CVPR 2025 Papers — Page 6
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2871 papers
Data-Free Group-Wise Fully Quantized Winograd Convolution via Learnable Scales
Shuokai Pan (Arm Inc.), Dibakar Gope (Arm Inc.)
GenerationData SynthesisConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: The paper studies group quantization of Winograd convolution under post-training quantization (PTQ) and achieves full 8-bit quantization on large diffusion models and ResNet.
Data-free Universal Adversarial Perturbation with Pseudo-semantic Prior
Chanhui Lee (Gwangju Institute of Science and Technology), Jeany Son (Gwangju Institute of Science and Technology)
Adversarial AttackConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a universal adversarial perturbation method without data, utilizing the generated UAP itself to extract pseudo-semantic priors to enhance the attack effectiveness.
Dataset Distillation with Neural Characteristic Function: A Minmax Perspective
Shaobo Wang (Shanghai Jiao Tong University), Linfeng Zhang (Shanghai Jiao Tong University)
OptimizationComputational EfficiencyKnowledge DistillationImage
🎯 What it does: A dataset distillation method based on Neural Characteristic Function (NCFM) is proposed, which transforms distribution matching into maximization and minimization of real and synthetic data distributions using minmax adversarial optimization.
DCEvo: Discriminative Cross-Dimensional Evolutionary Learning for Infrared and Visible Image Fusion
Jinyuan Liu (Dalian University of Technology), Xin Fan (Dalian University of Technology)
Image TranslationObject DetectionSegmentationConvolutional Neural NetworkImageMultimodality
🎯 What it does: The DCEvo framework is proposed, achieving the fusion of infrared and visible images while enhancing visual quality and downstream task accuracy.
De^2Gaze: Deformable and Decoupled Representation Learning for 3D Gaze Estimation
Yunfeng Xiao (Tianjin University), Erwei Yin (Academy of Military Sciences)
SegmentationRepresentation LearningTransformerImage
🎯 What it does: A lightweight 3D gaze estimation framework called De 2 Gaze is proposed, which can simultaneously output a 3D eyeball model, gaze vector, and 2D semantic segmentation results.
DEAL: Data-Efficient Adversarial Learning for High-Quality Infrared Imaging
Zhu Liu (Dalian University of Technology), Risheng Liu (Dalian University of Technology)
RestorationObject DetectionDepth EstimationSuper ResolutionSpiking Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a data-efficient adversarial learning framework (DAS) and a dual-interaction network for high-quality enhancement of infrared images under various degradation conditions.
Debiasing Multimodal Large Language Models via Noise-Aware Preference Optimization
Zefeng Zhang (Institute of Information Engineering, Chinese Academy of Sciences), Tingwen Liu (Institute of Information Engineering, Chinese Academy of Sciences)
OptimizationTransformerLarge Language ModelReinforcement LearningMultimodalityBenchmark
🎯 What it does: This paper proposes a multimodal large language model debiasing method based on preference optimization, first constructing the RLAIF-V-Bias dataset and designing the NaPO algorithm to suppress modal bias.
DeCafNet: Delegate and Conquer for Efficient Temporal Grounding in Long Videos
Zijia Lu (Northeastern University), Mei Chen (Microsoft)
RecognitionComputational EfficiencyTransformerContrastive LearningVideoText
🎯 What it does: This paper studies the task of temporal grounding in long video texts and proposes a 'Delegate-Conquer' strategy: using a low-cost sidekick encoder to first generate dense features for the entire video and create a saliency map, then processing only the salient segments with an expert encoder, and finally obtaining the final time segment through unified fusion and multi-scale temporal refinement.
Decentralized Diffusion Models
David McAllister (University of California), Angjoo Kanazawa (University of California)
GenerationData SynthesisOptimizationKnowledge DistillationMixture of ExpertsDiffusion modelFlow-based ModelImage
🎯 What it does: A decentralized diffusion model framework (DDM) has been constructed, which divides the training data into several subsets, trains expert models independently on separate GPU clusters, and uses a lightweight router to weight and fuse expert outputs during inference to complete global diffusion tasks.
Decision SpikeFormer: Spike-Driven Transformer for Decision Making
Wei Huang (Shanghai AI Laboratory), Nanyang Ye (Shanghai Jiao Tong University)
Spiking Neural NetworkTransformerReinforcement LearningSequential
🎯 What it does: This paper studies a temporal self-attention-based spiking-driven Transformer model for offline reinforcement learning, called DSFormer.
DeCLIP: Decoupled Learning for Open-Vocabulary Dense Perception
Junjie Wang (Harbin Institute of Technology), Zhuotao Tian (Harbin Institute of Technology)
Object DetectionSegmentationKnowledge DistillationRepresentation LearningTransformerContrastive LearningImage
🎯 What it does: The DeCLIP framework is proposed, which decouples the self-attention module of CLIP to separately learn local identification ability and spatial consistency, enhancing the dense prediction performance of open vocabulary.
DeClotH: Decomposable 3D Cloth and Human Body Reconstruction from a Single Image
Hyeongjin Nam (Seoul National University), Kyoung Mu Lee (Seoul National University)
RestorationSegmentationGenerationDiffusion modelImageMesh
🎯 What it does: A framework called DeClotH is proposed for decomposing and reconstructing 3D clothing and human models from a single image.
