ICCV 2023 Papers — Page 5
IEEE/CVF International Conference on Computer Vision · 2156 papers
Decoupled Iterative Refinement Framework for Interacting Hands Reconstruction from a Single RGB Image
Pengfei Ren (Beijing University of Posts and Telecommunications), Jianxin Liao (Beijing University of Posts and Telecommunications)
SegmentationPose EstimationGraph Neural NetworkTransformerImage
🎯 What it does: A decoupled iterative refinement framework is proposed to achieve pixel-level 3D reconstruction of interacting hands from a single RGB image.
DEDRIFT: Robust Similarity Search under Content Drift
Dmitry Baranchuk (Yandex Research), I. Zeki Yalniz (Meta AI)
RetrievalAnomaly DetectionVideo
🎯 What it does: This paper proposes an online index adaptive method called DEDRIFT to mitigate the content drift problem in large-scale vector retrieval.
Deep Active Contours for Real-time 6-DoF Object Tracking
Long Wang (SenseTime Research), Xiaowei Zhou (Zhejiang University)
Object TrackingPose EstimationConvolutional Neural NetworkVideo
🎯 What it does: A learning-based active contour model, DeepAC, has been developed for 6-DoF object tracking in real-time RGB video.
Deep Directly-Trained Spiking Neural Networks for Object Detection
Qiaoyi Su (University of Chinese Academy of Sciences), Guoqi Li (University of Chinese Academy of Sciences)
Object DetectionSpiking Neural NetworkImage
🎯 What it does: A deep pulse neural network framework EMS-YOLO based on direct training is proposed, utilizing the full pulse residual network EMS-ResNet to achieve low-energy target detection.
Deep Equilibrium Object Detection
Shuai Wang, Limin Wang
Object Detection
🎯 What it does: The content of this paper cannot be parsed, as it lacks recognizable text.
Deep Feature Deblurring Diffusion for Detecting Out-of-Distribution Objects
Aming Wu (Xidian University), Cheng Deng (Xidian University)
Object DetectionDiffusion modelImage
🎯 What it does: A deep feature diffusion method based on Gaussian blur for forward diffusion and reverse deblurring (DFDD) is proposed for unsupervised discrete distribution object detection.
Deep Fusion Transformer Network with Weighted Vector-Wise Keypoints Voting for Robust 6D Object Pose Estimation
Jun Zhou (Hong Kong Polytechnic University), Jing Qin (Hong Kong Polytechnic University)
Pose EstimationTransformerPoint Cloud
🎯 What it does: This paper proposes the Deep Fusion Transformer (DFTr) network, which fuses RGB and depth features through a cross-modal Transformer and designs a weighted vector-wise voting algorithm to achieve non-iterative localization of 3D keypoints, thereby enabling robust 6D object pose estimation.
Deep Geometrized Cartoon Line Inbetweening
Li Siyao (Nanyang Technological University), Chen Change Loy (Nanyang Technological University)
GenerationData SynthesisGraph Neural NetworkTransformerOptical FlowImageVideo
🎯 What it does: This paper proposes a method for generating intermediate frames of cartoon lines based on graph structures, called AnimeInbet, which geometrizes sparse black-and-white lines into a vertex graph and achieves intermediate frame synthesis through vertex correspondence and repositioning.
Deep Geometry-Aware Camera Self-Calibration from Video
Annika Hagemann (Bosch Research), Christoph Stiller (Karlsruhe Institute of Technology)
Pose EstimationDepth EstimationOptimizationConvolutional Neural NetworkSimultaneous Localization and MappingVideo
🎯 What it does: A deep learning-based camera self-calibration method (DroidCalib) is proposed, which incorporates a differentiable self-calibration Bundle Adjustment (SC-BA) layer into a deep visual SLAM system, enabling the simultaneous estimation of camera intrinsic parameters, pose, and depth from uncalibrated monocular video during application.
Deep Homography Mixture for Single Image Rolling Shutter Correction
Weilong Yan (University of Electronic Science and Technology of China), Shuaicheng Liu (University of Electronic Science and Technology of China)
RestorationConvolutional Neural NetworkOptical FlowImageVideo
🎯 What it does: This paper proposes a single-frame rolling shutter correction method based on a Deep Homography Mixture model, which can recover a global shutter image from just a single rolling shutter image.
Deep Image Harmonization with Globally Guided Feature Transformation and Relation Distillation
Li Niu (Shanghai Jiao Tong University), Liqing Zhang (Shanghai Jiao Tong University)
Image HarmonizationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: A global guidance-based feature transformation (GIFT) and relational distillation image harmonization network called GiftNet is proposed, along with the contribution of a new ccHarmony dataset.
Deep Image Harmonization with Learnable Augmentation
Li Niu (Shanghai Jiao Tong University), Liqing Zhang (Shanghai Jiao Tong University)
Image HarmonizationData SynthesisConvolutional Neural NetworkAuto EncoderImage
🎯 What it does: A learnable enhancement network SycoNet is proposed for automatically generating diverse synthetic images and dynamically augmenting training data, thereby improving the effect of image harmonization.
Deep Incubation: Training Large Models by Divide-and-Conquering
Zanlin Ni (Tsinghua University), Gao Huang (Tsinghua University)
Object DetectionSegmentationTransformerSupervised Fine-TuningImage
🎯 What it does: A deep learning framework based on modular training and lightweight shared meta-model incubation is proposed, achieving efficient and parallel training of large models.
