ICCV 2023 Papers — Page 6
IEEE/CVF International Conference on Computer Vision · 2156 papers
Diverse Inpainting and Editing with GAN Inversion
Ahmet Burak Yildirim (Bilkent University), Aysegul Dundar (Bilkent University)
RestorationGenerationGenerative Adversarial NetworkImage
🎯 What it does: A GAN-based framework is proposed that can achieve diverse inpainting and editing on an image with missing regions.
Divide and Conquer: 3D Point Cloud Instance Segmentation With Point-Wise Binarization
Weiguang Zhao (Duke Kunshan University), Kaizhu Huang (Xi'an Jiaotong-Liverpool University)
Object DetectionSegmentationConvolutional Neural NetworkPoint Cloud
🎯 What it does: An end-to-end 3D point cloud instance segmentation network PBNet is proposed, which utilizes point density binarization and local scene reconstruction to achieve finer instance segmentation.
Divide and Conquer: a Two-Step Method for High Quality Face De-identification with Model Explainability
Yunqian Wen (Shanghai Jiao Tong University), Li Song (Shanghai Jiao Tong University)
RestorationGenerationPose EstimationSafty and PrivacyConvolutional Neural NetworkNeural Radiance FieldGenerative Adversarial NetworkImage
🎯 What it does: A two-step facial de-identification method called IDeudemon is proposed, which first perturbs the identity code using 3D NeRF and then restores details using GAN.
Divide&Classify: Fine-Grained Classification for City-Wide Visual Geo-Localization
Gabriele Trivigno (Politecnico di Torino), Carlo Masone (Politecnico di Torino)
ClassificationRetrievalConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper proposes a framework that transforms the fine-grained urban visual localization task into a classification problem, designing a partitioning strategy and multi-classifier combination inference pipeline for densely sampled maps.
DLGSANet: Lightweight Dynamic Local and Global Self-Attention Networks for Image Super-Resolution
Xiang Li (Nanjing University of Science and Technology), Jinshan Pan (Nanjing University of Science and Technology)
RestorationSuper ResolutionTransformerImage
🎯 What it does: A lightweight dynamic local and global self-attention network, DLGSANet, is proposed for single image super-resolution.
DLT: Conditioned layout generation with Joint Discrete-Continuous Diffusion Layout Transformer
Elad Levi (Wix.com), Meir Perez (Wix.com)
GenerationTransformerDiffusion modelImage
🎯 What it does: A joint discrete-continuous diffusion Transformer (DLT) is designed to generate or edit graphic layouts based on any subset constraints (category, position, size) provided by the user.
DMNet: Delaunay Meshing Network for 3D Shape Representation
Chen Zhang (Huazhong University of Science and Technology), Wenbing Tao (Huazhong University of Science and Technology)
GenerationRepresentation LearningGraph Neural NetworkPoint CloudMesh
🎯 What it does: This paper proposes a learning-based surface reconstruction method DMNet based on Delaunay triangulation, which can achieve high-precision, closed mesh reconstruction of point clouds without the need for visibility constraints.
DNA-Rendering: A Diverse Neural Actor Repository for High-Fidelity Human-Centric Rendering
Wei Cheng (Shanghai AI Laboratory), Kwan-Yee Lin (CUHK)
GenerationPose EstimationNeural Radiance FieldImageVideoBenchmark
🎯 What it does: This paper presents DNA-Rendering, a large-scale dataset covering over 1500 human bodies, 5000 actions, 67.5M frames, and high resolution (4096×3000) multi-view data, accompanied by complete annotations (SMPLX, keypoints, foreground masks, etc.), providing a rich and diverse baseline for high-fidelity human rendering.
Do DALL-E and Flamingo Understand Each Other?
Hang Li (Ludwig Maximilian University of Munich), Volker Tresp (Siemens AG)
GenerationData SynthesisTransformerVision Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: This paper proposes a reconstruction task through image→text→image and text→image→text to examine the mutual understanding between image generation models and image description models, and builds a joint fine-tuning framework to enhance the performance of both types of models.
DocTr: Document Transformer for Structured Information Extraction in Documents
Haofu Liao (Amazon), Vijay Mahadevan (AWS AI Labs)
Object DetectionTransformerVision Language ModelTextMultimodality
🎯 What it does: This paper proposes an entity representation based on anchor words and bounding boxes, modeling structural information extraction as an anchor word detection and association problem, and develops a multimodal Transformer (DocTr) for end-to-end entity detection and linking.
Document Understanding Dataset and Evaluation (DUDE)
Jordy Van Landeghem (KU Leuven), Tomasz Stanislawek
TransformerSupervised Fine-TuningVision Language ModelImageTextMultimodalityBenchmarkFinance Related
🎯 What it does: A multi-page, multi-domain, multi-industry document visual question answering benchmark (DUDE) has been created, and a systematic evaluation of existing models on this benchmark has been conducted.
Does Physical Adversarial Example Really Matter to Autonomous Driving? Towards System-Level Effect of Adversarial Object Evasion Attack
Ningfei Wang (University of California), Qi Alfred Chen (University of California)
Autonomous DrivingAdversarial AttackVideo
🎯 What it does: In autonomous driving systems, a measurement study of the system-level effects of physical adversarial object evasion attacks is conducted, and a system model-based attack design (SysAdv) is proposed to enhance the system-level violation rate.
