CVPR 2024 Papers — Page 7
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2716 papers
Discriminative Probing and Tuning for Text-to-Image Generation
Leigang Qu (National University of Singapore), Tat-Seng Chua (National University of Singapore)
GenerationData SynthesisTransformerDiffusion modelImageText
🎯 What it does: By introducing a discriminative adapter and discriminative fine-tuning, the Stable Diffusion model undergoes two-stage optimization to enhance the alignment capability between text and images.
Discriminative Sample-Guided and Parameter-Efficient Feature Space Adaptation for Cross-Domain Few-Shot Learning
Rashindrie Perera (University of Melbourne), Saman Halgamuge (University of Melbourne)
ClassificationDomain AdaptationMeta LearningTransformerContrastive LearningImage
🎯 What it does: This paper studies the cross-domain few-shot classification problem and proposes a lightweight and parameter-efficient feature space adaptation method.
Disentangled Pre-training for Human-Object Interaction Detection
Zhuolong Li (South China University of Technology), Xiangmin Xu (South China University of Technology)
RecognitionObject DetectionTransformerContrastive LearningImageVideoText
🎯 What it does: A decoupled pre-training framework DP-HOI is proposed, which separately pre-trains the detection and interaction decoders of DETR using object detection and action recognition data, ultimately aimed at improving HOI detection performance.
Disentangled Prompt Representation for Domain Generalization
De Cheng (Xidian University), Xinbo Gao (Chongqing University of Posts and Telecommunications)
Domain AdaptationRepresentation LearningTransformerLarge Language ModelPrompt EngineeringVision Language ModelContrastive LearningImageText
🎯 What it does: This paper proposes a domain generalization framework (DPR) based on GPT-Assist text decoupling and visual prompt learning. It first divides text prompts into domain-invariant and domain-specific parts, then uses them to guide the visual encoder in learning corresponding visual representations, and combines domain-specific prototype learning during the inference phase to achieve better generalization.
Dispel Darkness for Better Fusion: A Controllable Visual Enhancer based on Cross-modal Conditional Adversarial Learning
Hao Zhang (Wuhan University), Jiayi Ma (Wuhan University)
Image TranslationRestorationObject DetectionSegmentationGenerative Adversarial NetworkImageMultimodality
🎯 What it does: A controllable visual enhancer DDBF is proposed, utilizing cross-modal conditional adversarial learning to achieve brightness enhancement of low-light visible images and infrared-visible fusion.
Dispersed Structured Light for Hyperspectral 3D Imaging
Suhyun Shin (POSTECH), Seung-Hwan Baek (Princeton University)
Depth EstimationImage
🎯 What it does: A low-cost, high-quality structured light DSL technology is proposed, achieved through a front-mounted film diffraction grating in front of a projector, to simultaneously acquire depth and high-resolution spectral information of the scene.
DiSR-NeRF: Diffusion-Guided View-Consistent Super-Resolution NeRF
Jie Long Lee (National University of Singapore), Gim Hee Lee (National University of Singapore)
RestorationGenerationSuper ResolutionDiffusion modelScore-based ModelNeural Radiance FieldImage
🎯 What it does: This paper proposes a perspective-consistent super-resolution NeRF (DiSR-NeRF) using knowledge from a 2D diffusion upsampler.
Distilled Datamodel with Reverse Gradient Matching
Jingwen Ye (National University of Singapore), Xinchao Wang (National University of Singapore)
ClassificationSafty and PrivacyComputational EfficiencyKnowledge DistillationImage
🎯 What it does: A two-stage Distilled Datamodel framework is proposed, which first compresses the influence of the training set into a small synthetic set using reverse gradient matching in the offline stage, and then quickly implements leave-one-out training and attribution matrix calculation using this set in the online stage.
Distilling CLIP with Dual Guidance for Learning Discriminative Human Body Shape Representation
Feng Liu (Michigan State University), Xiaoming Liu (Michigan State University)
RecognitionKnowledge DistillationRepresentation LearningConvolutional Neural NetworkContrastive LearningMesh
🎯 What it does: This paper proposes CLIP3DReID, which utilizes CLIP knowledge distillation and dual-guided learning of human shape features to enhance long-term person re-identification performance.
Distilling ODE Solvers of Diffusion Models into Smaller Steps
Sanghwan Kim (ETH Zurich), Fisher Yu (Carnegie Mellon University)
GenerationComputational EfficiencyKnowledge DistillationDiffusion modelImageOrdinary Differential Equation
🎯 What it does: A D-ODE sampler is proposed, which incorporates a single learnable parameter into existing ODE solvers. By using knowledge distillation, the trajectories of large-step sampling are transferred to small-step sampling, achieving faster and more accurate diffusion model sampling.
Distilling Semantic Priors from SAM to Efficient Image Restoration Models
Quan Zhang (Tsinghua University), Yunhe Wang (Huawei)
RestorationSegmentationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a framework that transfers the semantic prior of the Segment Anything Model (SAM) to existing image restoration models through Semantic Prior Fusion (SPF) and Semantic Prior Distillation (SPD) methods, ensuring that the computational load during the inference phase does not increase.
Distilling Vision-Language Models on Millions of Videos
Yue Zhao (Google Research), Liangzhe Yuan (Google Research)
RetrievalKnowledge DistillationRepresentation LearningTransformerLarge Language ModelVision Language ModelContrastive LearningVideoText
🎯 What it does: By transferring image-language models to the video domain and automatically generating millions of high-quality pseudo subtitles through a two-stage visual and language adaptation, followed by training a video-language alignment model with these subtitles, zero-shot retrieval and understanding of large-scale videos have been achieved.