Deep Multitask Learning with Progressive Parameter Sharing
Haosen Shi (Chinese University of Hong Kong), Sinno Jialin Pan (Chinese University of Hong Kong)
ClassificationSegmentationDepth EstimationConvolutional Neural NetworkReinforcement LearningImage
🎯 What it does: A progressive parameter sharing strategy (MPPS) is proposed, which gradually increases the sharing degree of the multi-task model by parameterizing, dynamically scheduling, and regularizing the sharing probability of each neuron, thereby enhancing the performance of multi-task learning.
Deep Multiview Clustering by Contrasting Cluster Assignments
Jie Chen (Sichuan University), Xi Peng (Sichuan University)
Representation LearningAuto EncoderContrastive LearningImage
🎯 What it does: A cross-view contrastive learning (CVCL) framework is proposed, which first learns view-related features using deep autoencoders, and then aligns the soft clustering assignments of different views through cluster-level contrastive loss, thereby achieving multi-view clustering.
Deep Optics for Video Snapshot Compressive Imaging
Ping Wang (Zhejiang University), Xin Yuan (Westlake University)
CompressionOptimizationTransformerVideo
🎯 What it does: A deep optical framework is proposed, which jointly optimizes the mask and reconstruction network in video snapshot compression imaging (SCI), and implements a structured mask and Res2former reconstruction network.
Deep Video Demoireing via Compact Invertible Dyadic Decomposition
Yuhui Quan, Ruotao Xu
RestorationConvolutional Neural NetworkVideo
🎯 What it does: This paper proposes a new method to address a specific problem in computer vision, with specific details not provided.
DeepChange: A Long-Term Person Re-Identification Benchmark with Clothes Change
Peng Xu (Tsinghua University), Xiatian Zhu (University of Surrey)
RecognitionRetrievalConvolutional Neural NetworkTransformerImageVideoMultimodalityBenchmark
🎯 What it does: This paper constructs a large-scale real long-term pedestrian re-identification dataset called DeepChange and conducts benchmark experiments on this dataset.
DeePoint: Visual Pointing Recognition and Direction Estimation
Shu Nakamura (Kyoto University), Ko Nishino (RIKEN)
RecognitionPose EstimationTransformerVideo
🎯 What it does: A model named DeePoint is proposed, which can automatically recognize pointing behavior in fixed-view RGB videos and estimate its 3D direction, and a large-scale pointing dataset called DP Dataset is constructed.
Deformable Model-Driven Neural Rendering for High-Fidelity 3D Reconstruction of Human Heads Under Low-View Settings
Baixin Xu (Nanyang Technological University), Ying He (Nanyang Technological University)
RestorationGenerationNeural Radiance FieldImage
🎯 What it does: Under low-view angle conditions, high-precision 3D reconstruction of human heads is achieved using neural implicit functions through geometric decomposition and two-stage coarse-to-fine training.
Deformable Neural Radiance Fields using RGB and Event Cameras
Qi Ma (ETH Zurich), Luc Van Gool (KU Leuven)
GenerationData SynthesisPose EstimationDepth EstimationNeural Radiance FieldMultimodality
🎯 What it does: This paper proposes a neural radiance field model that can learn deformable scenes under sparse RGB frames and continuous event streams by combining synchronized data from RGB cameras and event cameras.
Deformer: Dynamic Fusion Transformer for Robust Hand Pose Estimation
Qichen Fu (Carnegie Mellon University), Kris M. Kitani (Salesforce Research)
Pose EstimationTransformerGenerative Adversarial NetworkVideo
🎯 What it does: A Deformer framework is proposed for 3D hand pose estimation in videos, combining spatial and temporal Transformers with a dynamic fusion module, enabling robust estimation even when the hand is occluded or in motion blur.
DeFormer: Integrating Transformers with Deformable Models for 3D Shape Abstraction from a Single Image
Di Liu (Rutgers University), Dimitris N. Metaxas (Rutgers University)
GenerationOptimizationTransformerImagePoint Cloud
🎯 What it does: A framework based on a dual-channel Transformer and a deformable model (Deformer) is proposed, utilizing a small number of superquadrics and local differentiable deformations to reconstruct 3D shapes from a single 2D image while maintaining semantic correspondence.
DeformToon3D: Deformable Neural Radiance Fields for 3D Toonification
Junzhe Zhang (Nanyang Technological University), Chen Change Loy (Nanyang Technological University)
GenerationData SynthesisNeural Radiance FieldGenerative Adversarial NetworkImage
🎯 What it does: DEFORMTOON3D is proposed, a 3D cartoonization framework that does not require fine-tuning and separates geometry from texture, supporting multi-style 3D cartoonization of a single model;
Degradation-Resistant Unfolding Network for Heterogeneous Image Fusion
Chunming He (Shenzhen International Graduate School Tsinghua University), Xiu Li (Shenzhen International Graduate School Tsinghua University)
Image TranslationRestorationObject DetectionObject TrackingImageMultimodalityMagnetic Resonance Imaging
🎯 What it does: A degradation robust unfolding network (DeRUN) for heterogeneous image fusion is proposed, capable of generating high-quality fused images in degraded scenarios such as low light and heavy fog.
DELFlow: Dense Efficient Learning of Scene Flow for Large-Scale Point Clouds
Chensheng Peng (Shanghai Jiao Tong University), Hesheng Wang (Shanghai Jiao Tong University)
Depth EstimationAutonomous DrivingComputational EfficiencyOptical FlowPoint Cloud
🎯 What it does: For large-scale point cloud 3D scene flow estimation, a DELFlow framework is proposed, which can process complete point clouds at once and predict point-level motion.