Decoder Gradient Shield: Provable and High-Fidelity Prevention of Gradient-Based Box-Free Watermark Removal
Haonan An (City University of Hong Kong), Yuguang Fang (City University of Hong Kong)
Image TranslationAdversarial AttackConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper addresses a black-box watermarking method for image-to-image models, revealing the vulnerability of the decoder to gradient attacks, and proposes the Decoder Gradient Shield (DGS) mechanism, which redirects and scales gradients in a black-box API to prevent attackers from training a de-watermarking network.
Decompositional Neural Scene Reconstruction with Generative Diffusion Prior
Junfeng Ni (Tsinghua University), Siyuan Huang (Tsinghua University)
GenerationOptimizationDiffusion modelScore-based ModelImage
🎯 What it does: From a sparse perspective, the scene is decomposed into object-level neural implicit surface reconstruction using the generative prior of a pre-trained diffusion model, thereby obtaining complete geometry and fine textures.
Decouple Distortion from Perception: Region Adaptive Diffusion for Extreme-low Bitrate Perception Image Compression
Jinchang Xu, Xiaodong Xie
CompressionTransformerDiffusion modelImage
🎯 What it does: Research on image compression at extremely low bit rates is conducted, and a MRIDC framework based on vector quantization coding and diffusion decoding is proposed.
Decouple-Then-Merge: Finetune Diffusion Models as Multi-Task Learning
Qianli Ma (Shanghai Jiao Tong University), Linfeng Zhang (Shanghai Jiao Tong University)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: A framework called DeMe is proposed, which splits the diffusion model across different time step ranges, fine-tunes it, and then merges it in the parameter space to reduce gradient conflicts and improve generation quality.
Decoupled Distillation to Erase: A General Unlearning Method for Any Class-centric Tasks
Yu Zhou (Sun Yat-sen University), Wei-Shi Zheng (Sun Yat-sen University)
ClassificationRecognitionSegmentationKnowledge DistillationConvolutional Neural NetworkTransformerImage
🎯 What it does: A method called DELETE is proposed, which can achieve complete forgetting of any class center task without accessing the remaining data and without interfering with the pre-training process.
Decoupled Motion Expression Video Segmentation
Hao Fang (Shandong University), Wei Zhang (Shandong University)
SegmentationTransformerVideo
🎯 What it does: The DMVS framework is proposed, which splits the motion expression video segmentation task into two parts: video instance segmentation and motion expression understanding, and builds a lightweight module on VITA.
DecoupledGaussian: Object-Scene Decoupling for Physics-Based Interaction
Miaowei Wang (University of Edinburgh), Daniel Morris (Michigan State University)
RestorationSegmentationGaussian SplattingVideoPhysics Related
🎯 What it does: This study proposes a system for separating static objects from their contact surfaces in everyday videos and recovering complete 3D geometry and textures, enabling physics-based interactive simulation.
Decoupling Fine Detail and Global Geometry for Compressed Depth Map Super-Resolution
Huan Zheng, Jianbing Shen
RestorationDepth EstimationSuper ResolutionConvolutional Neural NetworkImage
🎯 What it does: A geometric decoupling network named GDNet is proposed for recovering high-quality depth maps from compressed depth images.
Decoupling Training-Free Guided Diffusion by ADMM
Youyuan Zhang (McGill University), Xujie Si (University of Toronto)
RestorationGenerationDiffusion modelImage
🎯 What it does: This paper proposes ADMMDiff, a framework for training unconditional generation using ADMM.
DeDe: Detecting Backdoor Samples for SSL Encoders via Decoders
Sizai Hou (Hong Kong University of Science and Technology), Duanyi Yao (Hong Kong University of Science and Technology)
Anomaly DetectionRepresentation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A decoder-based reverse mapping method called DeDe is proposed for detecting hidden backdoor samples in unsupervised self-supervised learning (SSL) encoders.
Deep Change Monitoring: A Hyperbolic Representative Learning Framework and a Dataset for Long-term Fine-grained Tree Change Detection
Yante Li, Guoying Zhao
Representation LearningConvolutional Neural NetworkImage
🎯 What it does: A UAVTC dataset for long-term fine-grained tree change detection is proposed, and a hypercurvature Siamese network (HSN) is designed to capture the hierarchical structure of tree changes.
Deep Fair Multi-View Clustering with Attention KAN
HaiMing Xu (Xidian University), Quanxue Gao (Beijing University of Technology)
Auto EncoderContrastive LearningTabular
🎯 What it does: This paper proposes a deep fair multi-view clustering framework (DFMVC-AKAN) that combines the attention mechanism with the Kolmogorov-Arnold network (KAN), capable of suppressing bias from sensitive attributes on clustering results while maintaining clustering accuracy.
DeepCompress-ViT: Rethinking Model Compression to Enhance Efficiency of Vision Transformers at the Edge
Sabbir Ahmed (Binghamton University), Adnan Siraj Rakin (Binghamton University)
ClassificationObject DetectionCompressionComputational EfficiencyKnowledge DistillationTransformerAuto EncoderImage
🎯 What it does: Designed and implemented DeepCompress-ViT, which significantly compresses the Vision Transformer using an encoder-decoder structure, and ensures accuracy is not lost through optimized inference-time decoding.