Distraction is All You Need: Memory-Efficient Image Immunization against Diffusion-Based Image Editing
Ling Lo (National Yang Ming Chiao Tung University), Wen-Huang Cheng (National Taiwan University)
Adversarial AttackTransformerDiffusion modelImage
🎯 What it does: A semantic attack targeting text-to-image diffusion models is proposed to achieve immunity against images and prevent malicious editing.
Distribution-aware Knowledge Prototyping for Non-exemplar Lifelong Person Re-identification
Kunlun Xu (Peking University), Jiahuan Zhou (Huazhong University of Science and Technology)
RecognitionRetrievalConvolutional Neural NetworkGaussian SplattingImage
🎯 What it does: A distribution-aware knowledge prototyping (DKP) framework is proposed, combining instance-level distribution modeling with distribution-guided prototype generation to achieve zero-shot memory lifelong face re-identification.
Distributionally Generative Augmentation for Fair Facial Attribute Classification
Fengda Zhang (Zhejiang University), Hanwang Zhang (Nanyang Technological University)
ClassificationRepresentation LearningGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: A two-stage label-free fair facial attribute classification method based on generative models is proposed. First, potential biased attributes are detected through a generative model, and then random amplitude generative augmentation is applied to the potential biased attributes of each image, training a representation network that is insensitive to these augmentations to achieve fair predictions.
DistriFusion: Distributed Parallel Inference for High-Resolution Diffusion Models
Muyang Li (Massachusetts Institute of Technology), Song Han (NVIDIA)
GenerationData SynthesisComputational EfficiencyDiffusion modelImage
🎯 What it does: DistriFusion has been developed, an untrained multi-GPU parallel inference algorithm designed to accelerate the generation of high-resolution diffusion models while maintaining image quality.
DITTO: Dual and Integrated Latent Topologies for Implicit 3D Reconstruction
Jaehyeok Shim (Ulsan National Institute of Science and Technology), Kyungdon Joo (Ulsan National Institute of Science and Technology)
RestorationGenerationConvolutional Neural NetworkTransformerPoint CloudMesh
🎯 What it does: This paper proposes DITTO, an implicit 3D reconstruction framework based on dual (point + grid) latent topology, specifically designed for noisy sparse point clouds.
DiVa-360: The Dynamic Visual Dataset for Immersive Neural Fields
Cheng-You Lu (Brown University), Srinath Sridhar (Brown University)
SegmentationGenerationNeural Radiance FieldImageVideoMultimodality
🎯 What it does: DiVa-360 is proposed and released, which is a real-world 360° multi-view dynamic desktop scene dataset, containing 17.4 million frames captured synchronously by 53 cameras, with sequences lasting up to 3 minutes and foreground-background segmentation;
DiVAS: Video and Audio Synchronization with Dynamic Frame Rates
Clara Fernandez-Labrador (Disney Research Studios), Christopher Schroers (Disney Research Studios)
TransformerContrastive LearningVideoMultimodalityAudio
🎯 What it does: Proposes the DiVAS model, which uses Transformer to directly process raw video and audio, automatically detecting and locating audio-video synchronization issues, and supports both face cropping and full-frame input, compatible with multiple frame rates and sampling rates.
DiverGen: Improving Instance Segmentation by Learning Wider Data Distribution with More Diverse Generative Data
Chengxiang Fan (Zhejiang University), Chunhua Shen (Zhejiang University)
Object DetectionSegmentationDiffusion modelImage
🎯 What it does: Proposes the DiverGen method, which generates data using multi-level categories, prompts, and model diversity, and conducts instance segmentation training on LVIS;
Diversified and Personalized Multi-rater Medical Image Segmentation
Yicheng Wu (Monash University), Jianfei Cai (University of Electronic Science and Technology of China)
SegmentationConvolutional Neural NetworkImageBiomedical Data
🎯 What it does: A dual-stage model D-Persona is proposed, which can generate diverse segmentation results and provide personalized segmentation for each expert.
Diversity-aware Channel Pruning for StyleGAN Compression
Jiwoo Chung (Sungkyunkwan University), Jae-Pil Heo (Sungkyunkwan University)
GenerationCompressionGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a channel pruning method based on the sensitivity of latent vector perturbation to channel pairs, aimed at compressing the StyleGAN generator while maintaining sample diversity.
DL3DV-10K: A Large-Scale Scene Dataset for Deep Learning-based 3D Vision
Lu Ling (Purdue University), Aniket Bera (Purdue University)
Data SynthesisRepresentation LearningNeural Radiance FieldVideoBenchmark
🎯 What it does: A large-scale real-world multi-view video dataset DL3DV-10K is proposed, containing 10,510 segments of 4K resolution videos, 51.2 million frames, covering 65 categories of POI locations, with fine-grained annotations on scene complexity (indoor/outdoor, reflections, transparency, texture frequency); a DL3DV-140 benchmark set is also constructed for systematic evaluation of existing NVS methods; and pre-training is conducted on this dataset to validate its forward transfer effect on general NeRF.
DMR: Decomposed Multi-Modality Representations for Frames and Events Fusion in Visual Reinforcement Learning
Haoran Xu (Sun Yat-sen University), Yonghong Tian (Peking University)
Autonomous DrivingReinforcement LearningContrastive LearningImageVideoMultimodality
🎯 What it does: The DMR framework is proposed, utilizing multimodal inputs from RGB frames and event cameras, explicitly decomposed into task-relevant common features and modality-specific noise, followed by using common features to drive RL decision-making.