DOLCE: A Model-Based Probabilistic Diffusion Framework for Limited-Angle CT Reconstruction
Jiaming Liu (Washington University in St. Louis), Hyojin Kim (Lawrence Livermore National Laboratory)
RestorationGenerationDiffusion modelImageBiomedical DataComputed Tomography
🎯 What it does: Proposed the DOLCE model, which combines conditional diffusion and forward measurement consistency to achieve limited-angle CT reconstruction;
Domain Adaptive Few-Shot Open-Set Learning
Debabrata Pal (Indian Institute of Technology Bombay), Biplab Banerjee (Indian Institute of Technology Bombay)
ClassificationDomain AdaptationMeta LearningConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes the Domain Adaptation Few-Shot Open Set Learning (DA-FSOS) task and designs an end-to-end model DAFOS-NET, which can simultaneously recognize known categories and reject unknown samples under the condition of fully supervised source domain and extremely few supervised target domain.
Domain Generalization Guided by Gradient Signal to Noise Ratio of Parameters
Mateusz Michalkiewicz (University of Queensland), Mahsa Baktashmotlagh (University of Queensland)
ClassificationDomain AdaptationMeta LearningConvolutional Neural NetworkImage
🎯 What it does: A DropBlock regularization method based on Gradient Signal-to-Noise Ratio (GSNR) is proposed, which automatically learns the dropout ratio for each network block through meta-learning to enhance the model's robustness in domain generalization tasks.
Domain Generalization of 3D Semantic Segmentation in Autonomous Driving
Jules Sanchez (Mines Paris - PSL, PSL University), François Goulette (ENSTA Paris, Institut Polytechnique de Paris)
SegmentationDomain AdaptationAutonomous DrivingConvolutional Neural NetworkPoint CloudBenchmark
🎯 What it does: A 3DLabelProp method based on the accumulation of LiDAR point cloud sequences and geometric label propagation is proposed, and the first domain generalization benchmark for 3D semantic segmentation is established.
Domain Generalization via Balancing Training Difficulty and Model Capability
Xueying Jiang (Nanyang Technological University), Shijian Lu (Nanyang Technological University)
Object DetectionSegmentationDomain AdaptationImage
🎯 What it does: A MoDify framework based on momentum difficulty is proposed, which dynamically adjusts the matching of training sample difficulty and model capability to address underfitting and overfitting issues in domain generalization training.
Domain Generalization via Rationale Invariance
Liang Chen (Tencent AI Lab), Lingqiao Liu (University of Adelaide)
ClassificationDomain AdaptationExplainability and InterpretabilityConvolutional Neural NetworkImageBenchmark
🎯 What it does: This paper introduces a 'rationale' matrix in domain generalization, enforcing consistent decision contributions among similar samples, thereby enhancing the model's robustness to unknown domains.
Domain Specified Optimization for Deployment Authorization
Haotian Wang (National University of Defense Technology), Dacheng Tao (University of Sydney)
Domain AdaptationOptimizationImageBenchmark
🎯 What it does: This paper proposes two new IP protection tasks for dataset centers: Source Domain Only Deployment Authorization (SDPA) and Target Combined Deployment Authorization (TDPA), and presents lightweight Domain-Specific Optimization (DSO) and Target Domain-Specific Optimization (TDSO) methods.
Domain-Specificity Inducing Transformers for Source-Free Domain Adaptation
Sunandini Sanyal (Indian Institute of Science), R Venkatesh Babu (Indian Institute of Science)
Domain AdaptationTransformerImage
🎯 What it does: The Domain-Specificity Inducing Transformer (DSiT) framework is proposed, utilizing the Query of the Transformer to separate and learn domain-specific and task-specific factors;
DomainAdaptor: A Novel Approach to Test-time Adaptation
Jian Zhang (Nanjing University), Yang Gao (Nanjing University)
Domain AdaptationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: In response to the domain shift between training and testing, a novel testing-time adaptive framework called DomainAdaptor is proposed, which can quickly adapt a pre-trained CNN using only the unlabeled test batch without accessing the source data.
DomainDrop: Suppressing Domain-Sensitive Channels for Domain Generalization
Jintao Guo (Nanjing University), Yinghuan Shi (Nanjing University)
Domain AdaptationConvolutional Neural NetworkImage
🎯 What it does: The DomainDrop framework is proposed, which suppresses channels in the source domain that are susceptible to domain shift through channel dropout guided by a domain discriminator during training, thereby enhancing the model's generalization ability to unseen target domains.
Doppelgangers: Learning to Disambiguate Images of Similar Structures
Ruojin Cai (Cornell University), Noah Snavely (Cornell University)
ClassificationRecognitionConvolutional Neural NetworkSimultaneous Localization and MappingImage
🎯 What it does: This paper proposes a visual ambiguity discrimination method based on a binary classification network, which automatically determines whether two similar images correspond to the same 3D surface and integrates it into the SfM process to avoid erroneous reconstructions.
DOT: A Distillation-Oriented Trainer
Borui Zhao (MEGVII Technology), Jiajun Liang (MEGVII Technology)
ClassificationOptimizationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: This study investigates the optimization process of knowledge distillation and finds that introducing distillation loss leads to a trade-off between task loss and distillation loss. It proposes the Distillation-Oriented Trainer (DOT), which sets different momenta for distillation loss and task loss, allowing the distillation loss to dominate the optimization, thereby reducing both losses simultaneously and achieving flatter, better generalizing minima.
