CVPR 2023 Papers — Page 6
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2353 papers
Depth Estimation From Indoor Panoramas With Neural Scene Representation
Wenjie Chang (University of Science and Technology of China), Zhiwei Xiong (University of Science and Technology of China)
Depth EstimationNeural Radiance FieldImage
🎯 What it does: A framework for indoor panoramic depth estimation based on neural implicit fields is proposed, using two networks (SDF network and color network) to learn scene geometry and color from a small number of panoramic images without the need for depth labels.
DeSTSeg: Segmentation Guided Denoising Student-Teacher for Anomaly Detection
Xuan Zhang (Tsinghua University), Ting Chen (Tsinghua University)
SegmentationAnomaly DetectionKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a segmentation-guided denoising model DeSTSeg based on a student-teacher framework for visual anomaly detection and localization.
DetCLIPv2: Scalable Open-Vocabulary Object Detection Pre-Training via Word-Region Alignment
Lewei Yao (Hong Kong University of Science and Technology), Hang Xu (Huawei Noah's Ark Lab)
Object DetectionTransformerVision Language ModelContrastive LearningImageText
🎯 What it does: This paper presents DetCLIPv2, an end-to-end open vocabulary object detection pre-training framework that directly learns fine-grained word-region alignment from large-scale image-text pairs and achieves unified optimization of detection, localization, and image-text alignment through joint training.
Detecting and Grounding Multi-Modal Media Manipulation
Rui Shao (Harbin Institute of Technology), Ziwei Liu (Nanyang Technological University)
ClassificationObject DetectionTransformerContrastive LearningMultimodality
🎯 What it does: This paper proposes a multi-modal media manipulation detection and localization task (DGM4), constructs a dataset containing 230k samples centered on human news, and introduces the HAMMER model, which incorporates hierarchical contrastive learning and cross-modal attention.
Detecting Backdoors During the Inference Stage Based on Corruption Robustness Consistency
Xiaogeng Liu (Huazhong University of Science and Technology), Chaowei Xiao (Arizona State University)
Anomaly DetectionConvolutional Neural NetworkImage
🎯 What it does: A backdoor trigger detection method based on Test-time Consistency evaluation of image corruption (TeCo) is proposed, which only utilizes black-box hard label outputs without the need for additional data or trigger assumptions.
Detecting Backdoors in Pre-Trained Encoders
Shiwei Feng (Purdue University), Xiangyu Zhang (Rutgers University)
Anomaly DetectionRepresentation LearningConvolutional Neural NetworkImage
🎯 What it does: This paper proposes DECREE, a backdoor detection method for pre-trained self-supervised encoders.
Detecting Everything in the Open World: Towards Universal Object Detection
Zhenyu Wang (Tsinghua University), Shengjin Wang (Tsinghua University)
Object DetectionConvolutional Neural NetworkVision Language ModelImage
🎯 What it does: A universal object detection framework named UniDetector is proposed, which can utilize multi-source heterogeneous label spaces for training and directly detect any category in the open world without training samples.
Detecting Human-Object Contact in Images
Yixin Chen (Beijing Institute of General Artificial Intelligence), Dimitrios Tzionas (University of Amsterdam)
Object DetectionSegmentationPose EstimationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a framework for detecting full-body human-object contact in a single color image, simultaneously outputting a contact heatmap and corresponding body part labels.
Detection Hub: Unifying Object Detection Datasets via Query Adaptation on Language Embedding
Lingchen Meng (Fudan University), Yu-Gang Jiang (Fudan University)
Object DetectionTransformerLarge Language ModelImage
🎯 What it does: Designed and implemented Detection Hub, a unified framework capable of training on multiple object detection datasets, addressing the taxonomic differences and annotation inconsistencies between datasets.
Detection of Out-of-Distribution Samples Using Binary Neuron Activation Patterns
Bartłomiej Olber (Warsaw University of Technology), Krystian Chachuła (NVIDIA)
Anomaly DetectionOptimizationHyperparameter SearchConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a method for out-of-distribution (OOD) detection of image data by calculating the Hamming distance between test samples and training samples using the binary activation pattern (NAP) in ReLU networks.
DETR With Additional Global Aggregation for Cross-Domain Weakly Supervised Object Detection
Zongheng Tang (Beihang University), Yi Yang (Zhejiang University)
Object DetectionDomain AdaptationTransformerImage
🎯 What it does: In the cross-domain weakly supervised object detection task, we propose DETR-GA: by adding class queries in the encoder of DETR and foreground queries in the decoder, we achieve simultaneous instance-level and image-level predictions, thereby utilizing weak supervision information for domain alignment.
DETRs With Hybrid Matching
Ding Jia (Peking University), Han Hu (Microsoft Research Asia)
Object DetectionObject TrackingPose EstimationTransformerImage
🎯 What it does: A hybrid matching strategy is proposed, which uses both one-to-one matching and one-to-many matching during the training phase of DETR to enhance the learning effect of positive samples, while maintaining the original one-to-one matching during the inference phase, preserving the advantages of end-to-end and no NMS.
Devil Is in the Queries: Advancing Mask Transformers for Real-World Medical Image Segmentation and Out-of-Distribution Localization
Mingze Yuan (Alibaba Group), Li Zhang (Peking University)
SegmentationAnomaly DetectionTransformerImageBiomedical DataComputed Tomography
🎯 What it does: This paper addresses the long-tail tumor problem in real-world medical image segmentation by proposing a MaxQuery method based on Mask Transformer, which achieves high-quality tumor segmentation and identifies out-of-view (OOV) abnormal regions through query responses; it also introduces a Query Distribution (QD) loss to guide queries to focus on the foreground and enhance boundary separation.