DNGaussian: Optimizing Sparse-View 3D Gaussian Radiance Fields with Global-Local Depth Normalization
Jiahe Li (Beihang University), Lin Gu (Beihang University)
Depth EstimationOptimizationNeural Radiance FieldGaussian SplattingPoint Cloud
🎯 What it does: For sparse view synthesis, the DNGaussian method is proposed, which combines 3D Gaussian Radiance Fields with depth regularization.
Do Vision and Language Encoders Represent the World Similarly?
Mayug Maniparambil (ML Labs Dublin City University), Noel E. O'Connor (ML Labs Dublin City University)
RetrievalRepresentation LearningTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: The study investigates the similarity of unaligned unimodal visual and language encoders in the semantic space and quantifies it using Centered Kernel Alignment (CKA).
Do You Remember? Dense Video Captioning with Cross-Modal Memory Retrieval
Minkuk Kim (Kyung Hee University), Seong Tae Kim (Kyung Hee University)
GenerationRetrievalTransformerVideoTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: A dense video captioning framework for cross-modal memory retrieval, CM2, is proposed, which enhances event localization and description using external text memory.
DocRes: A Generalist Model Toward Unifying Document Image Restoration Tasks
Jiaxin Zhang (South China University of Technology), Lianwen Jin (South China University of Technology)
RestorationTransformerPrompt EngineeringImage
🎯 What it does: This paper proposes DocRes, a unified model for five document image restoration tasks (dewarping, shadow removal, appearance enhancement, deblurring, and binarization).
Domain Gap Embeddings for Generative Dataset Augmentation
Yinong Oliver Wang (Carnegie Mellon University), Fernando De la Torre (Carnegie Mellon University)
ClassificationSegmentationGenerationData SynthesisDomain AdaptationDiffusion modelContrastive LearningImage
🎯 What it does: Proposes the DoGE framework, which extracts domain difference embeddings from the source and target domains, and utilizes Stable UnCLIP to generate cross-domain synthetic data under unsupervised and few-shot conditions, achieving task-agnostic data augmentation.
Domain Prompt Learning with Quaternion Networks
Qinglong Cao (Shanghai Jiao Tong University), Xiaokang Yang (Shanghai Jiao Tong University)
SegmentationDomain AdaptationPrompt EngineeringVision Language ModelContrastive LearningImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper proposes a domain prompt learning method based on quaternion networks, which utilizes the visual features of domain-specific foundational models to transfer the contextual embeddings of general VLMs to specialized domains in remote sensing and medical imaging.
Domain Separation Graph Neural Networks for Saliency Object Ranking
Zijian Wu (Nanjing University of Science and Technology), Siyang Song (University of Leicester)
Object DetectionSegmentationGraph Neural NetworkTransformerSupervised Fine-TuningImage
🎯 What it does: This paper proposes a Domain-Separated Graph Neural Network (DSGNN) for simultaneously ranking and segmenting multiple targets in images based on saliency.
Domain-Agnostic Mutual Prompting for Unsupervised Domain Adaptation
Zhekai Du (University of Electronic Science and Technology of China), Jingjing Li (Tongji University)
Domain AdaptationTransformerPrompt EngineeringContrastive LearningImage
🎯 What it does: This paper proposes a Domain-Agnostic Mutual Prompting (DAMP) method utilizing CLIP for unsupervised domain adaptation, aligning visual and textual prompts to enhance the transfer of source domain knowledge to the target domain.
Domain-Rectifying Adapter for Cross-Domain Few-Shot Segmentation
Jiapeng Su (Harbin Institute of Technology), Fanglin Chen (Nanjing University)
SegmentationDomain AdaptationMeta LearningContrastive LearningImage
🎯 What it does: This work proposes a lightweight Domain-Rectifying Adapter for cross-domain few-shot semantic segmentation.
Domain-Specific Block Selection and Paired-View Pseudo-Labeling for Online Test-Time Adaptation
Yeonguk Yu (Gwangju Institute of Science and Technology), Kyoobin Lee (Gwangju Institute of Science and Technology)
Domain AdaptationImage
🎯 What it does: The DPLOT framework is proposed, which first updates network blocks that only affect domain-specific features through domain-specific block selection before training, and then generates high-quality pseudo-labels during online testing using paired views with only horizontal flipping, performing entropy minimization and symmetric cross-entropy training.
Don't Drop Your Samples! Coherence-Aware Training Benefits Conditional Diffusion
Nicolas Dufour (Université Gustave Eiffel), David Picard (Valeo)
GenerationTransformerDiffusion modelImageText
🎯 What it does: A training method is proposed that uses 'consistency scores' to dynamically adjust the conditional influence in conditional diffusion models, enabling the model to generate images that are more conditionally aligned and of higher quality in noisy label environments.
Don't Look into the Dark: Latent Codes for Pluralistic Image Inpainting
Haiwei Chen (University of Southern California), Yajie Zhao (University of Southern California)
RestorationTransformerGenerative Adversarial NetworkImage
🎯 What it does: A diversified image inpainting method based on discrete latent codes is proposed, utilizing restricted encoding, bidirectional Transformer, and coupled decoders to complete large area occlusion repair.