Downscaled Representation Matters: Improving Image Rescaling with Collaborative Downscaled Images
Bingna Xu (South China University of Technology), Jian Chen (South China University of Technology)
RestorationSuper ResolutionOptimizationImage
🎯 What it does: Proposes a Hierarchical Collaborative Downsampling (HCD) method that directly optimizes gradients on low-resolution images to enhance image reconstruction quality.
Downstream-agnostic Adversarial Examples
Ziqi Zhou (Huazhong University of Science and Technology), Hai Jin (Huazhong University of Science and Technology)
Adversarial AttackGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: A downstream agnostic universal adversarial attack framework AdvEncoder based on a pre-trained self-supervised encoder is proposed, capable of generating perturbations or sticker attacks for any downstream task using the encoder.
DPF-Net: Combining Explicit Shape Priors in Deformable Primitive Field for Unsupervised Structural Reconstruction of 3D Objects
Qingyao Shuai (University of Science and Technology of China), Xuejin Chen (University of Science and Technology of China)
RestorationSegmentationGenerationConvolutional Neural NetworkPoint Cloud
🎯 What it does: This paper proposes an unsupervised structured 3D reconstruction framework called DPF-Net, which combines a Parameterized Primitive Field with an implicit deformation field to achieve part-level structured reconstruction and detail recovery of target objects.
DPM-OT: A New Diffusion Probabilistic Model Based on Optimal Transport
Zezeng Li (Dalian University of Technology), David Xianfeng Gu (State University of New York at Stony Brook)
GenerationData SynthesisComputational EfficiencyDiffusion modelImage
🎯 What it does: A fast diffusion probability model (DPM-OT) is proposed, which combines semi-discrete optimal transport. It directly maps the Gaussian prior to the intermediate latent space in one step, and then generates samples using a small number of reverse diffusion steps.
DPS-Net: Deep Polarimetric Stereo Depth Estimation
Chaoran Tian (Zhejiang University), Zhaopeng Cui (Zhejiang University)
Depth EstimationConvolutional Neural NetworkRecurrent Neural NetworkImage
🎯 What it does: This paper proposes an end-to-end DPS-Net for estimating depth from stereo polarized images and RGB images.
DQS3D: Densely-matched Quantization-aware Semi-supervised 3D Detection
Huan-ang Gao (Tsinghua University), Guyue Zhou (Tsinghua University)
Object DetectionConvolutional Neural NetworkPoint Cloud
🎯 What it does: A framework DQS3D is proposed for semi-supervised 3D detection using dense matching and online compensation for quantization errors.
DR-Tune: Improving Fine-tuning of Pretrained Visual Models by Distribution Regularization with Semantic Calibration
Nan Zhou (Beihang University), Di Huang (Beihang University)
ClassificationSupervised Fine-TuningImage
🎯 What it does: A new framework called DR-Tune is proposed, which utilizes pre-trained feature distribution for distribution regularization and semantic calibration during fine-tuning.
DRAW: Defending Camera-shooted RAW Against Image Manipulation
Xiaoxiao Hu (Fudan University), Xinpeng Zhang (Fudan University)
Safty and PrivacyConvolutional Neural NetworkImage
🎯 What it does: Embed an invisible protection signal in RAW images, so that the rendered RGB images carry this signal, allowing for accurate localization of tampered areas in subsequent modifications.
DREAM: Efficient Dataset Distillation by Representative Matching
Yanqing Liu (National University of Singapore), Yang You (National University of Singapore)
Computational EfficiencyKnowledge DistillationImage
🎯 What it does: This paper proposes the DREAM (Dataset distillation by REpresentative Matching) method, which utilizes representative samples for gradient matching to achieve efficient dataset distillation.
DreamBooth3D: Subject-Driven Text-to-3D Generation
Amit Raj, Varun Jampani (Google)
GenerationData SynthesisDiffusion modelNeural Radiance FieldImage
🎯 What it does: The DreamBooth3D method is proposed, which can generate personalized 3D assets using only 3-6 images without pose information, and supports text-driven modifications of pose, accessories, and color.
DreamPose: Fashion Video Synthesis with Stable Diffusion
Johanna Karras (University of Washington), Ira Kemelmacher-Shlizerman (NVIDIA)
GenerationData SynthesisPose EstimationDiffusion modelImageVideo
🎯 What it does: The DreamPose method based on diffusion models can synthesize realistic animated videos from a single fashion photo and a sequence of poses, maintaining both the details of the person and the clothing while achieving natural movement of the garments.
DreamTeacher: Pretraining Image Backbones with Deep Generative Models
Daiqing Li (University of Toronto), Sanja Fidler (University of Toronto)
Object DetectionSegmentationKnowledge DistillationRepresentation LearningConvolutional Neural NetworkDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes the DreamTeacher framework, which achieves unsupervised self-supervised pre-training by distilling knowledge from a pre-trained generative model to the target image backbone.
DREAMWALKER: Mental Planning for Continuous Vision-Language Navigation
Hanqing Wang (Beijing Institute of Technology), Wenguan Wang (Zhejiang University)
OptimizationExplainability and InterpretabilityRobotic IntelligenceRecurrent Neural NetworkGraph Neural NetworkReinforcement LearningVision Language ModelWorld ModelMultimodality
🎯 What it does: This paper studies a vision-language navigation (VLN-CE) agent called DREAMWALKER, which can perform 'mental planning' in an internally abstract discrete environment and then map the best plan to low-level actions in a real continuous environment.