Delicate Textured Mesh Recovery from NeRF via Adaptive Surface Refinement
Jiaxiang Tang (Peking University), Gang Zeng (Peking University)
RestorationGenerationNeural Radiance FieldImageMesh
🎯 What it does: Recover high-quality face meshes with diffuse and specular lighting textures from multi-view RGB images, supporting object-level and scene-level datasets;
DeLiRa: Self-Supervised Depth, Light, and Radiance Fields
Vitor Guizilini (Toyota Research Institute), Adrien Gaidon (Toyota Research Institute)
GenerationDepth EstimationTransformerNeural Radiance FieldAuto EncoderImage
🎯 What it does: In the self-supervised learning framework of multiple images with a limited view, we propose using multi-view photometric consistency as a regularization for volume rendering, and design a Transformer-based auto-decoder architecture that jointly learns depth fields, lighting fields, and radiance fields from a shared latent space;
Delta Denoising Score
Amir Hertz (Google Research), Daniel Cohen-Or (Tel Aviv University)
Image TranslationGenerationDiffusion modelImage
🎯 What it does: Proposes Delta Denoising Score (DDS), a scoring mechanism based on diffusion models for image editing and image-to-image translation;
Delving into Motion-Aware Matching for Monocular 3D Object Tracking
Kuan-Chih Huang (University of California, Merced), Yi-Hsuan Tsai (Yonsei University)
Object TrackingAutonomous DrivingTransformerContrastive LearningPoint Cloud
🎯 What it does: This paper proposes a monocular 3D multi-object tracking framework called MoMA-M3T based on motion-aware matching, which utilizes motion features and a motion Transformer to associate trajectories with detection boxes.
Democratising 2D Sketch to 3D Shape Retrieval Through Pivoting
Pinaki Nath Chowdhury (University of Surrey), Yi-Zhe Song (University of Surrey)
RetrievalConvolutional Neural NetworkContrastive LearningImagePoint Cloud
🎯 What it does: This paper proposes a zero-shot method for retrieving 3D shapes from 2D sketches without the need for training samples that correspond sketches to 3D shapes, allowing ordinary users to freely draw and search from any perspective.
Denoising Diffusion Autoencoders are Unified Self-supervised Learners
Weilai Xiang (Beihang University), Yunhong Wang (Beihang University)
ClassificationRepresentation LearningDiffusion modelAuto EncoderImage
🎯 What it does: This study investigates the representation learning capability of the diffusion model (DDAE) under unsupervised pre-training and verifies its direct applicability to image classification tasks.
Dense 2D-3D Indoor Prediction with Sound via Aligned Cross-Modal Distillation
Heeseung Yun (Seoul National University), Gunhee Kim (Seoul National University)
SegmentationDepth EstimationKnowledge DistillationConvolutional Neural NetworkContrastive LearningMultimodalityAudio
🎯 What it does: Utilizing audio input, dense 2D and 3D predictions of indoor environments are achieved through cross-modal distillation with a visual teacher model, including depth, semantic segmentation, and 3D reconstruction.
Dense Text-to-Image Generation with Attention Modulation
Yunji Kim (NAVER AI Lab), Jun-Yan Zhu (Carnegie Mellon University)
Object DetectionGenerationData SynthesisTransformerDiffusion modelImageText
🎯 What it does: A training-free attention modulation method is proposed, enabling pre-trained text-to-image diffusion models to accurately handle dense descriptions and achieve image layout control.
DenseShift: Towards Accurate and Efficient Low-Bit Power-of-Two Quantization
Xinlin Li (Huawei Technologies), Vahid Partovi Nia (Huawei Technologies)
ClassificationObject DetectionSegmentationImageAudio
🎯 What it does: Proposes the DenseShift network, which removes zero weights, employs signature-scale splitting, and low variance initialization, making low-bit Shift networks more efficient in both inference and training.
Density-invariant Features for Distant Point Cloud Registration
Quan Liu (Shanghai Jiao Tong University), Minyi Guo (Shanghai Jiao Tong University)
Autonomous DrivingRepresentation LearningContrastive LearningPoint Cloud
🎯 What it does: This paper proposes a density-invariant feature extraction method based on Group-wise Contrastive Learning (GCL) to address the density mismatch problem in long-range LiDAR point cloud registration.
Designing Phase Masks for Under-Display Cameras
Anqi Yang (Samsung Advanced Institute of Technology), Aswin C. Sankaranarayanan (Carnegie Mellon University)
RestorationOptimizationOptical FlowImage
🎯 What it does: Two micro-lens arrays (phase masks) are inserted under a Transparent OLED (TOLED) display to enhance light transmission and suppress diffraction blur, while maintaining display quality through polarization dependence.
DETA: Denoised Task Adaptation for Few-Shot Learning
Ji Zhang (University of Electronic Science and Technology of China), Jingkuan Song (Shenzhen Institute for Advanced Study, University of Electronic Science and Technology of China)
ClassificationMeta LearningContrastive LearningImage
🎯 What it does: In few-shot learning, a unified image and label denoising framework DETA is proposed for task adaptation during testing.
Detecting Objects with Context-Likelihood Graphs and Graph Refinement
Aritra Bhowmik (University of Amsterdam), Cees G. M. Snoek (University of Amsterdam)
Object DetectionGraph Neural NetworkImage
🎯 What it does: A 'context-likelihood graph' is constructed based on existing object detectors (such as Faster-RCNN and DETR), generating a graph structure through object relationship priors and initial class predictions, and using an energy model for iterative optimization. This approach enhances detection accuracy by utilizing graph context information, especially for rare categories.