Devil's on the Edges: Selective Quad Attention for Scene Graph Generation
Deunsol Jung (Pohang University of Science and Technology), Minsu Cho (Pohang University of Science and Technology)
Object DetectionGenerationTransformerGraph
🎯 What it does: The paper proposes a selective quadruple attention network named SQUAT for scene graph generation, which achieves efficient contextual reasoning through edge selection and quadruple attention.
DexArt: Benchmarking Generalizable Dexterous Manipulation With Articulated Objects
Chen Bao (Shanghai Jiao Tong University), Xiaolong Wang (UC San Diego)
Robotic IntelligenceReinforcement LearningContrastive LearningPoint CloudBenchmark
🎯 What it does: Proposed the DexArt benchmark, defining four fine manipulation tasks using multi-finger wrist operations on articulated objects, and trained a general policy using point cloud observations.
DF-Platter: Multi-Face Heterogeneous Deepfake Dataset
Kartik Narayan (Indian Institute of Technology Jodhpur), Richa Singh (Indian Institute of Technology Jodhpur)
GenerationData SynthesisGenerative Adversarial NetworkVideoMultimodalityBenchmark
🎯 What it does: A large multi-face multi-modal deepfake dataset, DF-Platter, has been proposed and released, covering low resolution, occlusion, and multi-person synthesis scenarios.
DiffCollage: Parallel Generation of Large Content With Diffusion Models
Qinsheng Zhang, Ming-Yu Liu
ClassificationRecognitionGenerationConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: This paper proposes a new computer vision algorithm to improve the accuracy of image recognition.
Differentiable Architecture Search With Random Features
Xuanyang Zhang (MEGVII Technology), Jian Sun (MEGVII Technology)
ClassificationNeural Architecture SearchImage
🎯 What it does: This paper proposes a novel variant of DARTS, RF-DARTS (and RF-PCDARTS), which addresses the performance collapse issue of traditional DARTS by training only the BatchNorm parameters (random features), and provides a theoretical analysis and experimental validation of its mechanism.
Differentiable Shadow Mapping for Efficient Inverse Graphics
Markus Worchel (Technische Universität Berlin), Marc Alexa (Technische Universität Berlin)
Pose EstimationComputational EfficiencyMesh
🎯 What it does: A differentiable shadow mapping method is proposed, which implements shadow computation in differentiable rasterization of triangular meshes and is differentiable, supporting inverse rendering tasks.
Difficulty-Based Sampling for Debiased Contrastive Representation Learning
Taeuk Jang (Purdue University), Xiaoqian Wang (Purdue University)
Representation LearningContrastive LearningImageTabular
🎯 What it does: A difficulty sampling-based debiased contrastive learning framework is proposed, which distinguishes easy negative samples and hard negative samples by training two encoders (bias amplification encoder and debiasing encoder) and dynamically weights negative samples based on their relative distances.
DiffPose: Toward More Reliable 3D Pose Estimation
Jia Gong (Singapore University of Technology and Design), Jun Liu (Singapore University of Technology and Design)
Pose EstimationGraph Neural NetworkTransformerDiffusion modelVideo
🎯 What it does: This paper proposes DiffPose, a monocular 3D human pose estimation framework based on diffusion models, treating 3D pose prediction as an inverse diffusion process from uncertain distribution to certain distribution.
DiffRF: Rendering-Guided 3D Radiance Field Diffusion
Norman Müller (Technical University of Munich), Matthias Nießner (Technical University of Munich)
GenerationData SynthesisDiffusion modelNeural Radiance FieldPoint Cloud
🎯 What it does: Proposes DiffRF, a 3D radiance field generation method based on diffusion models;
DiffSwap: High-Fidelity and Controllable Face Swapping via 3D-Aware Masked Diffusion
Wenliang Zhao (Tsinghua University), Jiwen Lu (Tsinghua University)
GenerationRetrievalDiffusion modelImageVideo
🎯 What it does: By reformulating the facial replacement task as a conditional filling task, high-fidelity and controllable face swapping is achieved using diffusion models.
DiffTalk: Crafting Diffusion Models for Generalized Audio-Driven Portraits Animation
Shuai Shen (Tsinghua University), Jiwen Lu (Tsinghua University)
GenerationDiffusion modelVideoAudio
🎯 What it does: A novel general-purpose speech-driven avatar animation framework called DiffTalk is proposed, which can generate high-quality, synchronized speaking videos without any fine-tuning.
Diffusion Art or Digital Forgery? Investigating Data Replication in Diffusion Models
Gowthami Somepalli (University of Maryland), Tom Goldstein (University of Maryland)
GenerationRetrievalTransformerDiffusion modelContrastive LearningImage
🎯 What it does: The study investigates and quantifies whether diffusion models replicate training set content when generating images, and constructs a copy detection system based on image retrieval.
Diffusion Probabilistic Model Made Slim
Xingyi Yang (National University of Singapore), Xinchao Wang (National University of Singapore)
RestorationGenerationData SynthesisComputational EfficiencyKnowledge DistillationDiffusion modelImage
🎯 What it does: A lightweight diffusion probability model called Spectral Diffusion (SD) is proposed, which achieves adaptive recovery of high-frequency details through waveform gating and spectral perception distillation, significantly reducing model size and computational load.