DReg-NeRF: Deep Registration for Neural Radiance Fields
Yu Chen (National University of Singapore), Gim Hee Lee (National University of Singapore)
GenerationData SynthesisTransformerNeural Radiance FieldPoint Cloud
🎯 What it does: This paper proposes a deep learning framework called DReg-NeRF, which requires no manual labeling or initialization, to align multiple NeRF blocks trained in different coordinate systems to a common global coordinate system.
DriveAdapter: Breaking the Coupling Barrier of Perception and Planning in End-to-End Autonomous Driving
Xiaosong Jia (Shanghai Jiao Tong University), Hongyang Li (Shanghai Jiao Tong University)
Autonomous DrivingConvolutional Neural NetworkReinforcement LearningImageVideoPoint Cloud
🎯 What it does: Proposes the DriveAdapter framework, which directly uses a frozen RL teacher model for planning in end-to-end autonomous driving, with the student model only responsible for perception. It aligns features through a learnable Adapter and incorporates masked feature alignment and action guidance loss to alleviate distribution differences and teacher errors.
DS-Fusion: Artistic Typography via Discriminated and Stylized Diffusion
Maham Tanveer (Simon Fraser University), Hao Zhang (Simon Fraser University)
GenerationConvolutional Neural NetworkDiffusion modelImageText
🎯 What it does: Automatically generate artistic fonts, integrating the semantics of the input words with the font shapes while maintaining readability;
Dual Aggregation Transformer for Image Super-Resolution
Zheng Chen (Shanghai Jiao Tong University), Fisher Yu (ETH Zürich)
RestorationSuper ResolutionTransformerImage
🎯 What it does: A Dual-Axis Aggregation Transformer (DAT) is proposed for single image super-resolution, capable of aggregating spatial and channel features simultaneously, achieving stronger representation ability.
Dual Learning with Dynamic Knowledge Distillation for Partially Relevant Video Retrieval
Jianfeng Dong (Zhejiang Gongshang University), Baolong Liu (Zhejiang Gongshang University)
RetrievalKnowledge DistillationConvolutional Neural NetworkTransformerVision Language ModelContrastive LearningVideoText
🎯 What it does: A dual-branch student network (inheritance branch and exploration branch) based on the large-scale visual language pre-training model CLIP is proposed for dynamic knowledge distillation to address the problem of Partial Relevant Video Retrieval (PRVR).
Dual Meta-Learning with Longitudinally Consistent Regularization for One-Shot Brain Tissue Segmentation Across the Human Lifespan
Yongheng Sun (Xi'an Jiaotong University), Chunfeng Lian (Xi'an Jiaotong University)
SegmentationMeta LearningConvolutional Neural NetworkImageBiomedical DataMagnetic Resonance ImagingAlzheimer's Disease
🎯 What it does: A dual meta-learning framework, DuMeta, is proposed to learn a brain tissue segmentation model that is consistent across the lifecycle, achieving high-precision segmentation across different age groups with just one labeled image.
Dual Pseudo-Labels Interactive Self-Training for Semi-Supervised Visible-Infrared Person Re-Identification
Jiangming Shi (Xiamen University), Yanyun Qu (Xiamen University)
RecognitionRetrievalConvolutional Neural NetworkContrastive LearningImageMultimodality
🎯 What it does: This paper proposes a Dual Pseudo-Label Interactive Self-Training (DPIS) framework for visible-infrared person re-identification (VI-ReID), which is trained in two semi-supervised scenarios: one with only visible images labeled (uni-semi) and the other with partial labeling in both modalities (bi-semi).
DVGaze: Dual-View Gaze Estimation
Yihua Cheng (Beihang University), Feng Lu (Beihang University)
RecognitionPose EstimationConvolutional Neural NetworkTransformerImage
🎯 What it does: A gaze estimation network called DV-Gaze is proposed, which can directly predict the gaze direction from a pair of camera images under a dual-camera perspective.
DVIS: Decoupled Video Instance Segmentation Framework
Tao Zhang (Wuhan University), Pengfei Wan (Kuaishou Technology)
Object TrackingSegmentationTransformerVideo
🎯 What it does: The DVIS framework is proposed, which breaks down the video instance segmentation task into three main sub-tasks: segmentation, tracking, and refinement. A lightweight referring tracker and temporal refiner are designed for each sub-task.
DyGait: Exploiting Dynamic Representations for High-performance Gait Recognition
Ming Wang, Xin Yu
RecognitionComputational EfficiencyConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper proposes a new method to address specific tasks in computer vision, aiming to improve the performance and efficiency of models.
Dynamic Hyperbolic Attention Network for Fine Hand-object Reconstruction
Zhiying Leng (Beihang University), Federico Tombari (Technical University of Munich)
Object DetectionPose EstimationGraph Neural NetworkImagePoint CloudMesh
🎯 What it does: Utilizing a dual-branch network to first roughly estimate the 3D mesh of the hand and object from a single RGB image, and then finely reconstructing the hand and object in hyperbolic space through dynamic hyperbolic graph convolution and image attention hyperbolic graph convolution, capturing geometric and multimodal information and modeling interactions.
Dynamic Mesh Recovery from Partial Point Cloud Sequence
Hojun Jang (Seoul National University), Young Min Kim (Seoul National University)
GenerationData SynthesisPose EstimationTransformerAuto EncoderPoint CloudMesh
🎯 What it does: By learning the spatiotemporal priors of large-scale motion data and utilizing CVAE+Transformer to generate complete 3D dynamic meshes, it is possible to recover human/hand meshes from noisy, locally or temporally missing point cloud sequences.