DeepLA-Net: Very Deep Local Aggregation Networks for Point Cloud Analysis
Ziyin Zeng (Wuhan University), Bijun Li (Wuhan University)
RecognitionSegmentationTransformerPoint Cloud
🎯 What it does: This paper proposes DeepLA-Net, a very deep local aggregation network for point cloud analysis;
DefectFill: Realistic Defect Generation with Inpainting Diffusion Model for Visual Inspection
Jaewoo Song (Seoul National University), Sungroh Yoon (Seoul National University)
GenerationData SynthesisAnomaly DetectionDiffusion modelImage
🎯 What it does: We propose DefectFill, which generates realistic and detail-rich defect images by fine-tuning a facial inpainting diffusion model with only a small number of defect samples, and uses these images for visual defect detection.
DefMamba: Deformable Visual State Space Model
Leiye Liu (Dalian University of Technology), Huchuan Lu (Dalian University of Technology)
ClassificationObject DetectionSegmentationConvolutional Neural NetworkTransformerImage
🎯 What it does: Proposes the DefMamba visual foundation model, which enhances the efficiency and performance of visual feature extraction by combining deformable scanning and a variable state space structure.
DEFOM-Stereo: Depth Foundation Model Based Stereo Matching
Hualie Jiang (Insta360 Research), Rui Huang (Chinese University of Hong Kong)
Depth EstimationConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper presents DEFOM-Stereo, a binocular matching framework integrated with Depth Anything V2 (a depth foundation model) on top of RAFT-Stereo, utilizing monocular relative depth estimation to initialize disparity and perform scale correction during the iterative update process, thereby achieving more robust depth estimation.
Deformable Radial Kernel Splatting
Yi-Hua Huang (University of Hong Kong), Xiaojuan Qi (University of Hong Kong)
Gaussian SplattingPoint Cloud
🎯 What it does: This paper proposes Deformable Radial Kernel (DRK) splatting, which replaces traditional Gaussian kernels with a single learnable planar kernel to achieve high-quality 3D scene reconstruction and real-time rendering.
DeformCL: Learning Deformable Centerline Representation for Vessel Extraction in 3D Medical Image
Ziwei Zhao (Yizhun Medical AI Co), Liwei Wang (Center for Machine Learning Research, Peking University)
SegmentationTransformerBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A continuous representation based on Deformable Centerline (DeformCL) is proposed for the extraction and segmentation of blood vessels in 3D medical images;
Degradation-Aware Feature Perturbation for All-in-One Image Restoration
Xiangpeng Tian (Sichuan University), Chao Ren (Sichuan University)
RestorationTransformerPrompt EngineeringImage
🎯 What it does: A global unified model DFPIR based on degradation-aware feature perturbation is designed, capable of handling multiple degradation tasks such as image denoising, dehazing, deraining, deblurring, and low-light enhancement in one go.
DEIM: DETR with Improved Matching for Fast Convergence
Shihua Huang (Intellindust AI Lab), Xi Shen
Object DetectionTransformerImage
🎯 What it does: Proposes the DEIM framework, improving the training speed and accuracy of the DETR real-time detector.
DejaVid: Encoder-Agnostic Learned Temporal Matching for Video Classification
Darryl Ho (Massachusetts Institute of Technology), Samuel Madden (Massachusetts Institute of Technology)
ClassificationTransformerVideo
🎯 What it does: An Encoder-agnostic DejaVid method is proposed, which encodes videos into variable-length time series embeddings (TSE) and uses Dynamic Time Warping (DTW) with learnable temporal feature weights to enhance video classification performance.
DELT: A Simple Diversity-driven EarlyLate Training for Dataset Distillation
Zhiqiang Shen (MBZUAI), Shitong Shao (MBZUAI)
Data SynthesisComputational EfficiencyKnowledge DistillationImage
🎯 What it does: Proposes the EarlyLate training strategy, which optimizes synthetic samples in stages to enhance the diversity and efficiency of dataset distillation under batch-to-global matching.
Denoising Functional Maps: Diffusion Models for Shape Correspondence
Aleksei Zhuravlev (University of Bonn), Vladislav Golyanik (MPI for Informatics)
GenerationData SynthesisDiffusion modelMesh
🎯 What it does: Using a denoising diffusion model to directly predict the functional mapping corresponding to shapes, addressing the issue of traditional functional mapping training struggling to achieve broad generalization.
Dense Dispersed Structured Light for Hyperspectral 3D Imaging of Dynamic Scenes
Suhyun Shin (POSTECH), Seung-Hwan Baek (POSTECH)
Depth EstimationOptimizationOptical FlowImageVideo
🎯 What it does: Using a stereo RGB camera and an RGB projector with a diffraction grating, a few-spectral-multiplexed structured light (DDSL) mode was designed to achieve high-precision spectral and depth joint reconstruction of dynamic scenes, with a capture speed of 6.6 fps, a depth error of about 4 mm, and a spectral FWHM of 15.5 nm.
Dense Match Summarization for Faster Two-view Estimation
Jonathan Astermark (Lund University), Viktor Larsson (Lund University)
Pose EstimationComputational EfficiencyImage
🎯 What it does: The research objective is to accelerate pose estimation of two-view cameras based on dense matching.
Dense-SfM: Structure from Motion with Dense Consistent Matching
JongMin Lee (Seoul National University), Sungjoo Yoo (Seoul National University)
Pose EstimationTransformerGaussian SplattingImage
🎯 What it does: A Dense-SfM framework based on dense matching and Gaussian Splatting for extended trajectories is proposed to achieve high-density and accurate 3D reconstruction and camera pose estimation from multi-view images.