Doodle Your 3D: From Abstract Freehand Sketches to Precise 3D Shapes
Hmrishav Bandyopadhyay (University of Surrey), Yi-Zhe Song (University of Surrey)
GenerationData SynthesisDiffusion modelImagePoint CloudMesh
🎯 What it does: A 3D shape generation framework based on latent diffusion models is proposed, capable of directly generating accurate and editable 3D models from users' abstract hand-drawn sketches (without the need for alignment or paired data).
Doubly Abductive Counterfactual Inference for Text-based Image Editing
Xue Song (Fudan University), Yu-Gang Jiang (Fudan University)
Image TranslationGenerationDiffusion modelImageText
🎯 What it does: This paper proposes a framework based on Doubly Abductive Counterfactual reasoning for text-driven single image editing.
DPHMs: Diffusion Parametric Head Models for Depth-based Tracking
Jiapeng Tang (Technical University of Munich), Matthias Nießner (Technical University of Munich)
Object TrackingPose EstimationDepth EstimationDiffusion modelVideoPoint Cloud
🎯 What it does: A head parameterization model based on diffusion models (DPHM) is proposed, which can achieve robust head reconstruction and expression tracking from monocular depth sequences.
DPMesh: Exploiting Diffusion Prior for Occluded Human Mesh Recovery
Yixuan Zhu (Tsinghua University), Jiwen Lu (Tsinghua University)
GenerationPose EstimationTransformerDiffusion modelContrastive LearningMesh
🎯 What it does: This paper proposes a framework (DPMesh) that utilizes a pre-trained diffusion model (Stable Diffusion) as a single-step feature extractor, combined with conditional injection and noise keypoint inference, to recover 3D human meshes under severe occlusion.
Dr. Bokeh: DiffeRentiable Occlusion-aware Bokeh Rendering
Yichen Sheng (Purdue University), Bedrich Benes (Purdue University)
ImageBenchmark
🎯 What it does: A differentiable and occlusion-aware realistic depth of field (Bokeh) rendering method called Dr.Bokeh is proposed;
Dr.Hair: Reconstructing Scalp-Connected Hair Strands without Pre-Training via Differentiable Rendering of Line Segments
Yusuke Takimoto (Huawei Technologies Japan), Bo Zheng (Huawei Technologies Japan)
GenerationOptimizationImageMesh
🎯 What it does: Using multi-view images and differentiable rasterization techniques, we optimize the strand bundle model to achieve complete 3D hair bundle reconstruction from the scalp to the surface without the need for pre-trained synthetic data.
Dr2Net: Dynamic Reversible Dual-Residual Networks for Memory-Efficient Finetuning
Chen Zhao (King Abdullah University of Science and Technology), Bernard Ghanem (King Abdullah University of Science and Technology)
RecognitionObject DetectionSegmentationTransformerSupervised Fine-TuningVideoPoint Cloud
🎯 What it does: This paper proposes Dr Net, a dynamic reversible dual residual network designed for low-memory end-to-end fine-tuning of pre-trained models without sacrificing accuracy.
Drag Your Noise: Interactive Point-based Editing via Diffusion Semantic Propagation
Haofeng Liu (Singapore Management University), Shengfeng He (Singapore Management University)
Diffusion modelImageBenchmark
🎯 What it does: This paper proposes DragNoise, a point-dragging interactive image editing method based on diffusion models, which utilizes the bottleneck features of U-Net for single-step semantic optimization and directly replaces the bottleneck features in subsequent time steps for efficient propagation.
DragDiffusion: Harnessing Diffusion Models for Interactive Point-based Image Editing
Yujun Shi (National University of Singapore), Song Bai (ByteDance Inc.)
GenerationOptimizationDiffusion modelImageBenchmark
🎯 What it does: This paper presents DragDiffusion, an interactive point-based image editing framework based on diffusion models, which supports precise drag-and-drop editing on both real images and diffusion-generated images.
Draw Step by Step: Reconstructing CAD Construction Sequences from Point Clouds via Multimodal Diffusion.
Weijian Ma (Fudan University), Xiangdong Zhou (Fudan University)
GenerationData SynthesisDiffusion modelMultimodalityPoint Cloud
🎯 What it does: Reconstructing CAD construction sequences from point clouds
DREAM: Diffusion Rectification and Estimation-Adaptive Models
Jinxin Zhou (Microsoft), Luming Liang (Microsoft)
RestorationSuper ResolutionDiffusion modelImage
🎯 What it does: The DREAM framework is proposed to address the training-sampling mismatch problem of conditional diffusion models in super-resolution tasks, achieving self-calibration of the model through diffusion correction and adaptive estimation in two steps.
DreamAvatar: Text-and-Shape Guided 3D Human Avatar Generation via Diffusion Models
Yukang Cao (University of Hong Kong), Kwan-Yee K. Wong (Tencent PCG)
GenerationData SynthesisDiffusion modelNeural Radiance FieldMesh
🎯 What it does: Proposes DreamAvatar, a 3D human avatar generation framework based on text and shape prompts.
DreamComposer: Controllable 3D Object Generation via Multi-View Conditions
Yunhan Yang (University of Hong Kong), Xihui Liu (Tsinghua University)
GenerationData SynthesisDiffusion modelPoint CloudMesh
🎯 What it does: The DreamComposer framework is proposed to achieve 3D object generation and zero-shot new view synthesis under multi-view conditions.
DreamControl: Control-Based Text-to-3D Generation with 3D Self-Prior
Tianyu Huang (Harbin Institute of Technology), Wangmeng Zuo (City University of Hong Kong)
GenerationData SynthesisDiffusion modelScore-based ModelNeural Radiance FieldTextPoint Cloud
🎯 What it does: A two-stage text-to-3D generation framework called DreamControl is designed, which first constructs a rough NeRF self-supervised 3D prior using SDS, and then generates fine textures through score distillation controlled by ControlNet, achieving high consistency in geometry and high-quality textures.
