
Explainability and InterpretabilityLarge Language ModelVision Language ModelContrastive LearningImageChain-of-Thought
🎯 What it does: Proposed an unsupervised method called DEAL that leverages large language models to generate discriminative visual concepts, and enhances the interpretability and predictive accuracy of Vision-Language Models for fine-grained concepts through contrastive learning, decoupling, and localization constraints.
Debiasing surgeon: fantastic weights and how to find them
Remi Nahon, Enzo Tartaglione (Télécom Paris)
Explainability and InterpretabilityImage
🎯 What it does: This paper proposes a method called Finding Fantastic Weights (FFW) to extract unbiased subnetworks from pre-trained deep networks without requiring model retraining or fine-tuning;
Deblur e-NeRF: NeRF from Motion-Blurred Events under High-speed or Low-light Conditions
Weng Fei Low (National University of Singapore), Gim Hee Lee (National University of Singapore)
RestorationGenerationData SynthesisNeural Radiance FieldVideo
🎯 What it does: Proposed a Deblur e-NeRF method that directly reconstructs a nearly blur-free NeRF from motion-blurred events.
Deblurring 3D Gaussian Splatting
Byeonghyeon Lee (Sungkyunkwan University), Eunbyung Park (Sungkyunkwan University)
RestorationNeural Radiance FieldGaussian SplattingImagePoint CloudBenchmark
🎯 What it does: This paper proposes a real-time deblurring framework called Deblurring 3D Gaussian Splatting, which adjusts the covariance of 3D Gaussian distributions during training using a small MLP, thereby restoring details in blurred images while maintaining the high frame rate of 3D-GS.
DECap: Towards Generalized Explicit Caption Editing via Diffusion Mechanism
Zhen Wang (Zhejiang University), Long Chen (HKUST)
GenerationTransformerVision Language ModelDiffusion modelImageText
🎯 What it does: Developed a discrete diffusion mechanism-based explicit title editing framework, DECap, which enables multi-step editing of reference titles and significantly improves editing effectiveness.
DecentNeRFs: Decentralized Neural Radiance Fields from Crowdsourced Images
Zaid Tasneem (Rice University), Ramesh Raskar (Rice University)
Federated LearningSafty and PrivacyNeural Radiance FieldImage
🎯 What it does: This paper proposes DecentNeRF, a decentralized and distributed neural radiance field learning framework that trains local and personal MLPs on user devices and securely aggregates the global model on the server, enabling large-scale 3D reconstruction from multiple viewpoints.
DECIDER: Leveraging Foundation Model Priors for Improved Model Failure Detection and Explanation
Rakshith Subramanyam (Axio.ai), Jayaraman J. Thiagarajan (Lawrence Livermore National Laboratory)
Anomaly DetectionExplainability and InterpretabilityConvolutional Neural NetworkLarge Language ModelVision Language ModelImageText
🎯 What it does: This paper proposes the DECIDER method, which generates task-related core attributes using a large language model (LLM) and constructs a 'bias-aware' classifier PIM with a vision-language model (VLM). It detects model failure by analyzing the prediction differences between the original classifier and PIM, and provides interpretable failure reasons through attribute ablation.
Deciphering the Role of Representation Disentanglement: Investigating Compositional Generalization in CLIP Models
Reza Abbasi (Sharif University of Technology), Mahdieh Soleymani Baghshah (Sharif University of Technology)
Explainability and InterpretabilityRepresentation LearningVision Language ModelContrastive LearningMultimodalityBenchmark
🎯 What it does: Constructed a novel OoD dataset, ImageNet-AO, specifically designed to evaluate the compositional generalization ability of CLIP models in single-object scenarios, and conducted zero-shot evaluation on multiple CLIP variants using this dataset.
DeCo: Decoupled Human-Centered Diffusion Video Editing with Motion Consistency
Xiaojing Zhong (South China University of Technology), Qingyao Wu (South China University of Technology)
Image HarmonizationGenerationData SynthesisPose EstimationDiffusion modelScore-based ModelVideo
🎯 What it does: Propose DeCo, a framework that separately edits humans and backgrounds in videos, utilizing the SMPL-X prior to ensure human motion consistency, while performing texture editing on the background through layered atlases.