DeNVeR: Deformable Neural Vessel Representations for Unsupervised Video Vessel Segmentation
Chun-Hung Wu (National Yang Ming Chiao Tung University), Yu-Lun Liu (National Yang Ming Chiao Tung University)
SegmentationOptical FlowVideo
🎯 What it does: This work proposes an unsupervised X-ray vascular video segmentation method that achieves precise segmentation of coronary vessels using test-time training, layer separation, and optical flow.
Depth Any Camera: Zero-Shot Metric Depth Estimation from Any Camera
Yuliang Guo (Bosch Research North America & Bosch Center for Artificial Intelligence), Liu Ren (Bosch Research North America & Bosch Center for Artificial Intelligence)
Depth EstimationConvolutional Neural NetworkImage
🎯 What it does: Train a depth estimation model that achieves zero-shot measurement accuracy on any camera (perspective, fisheye, 360°), and propose a complete framework that includes unified ERP space transformation, perspective-aware image to ERP conversion, FoV alignment, and multi-resolution training.
Depth-Guided Bundle Sampling for Efficient Generalizable Neural Radiance Field Reconstruction
Li Fang (Communication University of China), Zhan Ma (Nanjing University)
Data SynthesisComputational EfficiencyNeural Radiance FieldPoint Cloud
🎯 What it does: A depth-guided beam sampling strategy is proposed, which groups adjacent rays for joint sampling and adaptively controls the number of sampling points to improve the rendering speed and quality of general NeRF.
DepthCrafter: Generating Consistent Long Depth Sequences for Open-world Videos
Wenbo Hu (Tencent AI Lab), Ying Shan (Tencent AI Lab)
GenerationDepth EstimationDiffusion modelVideo
🎯 What it does: We propose DepthCrafter, which achieves open-world video depth estimation without the need for camera pose or optical flow through a conditional diffusion model, generating long sequences of temporally and spatially consistent depth.
DepthCues: Evaluating Monocular Depth Perception in Large Vision Models
Duolikun Danier (University of Edinburgh), Oisin Mac Aodha (University of Edinburgh)
Depth EstimationSupervised Fine-TuningImageBenchmark
🎯 What it does: A DepthCues benchmark was designed, which includes six types of human monocular depth perception cues (elevation, light and shadow, occlusion, perspective, size, texture gradient) for evaluation tasks, and these cues were used to probe the depth cue learning capabilities of 20 large pre-trained visual models without depth supervision.
DepthSplat: Connecting Gaussian Splatting and Depth
Haofei Xu (ETH Zurich), Marc Pollefeys (Microsoft)
Data SynthesisDepth EstimationTransformerGaussian SplattingPoint Cloud
🎯 What it does: This paper proposes DepthSplat, which combines multi-view depth estimation with 3D Gaussian splatting to achieve efficient reconstruction and view synthesis from sparse viewpoints.
Derivative-Free Diffusion Manifold-Constrained Gradient for Unified XAI
Won Jun Kim (KAIST), Jong Chul Ye (KAIST)
Explainability and InterpretabilityDiffusion modelImageBiomedical Data
🎯 What it does: This paper proposes a FreeMCG method based on diffusion models and EnKF for the unified implementation of black-box feature importance explanation and counterfactual explanation.
DeRS: Towards Extremely Efficient Upcycled Mixture-of-Experts Models
Yongqi Huang (Fudan University), Tao Chen (Fudan University)
OptimizationComputational EfficiencyMixture of ExpertsTextMultimodalityBiomedical Data
🎯 What it does: This paper proposes a Decompose‑Replace‑Synthesis (DeRS) paradigm, which disassembles the upper-cycle MoE experts into shared base weights and expert-specific incremental weights, achieving extremely high parameter efficiency by compressing during inference or efficiently constructing experts during training.
Descriptor-In-Pixel : Point-Feature Tracking For Pixel Processor Arrays
Laurie Bose (University of Manchester), Piotr Dudek (Visionchip Limited)
Object TrackingVideo
🎯 What it does: This paper proposes and implements a complete end-to-end system for point feature detection and tracking on a pixel processing array (PPA), using the Descriptor-In-Pixel scheme, with all computations performed inside the sensor.
Design2GarmentCode: Turning Design Concepts to Tangible Garments Through Program Synthesis
Feng Zhou (Zhejiang Sci-Tech University), Huamin Wang (Style3D Research)
GenerationData SynthesisTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: Using large multimodal models and program synthesis techniques, multimodal design concepts (images, text, sketches, etc.) are transformed into executable parametric sewing pattern programs.
DesignDiffusion: High-Quality Text-to-Design Image Generation with Diffusion Models
Zhendong Wang (University of Science and Technology of China), Houqiang Li (University of Science and Technology of China)
GenerationDiffusion modelImageText
🎯 What it does: An end-to-end one-stage diffusion model called DesignDiffusion is proposed to directly generate design images from text descriptions, balancing visual and textual elements.
DeSiRe-GS: 4D Street Gaussians for Static-Dynamic Decomposition and Surface Reconstruction for Urban Driving Scenes
Chensheng Peng (University of California Berkeley), Wei Zhan (University of California Berkeley)
Autonomous DrivingGaussian SplattingPoint Cloud
🎯 What it does: This paper proposes DeSiRe-GS, a self-supervised static-dynamic decomposition and surface reconstruction framework based on 4D Gaussian splatting.