DreamMatcher: Appearance Matching Self-Attention for Semantically-Consistent Text-to-Image Personalization
Jisu Nam (Korea University), Seunggyu Chang (NAVER Cloud)
Image TranslationGenerationTransformerDiffusion modelImage
🎯 What it does: A DreamMatcher plugin is proposed that does not require additional fine-tuning, utilizing semantic matching to inject the appearance information of reference images into the self-attention module, enhancing the personalized appearance of the subject from text to image while maintaining structural consistency.
DreamPropeller: Supercharge Text-to-3D Generation with Parallel Sampling
Linqi Zhou (Stanford University), Stefano Ermon (Stanford University)
GenerationData SynthesisComputational EfficiencyScore-based ModelGaussian SplattingTextPoint Cloud
🎯 What it does: A parallel sampling acceleration framework called DreamPropeller is proposed, which can be directly wrapped in any score distillation-based text-to-3D generation pipeline.
DreamSalon: A Staged Diffusion Framework for Preserving Identity-Context in Editable Face Generation
Haonan Lin (Xi'an Jiaotong University)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: This paper proposes DreamSalon, a noise-guided, staged editing framework for fine-grained text-driven facial editing while preserving facial identity and context.
DreamVideo: Composing Your Dream Videos with Customized Subject and Motion
Yujie Wei, Hongming Shan
GenerationData SynthesisDiffusion modelVideo
🎯 What it does: Proposes DreamVideo, which can customize videos of subjects and actions from a small number of images and videos.
DRESS: Instructing Large Vision-Language Models to Align and Interact with Humans via Natural Language Feedback
Yangyi Chen (SRI International), Ajay Divakaran (SRI International)
Reinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes DRESS, a framework that aligns and enhances the multi-turn interaction capabilities of large-scale visual language models through natural language feedback generated by large language models, categorized into critical and improvement feedback.
DriveTrack: A Benchmark for Long-Range Point Tracking in Real-World Videos
Arjun Balasingam (Massachusetts Institute of Technology), Hari Balakrishnan (Massachusetts Institute of Technology)
Object TrackingAutonomous DrivingVideoPoint CloudBenchmark
🎯 What it does: We propose DriveTrack, a large-scale long-distance keypoint tracking benchmark based on real driving videos, which automatically generates approximately 1B point trajectories.
DriveWorld: 4D Pre-trained Scene Understanding via World Models for Autonomous Driving
Chen Min (Peking University), Bin Dai (Unmanned Systems Technology Research Center, Defense Innovation Institute)
Object DetectionObject TrackingAutonomous DrivingRecurrent Neural NetworkPrompt EngineeringWorld ModelVideoPoint Cloud
🎯 What it does: This paper studies a world model-based 4D pre-training framework called DriveWorld, which learns a unified spatiotemporal BEV representation from multi-camera videos.
Driving Everywhere with Large Language Model Policy Adaptation
Boyi Li (NVIDIA), Marco Pavone (Stanford University)
Autonomous DrivingTransformerLarge Language ModelVideo
🎯 What it does: LLaDA is proposed, a driving strategy adaptation framework based on large language models, which utilizes traffic rule manuals and LLMs for zero-shot real-time adjustments of driving plans, enabling vehicles to comply with local traffic regulations in different regions.
Driving into the Future: Multiview Visual Forecasting and Planning with World Model for Autonomous Driving
Yuqi Wang, Zhaoxiang Zhang
GenerationAutonomous DrivingDiffusion modelWorld ModelVideo
🎯 What it does: We propose Drive-WM, a world model capable of generating high-quality, controllable consistent videos from multiple perspectives, and apply it to end-to-end autonomous driving planning.
Driving-Video Dehazing with Non-Aligned Regularization for Safety Assistance
Junkai Fan (Nanjing University of Science and Technology), Jian Yang (Nanjing University of Science and Technology)
RestorationAutonomous DrivingSafty and PrivacyOptical FlowVideo
🎯 What it does: A framework based on non-aligned reference frame matching and video dehazing is proposed, capable of performing dehazing on real driving videos.
DrivingGaussian: Composite Gaussian Splatting for Surrounding Dynamic Autonomous Driving Scenes
Xiaoyu Zhou (Peking University), Ming-Hsuan Yang (University of California)
Autonomous DrivingGaussian SplattingImagePoint Cloud
🎯 What it does: A framework based on Composite Gaussian Splatting, called DrivingGaussian, is proposed for efficiently and accurately reconstructing large-scale, dynamic autonomous driving scenes, supporting panoramic view synthesis and dynamic object editing.
DS-NeRV: Implicit Neural Video Representation with Decomposed Static and Dynamic Codes
Hao Yan (Tianjin University), Dadong Jiang (Tianjin University)
RestorationCompressionNeural Radiance FieldVideo
🎯 What it does: This paper proposes DS-NeRV, an implicit neural video representation that decomposes videos into sparse, learnable static and dynamic codes. It decouples static and dynamic information using different sampling rates and weighted/interpolation methods, and achieves video reconstruction through cross-channel attention fusion.
DSGG: Dense Relation Transformer for an End-to-end Scene Graph Generation
Zeeshan Hayder (Data61 CSIRO), Xuming He (ShanghaiTech University)
Object DetectionSegmentationGenerationKnowledge DistillationTransformerGraph
🎯 What it does: An end-to-end dense relationship Transformer network called DSGG is designed, which utilizes graph-aware queries to predict a complete scene graph in one go, including object categories, bounding boxes, pixel segmentation, and all relationships between objects.