Decomposed Vector-Quantized Variational Autoencoder for Human Grasp Generation
zhao zhe, Huadong Ma (State Key Laboratory of Networking and Switching Technology)
GenerationAuto EncoderPoint Cloud
🎯 What it does: Proposed a decomposition-based vector-quantized variational autoencoder (DVQ-VAE) to generate realistic human grasps that conform to objects.
Decomposition Betters Tracking Everything Everywhere
Rui Li (University of Science and Technology of China), Dong Liu (University of Science and Technology of China)
Object TrackingSegmentationFlow-based ModelNeural Radiance FieldOptical FlowVideo
🎯 What it does: Proposes a decomposition-based test-time optimization method called DecoMotion for precise long-range motion tracking of each pixel in videos.
Decomposition of Neural Discrete Representations for Large-Scale 3D Mapping
Minseong Park (Yonsei University), Euntai Kim (Yonsei University)
Autonomous DrivingComputational EfficiencyRepresentation LearningPoint Cloud
🎯 What it does: Proposed a decomposition-based discrete neural mapping method (DNMap), which constructs efficient 3D environment representations by learning combinable discrete embeddings and low-resolution continuous embeddings in sparse octrees;
Decoupling Common and Unique Representations for Multimodal Self-supervised Learning
Yi Wang (Technical University of Munich), Xiao Xiang Zhu (Technical University of Munich)
ClassificationSegmentationRepresentation LearningConvolutional Neural NetworkContrastive LearningMultimodality
🎯 What it does: This paper proposes a multi-modal self-supervised learning framework called DeCUR, which learns cross-modal consistency and modality-specific information by dividing the feature dimension into common and unique parts, and combines cross-modal and single-modal redundancy reduction loss;
Deep Companion Learning: Enhancing Generalization Through Historical Consistency
Ruizhao Zhu (Boston University), Venkatesh Saligrama (Boston University)
OptimizationComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: Propose the Deep Companion Learning (DCL) method, which achieves historical consistency regularization by training an isomorphic companion model to the main model, thereby enhancing the generalization performance of deep neural networks in supervised learning.
Deep Cost Ray Fusion for Sparse Depth Video Completion
Jungeon Kim (POSTECH), Seungyong Lee (POSTECH)
Depth EstimationTransformerVideoPoint Cloud
🎯 What it does: Propose a learning-based sparse depth video completion framework called RayFusion, which constructs a cost volume using sparse depth and color information from multi-frame RGB-D videos, and achieves temporal depth completion through ray-level attention fusion.
Deep Diffusion Image Prior for Efficient OOD Adaptation in 3D Inverse Problems
Hyungjin Chung (KAIST), Jong Chul Ye (KAIST)
RestorationDomain AdaptationDiffusion modelBiomedical DataMagnetic Resonance ImagingComputed TomographyOrdinary Differential Equation
🎯 What it does: This paper proposes the Deep Diffusion Image Prior (DDIP) framework and designs an efficient 3D OOD adaptation method, D3IP, addressing the high computational cost and memory consumption of traditional SCD in 3D inverse problems, while achieving cross-slice joint optimization to improve reconstruction consistency and speed.
Deep Feature Surgery: Towards Accurate and Efficient Multi-Exit Networks
Cheng Gong (Nankai University), Le Zhang (University of Electronic Science and Technology of China)
Computational EfficiencyImage
🎯 What it does: Studied a training method for multi-output networks that resolves gradient conflicts and enhances inference efficiency.
Deep Nets with Subsampling Layers Unwittingly Discard Useful Activations at Test-Time
Chiao-An Yang (Purdue University), Raymond Yeh
ClassificationSegmentationConvolutional Neural NetworkTransformerImage
🎯 What it does: Propose to utilize the discarded activation information from downsampling layers in deep networks during testing, designing a search and aggregation mechanism to improve prediction results in image classification and semantic segmentation.