DeSplat: Decomposed Gaussian Splatting for Distractor-Free Rendering
Yihao Wang (Aalto University), Arno Solin (Aalto University)
RestorationOptimizationGaussian SplattingImage
🎯 What it does: This paper proposes a method called DeSplat, which utilizes 3D Gaussian splatting to decompose scenes, modeling static scenes and view-specific distractors separately, thereby achieving interference-free view synthesis.
Detail-Preserving Latent Diffusion for Stable Shadow Removal
Jiamin Xu (Hangzhou Dianzi University), Gang Xu (Hangzhou Dianzi University)
RestorationDiffusion modelImage
🎯 What it does: Using a pre-trained Stable Diffusion model, a two-stage detail-preserving latent diffusion method is proposed to achieve shadow removal without masking.
Detect Any Mirrors: Boosting Learning Reliability on Large-Scale Unlabeled Data with an Iterative Data Engine
Zhaohu Xing (Hong Kong University of Science and Technology), Lei Zhu (Hong Kong University of Science and Technology)
Object DetectionSegmentationAnomaly DetectionKnowledge DistillationTransformerLarge Language ModelImageVideoMultimodality
🎯 What it does: By constructing a semi-supervised mirror detection framework and collecting approximately 400,000 unlabeled mirror images, the quality of pseudo-labels is improved using an iterative data engine, achieving more robust mirror detection.
Detect-and-Guide: Self-regulation of Diffusion Models for Safe Text-to-Image Generation via Guideline Token Optimization
Feifei Li (Fudan University), Min Yang (Fudan University)
GenerationDiffusion modelImageText
🎯 What it does: This paper proposes the Detect-and-Guide (DAG) framework, which performs self-inspection in text-to-image diffusion models by optimizing prompt words and extracting Class Activation Maps (CAM). It then employs adaptive safety guidance during the sampling process to finely eliminate detected unsafe areas, achieving safe generation of sexual content.
Detecting Adversarial Data Using Perturbation Forgery
Qian Wang (Huazhong University of Science and Technology), Ning Yu (Netflix Eyeline)
Anomaly DetectionAdversarial AttackDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: Proposes the Perturbation Forgery method, which perturbs the noise distribution of known attacks, generates sparse masks, and synthesizes pseudo-adversarial samples to train a binary classifier for detecting unknown adversarial attacks.
Detecting Backdoor Attacks in Federated Learning via Direction Alignment Inspection
Jiahao Xu (University of Nevada), Rui Hu (University of Nevada)
Anomaly DetectionFederated LearningImage
🎯 What it does: A backdoor defense method based on directional alignment detection, AlignIns, is proposed to identify and filter malicious model updates in federated learning.
Detecting Open World Objects via Partial Attribute Assignment
Muli Yang (Institute for Infocomm Research), Hongyuan Zhu (Institute for Infocomm Research)
Object DetectionContrastive LearningImageBenchmark
🎯 What it does: This paper proposes the PASS method, which automatically selects and optimizes a small subset of attributes from a large attribute pool through partial optimal transport and curriculum learning, to achieve simultaneous detection of known and unknown objects in open-world object detection (OWOD).
Detecting Out-of-Distribution Through the Lens of Neural Collapse
Litian Liu (Massachusetts Institute of Technology), Yao Qin (University of California Santa Barbara)
ClassificationAnomaly DetectionConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes a post-hoc OOD detection method based on the theory of Neural Collapse, utilizing the alignment of ID sample features with the last layer weight vectors and the L1 norm of the features to distinguish OOD samples.
Detection-Friendly Nonuniformity Correction: A Union Framework for Infrared UAV Target Detection
Houzhang Fang (Xidian University), Luxin Yan (Huazhong University of Science and Technology)
RestorationObject DetectionTransformerImageBenchmark
🎯 What it does: A joint, detection-friendly end-to-end framework called UniCD is proposed for simultaneously performing non-uniformity correction and target detection of infrared drone images under low-frequency non-uniformity conditions.
Deterministic Certification of Graph Neural Networks against Graph Poisoning Attacks with Arbitrary Perturbations
Jiate Li (Milwaukee School of Engineering), Binghui Wang (Illinois Institute of Technology)
ClassificationAdversarial AttackGraph Neural NetworkGraph
🎯 What it does: Designed and implemented PGNNCert, a deterministic provably robust defense method against poisoning attacks during the training of graph neural networks.
Deterministic Image-to-Image Translation via Denoising Brownian Bridge Models with Dual Approximators
Bohan Xiao (Wayne State University), Ming Dong (Wayne State University)
Image TranslationGenerationScore-based ModelImageStochastic Differential Equation
🎯 What it does: A dual-approximation denoising Brownian bridge model (Dual‑approx Bridge) is proposed, achieving deterministic image-to-image translation.
Deterministic-to-Stochastic Diverse Latent Feature Mapping for Human Motion Synthesis
Yu Hua (Nanyang Technological University), Qiang Zhang (Dalian University of Technology)
GenerationData SynthesisTransformerDiffusion modelAuto EncoderVideoSequentialStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: A new two-stage human action generation framework, DSDFM, is proposed, which learns the latent space through VQ-VAE and then uses Deterministic Ordinary Equation (DerODE) to map from Gaussian distribution to latent distribution, followed by enhancing diversity during sampling using Diverse Stochastic Differential Equation (DivSDE).