Detection Transformer with Stable Matching
Shilong Liu (Tsinghua University), Lei Zhang (International Digital Economy Academy)
Object DetectionSegmentationTransformerImage
🎯 What it does: To address the matching instability issue that arises between multiple decoder layers in the DETR series detectors, two improvement schemes are proposed: position-supervised loss and position-modulated cost. Additionally, dense memory fusion is introduced to accelerate convergence, ultimately resulting in the construction of the Stable-DINO model.
DetermiNet: A Large-Scale Diagnostic Dataset for Complex Visually-Grounded Referencing using Determiners
Clarence Lee (Singapore University of Technology and Design), Cheston Tan (A*STAR)
Object DetectionData SynthesisVision Language ModelImageBenchmark
🎯 What it does: Constructed and released the DetermiNet diagnostic dataset, which includes 250,000 synthetic images and brief titles across 25 types of determiners (article, possessive, demonstrative, quantifier), and provides benchmark evaluations.
DETR Does Not Need Multi-Scale or Locality Design
Yutong Lin (Xi'an Jiaotong University), Han Hu (Microsoft Research Asia)
Object DetectionTransformerImage
🎯 What it does: Improved the original DETR, allowing it to achieve competitive performance with single-scale features and global cross-attention.
DETRDistill: A Universal Knowledge Distillation Framework for DETR-families
Jiahao Chang (University of Science and Technology of China), Feng Zhao (University of Science and Technology of China)
Object DetectionComputational EfficiencyKnowledge DistillationTransformerImage
🎯 What it does: Provides a universal knowledge distillation framework for Transformer-based detectors (DETR series), allowing smaller models to reduce inference costs while maintaining or exceeding the performance of larger models.
DETRs with Collaborative Hybrid Assignments Training
Zhuofan Zong (SenseTime Research), Yu Liu (SenseTime Research)
Object DetectionTransformerImage
🎯 What it does: A new collaborative hybrid allocation training scheme called C o-DETR is proposed to improve the efficiency and effectiveness of DETR-based detectors.
DetZero: Rethinking Offboard 3D Object Detection with Long-term Sequential Point Clouds
Tao Ma (Chinese University of Hong Kong), Hongsheng Li (Chinese University of Hong Kong)
Object DetectionObject TrackingAutonomous DrivingTransformerPoint CloudBenchmark
🎯 What it does: The DetZero framework is proposed, utilizing complete long sequences of LiDAR point clouds for offline 3D object detection, and generating complete trajectories through multi-frame detection and offline tracking, followed by an attention-based attribute-level refinement module (geometry, position, confidence) to further enhance detection accuracy.
DFA3D: 3D Deformable Attention For 2D-to-3D Feature Lifting
Hongyang Li (South China University of Technology), Lei Zhang (South China University of Technology)
Object DetectionDepth EstimationAutonomous DrivingTransformerImagePoint Cloud
🎯 What it does: Designed and implemented 3D Deformable Attention (DFA3D) to elevate multi-view 2D image features into a unified 3D space, refining features layer by layer through a hierarchical Transformer;
DG-Recon: Depth-Guided Neural 3D Scene Reconstruction
Jihong Ju (Qualcomm Technologies), Mohsen Ghafoorian (Qualcomm Technologies)
Object DetectionSegmentationDepth EstimationConvolutional Neural NetworkRecurrent Neural NetworkSimultaneous Localization and MappingPoint Cloud
🎯 What it does: A 3D scene reconstruction framework DG-Recon based on monocular depth priors is proposed, utilizing depth-guided feature reprojection and variable/cross attention fusion to achieve online real-time reconstruction.
DG3D: Generating High Quality 3D Textured Shapes by Learning to Discriminate Multi-Modal Diffusion-Renderings
Qi Zuo (Alibaba Group), Liefeng Bo (Alibaba Group)
GenerationData SynthesisDiffusion modelGenerative Adversarial NetworkPoint CloudMesh
🎯 What it does: Generate high-quality, texture-complete 3D mesh models
DiFaReli: Diffusion Face Relighting
Puntawat Ponglertnapakorn (Vistec), Supasorn Suwajanakorn
Image TranslationRestorationGenerationDiffusion modelAuto EncoderImage
🎯 What it does: Single-view face relighting using conditional DDIM and mask modulation networks
Diff-Retinex: Rethinking Low-light Image Enhancement with A Generative Diffusion Model
Xunpeng Yi (Wuhan University), Jiayi Ma (Wuhan University)
RestorationGenerationTransformerDiffusion modelImage
🎯 What it does: This paper proposes a low-light image enhancement method called Diff-Retinex, which combines Retinex decomposition with a diffusion model to achieve a physical explanation and generative completion for low-light images.
DiffCloth: Diffusion Based Garment Synthesis and Manipulation via Structural Cross-modal Semantic Alignment
Xujie Zhang (Sun Yat-Sen University), Xiaodan Liang (Sun Yat-Sen University)
GenerationData SynthesisDiffusion modelImageText
🎯 What it does: This paper proposes DiffCloth, a clothing image generation and text-driven local editing framework based on diffusion models. It utilizes bidirectional matching of visual segmentation and text attribute phrases to achieve cross-modal part-level semantic alignment, and supports precise local editing by modifying text.