Dynamic Mesh-Aware Radiance Fields
Yi-Ling Qiao (University of Maryland), Ming C. Lin (University of Maryland)
GenerationComputational EfficiencyNeural Radiance FieldMesh
🎯 What it does: A unified rendering and simulation pipeline has been constructed, enabling the coupling of NeRF scenes and polygon meshes within the same physical space, and achieving real-time rendering and dynamic simulation on the GPU.
Dynamic Perceiver for Efficient Visual Recognition
Yizeng Han (Tsinghua University), Gao Huang (Tsinghua University)
ClassificationRecognitionObject DetectionComputational EfficiencyTransformerImage
🎯 What it does: A dual-branch Dynamic Perceiver framework is designed to achieve dynamic early exit to enhance the inference efficiency of visual models.
Dynamic PlenOctree for Adaptive Sampling Refinement in Explicit NeRF
Haotian Bai (Hong Kong University of Science and Technology), Lin Wang (Hong Kong University of Science and Technology)
CompressionOptimizationNeural Radiance FieldImage
🎯 What it does: Improved the existing PlenOctree (POT) model and proposed the Dynamic PlenOctree (DOT), which dynamically optimizes the octree structure through training signal-driven sampling and pruning, combined with hierarchical feature fusion.
Dynamic Point Fields
Sergey Prokudin (ETH Zurich), Siyu Tang (ETH Zurich)
GenerationOptimizationPoint Cloud
🎯 What it does: This paper proposes a dynamic point field model that combines explicit point clouds with implicit deformation networks to achieve efficient non-rigid 3D surface modeling.
Dynamic Residual Classifier for Class Incremental Learning
Xiuwei Chen (Sun Yat-sen University), Xiaobin Chang (Sun Yat-sen University)
ClassificationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a Dynamic Residual Classifier (DRC) to address the dynamically worsening sample imbalance problem in Class Incremental Learning (CIL) as tasks increase, and integrates it with three mainstream CIL processes (MDT, MEC, MAF);
Dynamic Snake Convolution Based on Topological Geometric Constraints for Tubular Structure Segmentation
Yaolei Qi (Southeast University), Guanyu Yang (Southeast University)
SegmentationConvolutional Neural NetworkBiomedical Data
🎯 What it does: This paper proposes a segmentation network specifically designed for tubular structures (such as blood vessels and roads) called DSCNet. It combines dynamic snake-like convolution, multi-view feature fusion, and topology continuity constraint loss based on persistent homology, significantly improving the segmentation accuracy and connectivity of tubular structures.
Dynamic Token Pruning in Plain Vision Transformers for Semantic Segmentation
Quan Tang (South China University of Technology), Yifan Liu (University of Adelaide)
SegmentationComputational EfficiencyTransformerImage
🎯 What it does: A semantic segmentation visual Transformer method based on Dynamic Token Pruning (DToP) is proposed, which evaluates the difficulty of each token through an intermediate auxiliary head and terminates the forward propagation of easy tokens in advance, thereby reducing computational power.
DynamicISP: Dynamically Controlled Image Signal Processor for Image Recognition
Masakazu Yoshimura (Sony Group Corporation), Takeshi Ohashi (Sony Group Corporation)
RecognitionObject DetectionImage
🎯 What it does: This paper proposes DynamicISP, a framework that dynamically adjusts ISP parameters based on feedback from downstream recognition models to improve image recognition accuracy.
DynaMITe: Dynamic Query Bootstrapping for Multi-object Interactive Segmentation Transformer
Amit Kumar Rana (RWTH Aachen University), Bastian Leibe (RWTH Aachen University)
Object DetectionSegmentationTransformerImage
🎯 What it does: This paper proposes a Transformer-based interactive segmentation framework called DynaMITe, which can handle multiple targets at once and dynamically generate queries through user clicks, achieving efficient iterative refinement for multiple instances.
E^2VPT: An Effective and Efficient Approach for Visual Prompt Tuning
Cheng Han (Purdue University), Dongfang Liu (Rochester Institute of Technology)
ClassificationRecognitionOptimizationComputational EfficiencyTransformerPrompt EngineeringImage
🎯 What it does: This paper proposes a new visual prompt tuning method called E-VPT, which adds visual prompts and key-value prompts to the input layer and self-attention layer of the Transformer model, respectively, and prunes the prompts layer by layer to achieve efficient parameter tuning.
E2E-LOAD: End-to-End Long-form Online Action Detection
Shuqiang Cao (Shandong University), Lin Ma (Meituan)
RecognitionObject DetectionTransformerVideo
🎯 What it does: An end-to-end Transformer framework called E2E-LOAD is proposed for online action detection in long videos.
E2NeRF: Event Enhanced Neural Radiance Fields from Blurry Images
Yunshan Qi (Beihang University), Jia Li (Beihang University)
RestorationPose EstimationNeural Radiance FieldImageVideo
🎯 What it does: Combine event cameras and RGB cameras to train a model that can recover clear 3D neural radiance fields (NeRF) from blurry inputs using event streams and blurred images.
E3Sym: Leveraging E(3) Invariance for Unsupervised 3D Planar Reflective Symmetry Detection
Ren-Wu Li (Institute of Computing Technology, Chinese Academy of Sciences), Lin Gao (Institute of Computing Technology, Chinese Academy of Sciences)
RecognitionSegmentationPoint Cloud
🎯 What it does: This paper proposes an unsupervised, end-to-end 3D global planar symmetry detection method called E3Sym, which constructs correspondences using E(3) invariant features and then obtains symmetric planes through differentiable clustering.