Deep Online Probability Aggregation Clustering
Yuxuan Yan (Xi'an Jiaotong University), Ruofan Yan (Xi'an Jiaotong University)
OptimizationRepresentation LearningConvolutional Neural NetworkContrastive LearningImageBenchmark
🎯 What it does: Propose a decentralized probabilistic aggregation clustering algorithm called PAC, extend it to an online probabilistic aggregation module OPA, and construct an end-to-end deep clustering framework DPAC, which can achieve stable online clustering without relying on center updates;
Deep Patch Visual SLAM
Lahav Lipson (Princeton University), Jia Deng (Princeton University)
Depth EstimationConvolutional Neural NetworkRecurrent Neural NetworkSimultaneous Localization and MappingOptical FlowImageVideo
🎯 What it does: Proposes DPV-SLAM, a monocular visual SLAM system based on DPVO, incorporating approximate loop closure mechanisms and classical image retrieval loop closure to improve accuracy and robustness.
Deep Polarization Cues for Single-shot Shape and Subsurface Scattering Estimation
Chenhao Li (Osaka University), Hajime Nagahara (Osaka University)
Depth EstimationConvolutional Neural NetworkImage
🎯 What it does: Proposes a method that jointly estimates the shape (normal/depth) and subsurface scattering (SSS) parameters of semi-transparent objects using four linearly polarized images in a single shot.
Deep Reward Supervisions for Tuning Text-to-Image Diffusion Models
Xiaoshi Wu (CUHK MMLab), Hongsheng Li (Avolution AI)
GenerationSupervised Fine-TuningReinforcement LearningDiffusion modelImageText
🎯 What it does: A depth reward supervision method called DRTune is studied, which directly applies gradient supervision to the final output image of a text-to-image diffusion model, and efficiently trains early denoising steps by stopping the input gradient of the denoising network and sampling isometric substeps.
Defect Spectrum: A Granular Look of Large-scale Defect Datasets with Rich Semantics
Shuai Yang (Hong Kong University of Science and Technology), Yingcong Chen (SmartMore Corp)
SegmentationData SynthesisAnomaly DetectionConvolutional Neural NetworkDiffusion modelImageBenchmark
🎯 What it does: Propose the Defect Spectrum benchmark dataset and develop the Defect-Gen two-stage diffusion generative model to enhance industrial defect detection and segmentation performance;
Delving Deep into Engagement Prediction of Short Videos
dasong Li, Jian Wang (Snap Inc.)
Recommendation SystemTransformerLarge Language ModelVideoTextMultimodalityAudio
🎯 What it does: This paper constructs a large-scale dataset named SnapUGC containing 90k real short videos and proposes two specialized metrics for measuring short video engagement: Normalized Average Watch Percentage (NAWP) and Watch Continuation Rate (ECR). Based on this, the authors design a cross-modal attention network that integrates multi-modal features including visual, audio, title/description, captions, and sentiment to predict engagement for cold-start short videos.
Delving into Adversarial Robustness on Document Tampering Localization
Huiru Shao (Xi'an Jiaotong-Liverpool University), Qiufeng Wang (Xi'an Jiaotong-Liverpool University)
SegmentationAdversarial AttackGenerative Adversarial NetworkImage
🎯 What it does: Investigated the vulnerability of document tampering localization (DTL) models under adversarial attacks, and proposed a latent manifold-based adversarial training method to enhance their robustness.
Denoising Vision Transformers
Jiawei Yang (University of Southern California), Yue Wang (University of Southern California)
RestorationObject DetectionSegmentationDepth EstimationTransformerImage
🎯 What it does: This paper addresses the widespread noise/grid-like artifact problem in pre-trained Vision Transformers (ViT) features by proposing a two-stage denoising framework called Denoising Vision Transformers (DVT).
denoiSplit: a method for joint microscopy image splitting and unsupervised denoising
Ashesh Ashesh (Fondazione Human Technopole), Florian Jug (Fondazione Human Technopole)
RestorationSegmentationAuto EncoderBiomedical Data
🎯 What it does: Proposes denoiSplit, a variational segmentation encoder-decoder network capable of simultaneously performing semantic image splitting and denoising under unsupervised conditions.
Dense Hand-Object(HO) GraspNet with Full Grasping Taxonomy and Dynamics
Woojin Cho (KAIST), Tae-Kyun (T-K) Kim
ClassificationData SynthesisPose EstimationImagePoint CloudBenchmark
🎯 What it does: Built and annotated a comprehensive hand-object interaction dataset called HOGraspNet, covering all 33 grasp categories (later merged into 28), and providing high-quality annotations for each sample, including 3D hand models, object poses, contact maps, and grasp labels.