Devil is in the Detail: Towards Injecting Fine Details of Image Prompt in Image Generation via Conflict-free Guidance and Stratified Attention
Kyungmin Jo (Korea Advanced Institute of Science and Technology), Jaegul Choo (Korea Advanced Institute of Science and Technology)
Image TranslationGenerationDiffusion modelImage
🎯 What it does: A training-independent image prompting method is proposed, which utilizes conflict-free guidance and hierarchical attention to enable diffusion models to more accurately reflect the details of image prompts.
Devils in Middle Layers of Large Vision-Language Models: Interpreting, Detecting and Mitigating Object Hallucinations via Attention Lens
Zhangqi Jiang (National University of Defense Technology), Xu Yang (Southeast University)
Object DetectionGenerationExplainability and InterpretabilityTransformerVision Language ModelImage
🎯 What it does: This paper conducts an in-depth study of the object hallucination mechanism in large visual-language models (LVLM) from the perspective of attention and proposes a no-training-cost hallucination suppression method based on intermediate layer attention modulation.
DexGrasp Anything: Towards Universal Robotic Dexterous Grasping with Physics Awareness
Yiming Zhong (ShanghaiTech University), Yuexin Ma (ShanghaiTech University)
Robotic IntelligenceLarge Language ModelDiffusion modelPoint Cloud
🎯 What it does: A physical perception hand grasping method based on diffusion models, DexGrasp Anything, is proposed, which can generate diverse and feasible eight-finger grasping postures suitable for different objects.
DexHandDiff: Interaction-aware Diffusion Planning for Adaptive Dexterous Manipulation
Zhixuan Liang (University of Hong Kong), Mingyu Ding (University of California Berkeley)
OptimizationRobotic IntelligenceReinforcement Learning from Human FeedbackLarge Language ModelReinforcement LearningDiffusion modelMultimodality
🎯 What it does: A framework for interactive perception diffusion planning for multi-finger robotic hands, called DexHandDiff, is proposed to achieve adaptive fine manipulation.
dFLMoE: Decentralized Federated Learning via Mixture of Experts for Medical Data Analysis
Luyuan Xie (Peking University), Junsong Yuan (State University of New York at Buffalo)
ClassificationSegmentationSuper ResolutionFederated LearningMixture of ExpertsImageTime SeriesBiomedical Data
🎯 What it does: A decentralized federated learning framework dFLMoE is proposed, where clients directly exchange lightweight head models and implement knowledge fusion using an attention-based Mixture of Experts (MoE) at each client.
DFM: Differentiable Feature Matching for Anomaly Detection
Sheng Wu (Beihang University), Baochang Zhang (Beihang University)
Anomaly DetectionTransformerImage
🎯 What it does: A differentiable feature matching framework is proposed, achieving end-to-end joint optimization of feature extraction and matching modules.
DFormerv2: Geometry Self-Attention for RGBD Semantic Segmentation
Bo-Wen Yin (Nankai University), Qibin Hou (Nankai University)
SegmentationTransformerMultimodality
🎯 What it does: DFormerv2 is proposed, a RGB-D semantic segmentation model that directly guides self-attention using the geometric priors of depth maps, without the need for an additional depth encoder, significantly improving segmentation performance.
DH-Set: Improving Vision-Language Alignment with Diverse and Hybrid Set-Embeddings Learning
Kun Zhang (University of Science and Technology of China), S.Kevin Zhou (University of Science and Technology of China)
RetrievalTransformerVision Language ModelContrastive LearningImageText
🎯 What it does: The DH-Set framework is proposed, which improves visual-text alignment by analyzing key local dimensions within multiple subspaces and merging them into a single mixed embedding during the inference phase.
DI-PCG: Diffusion-based Efficient Inverse Procedural Content Generation for High-quality 3D Asset Creation
Wang Zhao (Tencent), Ying Shan (Tencent)
GenerationData SynthesisComputational EfficiencyTransformerDiffusion modelImage
🎯 What it does: DI-PCG achieves efficient generation of corresponding 3D assets by inferring procedural generator parameters from a single image using a diffusion model.
DiC: Rethinking Conv3x3 Designs in Diffusion Models
Yuchuan Tian (Peking University), Hanting Chen (Huawei)
GenerationData SynthesisConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: A diffusion model DiC based on pure 3x3 convolutions is proposed, achieving image generation under the Encoder-Decoder hourglass architecture.
DiET-GS: Diffusion Prior and Event Stream-Assisted Motion Deblurring 3D Gaussian Splatting
Seungjun Lee (National University of Singapore), Gim Hee Lee (National University of Singapore)
RestorationDiffusion modelGaussian SplattingImage
🎯 What it does: This paper proposes a two-stage 3D Gaussian Splatting framework (DiET-GS) based on event cameras and diffusion models, aimed at recovering clear 3D representations from motion-blurred multi-view images.
Diff-Palm: Realistic Palmprint Generation with Polynomial Creases and Intra-Class Variation Controllable Diffusion Models
Jianlong Jin (Hefei University of Technology), Yunsheng Wu (Tencent Youtu Lab)
RecognitionGenerationData SynthesisDiffusion modelImage
🎯 What it does: A Diff-Palm model based on polynomial line drawing and K-step noise sharing sampling is proposed to generate high-quality, controllable variations of palm print images on a large scale, and to directly train recognition networks without fine-tuning on real data.