Diffusion Video Autoencoders: Toward Temporally Consistent Face Video Editing via Disentangled Video Encoding
Gyeongman Kim (Korea Advanced Institute of Science and Technology), Eunho Yang (Korea Advanced Institute of Science and Technology)
GenerationData SynthesisDiffusion modelAuto EncoderVideo
🎯 What it does: A facial video editing framework based on diffusion autoencoders is proposed, which can decompose videos into three features: identity, action, and background, and achieve temporally consistent facial attribute changes by editing identity features.
Diffusion-Based Generation, Optimization, and Planning in 3D Scenes
Siyuan Huang (National Key Laboratory of General Artificial Intelligence, BIGAI), Song-Chun Zhu (National Key Laboratory of General Artificial Intelligence, BIGAI)
GenerationOptimizationRobotic IntelligenceDiffusion modelPoint CloudMesh
🎯 What it does: This paper presents SceneDiffuser, a unified conditional generation framework based on diffusion models that can simultaneously perform generation, physical optimization, and planning tasks in 3D scenes.
Diffusion-Based Signed Distance Fields for 3D Shape Generation
Jaehyeok Shim (Ulsan National Institute of Science and Technology), Kyungdon Joo (Ulsan National Institute of Science and Technology)
GenerationData SynthesisSuper ResolutionConvolutional Neural NetworkDiffusion modelPoint CloudMesh
🎯 What it does: A two-stage SDF generation framework based on diffusion models (SDF-Diffusion) is proposed, which first generates low-resolution SDF and then generates high-resolution SDF through super-resolution, and can directly output meshes.
Diffusion-SDF: Text-To-Shape via Voxelized Diffusion
Muheng Li (Tsinghua University), Jiwen Lu (Tsinghua University)
GenerationData SynthesisDiffusion modelAuto EncoderTextPoint Cloud
🎯 What it does: A two-stage text-to-shape generation framework called Diffusion-SDF based on voxelized SDF is proposed, supporting text generation, completion, and editing.
DiffusioNeRF: Regularizing Neural Radiance Fields With Denoising Diffusion Models
Jamie Wynn (Niantic), Daniyar Turmukhambetov (Niantic)
GenerationData SynthesisDepth EstimationDiffusion modelNeural Radiance FieldImagePoint Cloud
🎯 What it does: This paper utilizes a Denoising Diffusion Model (DDM) to learn the log probability gradient of RGBD patches, and uses this gradient as a regularization term to directly constrain the color and geometry fields of NeRF, in order to improve view synthesis and 3D reconstruction quality under limited viewpoints.
DiffusionRig: Learning Personalized Priors for Facial Appearance Editing
Zheng Ding (UC San Diego), Xiuming Zhang (Adobe)
Image TranslationGenerationDiffusion modelImage
🎯 What it does: We constructed a personalized facial appearance editing system called DiffusionRig based on diffusion models, which can accurately edit lighting, expressions, and poses while maintaining the identity of the person and high-frequency details.
DIFu: Depth-Guided Implicit Function for Clothed Human Reconstruction
Dae-Young Song (Electronics and Telecommunications Research Institute), Donghyeon Cho (Chungnam National University)
GenerationDepth EstimationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a deep-guided implicit function (DIFu) for 3D reconstruction of clothed humans from a single image.
DiGA: Distil To Generalize and Then Adapt for Domain Adaptive Semantic Segmentation
Fengyi Shen (Technical University of Munich), Alois Knoll (Technical University of Munich)
SegmentationDomain AdaptationKnowledge DistillationConvolutional Neural NetworkTransformerImage
🎯 What it does: An end-to-end domain adaptive semantic segmentation framework DiGA is proposed through symmetric knowledge distillation, cross-domain mixed data augmentation, and threshold-free bidirectional consensus pseudo-labeling.
DiGeo: Discriminative Geometry-Aware Learning for Generalized Few-Shot Object Detection
Jiawei Ma (Columbia University), Shih-Fu Chang (Columbia University)
Object DetectionKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: For general few-shot object detection, a DiGeo framework is proposed, which achieves a balance in detection performance between base classes and novel classes through geometric-aware discriminative feature learning.
Dimensionality-Varying Diffusion Process
Han Zhang (Shanghai Jiao Tong University), Fan Cheng (Shanghai Jiao Tong University)
GenerationData SynthesisComputational EfficiencyDiffusion modelImageStochastic Differential Equation
🎯 What it does: A variable-dimensional diffusion process (DVDP) is proposed, which dynamically reduces the signal dimension during the diffusion process to accelerate training and sampling.
DINER: Depth-Aware Image-Based NEural Radiance Fields
Malte Prinzler (Max Planck Institute for Intelligent Systems), Justus Thies (Max Planck Institute for Intelligent Systems)
GenerationDepth EstimationNeural Radiance FieldImage
🎯 What it does: This paper proposes a depth-aware image-driven neural radiance field (DINER) based on sparse RGB views, capable of reconstructing high-quality 3D scenes and generating novel view images using only four large-parallax views.
DINER: Disorder-Invariant Implicit Neural Representation
Shaowen Xie (Nanjing University), Zhan Ma (Tencent Company)
RestorationData SynthesisImageVideo
🎯 What it does: This paper proposes a model that incorporates a hash table (DINER) at the front end of implicit neural representations (INR), which reduces spectral bias and enhances representation capability by remapping input coordinates.