DiffDis: Empowering Generative Diffusion Model with Cross-Modal Discrimination Capability
Runhui Huang (Shenzhen Campus of Sun Yat-sen University), Hang Xu (Huawei Noah's Ark Lab)
GenerationRetrievalTransformerDiffusion modelContrastive LearningImageTextMultimodality
🎯 What it does: A unified model named DiffDis is proposed, capable of simultaneously performing image generation and image-text discrimination (alignment) tasks under the diffusion process.
DiffDreamer: Towards Consistent Unsupervised Single-view Scene Extrapolation with Conditional Diffusion Models
Shengqu Cai (Stanford University), Gordon Wetzstein (Stanford University)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: Under the condition of a single image input, an unsupervised conditional diffusion model is used to generate multi-view consistent scenes with long-distance camera trajectories, addressing the consistency and drift issues in scene extrapolation.
Differentiable Transportation Pruning
Yunqiang Li (Axelera AI), Bram-Ernst Verhoef (Axelera AI)
CompressionOptimizationConvolutional Neural NetworkImage
🎯 What it does: A differentiable sparse learning method is proposed, achieving precise control of network sparsity through pruning via optimal transport with entropy regularization.
DiffFacto: Controllable Part-Based 3D Point Cloud Generation with Cross Diffusion
George Kiyohiro Nakayama, Leonidas Guibas
SegmentationGenerationData SynthesisDiffusion modelFlow-based ModelAuto EncoderPoint Cloud
🎯 What it does: Proposes the DiffFacto model, which uses factorized part styles and transformation distributions along with cross-diffusion networks to achieve controllable generation and editing of 3D point clouds based on parts.
DiffFit: Unlocking Transferability of Large Diffusion Models via Simple Parameter-efficient Fine-Tuning
Enze Xie (Huawei Noah's Ark Lab), Zhenguo Li (Huawei Noah's Ark Lab)
GenerationData SynthesisTransformerSupervised Fine-TuningDiffusion modelImage
🎯 What it does: This paper proposes DiffFit, a lightweight fine-tuning strategy that only adjusts biases and scaling factors while freezing most parameters, to efficiently adapt large diffusion models to downstream tasks.
DIFFGUARD: Semantic Mismatch-Guided Out-of-Distribution Detection Using Pre-Trained Diffusion Models
Ruiyuan Gao (Chinese University of Hong Kong), Qiang Xu (Chinese University of Hong Kong)
Anomaly DetectionDiffusion modelImage
🎯 What it does: A semantic mismatch detection framework called DIFFGUARD is proposed, which utilizes a pre-trained diffusion model to identify OOD samples in image classifiers.
DiffIR: Efficient Diffusion Model for Image Restoration
Bin Xia (Tsinghua University), Luc Van Gool (ETH Zurich)
RestorationSuper ResolutionTransformerDiffusion modelImage
🎯 What it does: DiffIR proposes a two-stage diffusion model framework for efficient image restoration, first extracting a compact recovery prior representation through the CPEN network, then denoising only that representation during the diffusion process, and finally reconstructing the image using DIRformer;
DiffPose: Multi-hypothesis Human Pose Estimation using Diffusion Models
Karl Holmquist (Linkoping University), Bastian Wandt (Linkoping University)
Pose EstimationTransformerDiffusion modelImage
🎯 What it does: This paper proposes a multi-hypothesis 3D human pose estimation method called DiffPose, which is based on a conditional diffusion model. It generates 3D poses that conform to the distribution by sampling multiple joint candidates from 2D heatmaps and embedding them into a Transformer.
DiffPose: SpatioTemporal Diffusion Model for Video-Based Human Pose Estimation
Runyang Feng (Jilin University), Hyung Jin Chang (University of Birmingham)
Pose EstimationTransformerDiffusion modelVideo
🎯 What it does: This paper proposes DiffPose, modeling video human pose estimation as a task of generating keypoint heatmaps using a conditional diffusion model.
DiffRate : Differentiable Compression Rate for Efficient Vision Transformers
Mengzhao Chen (Xiamen University), Ping Luo (Shanghai AI Laboratory)
CompressionComputational EfficiencyTransformerImage
🎯 What it does: A differentiable compression rate (DiffRate) framework is proposed, which jointly uses token pruning and merging to automatically learn the compression ratio for each layer.
DiffTAD: Temporal Action Detection with Proposal Denoising Diffusion
Sauradip Nag (University of Surrey), Tao Xiang (University of Surrey)
RecognitionObject DetectionTransformerDiffusion modelVideo
🎯 What it does: Treating temporary action detection as a denoising diffusion process from noisy proposals to action boundaries, the DiffTAD model is proposed.
DiffuMask: Synthesizing Images with Pixel-level Annotations for Semantic Segmentation Using Diffusion Models
Weijia Wu (Zhejiang University), Chunhua Shen (Zhejiang University)
SegmentationGenerationData SynthesisPrompt EngineeringDiffusion modelImage
🎯 What it does: Utilize the cross-attention maps of the diffusion model (Stable Diffusion) to automatically generate a large number of synthetic images and corresponding pixel-level semantic masks, constructing training data without manual annotation;
Diffuse3D: Wide-Angle 3D Photography via Bilateral Diffusion
Yutao Jiang (South China University of Technology), Shengfeng He (Singapore Management University)
GenerationData SynthesisDepth EstimationDiffusion modelImage
🎯 What it does: A 3D photography method based on bilateral diffusion is proposed, which can achieve wide-angle new perspective synthesis from a single image.