Diff2Flow: Training Flow Matching Models via Diffusion Model Alignment
Johannes Schusterbauer (CompVis), Björn Ommer (CompVis)
GenerationDepth EstimationDiffusion modelFlow-based ModelImage
🎯 What it does: This paper proposes the Diff2Flow framework, which efficiently transfers knowledge from pre-trained diffusion models to flow matching models, achieving fast inference and better performance.
DiffCAM: Data-Driven Saliency Maps by Capturing Feature Differences
Xingjian Li (Carnegie Mellon University), Min Xu (Carnegie Mellon University)
Object DetectionExplainability and InterpretabilityConvolutional Neural NetworkImageMagnetic Resonance Imaging
🎯 What it does: Proposes DiffCAM, an explanation method that generates attention heatmaps by capturing the differences between target samples and reference distributions;
DIFFER: Disentangling Identity Features via Semantic Cues for Clothes-Changing Person Re-ID
Xin Liang (Center for Research in Computer Vision University of Central Florida), Yogesh S Rawat (Center for Research in Computer Vision University of Central Florida)
RecognitionRetrievalTransformerPrompt EngineeringVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: This paper proposes the NBDetach module, which uses semantic descriptions generated by visual language models as pseudo-labels to achieve the decoupling of identity (biometric) and non-identity (non-biometric) information in human image features for clothing transformation-based person re-identification.
Difference Inversion: Interpolate and Isolate the Difference with Token Consistency for Image Analogy Generation
Hyunsoo Kim (Korea University), Suhyun Kim (Kyung Hee University)
GenerationData SynthesisVision Language ModelDiffusion modelImage
🎯 What it does: This paper proposes a Difference Inversion method that generates a target image B′ satisfying A:A′ :: B:B′ using image triplets {A, A′, B}, relying solely on image input without the need for text instructions.
Differentiable Inverse Rendering with Interpretable Basis BRDFs
Hoon-Gyu Chung (POSTECH), Seung-Hwan Baek (POSTECH)
RestorationOptimizationExplainability and InterpretabilityImage
🎯 What it does: A differentiable inverse rendering method is proposed, using 2D Gaussian representations for geometry and jointly learning interpretable reference BRDFs, dynamically adjusting the number of references and achieving spatially separated SVBRDF reconstruction through sparse regularization.
DiffFNO: Diffusion Fourier Neural Operator
Xiaoyi Liu (Washington University in St. Louis), Hao Tang (Peking University)
RestorationSuper ResolutionDiffusion modelImageStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: This paper proposes the Diffusion Fourier Neural Operator (DiffFNO), which combines diffusion models, weighted Fourier neural operators, and attention operators to achieve super-resolution of images at arbitrary scales.
DiffLO: Semantic-Aware LiDAR Odometry with Diffusion-Based Refinement
Yongshu Huang (Xiamen University), Cheng Wang (Xiamen University)
Pose EstimationAutonomous DrivingKnowledge DistillationDiffusion modelPoint Cloud
🎯 What it does: A LiDAR pose estimation network called DiffLO is proposed, which integrates semantic awareness and diffusion models to improve the accuracy and robustness of large-scale LiDAR pose estimation.
DiffLocks: Generating 3D Hair from a Single Image using Diffusion Models
Radu Alexandru Rosu (Meshcapade), Michael J. Black (Max Planck Institute for Intelligent Systems)
GenerationData SynthesisTransformerDiffusion modelImage
🎯 What it does: By constructing a synthetic hair dataset of over 40,000 samples and training the DiffLocks model based on Diffusion Transformer, high-detail 3D hair can be generated directly from a single RGB image.
DiffPortrait360: Consistent Portrait Diffusion for 360 View Synthesis
Yuming Gu (University of Southern California), Hao Li (The Chinese University of Hong Kong)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: DiffPortrait360, based on diffusion models, generates head images with consistent 360° views from a single facial photo.
DiffSensei: Bridging Multi-Modal LLMs and Diffusion Models for Customized Manga Generation
Jianzong Wu (Peking University), Yunhai Tong (Peking University)
GenerationData SynthesisLarge Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: This paper proposes a custom comic generation task and constructs the DiffSensei framework, which combines diffusion models with multimodal large language models (MLLM) to achieve dynamic generation and layout control of multi-character comic panels.
Diffusion Bridge: Leveraging Diffusion Model to Reduce the Modality Gap Between Text and Vision for Zero-Shot Image Captioning
Jeong Ryong Lee (Yonsei University), Dosik Hwang (Korea Institute of Science and Technology)
GenerationRetrievalTransformerVision Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: A zero-shot image captioning method (Diffusion Bridge) is proposed that bridges the gap between visual and textual modalities in the CLIP embedding space using diffusion models.
Diffusion Model is Effectively Its Own Teacher
Xinyin Ma (National University of Singapore), Xinchao Wang (National University of Singapore)
GenerationKnowledge DistillationTransformerDiffusion modelImageOrdinary Differential Equation
🎯 What it does: This paper proposes a Self Step-Distillation (SSD/iSSD) framework that utilizes diffusion models to fuse predictions or features at different time steps, further enhancing the generation quality of the model in N-step to N-step distillation.
Diffusion Renderer: Neural Inverse and Forward Rendering with Video Diffusion Models
Ruofan Liang (NVIDIA), Zian Wang (NVIDIA)
GenerationDomain AdaptationDiffusion modelVideo
🎯 What it does: This paper presents DIFFUSIONRENDERER, a unified video diffusion model framework that can both reverse render to estimate the G-buffer and forward render to generate images under realistic lighting.