DINN360: Deformable Invertible Neural Network for Latitude-Aware 360deg Image Rescaling
Yichen Guo (Beihang University), Yunjin Chen (University of British Columbia)
RestorationSuper ResolutionTransformerFlow-based ModelImage
🎯 What it does: A 360° image rescaling method called DINN360 is proposed, utilizing reversible networks to achieve deformable downsampling and latitude-aware high-frequency projection, recovering high-resolution 360° scenes from low-resolution images.
Dionysus: Recovering Scene Structures by Dividing Into Semantic Pieces
Likang Wang (Hong Kong University of Science and Technology), Lei Chen (Hong Kong University of Science and Technology)
Depth EstimationComputational EfficiencyConvolutional Neural NetworkSupervised Fine-TuningVideo
🎯 What it does: A real-time efficient 3D reconstruction framework called Dionysus is proposed, which utilizes cross-view semantic consistency to filter the most valuable depth candidates and reallocates them, thereby improving computational efficiency while maintaining high reconstruction quality.
DIP: Dual Incongruity Perceiving Network for Sarcasm Detection
Changsong Wen (Nankai University), Jufeng Yang (Nankai University)
ClassificationRecognitionTransformerContrastive LearningImageTextMultimodality
🎯 What it does: This paper studies the problem of multimodal sarcasm detection and proposes a Dual Incongruity Perceiving (DIP) network that identifies sarcastic image-text pairs by explicitly modeling the incongruities at the factual and emotional levels.
Directional Connectivity-Based Segmentation of Medical Images
Ziyun Yang (Duke University), Sina Farsiu (Duke University)
SegmentationImageBiomedical Data
🎯 What it does: A medical image segmentation network DconnNet based on directional pixel connectivity decoupling is proposed.
DISC: Learning From Noisy Labels via Dynamic Instance-Specific Selection and Correction
Yifan Li (Chinese Academy of Sciences), Xilin Chen (Chinese Academy of Sciences)
ClassificationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: The research proposes a selection and correction method based on dynamic instance-specific thresholds (DISC), which utilizes dual-view learning to partition and correct data with noisy labels, thereby improving classification performance in noisy environments.
DisCo-CLIP: A Distributed Contrastive Loss for Memory Efficient CLIP Training
Yihao Chen (International Digital Economy Academy), Lei Zhang (International Digital Economy Academy)
Computational EfficiencyContrastive LearningImage
🎯 What it does: Designed and implemented DisCo-CLIP, a distributed contrastive loss that significantly reduces the memory requirements for CLIP training.
DisCoScene: Spatially Disentangled Generative Radiance Fields for Controllable 3D-Aware Scene Synthesis
Yinghao Xu (Chinese University of Hong Kong), Sergey Tulyakov (Snap Inc.)
GenerationData SynthesisAutonomous DrivingNeural Radiance FieldGenerative Adversarial NetworkImage
🎯 What it does: A 3D perception scene generation framework called DisCoScene is proposed, which can learn controllable high-quality 3D scene synthesis from single-view 2D image data based on abstract layout priors.
Discovering the Real Association: Multimodal Causal Reasoning in Video Question Answering
Chuanqi Zang (Beijing Institute of Technology), Wei Liang (Beijing Institute of Technology)
RecognitionObject DetectionVision Language ModelVideoTextMultimodality
🎯 What it does: This paper proposes a multi-modal causal reasoning framework MCR, aimed at eliminating the co-occurrence bias between visual and textual information in video question answering tasks, thereby enhancing the model's robustness and generalization ability.
Discrete Point-Wise Attack Is Not Enough: Generalized Manifold Adversarial Attack for Face Recognition
Qian Li (Eastern Institute for Advanced Study), Yuntian Chen (Eastern Institute for Advanced Study)
RecognitionAdversarial AttackGenerative Adversarial NetworkImage
🎯 What it does: A framework for facial recognition generalization manifold adversarial attack (GMAA) is proposed, which expands the attack targets to a multi-state set and enhances the attack space from discrete points to a continuous manifold through facial action coding, thereby generating more generalized adversarial samples.
Discriminating Known From Unknown Objects via Structure-Enhanced Recurrent Variational AutoEncoder
Aming Wu (Xidian University), Cheng Deng (Xidian University)
Object DetectionAnomaly DetectionRecurrent Neural NetworkAuto EncoderImage
🎯 What it does: This paper proposes a Structure-Enhanced Recursive Variational Autoencoder (SR-VAE) for unsupervised OOD (Out-of-Distribution) object detection, integrating LoG structural enhancement, recursive VAE to generate diverse classification features, and cyclic consistency conditional VAE to synthesize virtual OOD features.
Discriminative Co-Saliency and Background Mining Transformer for Co-Salient Object Detection
Long Li (Northwestern Polytechnical University), Fahad Shahbaz Khan (CVL Linkoping University)
Object DetectionTransformerImage
🎯 What it does: A Transformer-based DMT framework is proposed to achieve co-salient object detection while explicitly mining co-salient and background information.
Discriminator-Cooperated Feature Map Distillation for GAN Compression
Tie Hu (Xiamen University), Rongrong Ji (Xiamen University)
Image TranslationCompressionKnowledge DistillationGenerative Adversarial NetworkImage
🎯 What it does: A feature map distillation method based on a teacher discriminator (DCD) is proposed, combined with collaborative adversarial training to achieve efficient compression of generators such as CycleGAN and Pix2Pix.