Diffusion Action Segmentation
Daochang Liu (University of Sydney), Chang Xu (University of Sydney)
SegmentationGenerationTransformerDiffusion modelVideo
🎯 What it does: A time action segmentation method based on diffusion models, DiffAct, is proposed, treating the frame-level action label sequence as a generative task obtained by gradually denoising from random noise.
Diffusion in Style
Martin Nicolas Everaert (Ecole Polytechnique Fédérale de Lausanne), Radhakrishna Achanta (Ecole Polytechnique Fédérale de Lausanne)
Image TranslationGenerationDiffusion modelImage
🎯 What it does: This paper proposes a method for adaptive Stable Diffusion using a small number of target style images—Diffusion in Style. By modifying the initial latent distribution and performing lightweight fine-tuning, it achieves fast and low-cost style transfer.
Diffusion Model as Representation Learner
Xingyi Yang (National University of Singapore), Xinchao Wang (National University of Singapore)
ClassificationSegmentationPose EstimationKnowledge DistillationRepresentation LearningReinforcement LearningDiffusion modelAuto EncoderImage
🎯 What it does: This paper views the diffusion probability model as a denoising autoencoder, analyzes its representation characteristics that change over time steps, and proposes the RepFusion method, which dynamically selects the optimal time step through reinforcement learning to distill the intermediate representations of the pre-trained DPM into a student network for tasks such as image classification, semantic segmentation, and keypoint detection.
Diffusion Models as Masked Autoencoders
Chen Wei (Meta AI), Christoph Feichtenhofer (Meta AI)
RestorationSegmentationGenerationRepresentation LearningTransformerDiffusion modelAuto EncoderImageVideo
🎯 What it does: This paper proposes the Diffusion Masked Autoencoder (DiffMAE), which combines diffusion models with masked autoencoders for visual representation pre-training, image/video generation, and restoration.
Diffusion-Based 3D Human Pose Estimation with Multi-Hypothesis Aggregation
Wenkang Shan (National Engineering Research Center of Visual Technology Peking University), Wen Gao (National Engineering Research Center of Visual Technology Peking University)
Pose EstimationDiffusion modelImage
🎯 What it does: A 3D human pose estimation framework based on diffusion models (D3DP) is proposed, which generates multiple 3D pose hypotheses conditioned on 2D keypoints and uses a joint-level multi-hypothesis aggregation method based on projection error (JPMA) to obtain a final single high-quality 3D pose.
Diffusion-based Image Translation with Label Guidance for Domain Adaptive Semantic Segmentation
Duo Peng (Singapore University of Technology and Design), Jun Liu (Singapore University of Technology and Design)
Image TranslationSegmentationDomain AdaptationDiffusion modelImage
🎯 What it does: This paper proposes an image translation framework based on diffusion models, guided by source domain pixel labels to achieve adaptive semantic segmentation in the target domain.
Diffusion-Guided Reconstruction of Everyday Hand-Object Interaction Clips
Yufei Ye (Carnegie Mellon University), Shubham Tulsiani (Carnegie Mellon University)
Diffusion modelVideo
🎯 What it does: This paper proposes a template-free 3D reconstruction method for hand-object interaction, utilizing video optimization combined with data-driven geometric priors to achieve the reconstruction of hand and object shapes and motions in short video clips.
Diffusion-SDF: Conditional Generative Modeling of Signed Distance Functions
Gene Chou, Felix Heide (Princeton University)
GenerationData SynthesisDiffusion modelAuto EncoderPoint CloudMesh
🎯 What it does: This paper proposes Diffusion-SDF, which utilizes diffusion models to achieve 3D shape generation, single-view reconstruction, point cloud scanning reconstruction, and sparse point cloud completion based on neural SDF representations.
DiffusionDet: Diffusion Model for Object Detection
Shoufa Chen (University of Hong Kong), Ping Luo (University of Hong Kong)
Object DetectionDiffusion modelImage
🎯 What it does: This paper proposes DiffusionDet, a framework that views object detection as a denoising diffusion process from noisy boxes to real boxes.
DiffusionRet: Generative Text-Video Retrieval with Diffusion Model
Peng Jin (Peking University), Jie Chen (Peking University)
RetrievalDiffusion modelContrastive LearningVideoText
🎯 What it does: A text-video retrieval framework called DiffusionRet based on diffusion models is proposed, treating the retrieval task as gradually generating a joint distribution of text and video from noise.
DiffV2S: Diffusion-Based Video-to-Speech Synthesis with Vision-Guided Speaker Embedding
Jeongsoo Choi (Korea Advanced Institute of Science and Technology), Yong Man Ro (Korea Advanced Institute of Science and Technology)
GenerationData SynthesisPrompt EngineeringDiffusion modelVideoAudio
🎯 What it does: A video-to-speech (V2S) synthesis system called DiffV2S has been developed, which completes speech reconstruction under no audio conditions through vision-guided speaker embedding.
DiLiGenT-Pi: Photometric Stereo for Planar Surfaces with Rich Details - Benchmark Dataset and Beyond
Feishi Wang (Peking University), Boxin Shi (Peking University)
RestorationData SynthesisConvolutional Neural NetworkSupervised Fine-TuningImageBenchmark
🎯 What it does: Created and released the real-world dataset DiLiGenTΠ, focused on nearly planar surfaces (with rich details), and systematically evaluated various known and unknown photometric stereo algorithms on this dataset under different lighting conditions.