EdaDet: Open-Vocabulary Object Detection Using Early Dense Alignment
Cheng Shi (ShanghaiTech University), Sibei Yang (ShanghaiTech University)
Object DetectionTransformerVision Language ModelImage
🎯 What it does: This paper proposes an open vocabulary object detection framework called EdaDet based on Early Dense Alignment, which can detect new category objects with training only on base class labels.
EDAPS: Enhanced Domain-Adaptive Panoptic Segmentation
Suman Saha (ETH Zurich), Luc Van Gool (ETH Zurich)
SegmentationDomain AdaptationTransformerImage
🎯 What it does: A framework called EDAPS specifically designed for domain adaptive panoptic segmentation is proposed, achieving end-to-end training on a dataset transitioning from synthetic to real scenes.
Editable Image Geometric Abstraction via Neural Primitive Assembly
Ye Chen (Shanghai Jiao Tong University), Zhangli Hu (Shanghai Jiao Tong University)
GenerationRetrievalTransformerContrastive LearningImage
🎯 What it does: This paper proposes an unsupervised image geometric abstraction framework that utilizes only four basic parametric primitives (triangle, rectangle, circle, semicircle) to predict primitive types, transformations, and colors through a Transformer, and achieves primitive assignment and photometric fusion via neural soft assignment.
Editing Implicit Assumptions in Text-to-Image Diffusion Models
Hadas Orgad (Technion), Yonatan Belinkov (Technion)
GenerationData SynthesisTransformerDiffusion modelText
🎯 What it does: Edit the implicit assumptions of the pre-trained text-to-image diffusion model to better align with user needs.
Effective Real Image Editing with Accelerated Iterative Diffusion Inversion
Zhihong Pan (Oppo Mobile Telecommunications Corporation), Stephen Huang (Oppo Mobile Telecommunications Corporation)
Image TranslationRestorationGenerationDiffusion modelImage
🎯 What it does: An accelerated iterative diffusion inversion method (AIDI) is proposed, which achieves high-quality real image editing by combining mixed guidance and random editing.
Efficient 3D Semantic Segmentation with Superpoint Transformer
Damien Robert (ENGIE Lab CRIGEN), Loic Landrieu (Universite Gustave Eiffel)
SegmentationAutonomous DrivingComputational EfficiencyTransformerPoint Cloud
🎯 What it does: A Transformer network based on a hierarchical superpoint structure (Superpoint Transformer, SPT) has been designed and implemented for semantic segmentation of large-scale point clouds. By utilizing fast hierarchical superpoint segmentation and sparse self-attention, the model significantly reduces the number of parameters and training/inference time while maintaining high accuracy.
Efficient Adaptive Human-Object Interaction Detection with Concept-guided Memory
Ting Lei (Wangxuan Institute of Computer Technology Peking University), Yang Liu (Wangxuan Institute of Computer Technology Peking University)
RecognitionObject DetectionTransformerContrastive LearningImage
🎯 What it does: A training-free and lightweight fine-tuning framework for person-object interaction detection, ADA-CM, is proposed, utilizing a concept-guided memory module and instance-aware adapter for efficient detection.
Efficient Computation Sharing for Multi-Task Visual Scene Understanding
Sara Shoouri (University of Michigan), Hun-Seok Kim (University of Michigan)
SegmentationDepth EstimationComputational EfficiencyTransformerDiffusion modelImageVideo
🎯 What it does: This paper proposes a Transformer-based multi-task visual scene understanding framework that utilizes the weights and activations of single-task models for cross-task and cross-time sharing to achieve efficient multi-task inference.
Efficient Controllable Multi-Task Architectures
Abhishek Aich (University of California), Yumin Suh (University of California)
SegmentationDepth EstimationComputational EfficiencyKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: A controllable multi-task network trained once that can dynamically extract sub-networks during inference based on user computation budget and task priority.
Efficient Converted Spiking Neural Network for 3D and 2D Classification
Yuxiang Lan (Xiamen University), Yun Fu (Northeastern University)
ClassificationComputational EfficiencySpiking Neural NetworkImagePoint Cloud
🎯 What it does: This paper proposes an efficient unified ANN-SNN conversion method for 3D point cloud and 2D image classification, achieving lossless conversion in a very short time step through adaptive dynamic thresholds and an adaptive emission mechanism.
Efficient Decision-based Black-box Patch Attacks on Video Recognition
Kaixun Jiang (Fudan University), Wenqiang Zhang (Fudan University)
RecognitionAdversarial AttackVideo
🎯 What it does: This study investigates adversarial patch attacks under decision-based black-box conditions in video recognition models.
Efficient Deep Space Filling Curve
Wanli Chen (Chinese University of Hong Kong), Bei Yu (Chinese University of Hong Kong)
GenerationOptimizationComputational EfficiencyConvolutional Neural NetworkGraph Neural NetworkImage
🎯 What it does: A deep space-filling curve generation framework based on efficient GCN is proposed, which can adaptively generate linear sequences of images.
Efficient Diffusion Training via Min-SNR Weighting Strategy
Tiankai Hang (Southeast University), Baining Guo (Microsoft Research Asia)
GenerationData SynthesisOptimizationComputational EfficiencyTransformerDiffusion modelAuto EncoderImage
🎯 What it does: A loss weighting strategy named Min‑SNRγ is proposed to adjust the gradient conflicts at different time steps during the training of diffusion models, thereby accelerating convergence and improving generation quality.