Disentangled Representation Learning for Unsupervised Neural Quantization
Haechan Noh (Sungkyunkwan University), Jae-Pil Heo (Sungkyunkwan University)
RetrievalCompressionRepresentation LearningAuto EncoderImage
🎯 What it does: A discrete representation learning method is proposed to achieve the separation of cluster center information in Unsupervised Neural Quantization (UNQ), allowing the deep quantizer to be compatible with inverted indexing and enabling efficient non-exhaustive approximate nearest neighbor retrieval.
Disentangling Orthogonal Planes for Indoor Panoramic Room Layout Estimation With Cross-Scale Distortion Awareness
Zhijie Shen (Beijing Jiaotong University), Yao Zhao (Beijing Jiaotong University)
SegmentationDepth EstimationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes an indoor panoramic layout estimation framework based on separating vertical and horizontal planes, soft flip fusion, and cross-scale distortion-aware assembly, which can significantly improve the accuracy of 3D spatial layout reconstruction.
Disentangling Writer and Character Styles for Handwriting Generation
Gang Dai (South China University of Technology), Shuangping Huang (Hong Kong Polytechnic University)
GenerationTransformerContrastive LearningImage
🎯 What it does: Proposes the Style-Disentangled Transformer (SDT), which achieves online Chinese handwritten character generation through writer-level and character-level style representation, and extends to offline handwritten character generation.
Distilling Cross-Temporal Contexts for Continuous Sign Language Recognition
Leming Guo (Tianjin University of Technology), Shengyong Chen (Tianjin University of Technology)
RecognitionKnowledge DistillationRecurrent Neural NetworkVideo
🎯 What it does: This paper proposes a Cross-Temporal Context Aggregation (CTCA) framework that enhances the shallow temporal aggregation module using a deep dual-path network and cross-context knowledge distillation, enabling the simultaneous capture of local and global temporal and linguistic priors for continuous sign language recognition.
Distilling Focal Knowledge From Imperfect Expert for 3D Object Detection
Jia Zeng (OpenDriveLab, Shanghai AI Lab), Hongyang Li (OpenDriveLab, Shanghai AI Lab)
Object DetectionAutonomous DrivingKnowledge DistillationTransformerSupervised Fine-TuningPoint Cloud
🎯 What it does: This paper proposes a query-based focal knowledge distillation framework, FD3D, which automatically locates instance-level focal regions in multi-camera 3D object detection using queries. It first performs masked generative distillation in these regions, and then conducts fine-grained distillation of local features through deformable attention alignment, thereby enhancing the detection performance of lightweight models.
Distilling Neural Fields for Real-Time Articulated Shape Reconstruction
Jeff Tan (Carnegie Mellon University), Deva Ramanan (Carnegie Mellon University)
Object DetectionPose EstimationKnowledge DistillationNeural Radiance FieldVideo
🎯 What it does: This paper proposes a method for training a category-specific forward network that can predict pose, shape, and appearance in real-time for deformable objects in videos by using dynamic NeRF as a teacher model for knowledge distillation, achieving unsupervised 3D shape reconstruction.
Distilling Self-Supervised Vision Transformers for Weakly-Supervised Few-Shot Classification & Segmentation
Dahyun Kang (Meta AI), Naila Murray (Meta AI)
ClassificationSegmentationKnowledge DistillationTransformerContrastive LearningImage
🎯 What it does: This paper proposes a joint model for weakly supervised few-shot classification and segmentation by generating pseudo-pixel labels using self-supervised Vision Transformer (ViT) attention maps.
Distilling Vision-Language Pre-Training To Collaborate With Weakly-Supervised Temporal Action Localization
Chen Ju (Shanghai Jiao Tong University), Yanfeng Wang (Shanghai Jiao Tong University)
RecognitionKnowledge DistillationTransformerPrompt EngineeringContrastive LearningVideoMultimodality
🎯 What it does: A dual-branch distillation collaborative framework is proposed, utilizing the complementarity of Classification-Based Pretraining (CBP) and Vision-Language Pretraining (VLP) in weakly supervised temporal action localization, with alternating training to achieve more complete and accurate localization results.
DistilPose: Tokenized Pose Regression With Heatmap Distillation
Suhang Ye (Xiamen University), Rongrong Ji (Xiamen University)
Pose EstimationComputational EfficiencyKnowledge DistillationTransformerGaussian SplattingImage
🎯 What it does: A human pose estimation framework named DistilPose is proposed, which utilizes the knowledge of a heatmap teacher model to enhance the accuracy and efficiency of a regression-based student model.
DistractFlow: Improving Optical Flow Estimation via Realistic Distractions and Pseudo-Labeling
Jisoo Jeong (Qualcomm AI Research), Fatih Porikli (Qualcomm AI Research)
Data-Centric LearningOptical FlowImageVideo
🎯 What it does: A data augmentation method for optical flow training called DistractFlow is proposed, which uses real scene images as distractions to enhance model robustness.
Distribution Shift Inversion for Out-of-Distribution Prediction
Runpeng Yu (National University of Singapore), Xinchao Wang (National University of Singapore)
Domain AdaptationDiffusion modelImage
🎯 What it does: This paper studies a Distribution Shift Inversion (DSI) algorithm that maps test samples back to the training distribution under unseen distributions to improve Out-of-Distribution (OoD) prediction performance.
DisWOT: Student Architecture Search for Distillation WithOut Training
Peijie Dong (National University of Defense Technology), Zimian Wei (National University of Defense Technology)
ClassificationKnowledge DistillationNeural Architecture SearchImage
🎯 What it does: A training-free student architecture search method DisWOT is proposed to find the optimal student network under a given teacher model to enhance knowledge distillation effectiveness.