Dense Multimodal Alignment for Open-Vocabulary 3D Scene Understanding
Ruihuang Li (Hong Kong Polytechnic University), Lei Zhang (Joins Hopkins University)
SegmentationAutonomous DrivingConvolutional Neural NetworkTransformerLarge Language ModelVision Language ModelContrastive LearningImageMultimodalityPoint Cloud
🎯 What it does: Propose the Dense Multimodal Alignment (DMA) framework, achieving open-vocabulary 3D scene understanding by establishing dense correspondences among 3D points, 2D pixels, and text.
DenseNets Reloaded: Paradigm Shift Beyond ResNets and ViTs
DongHyun Kim, Dongyoon Han (NAVER AI Lab)
ClassificationObject DetectionSegmentationConvolutional Neural NetworkTransformerImageMultimodality
🎯 What it does: Revive DenseNet, propose RDNet, and train/evaluate on ImageNet-1K, ADE20K, COCO, etc.
Dependency-aware Differentiable Neural Architecture Search
Buang Zhang (East China Normal University), Chenjuan Guo (East China Normal University)
Neural Architecture SearchImageBenchmark
🎯 What it does: Propose a new differentiable neural architecture search method called DaNAS, which models the architecture weights of each edge as random variables in a multivariate normal distribution. It learns the mean vector and covariance matrix, explicitly modeling dependencies between edges. The method samples architecture weights and alternately updates distribution parameters and network weights during training. It also dynamically prunes the search space using the learned covariance matrix and identifies general high-performance modules by analyzing the covariance matrix.
DEPICT: Diffusion-Enabled Permutation Importance for Image Classification Tasks
Sarah Jabbour (University of Michigan), Jenna Wiens (University of Michigan)
ClassificationExplainability and InterpretabilityDiffusion modelImageTextBiomedical Data
🎯 What it does: Propose the DEPICT method, which generates images using text-conditioned diffusion models after swapping concepts in the text space, and then evaluates the importance of concepts through model performance degradation to provide a global concept importance ranking.
Depicting Beyond Scores: Advancing Image Quality Assessment through Multi-modal Language Models
Zhiyuan You (Chinese University of Hong Kong), Chao Dong (Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences)
Explainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: Propose a multimodal large language model-based image quality assessment method, DepictQA, capable of descriptive quality evaluation and comparison of multiple images, outputting interpretable text results.
Depth on Demand: Streaming Dense Depth from a Low Frame Rate Active Sensor
Andrea Conti (University of Bologna), Stefano Mattoccia (University of Bologna)
Depth EstimationConvolutional Neural NetworkRecurrent Neural NetworkImagePoint Cloud
🎯 What it does: Propose a framework that realizes real-time dense depth flow using a high-frame-rate RGB camera based on a low-frame-rate sparse active depth sensor.
Depth-Aware Blind Image Decomposition for Real-World Adverse Weather Recovery
Chao Wang (University of Technology Sydney), Yi Yang (Zhejiang University)
RestorationDepth EstimationNeural Architecture SearchConvolutional Neural NetworkTransformerImageBenchmark
🎯 What it does: Proposed the deep perceptual blind image decomposition network DeBNet for restoring clear images in real-world scenarios with mixed adverse weather.
Depth-guided NeRF Training via Earth Mover’s Distance
Anita Rau (Stanford University), Serena Yeung-Levy (Stanford University)
Depth EstimationDiffusion modelNeural Radiance FieldImage
🎯 What it does: Propose a depth-guided NeRF training method based on Earth Mover's Distance (EMD), leveraging depth priors and uncertainty information provided by diffusion models to guide the ray termination distance distribution, thereby improving geometric reconstruction accuracy in indoor scenes with limited viewpoints.
DetailSemNet: Elevating Signature Verification through Detail-Semantic Integration
Meng-Cheng Shih (National Yang Ming Chiao Tung University), Ching-Chun Huang (E.SUN Financial Holding Co Ltd)
RecognitionConvolutional Neural NetworkTransformerContrastive LearningImage
🎯 What it does: Propose the DetailSemNet model, achieving offline signature verification through local structural matching and detail semantic fusion.