DSL-FIQA: Assessing Facial Image Quality via Dual-Set Degradation Learning and Landmark-Guided Transformer
Wei-Ting Chen (National Taiwan University), Jian Wang (Snap Inc.)
TransformerContrastive LearningImage
🎯 What it does: This paper proposes a Transformer-based solution for General Face Image Quality Assessment (GFIQA) called DSL-FIQA, and constructs a larger and more balanced CGFIQA-40k dataset.
Dual DETRs for Multi-Label Temporal Action Detection
Yuhan Zhu (Nanjing University), Limin Wang (Nanjing University)
RecognitionObject DetectionTransformerVideo
🎯 What it does: A dual-layer query framework called DualDETR is proposed for precise temporal action detection in multi-label videos.
Dual Memory Networks: A Versatile Adaptation Approach for Vision-Language Models
Yabin Zhang (Hong Kong Polytechnic University), Lei Zhang (Hong Kong Polytechnic University)
ClassificationDomain AdaptationTransformerVision Language ModelContrastive LearningImage
🎯 What it does: Proposes Dual Memory Networks (DMN), which achieve zero-shot, few-shot, and training-free few-shot adaptation of pre-trained visual language models through dynamic and static memory networks;
Dual Pose-invariant Embeddings: Learning Category and Object-specific Discriminative Representations for Recognition and Retrieval
Rohan Sarkar (Purdue University), Avinash Kak (Purdue University)
RecognitionRetrievalConvolutional Neural NetworkContrastive LearningPoint Cloud
🎯 What it does: A dual-space network is proposed to simultaneously learn category embeddings and object identity embeddings for pose-invariant object recognition and retrieval.
Dual Prior Unfolding for Snapshot Compressive Imaging
Jiancheng Zhang (Northwestern Polytechnical University), Yin-Ping Zhao (Northwestern Polytechnical University)
RestorationCompressionTransformerImage
🎯 What it does: Proposes the Dual Prior Unfolding (DPU) model for efficient reconstruction of three-dimensional hyperspectral images from spectral snapshot compressed imaging (SCI).
Dual Prototype Attention for Unsupervised Video Object Segmentation
Suhwan Cho (Yonsei University), Sangyoun Lee (Yonsei University)
SegmentationConvolutional Neural NetworkOptical FlowVideoMultimodality
🎯 What it does: This paper proposes a Dual Prototype Attention (DPA) framework for unsupervised video object segmentation, primarily achieved through two modules: Inter-Modality Attention (IMA) and Inter-Frame Attention (IFA), which facilitate the dense fusion of RGB and optical flow information and the effective propagation of global video context.
Dual-Consistency Model Inversion for Non-Exemplar Class Incremental Learning
Zihuan Qiu (University of Electronic Science and Technology of China), Qingbo Wu (University of Electronic Science and Technology of China)
Domain AdaptationKnowledge DistillationGenerative Adversarial NetworkImage
🎯 What it does: Old class samples are generated through the Dual-Consistency Model Inversion, and these samples are used for knowledge distillation in class-incremental learning without samples, enhancing the model's memory of old classes.
Dual-Enhanced Coreset Selection with Class-wise Collaboration for Online Blurry Class Incremental Learning
Yutian Luo (Renmin University of China), Zhiwu Lu (Renmin University of China)
ClassificationData-Centric LearningImage
🎯 What it does: Proposes the DECO method to address data imbalance and category variation issues in online fuzzy category incremental learning, constructing real-time category balanced memory and dual-enhanced core subset selection.
Dual-Scale Transformer for Large-Scale Single-Pixel Imaging
Gang Qu (Westlake University), Xin Yuan (Westlake University)
RestorationTransformerImage
🎯 What it does: This paper proposes a deep unfolding network HATNet for large-scale single-pixel imaging (SPI) reconstruction, utilizing the Kronecker SPI model to achieve direct sampling and reconstruction of the entire image.
Dual-View Visual Contextualization for Web Navigation
Jihyung Kil (Ohio State University), Wei-Lun Chao (Ohio State University)
TransformerLarge Language ModelVision-Language-Action ModelImageTextBenchmark
🎯 What it does: Design a Dual-View Contextualized Representation (DUAL-VCR) that adds visual neighborhood context to each HTML element in webpage screenshots to enhance its representation;
DualAD: Disentangling the Dynamic and Static World for End-to-End Driving
Simon Doll (Mercedes-Benz AG), Hendrik P. A. Lensch
Object DetectionObject TrackingSegmentationAutonomous DrivingTransformerMultimodality
🎯 What it does: This paper proposes a dual-stream architecture that models dynamic entities and static scenes separately, achieving end-to-end perception, prediction, and planning for vehicles and the environment.
DUDF: Differentiable Unsigned Distance Fields with Hyperbolic Scaling
Miguel Fainstein (Universidad Torcuato Di Tella), Emmanuel Iarussi (CONICET)
GenerationOptimizationPoint CloudMesh
🎯 What it does: A differentiable unsigned distance field (DUDF) is proposed, which reconstructs open surfaces through hyperbolic scaling, supports direct training using differentiable neural networks, and achieves high-quality, rapid 3D reconstruction.