DivClust: Controlling Diversity in Deep Clustering
Ioannis Maniadis Metaxas (Queen Mary University of London), Ioannis Patras (Queen Mary University of London)
OptimizationImage
🎯 What it does: The DivClust framework is proposed, which can explicitly control the diversity of multiple clustering results in deep clustering and generate a diverse set of base clusters.
Diverse 3D Hand Gesture Prediction From Body Dynamics by Bilateral Hand Disentanglement
Xingqun Qi (University of Technology Sydney), Xin Yu (University of Queensland)
GenerationPose EstimationTransformerAuto EncoderVideo
🎯 What it does: Predicting natural and diverse 3D hand movements based on upper body skeleton
Diverse Embedding Expansion Network and Low-Light Cross-Modality Benchmark for Visible-Infrared Person Re-Identification
Yukang Zhang (Xiamen University), Hanzi Wang (Xiamen University)
RecognitionRetrievalConvolutional Neural NetworkImageMultimodalityBenchmark
🎯 What it does: A DEEN network is proposed to generate diverse embeddings in the embedding space and aggregate multi-layer features, enhancing visible-infrared portrait recognition performance.
Diversity-Aware Meta Visual Prompting
Qidong Huang (University of Science and Technology of China), Nenghai Yu (University of Science and Technology of China)
ClassificationMeta LearningTransformerPrompt EngineeringImage
🎯 What it does: A Diversity-Aware Meta Visual Prompting (DAM-VP) method is proposed, which utilizes clustering to adaptively partition downstream datasets and learn independent prompts for each subset, while using meta-learning pre-trained meta-prompts to accelerate and enhance the transfer performance of frozen pre-trained models on visual tasks.
Diversity-Measurable Anomaly Detection
Wenrui Liu (Institute of Computing Technology Chinese Academy of Sciences), Xilin Chen (Institute of Computing Technology Chinese Academy of Sciences)
Anomaly DetectionAuto EncoderImageVideo
🎯 What it does: A diversity-measurable anomaly detection framework based on reconstruction (DMAD) is proposed, which enhances the reconstruction effect for diverse normal patterns by separating prototype compression from measurable deformation, while accurately assessing the degree of anomalies.
Divide and Adapt: Active Domain Adaptation via Customized Learning
Duojun Huang (Sun Yat-sen University), Guanbin Li (Sun Yat-sen University)
Domain AdaptationImage
🎯 What it does: A Divide and Adapt (DiaNA) framework is proposed, which classifies target domain samples into four categories based on uncertainty and domain discrepancy, and designs customized learning strategies for each category to achieve efficient label utilization in active domain adaptation.
Divide and Conquer: Answering Questions With Object Factorization and Compositional Reasoning
Shi Chen (University of Minnesota), Qi Zhao (University of Minnesota)
ClassificationObject DetectionExplainability and InterpretabilityVision Language ModelImageMultimodality
🎯 What it does: Proposes a visual question answering framework based on object factorization and combinatorial reasoning;
DKM: Dense Kernelized Feature Matching for Geometry Estimation
Johan Edstedt (Linkoping University), Michael Felsberg (Linkoping University)
Depth EstimationConvolutional Neural NetworkImage
🎯 What it does: A fully dense feature matching method DKM is proposed for two-view geometry estimation.
DKT: Diverse Knowledge Transfer Transformer for Class Incremental Learning
Xinyuan Gao (Xi'an Jiaotong University), Yihong Gong (Xi'an Jiaotong University)
ClassificationKnowledge DistillationTransformerImage
🎯 What it does: A Diversified Knowledge Transfer Transformer (DKT) is proposed, which achieves knowledge transfer and catastrophic forgetting suppression in class-incremental learning through task-general and task-specific attention modules and a dual classifier.
DLBD: A Self-Supervised Direct-Learned Binary Descriptor
Bin Xiao (Chongqing University of Posts and Telecommunications), Xinbo Gao (Chongqing University of Posts and Telecommunications)
ClassificationRetrievalContrastive LearningImage
🎯 What it does: A model-agnostic Binary Transformation Layer (BTL) is proposed, which directly learns binary descriptors (DLBD) under a self-supervised framework, and a wide-temperature calibrated cross-entropy loss is designed to enhance the diversity of descriptor distributions.
DNeRV: Modeling Inherent Dynamics via Difference Neural Representation for Videos
Qi Zhao (Nanjing University), Zhan Ma (Nanjing University)
RestorationGenerationCompressionConvolutional Neural NetworkVideo
🎯 What it does: A dual-stream implicit neural representation method based on frame differencing (DNeRV) is proposed, which learns the spatiotemporal dynamics of videos through the collaborative learning of content flow and differential flow;
DNF: Decouple and Feedback Network for Seeing in the Dark
Xin Jin (Nankai University), Chongyi Li (Nanyang Technological University)
RestorationConvolutional Neural NetworkImage
🎯 What it does: A DNF (Decouple and Feedback Network) framework is proposed for low-light enhancement of RAW images, utilizing domain-specific decoupling and feature-level feedback to achieve more accurate denoising and color recovery.
Document Image Shadow Removal Guided by Color-Aware Background
Ling Zhang (Wuhan University of Science and Technology), Chunxia Xiao (Wuhan University)
RestorationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: A document image shadow removal method is proposed, which first estimates the spatially varying background using CBENet, and then uses BGShadowNet to complete shadow removal by combining background information.