DIME-FM : DIstilling Multimodal and Efficient Foundation Models
Ximeng Sun (Boston University), Xide Xia (Meta AI)
Domain AdaptationKnowledge DistillationTransformerContrastive LearningImageTextMultimodality
🎯 What it does: Using large-scale publicly available unpaired images and text, the DIME-FM mechanism distills the knowledge of CLIP-ViT-L/14 into smaller models such as ViT-B/32 and Swin-Tiny, resulting in a customized vision-language foundation model that can compete with the original CLIP-ViT-B/32 on zero-shot and linear probe tasks, and exhibits stronger robustness to distribution shifts.
DINAR: Diffusion Inpainting of Neural Textures for One-Shot Human Avatars
David Svitov (Samsung AI Center), Victor Lempitsky (Cinemersive Labs)
RestorationGenerationDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a method for generating animatable full-body human avatars from a single RGB image, utilizing the SMPL-X parametric model combined with neural textures, and completing the textures through a diffusion model.
DIRE for Diffusion-Generated Image Detection
Zhendong Wang (University of Science and Technology of China), Houqiang Li (University of Science and Technology of China)
Anomaly DetectionConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: A method called DIRE, based on the reconstruction error of diffusion models, is proposed for detecting images generated by diffusion models.
Discovering Spatio-Temporal Rationales for Video Question Answering
Yicong Li (National University of Singapore), Tat-Seng Chua (National University of Singapore)
RecognitionRetrievalExplainability and InterpretabilityTransformerVideo
🎯 What it does: A differentiable spatiotemporal rationalization (STR) module and the TranSTR network are proposed to automatically select key frames and key objects in long videos to support question answering.
Discrepant and Multi-Instance Proxies for Unsupervised Person Re-Identification
Chang Zou (Xi'an Jiaotong University), Chi Zhang (Xi'an Jiaotong University)
RecognitionRetrievalContrastive LearningImage
🎯 What it does: This paper proposes an unsupervised pedestrian re-identification method based on inconsistent clustering proxies and multi-instance proxies, utilizing contrastive learning to enhance feature representation.
Discriminative Class Tokens for Text-to-Image Diffusion Models
Idan Schwartz (Tel Aviv University), Sagie Benaim (University of Copenhagen)
GenerationData SynthesisTransformerDiffusion modelImageText
🎯 What it does: A new class-specific word embedding (discriminative class token) is trained for text-to-image diffusion models, which enhances the category accuracy and detail representation of generated images without the need for additional labeled images.
Disentangle then Parse: Night-time Semantic Segmentation with Illumination Disentanglement
Zhixiang Wei (University of Science and Technology of China), Yi Jin (University of Science and Technology of China)
SegmentationAutonomous DrivingImage
🎯 What it does: A new paradigm for nighttime semantic segmentation called 'Decomposition and Parsing (DTP)' is proposed, which decomposes nighttime images into illumination-invariant reflectance components and illumination-specific lighting components, and performs segmentation based on this.
Disentangling Spatial and Temporal Learning for Efficient Image-to-Video Transfer Learning
Zhiwu Qing (Huazhong University of Science and Technology), Nong Sang (Huazhong University of Science and Technology)
Domain AdaptationComputational EfficiencyTransformerContrastive LearningVideo
🎯 What it does: Proposes the DiST framework, freezing CLIP ViT as a spatial encoder, combined with a lightweight temporal encoder and fusion branch to achieve efficient video transfer learning.
DISeR: Designing Imaging Systems with Reinforcement Learning
Tzofi Klinghoffer (Massachusetts Institute of Technology), Ramesh Raskar (Massachusetts Institute of Technology)
Depth EstimationAutonomous DrivingOptimizationTransformerReinforcement LearningImage
🎯 What it does: This paper proposes a search space based on Context-Free Grammar (CFG) and utilizes Reinforcement Learning (PPO) to automatically co-design camera systems and perception models, achieving task-specific imaging system configurations.
Disposable Transfer Learning for Selective Source Task Unlearning
Seunghee Koh (Korea Advanced Institute of Science and Technology), Junmo Kim (Korea Advanced Institute of Science and Technology)
Knowledge DistillationData-Centric LearningConvolutional Neural NetworkImage
🎯 What it does: A Disposable Transfer Learning (DTL) framework is proposed, which combines a new Gradient Collision loss (GC loss) to eliminate source task knowledge after completing transfer learning while maintaining target task performance, and introduces Piggyback Learning Accuracy as an evaluation metric.
DistillBEV: Boosting Multi-Camera 3D Object Detection with Cross-Modal Knowledge Distillation
Zeyu Wang (QCraft), Xiaodong Yang (QCraft)
Object DetectionAutonomous DrivingKnowledge DistillationPoint Cloud
🎯 What it does: By using cross-modal knowledge distillation, the features of the LiDAR-based detector are transferred to the multi-camera BEV detector, thereby enhancing its 3D detection performance.
Distilled Reverse Attention Network for Open-world Compositional Zero-Shot Learning
Yun Li (CSIRO Data61), Lina Yao (University of New South Wales)
RecognitionKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes the Distilled Reverse Attention Network (DRANet) for Open World Compositional Zero-Shot Learning (OW-CZSL), which enhances the recognition ability of unseen combinations by separately learning visual primitives of attributes and objects and achieving feature decoupling.
Distilling Coarse-to-Fine Semantic Matching Knowledge for Weakly Supervised 3D Visual Grounding
Zehan Wang (Zhejiang University), Zhou Zhao (Zhejiang University)
Object DetectionComputational EfficiencyKnowledge DistillationTransformerContrastive LearningTextPoint Cloud
🎯 What it does: A weakly supervised 3D visual localization framework is proposed, which learns the localization of target objects using only the correspondence between scene and sentence-level information. A coarse-to-fine semantic matching model is employed to first filter candidates roughly, and then semantic reconstruction is used to evaluate fine similarity. Knowledge distillation is applied to transfer matching knowledge to a conventional two-stage 3D visual localization network, significantly reducing inference costs and improving performance.