DuPL: Dual Student with Trustworthy Progressive Learning for Robust Weakly Supervised Semantic Segmentation
Yuanchen Wu (Shanghai University), Xiaoqiang Li (Shanghai University)
SegmentationKnowledge DistillationImage
🎯 What it does: Proposes a dual-student framework and trustworthy evolutionary learning (DuPL) to achieve weakly supervised semantic segmentation.
DUSt3R: Geometric 3D Vision Made Easy
Shuzhe Wang (Aalto University), Jerome Revaud (Naver Labs Europe)
Pose EstimationDepth EstimationTransformerImage
🎯 What it does: DUSt3R is proposed, an end-to-end model that can directly generate dense 3D reconstruction and camera parameters from uncalibrated, unposed multi-view images.
DVMNet: Computing Relative Pose for Unseen Objects Beyond Hypotheses
Chen Zhao (École Polytechnique Fédérale de Lausanne), Mathieu Salzmann (École Polytechnique Fédérale de Lausanne)
Pose EstimationTransformerImage
🎯 What it does: A network called DVMNet based on deep voxel matching is proposed for relative pose estimation between single reference images.
DyBluRF: Dynamic Neural Radiance Fields from Blurry Monocular Video
Huiqiang Sun (Huazhong University of Science and Technology), Zhiguo Cao (Huazhong University of Science and Technology)
RestorationGenerationNeural Radiance FieldOptical FlowVideo
🎯 What it does: A dynamic neural radiance field model named DyBluRF is proposed, which can directly generate clear new perspective images from monocular motion blur videos.
Dynamic Adapter Meets Prompt Tuning: Parameter-Efficient Transfer Learning for Point Cloud Analysis
Xin Zhou (Huazhong University of Science and Technology), Xiang Bai (Huazhong University of Science and Technology)
ClassificationObject DetectionPrompt EngineeringPoint Cloud
🎯 What it does: Proposes a method for dynamic adapters and internal prompt tuning to achieve efficient transfer learning of parameters for pre-trained point cloud models.
Dynamic Cues-Assisted Transformer for Robust Point Cloud Registration
Hong Chen (Huazhong University of Science and Technology), Yihua Tan (Huazhong University of Science and Technology)
Pose EstimationAutonomous DrivingTransformerPoint Cloud
🎯 What it does: This paper proposes a Dynamic Cue-Assisted Transformer (DCATr) for point cloud registration, which utilizes consistency cues to constrain attention, reducing interference and improving matching accuracy.
Dynamic Graph Representation with Knowledge-aware Attention for Histopathology Whole Slide Image Analysis
Jiawen Li (Shenzhen International Graduate School, Tsinghua University), Yonghong He (Shenzhen International Graduate School, Tsinghua University)
ClassificationGraph Neural NetworkTransformerImageBiomedical Data
🎯 What it does: A full-slide image analysis framework WiKG based on dynamic graphs and knowledge-aware attention is proposed.
Dynamic Inertial Poser (DynaIP): Part-Based Motion Dynamics Learning for Enhanced Human Pose Estimation with Sparse Inertial Sensors
Yu Zhang (Shanghai Jiao Tong University), Ling Pei (Shanghai Jiao Tong University)
Pose EstimationRecurrent Neural NetworkTime Series
🎯 What it does: This paper proposes DynaIP, a real-time full-body pose estimation framework based on six sparse IMUs, which achieves pseudo-velocity regression and pose regression through a two-stage network, and employs part-based learning and global feature fusion.
Dynamic LiDAR Re-simulation using Compositional Neural Fields
Hanfeng Wu (ETH Zurich), Shengyu Huang (ETH Zurich)
Object DetectionSegmentationData SynthesisAutonomous DrivingNeural Radiance FieldPoint Cloud
🎯 What it does: In dynamic driving scenarios, high-fidelity LiDAR re-simulation is achieved using editable neural fields (DyNFL), supporting object insertion, deletion, and trajectory editing.
Dynamic Policy-Driven Adaptive Multi-Instance Learning for Whole Slide Image Classification
Tingting Zheng (Harbin Institute of Technology), Hongxun Yao (Harbin Institute of Technology)
ClassificationTransformerReinforcement LearningContrastive LearningImageBiomedical Data
🎯 What it does: A dynamic strategy-driven adaptive multi-instance learning framework (PAMIL) is proposed, which optimizes the classification of whole slide images (WSI) through historical instance information.
Dynamic Prompt Optimizing for Text-to-Image Generation
Wenyi Mo (Renmin University of China), Qing Yang (Du Xiaoman Technology)
GenerationTransformerLarge Language ModelReinforcement LearningPrompt EngineeringImageText
🎯 What it does: This paper proposes an automated prompt editing method (Prompt Auto-Editing, PAE), which generates dynamically fine-grained control prompts (DF-Prompt) with weights and temporal ranges through pre-trained language models and reinforcement learning, aiming to enhance the visual aesthetics and semantic consistency of text-to-image generation.
Dynamic Support Information Mining for Category-Agnostic Pose Estimation
Pengfei Ren (Beijing University of Posts and Telecommunications), Jianxin Liao (Beijing University of Posts and Telecommunications)
Pose EstimationGraph Neural NetworkTransformerImage
🎯 What it does: This paper proposes a Support Information Dynamic Perception Network (SDPNet), which achieves more accurate category-independent pose estimation by fully exploiting the semantic and structural information of support samples in the similarity matching and iterative optimization of query samples.