Domain Expansion of Image Generators
Yotam Nitzan, Eli Shechtman
GenerationDomain AdaptationGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes the Domain Expansion task, which expands a pre-trained image generator to multiple new domains while maintaining the original generator's performance, achieving multi-domain coexistence.
Domain Generalized Stereo Matching via Hierarchical Visual Transformation
Tianyu Chang (University of Science and Technology of China), Meng Wang (Hefei University of Technology)
Depth EstimationDomain AdaptationConvolutional Neural NetworkImage
🎯 What it does: By applying global, local, and pixel-level visual transformations to synthetic data, the diversity of the training domain is enhanced, thereby improving the generalization ability of the stereo matching network on unseen real domains.
Don't Lie to Me! Robust and Efficient Explainability With Verified Perturbation Analysis
Thomas Fel (Brown University), Thomas Serre (Brown University)
Explainability and InterpretabilityComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: This paper proposes EVA (Explaining using Verified Perturbation Analysis) — a method that utilizes formal error constraints (verified perturbation analysis) to conduct a complete search of the model input space to generate explanations.
DoNet: Deep De-Overlapping Network for Cytology Instance Segmentation
Hao Jiang (Hong Kong University of Science and Technology), Hao Chen (Hong Kong University of Science and Technology)
Object DetectionSegmentationConvolutional Neural NetworkImage
🎯 What it does: A deep de-overlapping network called DoNet is proposed, which implements cell instance segmentation using a decomposition-recombination strategy to address overlapping cells and artifact interference.
Doubly Right Object Recognition: A Why Prompt for Visual Rationales
Chengzhi Mao (Columbia University), Carl Vondrick (Columbia University)
RecognitionObject DetectionRetrievalExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageTextBenchmarkChain-of-Thought
🎯 What it does: This paper proposes the 'doubly right' object recognition task, which requires the model to provide both the correct category and a corresponding rationalization. It constructs a large-scale benchmark through automated generation of reasons and visual retrieval.
DP-NeRF: Deblurred Neural Radiance Field With Physical Scene Priors
Dogyoon Lee (Yonsei University), Sangyoun Lee (Korea Institute of Science and Technology)
RestorationNeural Radiance FieldImage
🎯 What it does: A DP-NeRF framework based on physical scene priors is proposed to recover consistent 3D neural radiance fields from blurred images.
DPE: Disentanglement of Pose and Expression for General Video Portrait Editing
Youxin Pang (Institute of Automation, Chinese Academy of Sciences), Dong-Ming Yan (Institute of Automation, Chinese Academy of Sciences)
Image TranslationGenerationPose EstimationGenerative Adversarial NetworkVideo
🎯 What it does: A novel unsupervised self-supervised framework is proposed, which can decouple pose and expression in the latent space of videos, enabling independent editing of pose and expression in portrait videos.
DPF: Learning Dense Prediction Fields With Weak Supervision
Xiaoxue Chen (Tsinghua University), Ya-Qin Zhang (Tsinghua University)
SegmentationConvolutional Neural NetworkTransformerImage
🎯 What it does: A dense prediction field (DPF) model based on implicit neural functions is proposed, utilizing point-level weakly supervised learning for dense prediction tasks such as semantic segmentation and intrinsic image decomposition.
DR2: Diffusion-Based Robust Degradation Remover for Blind Face Restoration
Zhixin Wang (Shanghai Jiao Tong University), Yanfeng Wang (Shanghai Jiao Tong University)
RestorationDiffusion modelImage
🎯 What it does: A robust degradation remover DR2 based on a diffusion model is proposed for blind face restoration, which first converts the degraded image into a degradation-insensitive rough result, and then uses an enhancement module to restore a high-quality image.
DrapeNet: Garment Generation and Self-Supervised Draping
Luca De Luigi (University of Bologna), Pascal Fua (EPFL)
GenerationGraph Neural NetworkGenerative Adversarial NetworkMesh
🎯 What it does: A full-process pipeline based on physical self-supervision is proposed, capable of generating various garments through a single network and fitting fabrics on different body shapes and poses.
Dream3D: Zero-Shot Text-to-3D Synthesis Using 3D Shape Prior and Text-to-Image Diffusion Models
Jiale Xu, Shenghua Gao (Tencent)
GenerationData SynthesisOptimizationDiffusion modelNeural Radiance FieldPoint CloudMesh
🎯 What it does: This paper proposes a zero-shot text-to-3D synthesis framework named Dream3D, which first generates rendered style images using a fine-tuned text-to-image diffusion model, and then obtains high-quality 3D shapes as priors through an image-to-shape network. This shape is then used to initialize a neural radiance field, which is optimized under the guidance of CLIP, ultimately generating 3D content that is consistent with the text and accurately shaped.
DreamBooth: Fine Tuning Text-to-Image Diffusion Models for Subject-Driven Generation
Nataniel Ruiz (Google Research), Kfir Aberman (Google Research)
GenerationData SynthesisSupervised Fine-TuningDiffusion modelImage
🎯 What it does: Given a small number (3-5 images) of subject images, the text-to-image diffusion model is fine-tuned to generate high-quality new images of the subject in various scenes, poses, and lighting conditions.
DropKey for Vision Transformer
Bonan Li (University of Chinese Academy of Sciences), Luoqi Liu (Meitu Inc.)