Diffusion Self-Distillation for Zero-Shot Customized Image Generation
Shengqu Cai (Stanford University), Gordon Wetzstein (Stanford University)
GenerationData SynthesisVision Language ModelDiffusion modelImage
🎯 What it does: Proposes the Diffusion Self-Distillation method, which utilizes a pre-trained text-image diffusion model to self-generate identity-preserving image pairs, achieving zero-shot instant customization generation.
Diffusion-4K: Ultra-High-Resolution Image Synthesis with Latent Diffusion Models
Jinjin Zhang (Beihang University), Di Huang (Beihang University)
GenerationData SynthesisDiffusion modelImageTextBenchmark
🎯 What it does: Proposed the Diffusion-4K framework, achieving direct generation of 4K level text-to-image, and constructed the Aesthetic-4K evaluation benchmark.
Diffusion-based Event Generation for High-Quality Image Deblurring
Xinan Xie (Sun Yat-sen University), Wei-Shi Zheng (Sun Yat-sen University)
RestorationGenerationDiffusion modelImage
🎯 What it does: This paper proposes an event-guided image deblurring framework (EGDeblurring) based on a diffusion model, which can perform deblurring using only blurred images.
Diffusion-based Realistic Listening Head Generation via Hybrid Motion Modeling
Yinuo Wang (Xi'an Jiaotong University), Fei Wang (Xi'an Jiaotong University)
GenerationData SynthesisTransformerDiffusion modelVideoMultimodalityAudio
🎯 What it does: A listener head generation framework based on diffusion models is proposed, achieving high-quality and expressive response head video generation through a mixture of explicit and implicit motion modeling.
DiffusionDrive: Truncated Diffusion Model for End-to-End Autonomous Driving
Bencheng Liao (Huazhong University of Science and Technology), Xinggang Wang (Huazhong University of Science and Technology)
Autonomous DrivingTransformerDiffusion modelMultimodality
🎯 What it does: A Truncated Diffusion Policy is proposed for end-to-end autonomous driving, capable of generating multimodal, drivable trajectories within just two steps of inference.
DiffusionSfM: Predicting Structure and Motion via Ray Origin and Endpoint Diffusion
Qitao Zhao (Carnegie Mellon University), Shubham Tulsiani (Carnegie Mellon University)
Pose EstimationDepth EstimationTransformerDiffusion modelImage
🎯 What it does: An end-to-end multi-view method based on a denoising diffusion model is proposed, which directly predicts pixel-level ray origins and endpoints, thereby obtaining both scene geometry and camera poses simultaneously.
DiffVsgg: Diffusion-Driven Online Video Scene Graph Generation
Mu Chen (Zhejiang University), Yi Yang (Zhejiang University)
Object DetectionGenerationDiffusion modelVideo
🎯 What it does: This paper proposes DIFFVSGG, a method for generating online video scene graphs based on latent diffusion models, which can update object categories, bounding boxes, and relationships in real-time for each frame, supporting continuous temporal reasoning.
DifIISR: A Diffusion Model with Gradient Guidance for Infrared Image Super-Resolution
Xingyuan Li (Dalian University of Technology), Jinyuan Liu (Dalian University of Technology)
RestorationObject DetectionSegmentationSuper ResolutionDiffusion modelImage
🎯 What it does: This paper proposes a diffusion model-based infrared image super-resolution method called DifIISR, which achieves high-quality infrared image reconstruction by injecting visual and perceptual gradients during the reverse diffusion process.
DIFIX3D+: Improving 3D Reconstructions with Single-Step Diffusion Models
Jay Zhangjie Wu (NVIDIA), Huan Ling (NVIDIA)
GenerationData SynthesisAutonomous DrivingDiffusion modelNeural Radiance FieldGaussian SplattingImage
🎯 What it does: This paper proposes a complete pipeline called DIFIX3D+ that utilizes a single-step diffusion model (DIFIX) to improve NeRF and 3D Gaussian Splatting for reconstruction and novel view synthesis.
DiG: Scalable and Efficient Diffusion Models with Gated Linear Attention
Lianghui Zhu (Huazhong University of Science and Technology), Xinggang Wang (Huazhong University of Science and Technology)
GenerationData SynthesisComputational EfficiencyTransformerDiffusion modelImage
🎯 What it does: The Diffusion GLA (DiG) model is proposed, which combines the Gated Linear Attention Transformer (GLA) with diffusion models, providing two efficient image generation backbones: plain and U-shape.
DiGIT: Multi-Dilated Gated Encoder and Central-Adjacent Region Integrated Decoder for Temporal Action Detection Transformer
Ho-Joong Kim (Korea University), Seong-Whan Lee (Korea University)
RecognitionObject DetectionTransformerVideo
🎯 What it does: A time action detection framework based on Transformer, called DiGIT, is proposed, which includes a multi-dilated gated encoder and a center-adjacent region integrated decoder.
Digital Twin Catalog: A Large-Scale Photorealistic 3D Object Digital Twin Dataset
Zhao Dong (Meta Reality Labs Research), Richard Newcombe (Meta Reality Labs Research)
Data SynthesisRobotic IntelligenceNeural Radiance FieldImagePoint CloudMesh
🎯 What it does: A Digital Twin Catalog (DTC) dataset has been proposed and released, containing high-precision 3D digital twin models of 2000 real objects, with millimeter-level geometric accuracy, photorealistic PBR materials, and multi-view evaluation data collected using DSLR and AR glasses.