DynVideo-E: Harnessing Dynamic NeRF for Large-Scale Motion- and View-Change Human-Centric Video Editing
Jia-Wei Liu (Show Lab), Mike Zheng Shou (National University of Singapore)
GenerationData SynthesisScore-based ModelNeural Radiance FieldVideo
🎯 What it does: This paper proposes DynVideo-E, which achieves large-scale human video editing with dynamic NeRF representation to handle motion and viewpoint changes.
Dysen-VDM: Empowering Dynamics-aware Text-to-Video Diffusion with LLMs
Hao Fei (National University of Singapore), Tat-Seng Chua (National University of Singapore)
GenerationData SynthesisTransformerLarge Language ModelDiffusion modelVideoText
🎯 What it does: A dynamic scene manager (Dysen) is proposed in combination with a dynamic scene graph (DSG) to enhance the temporal dynamic modeling of video diffusion models, achieving higher quality text-to-video generation.
DYSON: Dynamic Feature Space Self-Organization for Online Task-Free Class Incremental Learning
Yuhang He (Xi'an Jiaotong University), Yihong Gong (Xi'an Jiaotong University)
ClassificationRepresentation LearningImageBenchmark
🎯 What it does: The DYSON framework is proposed for online task-free boundary category incremental learning, which first calculates the optimal geometric structure and aligns the feature space to achieve buffer-free incremental learning.
E-GPS: Explainable Geometry Problem Solving via Top-Down Solver and Bottom-Up Generator
Wenjun Wu (Xi'an Jiaotong University), Qianying Wang (Lenovo Research)
GenerationData SynthesisOptimizationExplainability and InterpretabilityTransformerTextBenchmark
🎯 What it does: This paper proposes the E-GPS method, which first translates geometric figures and problem statements into a unified formal language, then uses a top-down interpretable solver to generate answers and complete reasoning steps, and enhances the model's understanding of theorem knowledge through a bottom-up question generator that augments the dataset.
Each Test Image Deserves A Specific Prompt: Continual Test-Time Adaptation for 2D Medical Image Segmentation
Ziyang Chen (Northwestern Polytechnical University), Yong Xia (Northwestern Polytechnical University)
SegmentationDomain AdaptationPrompt EngineeringImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A medical image segmentation method is proposed that achieves continuous testing adaptation by learning low-frequency visual prompts for each test image without updating model parameters.
EAGLE: Eigen Aggregation Learning for Object-Centric Unsupervised Semantic Segmentation
Chanyoung Kim (Yonsei University), Seong Jae Hwang (Yonsei University)
SegmentationTransformerContrastive LearningImage
🎯 What it does: This paper proposes the EAGLE method, which utilizes the EiCue pointers obtained from spectral clustering and object-level contrastive learning to achieve unsupervised semantic segmentation.
EarthLoc: Astronaut Photography Localization by Indexing Earth from Space
Gabriele Berton (Politecnico di Torino), Carlo Masone (Politecnico di Torino)
RetrievalConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper redefines the task of locating Earth images taken by astronauts in space capsules as a large-scale image retrieval problem, retrieving the most similar satellite images from a database and using their geographic tags for localization.
EASE-DETR: Easing the Competition among Object Queries
Yulu Gao (Beihang University), Si Liu (Beihang University)
Object DetectionTransformerImage
🎯 What it does: EASE-DETR improves detection accuracy by alleviating query competition through the introduction of 'leading/lagging' query relationships in the self-attention and cross-attention layers of the DETR decoder.
EasyDrag: Efficient Point-based Manipulation on Diffusion Models
Xingzhong Hou (Chinese Academy of Sciences), Haihang You (Chinese Academy of Sciences)
Image TranslationGenerationDiffusion modelImage
🎯 What it does: We propose EasyDrag, a point-controlled interactive image editing framework based on a pre-trained diffusion model, capable of accurately dragging key points in images without the need for LoRA fine-tuning or hand-drawn masks.
ECLIPSE: A Resource-Efficient Text-to-Image Prior for Image Generations
Maitreya Patel (Arizona State University), Yezhou Yang (Arizona State University)
GenerationData SynthesisComputational EfficiencyTransformerVision Language ModelContrastive LearningImageText
🎯 What it does: This paper studies and proposes ECLIPSE, a strategy for training text-to-image (T2I) prior models using contrastive learning, significantly reducing model parameters and training data requirements while maintaining or improving generation quality and compositional ability.
Eclipse: Disambiguating Illumination and Materials using Unintended Shadows
Dor Verbin (Google Research), Pratul P. Srinivasan (Google Research)
RestorationOptimizationNeural Radiance FieldImage
🎯 What it does: Utilizing unexpected shadows cast by unobserved moving occluders (such as photographers or camera operators) to jointly recover the spatially varying material of objects, environmental lighting, and the shape of the occluders from a series of images.
ECLIPSE: Efficient Continual Learning in Panoptic Segmentation with Visual Prompt Tuning
Beomyoung Kim (NAVER Cloud), Sung Ju Hwang (KAIST)
SegmentationTransformerPrompt EngineeringImage
🎯 What it does: This paper proposes a panoptic segmentation method for continuous generalization called ECLIPSE, which utilizes visual prompt tuning to fine-tune a frozen base model, achieving continual learning without distillation by only fine-tuning a small number of prompts.
ECoDepth: Effective Conditioning of Diffusion Models for Monocular Depth Estimation
Suraj Patni (Indian Institute of Technology Delhi), Chetan Arora (Indian Institute of Technology Delhi)
Depth EstimationTransformerDiffusion modelImage
🎯 What it does: This paper studies a monocular depth estimation method ECoDepth based on a conditional diffusion model and using ViT features.