ClassificationObject DetectionPose EstimationTransformerSupervised Fine-TuningImage
🎯 What it does: A new dropout method for self-attention layers called DropKey is proposed and implemented, which randomly drops keys before calculating the attention matrix, and provides a layer-wise decreasing dropout ratio schedule, while verifying that structured dropout does not significantly improve ViT.
DropMAE: Masked Autoencoders With Spatial-Attention Dropout for Tracking Tasks
Qiangqiang Wu (City University of Hong Kong), Antoni B. Chan (City University of Hong Kong)
Object TrackingSegmentationTransformerAuto EncoderVideo
🎯 What it does: This paper proposes self-supervised pre-training of Masked Autoencoder (MAE) on videos, aiming to provide better temporal matching representations for Visual Object Tracking (VOT) and Video Object Segmentation (VOS);
DSFNet: Dual Space Fusion Network for Occlusion-Robust 3D Dense Face Alignment
Heyuan Li (National University of Singapore), Robby T. Tan (National University of Singapore)
RecognitionPose EstimationConvolutional Neural NetworkImage
🎯 What it does: Proposes DSFNet, which combines predictions from both image space and model space to achieve occlusion-robust 3D dense face alignment.
DSVT: Dynamic Sparse Voxel Transformer With Rotated Sets
Haiyang Wang (Peking University), Liwei Wang (Peking University)
Object DetectionAutonomous DrivingTransformerPoint Cloud
🎯 What it does: A deployable Transformer backbone DSVT for sparse point clouds is proposed, supporting efficient dynamic sparse window attention and 3D attention-based pooling.
Dual Alignment Unsupervised Domain Adaptation for Video-Text Retrieval
Xiaoshuai Hao (Institute of Information Engineering, Chinese Academy of Sciences), Bo Li (Institute of Information Engineering, Chinese Academy of Sciences)
RetrievalDomain AdaptationContrastive LearningVideoText
🎯 What it does: Proposes the DADA framework for unsupervised domain adaptive video-text retrieval;
Dual-Bridging With Adversarial Noise Generation for Domain Adaptive rPPG Estimation
Jingda Du (Hong Kong Baptist University), Pong C. Yuen (Hong Kong Baptist University)
Domain AdaptationGenerative Adversarial NetworkVideo
🎯 What it does: This paper proposes a dual-bridging network for domain-adaptive remote photoplethysmography (rPPG) estimation in the target domain, achieved through a noise reducer and a noise generator.
Dual-Path Adaptation From Image to Video Transformers
Jungin Park (Yonsei University), Kwanghoon Sohn (Korea Institute of Science and Technology)
RecognitionDomain AdaptationTransformerPrompt EngineeringVideo
🎯 What it does: A dual-path parameter-efficient transfer framework called DualPath is proposed, utilizing a frozen pre-trained image Transformer (ViT/Swin) to achieve video action recognition through spatial and temporal path adaptation.
DualRefine: Self-Supervised Depth and Pose Estimation Through Iterative Epipolar Sampling and Refinement Toward Equilibrium
Antyanta Bangunharcana (Korea Advanced Institute of Science and Technology), Kyung-Soo Kim (Korea Advanced Institute of Science and Technology)
Pose EstimationDepth EstimationAutonomous DrivingSupervised Fine-TuningSimultaneous Localization and MappingImage
🎯 What it does: This work proposes a self-supervised multi-frame depth and pose estimation framework called DualRefine, which refines both depth and camera pose simultaneously using local epipolar line sampling matching and iterative updates.
DualRel: Semi-Supervised Mitochondria Segmentation From a Prototype Perspective
Huayu Mai (University of Science and Technology of China), Feng Wu (University of Science and Technology of China)
SegmentationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes DualRel, a dual-reliable network for semi-supervised mitochondrial segmentation, which constructs reliable prototype-level consistency regularization from a prototype perspective, significantly reducing the misleading effects of unlabeled data.
DualVector: Unsupervised Vector Font Synthesis With Dual-Part Representation
Ying-Tian Liu (Tsinghua University), Song-Hai Zhang (Adobe)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: This paper studies an unsupervised vector font synthesis method called DualVector, which can generate high-quality vector glyphs solely through bitmap training, enabling font reconstruction and few-shot font generation.
DyLiN: Making Light Field Networks Dynamic
Heng Yu (Carnegie Mellon University), László A. Jeni (Carnegie Mellon University)
GenerationComputational EfficiencyKnowledge DistillationNeural Radiance FieldVideo
🎯 What it does: Two dynamic light field networks (DyLiN and CoDyLiN) are proposed, achieving high-quality real-time rendering of non-rigid deformation and topological change scenes by learning ray deformation fields and high-dimensional space encoding.
DynaFed: Tackling Client Data Heterogeneity With Global Dynamics
Renjie Pi (Hong Kong University of Science and Technology), Qifeng Chen
Federated LearningImage
🎯 What it does: The DYNAFED framework is proposed, which synthesizes pseudo data on the server side using the training trajectory of the global model to correct model drift caused by client data heterogeneity, thereby improving the convergence speed and final performance of federated learning.
DynaMask: Dynamic Mask Selection for Instance Segmentation
Ruihuang Li (Hong Kong Polytechnic University), Lei Zhang (Hong Kong Polytechnic University)
Object DetectionSegmentationConvolutional Neural NetworkImage
🎯 What it does: A dynamic mask selection framework called DynaMask is designed to enhance instance segmentation quality through a dual-layer FPN, dynamically selecting the most suitable mask resolution for each instance.