Efficient Discovery and Effective Evaluation of Visual Perceptual Similarity: A Benchmark and Beyond
Oren Barkan (Open University), Noam Koenigstein (Tel Aviv University)
RetrievalSupervised Fine-TuningContrastive LearningImageBenchmark
🎯 What it does: An efficient annotation method for discovering similar images, EDS, has been developed, and based on this, a benchmark dataset of 110K pairs of fashion visual similarity has been constructed.
Efficient Emotional Adaptation for Audio-Driven Talking-Head Generation
Yuan Gan (Zhejiang University), Yi Yang (Zhejiang University)
GenerationTransformerVideoAudio
🎯 What it does: This paper proposes a two-stage emotional adaptation framework EAT, which efficiently transfers a pre-trained emotion-agnostic speaker image generation model to the emotional speaker generation task.
Efficient Joint Optimization of Layer-Adaptive Weight Pruning in Deep Neural Networks
Kaixin Xu (Agency for Science Technology and Research), Weisi Lin (Nanyang Technological University)
OptimizationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a layer adaptive weight pruning joint optimization method aimed at minimizing network output distortion while satisfying overall pruning ratio constraints.
Efficient LiDAR Point Cloud Oversegmentation Network
Le Hui (Northwestern Polytechnical University), Jian Yang (Nanjing University of Science and Technology)
SegmentationAutonomous DrivingConvolutional Neural NetworkPoint Cloud
🎯 What it does: The SuperLiDAR network is proposed for end-to-end over-segmentation of LiDAR point clouds, generating superpoints that are uniform in both semantics and geometry.
Efficient Model Personalization in Federated Learning via Client-Specific Prompt Generation
Fu-En Yang (National Taiwan University), Yu-Chiang Frank Wang (NVIDIA)
Federated LearningTransformerPrompt EngineeringImage
🎯 What it does: In the federated learning framework, a model personalization method is proposed through a server-side personalized prompt generator and a client-side prompt adapter.
Efficient Neural Supersampling on a Novel Gaming Dataset
Antoine Mercier (Qualcomm AI Research), Guillaume Berger (Qualcomm AI Research)
Super ResolutionComputational EfficiencyConvolutional Neural NetworkImageVideoMultimodality
🎯 What it does: Developed an efficient neural network game supersampling algorithm that utilizes viewport jitter, negative mipmap offset, and multimodal (color, depth, motion vector) data to achieve real-time high-resolution rendering.
Efficient Region-Aware Neural Radiance Fields for High-Fidelity Talking Portrait Synthesis
Jiahe Li (Beihang University), Lin Gu (Griffith University)
GenerationData SynthesisNeural Radiance FieldVideoMultimodalityAudio
🎯 What it does: This paper proposes ER-NeRF, a method for high-fidelity, real-time speaker avatar synthesis achieved through region-aware neural radiance fields.
Efficient Transformer-based 3D Object Detection with Dynamic Token Halting
Mao Ye (Cruise LLC), Qiang Liu (University of Texas at Austin)
Object DetectionAutonomous DrivingComputational EfficiencyTransformerPoint Cloud
🎯 What it does: This paper proposes a dynamic token stagnation mechanism that evaluates and stagnates unimportant tokens layer by layer in a Transformer-based 3D object detection model to significantly reduce computational load.
Efficient Unified Demosaicing for Bayer and Non-Bayer Patterned Image Sensors
Haechang Lee (Seoul National University), Se Young Chun (Seoul National University)
RestorationMeta LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A unified deep demosaicing model KLAP is proposed for Bayer and various non-Bayer CFA (Quad, Nona, Q×Q) image sensors, incorporating meta-learning (KLAP-M) during the inference stage to reduce unknown sensor artifacts in real RAW images.
Efficient Video Action Detection with Token Dropout and Context Refinement
Lei Chen (Nanjing University), Limin Wang (Nanjing University)
RecognitionObject DetectionComputational EfficiencyTransformerVideo
🎯 What it does: This paper proposes an efficient video action detection framework EVAD based on ViT, achieving efficient inference through spatiotemporal token dropping centered on key frames and context refinement.
Efficient Video Prediction via Sparsely Conditioned Flow Matching
Aram Davtyan (University of Bern), Paolo Favaro (University of Bern)
GenerationData SynthesisComputational EfficiencyFlow-based ModelGenerative Adversarial NetworkVideoOrdinary Differential Equation
🎯 What it does: We propose RIVER, a sparse conditional video prediction and generation model based on flow matching, which utilizes VQGAN latent space encoding and randomly selects past frames as conditions, supporting high-resolution video generation and visual planning.
Efficient View Synthesis with Neural Radiance Distribution Field
Yushuang Wu (Chinese University of Hong Kong Shenzhen), Yan Lu (Microsoft Research)
GenerationData SynthesisComputational EfficiencyKnowledge DistillationNeural Radiance FieldImage
🎯 What it does: This paper proposes NeRDF (Neural Radiance Distribution Field) representation, which obtains the radiation distribution of rays through a single forward inference of the network and generates images using volume rendering; it also trains an efficient small MLP using knowledge distillation from a teacher NeRF, online view sampling, and volume density constraints.
Efficient-VQGAN: Towards High-Resolution Image Generation with Efficient Vision Transformers
Shiyue Cao (University of Chinese Academy of Sciences), Kaigi Huang
GenerationData SynthesisTransformerAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes Efficient-VQGAN, an efficient two-stage vector quantization model that improves the first stage quantization with local attention, while the second stage uses a multi-granularity attention Transformer to generate high-resolution images, combining autoencoding training with autoregressive inference.