Distilling DETR with Visual-Linguistic Knowledge for Open-Vocabulary Object Detection
Liangqi Li (Hikvision Research Institute), Shiliang Pu (Hikvision Research Institute)
Object DetectionKnowledge DistillationTransformerVision Language ModelImage
🎯 What it does: This paper proposes an end-to-end open vocabulary object detection framework DK-DETR based on Deformable DETR, and utilizes visual-language model (VLM) knowledge distillation to enhance the detection performance of new categories.
Distilling from Similar Tasks for Transfer Learning on a Budget
Kenneth Borup (Aarhus University), Bharath Hariharan (Cornell University)
ClassificationObject DetectionKnowledge DistillationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper proposes training efficient and accurate object recognition models through cross-task knowledge distillation from different source models (without source data) under the constraints of limited labels and computational resources.
Distilling Large Vision-Language Model with Out-of-Distribution Generalizability
Xuanlin Li (University of California San Diego), Hao Su (University of California San Diego)
ClassificationKnowledge DistillationConvolutional Neural NetworkTransformerLarge Language ModelVision Language ModelContrastive LearningImage
🎯 What it does: Utilizing the visual space and visual-language alignment structure of a large-scale teacher visual-language model for knowledge distillation of the student model, with a focus on enhancing the student's zero-shot and few-shot performance in open vocabulary and out-of-distribution (OOD) classification tasks.
Distracting Downpour: Adversarial Weather Attacks for Motion Estimation
Jenny Schmalfuss (Institute for Visualization and Interactive Systems, University of Stuttgart), Andrés Bruhn (Institute for Visualization and Interactive Systems, University of Stuttgart)
Autonomous DrivingAdversarial AttackOptical FlowImageVideo
🎯 What it does: A differentiable particle rendering framework is designed to generate realistic, three-dimensional consistent, and temporally coherent weather effects. This framework is used to conduct adversarial attacks on optical flow estimation while evaluating and enhancing the network's robustness under weather conditions.
Distributed Bundle Adjustment with Block-Based Sparse Matrix Compression for Super Large Scale Datasets
Maoteng Zheng (China University of Geosciences), Hao Qu (Mirauge3D Technology)
OptimizationSimultaneous Localization and MappingImage
🎯 What it does: A distributed precise Levenberg–Marquardt algorithm is used to implement global Bundle Adjustment for extremely large-scale data.
Distribution Shift Matters for Knowledge Distillation with Webly Collected Images
Jialiang Tang (Nanjing University of Science and Technology), Chen Gong (Nanjing University of Science and Technology)
ClassificationKnowledge DistillationContrastive LearningImage
🎯 What it does: This paper proposes a KD3 method for knowledge distillation using web images collected from the internet in the absence of original training data, explicitly addressing the distribution shift between web images and original data.
Distribution-Aligned Diffusion for Human Mesh Recovery
Lin Geng Foo (Singapore University of Technology and Design), Jun Liu (Singapore University of Technology and Design)
RestorationPose EstimationTransformerDiffusion modelMesh
🎯 What it does: A method for recovering 3D human meshes from a single RGB image based on diffusion models (Human Mesh Diffusion, HMDiff) is designed, and a Distribution Alignment Technique (DAT) is proposed to inject input-specific prior information during the reverse diffusion process.
Distribution-Aware Prompt Tuning for Vision-Language Models
Eulrang Cho (Korea University), Hyunwoo J Kim
Domain AdaptationTransformerPrompt EngineeringVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: A distribution-aware prompt tuning method DAPT is proposed to enhance the performance of pre-trained vision-language models in few-shot learning and domain transfer tasks.
Distribution-Consistent Modal Recovering for Incomplete Multimodal Learning
Yuanzhi Wang (Nanjing University of Science and Technology), Yong Li (Nanjing University of Science and Technology)
ClassificationRestorationTransformerFlow-based ModelMultimodality
🎯 What it does: This paper proposes a distribution-consistent modal recovery framework called DiCMoR, which first maps observable modalities to a Gaussian distribution space, then transfers the distribution to the missing modality through a cross-modal flow model to generate corresponding features, and finally inputs the recovered modality along with the existing modalities into a multimodal Transformer for downstream classification.
Diverse Cotraining Makes Strong Semi-Supervised Segmentor
Yijiang Li (Johns Hopkins University), Ying Gao (University of Hong Kong)
SegmentationConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper systematically explores the diversity in three dimensions: input domain, data augmentation, and model architecture, by re-examining the core assumptions of deep co-training and theoretically proving the impact of network homogenization on generalization error. It proposes the Diverse Co-training framework to enhance semi-supervised semantic segmentation performance.
Diverse Data Augmentation with Diffusions for Effective Test-time Prompt Tuning
Chun-Mei Feng (Institute of High Performance Computing Agency for Science Technology and Research), Wangmeng Zuo (Institute of High Performance Computing Agency for Science Technology and Research)
Data SynthesisDomain AdaptationOptimizationTransformerPrompt EngineeringVision Language ModelDiffusion modelImage
🎯 What it does: This paper proposes DiffTPT, a method for adaptive prompt tuning on a single image during testing, which generates diverse augmented images using a stable diffusion model and employs cosine similarity filtering to ensure semantic consistency.