Efficiently Robustify Pre-Trained Models
Nishant Jain (Indian Institute of Technology Roorkee), Vibhav Vineet (Microsoft Research)
Domain AdaptationComputational EfficiencyKnowledge DistillationImage
🎯 What it does: By inserting a small robust teacher model into a large-scale pre-trained model for knowledge distillation, and using a multi-head structure with an uncertainty-aware head selection mechanism, only a small number of parameters need to be fine-tuned to enhance the model's robustness under various distribution shifts, while maintaining the original clean accuracy and transfer learning capability.
EfficientTrain: Exploring Generalized Curriculum Learning for Training Visual Backbones
Yulin Wang (Tsinghua University), Gao Huang (Tsinghua University)
Computational EfficiencyRepresentation LearningConvolutional Neural NetworkImage
🎯 What it does: EfficientTrain is proposed, a training strategy that initially exposes only low-frequency information and gradually introduces high-frequency information while progressively increasing the intensity of data augmentation, making the training of visual foundation models more efficient.
EfficientViT: Lightweight Multi-Scale Attention for High-Resolution Dense Prediction
Han Cai (Massachusetts Institute of Technology), Song Han (Massachusetts Institute of Technology)
ClassificationSegmentationSuper ResolutionTransformerSupervised Fine-TuningImage
🎯 What it does: This paper proposes the EfficientViT model, which achieves high-resolution dense prediction through lightweight multi-scale attention.
EGC: Image Generation and Classification via a Diffusion Energy-Based Model
Qiushan Guo (University of Hong Kong), Ping Luo (University of Hong Kong)
ClassificationGenerationDiffusion modelImage
🎯 What it does: A unified energy-based model EGC is proposed, which can perform image classification and generate images through a diffusion reverse process. The forward process estimates the energy of the joint distribution p(x, y), while the backward process computes the gradient of the joint distribution for sample reconstruction and generation.
EGformer: Equirectangular Geometry-biased Transformer for 360 Depth Estimation
Ilwi Yun (Seoul National University), Chae Eun Rhee (Inha University)
Depth EstimationTransformerImage
🎯 What it does: EGformer is proposed, a geometric bias Transformer for depth estimation of 360-degree panoramic images, achieving more accurate depth predictions by incorporating isometric rectangular projection geometric information into local attention.
Ego-Humans: An Ego-Centric 3D Multi-Human Benchmark
Rawal Khirodkar, Kris Kitani
Object TrackingData SynthesisPose EstimationTransformerSimultaneous Localization and MappingImageVideoBenchmark
🎯 What it does: A multi-view, outdoor multi-person 3D dataset called EgoHumans was constructed based on wearable glasses and fixed cameras, and the EgoFormer algorithm was proposed to achieve 3D human tracking from a first-person perspective.
Ego-Only: Egocentric Action Detection without Exocentric Transferring
Huiyu Wang (Meta AI), Lorenzo Torresani (Meta AI)
RecognitionSegmentationTransformerAuto EncoderVideo
🎯 What it does: An action detection model was trained using only egocentric video data without any exocentric pre-training.
EgoLoc: Revisiting 3D Object Localization from Egocentric Videos with Visual Queries
Jinjie Mai (King Abdullah University of Science and Technology), Bernard Ghanem (King Abdullah University of Science and Technology)
Object DetectionPose EstimationRetrievalSimultaneous Localization and MappingVideo
🎯 What it does: An end-to-end visual query 3D localization (VQ3D) pipeline called EgoLoc is proposed to locate the most recent appearance of the queried object in first-person videos and provide a relative 3D displacement vector.
EgoObjects: A Large-Scale Egocentric Dataset for Fine-Grained Object Understanding
Chenchen Zhu (Meta AI), Zhicheng Yan (Meta AI)
Object DetectionFederated LearningVideo
🎯 What it does: Created a large-scale first-person perspective EgoObjects dataset, containing hundreds of thousands of video frames and target boxes, annotated with category and instance IDs;
EgoPCA: A New Framework for Egocentric Hand-Object Interaction Understanding
Yue Xu (Shanghai Jiao Tong University), Chi-Keung Tang (Hong Kong University of Science and Technology)
RecognitionObject DetectionPose EstimationTransformerContrastive LearningOptical FlowVideo
🎯 What it does: This paper proposes the EgoPCA framework, which includes a balanced pre-training set, the One4All baseline model, and a new test set to enhance learning of first-person hand-object interaction (Ego-HOI);
EgoTV: Egocentric Task Verification from Natural Language Task Descriptions
Rishi Hazra (Orebro University), Ruta Desai (Meta)
Data SynthesisExplainability and InterpretabilityRecurrent Neural NetworkVision Language ModelVideoTextBenchmark
🎯 What it does: Created the Egocentric Task Verification (EgoTV) benchmark and synthetic dataset, and proposed the neural-symbolic framework NSG to perform task verification based on natural language task descriptions in perspective videos.
EgoVLPv2: Egocentric Video-Language Pre-training with Fusion in the Backbone
Shraman Pramanick (Johns Hopkins University), Pengchuan Zhang (Meta AI)
RetrievalRepresentation LearningTransformerContrastive LearningVideoTextMultimodality
🎯 What it does: We propose EgoVLPv2, a second-generation self-centered video-language pre-training framework that integrates cross-modal attention into the video and text Transformer backbone.