ECCV 2024 Papers — Page 2
European Conference on Computer Vision · 2387 papers
AdaDistill: Adaptive Knowledge Distillation for Deep Face Recognition
Fadi Boutros (Fraunhofer Institute for Computer Graphics Research IGD), Naser Damer (Fraunhofer Institute for Computer Graphics Research IGD)
RecognitionKnowledge DistillationImage
🎯 What it does: Proposed an adaptive knowledge distillation method called AdaDistill, applicable to deep face recognition;
AdaGlimpse: Active Visual Exploration with Arbitrary Glimpse Position and Scale
Adam Pardyl (IDEAS NCBR), Bartosz Zieliński (IDEAS NCBR)
SegmentationRobotic IntelligenceTransformerReinforcement LearningAuto EncoderImage
🎯 What it does: Develop AdaGlimpse, proposing an active visual exploration framework that can adaptively select arbitrary position and scale windows in a continuous action space to achieve efficient environmental perception.
AdaIFL: Adaptive Image Forgery Localization via a Dynamic and Importance-aware Transformer Network
Yuxi Li (Peking University), Yuesheng Zhu (Peking University)
SegmentationTransformerMixture of ExpertsImage
🎯 What it does: Designed and implemented a Transformer network called AdaIFL based on dynamic routing and importance-aware mechanisms for high-precision image forgery localization.
AdaLog: Post-Training Quantization for Vision Transformers with Adaptive Logarithm Quantizer
Zhuguanyu Wu (Beihang University), Yunhong Wang (Beihang University)
ClassificationObject DetectionSegmentationComputational EfficiencyTransformerImage
🎯 What it does: This paper proposes the AdaLog quantizer, achieving adaptive log-based quantization for Vision Transformers in post-training quantization, and extends it to activation layers after Softmax and GELU through bias reparameterization.
AdaNAT: Exploring Adaptive Policy for Token-Based Image Generation
Zanlin Ni (Tsinghua University), Gao Huang (Tsinghua University)
GenerationTransformerReinforcement LearningAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: Propose AdaNAT, which improves the image generation quality of non-autoregressive Transformers (NAT) by learning adaptive generation strategies.
Adapt without Forgetting: Distill Proximity from Dual Teachers in Vision-Language Models
Mengyu Zheng (University of Sydney), Chang Xu (Huawei)
Knowledge DistillationRepresentation LearningGraph Neural NetworkTransformerVision Language ModelMultimodalityBenchmark
🎯 What it does: Studied a method to maintain and enhance the zero-shot transferability of the CLIP vision-language model in continual learning scenarios.
Adapt2Reward: Adapting Video-Language Models to Generalizable Robotic Rewards via Failure Prompts
Yanting Yang (Zhejiang University), Xiaofei He (Xidian University)
Domain AdaptationRobotic IntelligenceReinforcement Learning from Human FeedbackPrompt EngineeringVision Language ModelContrastive LearningVideoText
🎯 What it does: Investigate how to adapt video-language models into a general language-conditioned reward function, and enhance generalization capabilities using a few robot failure samples.
Adapting Fine-Grained Cross-View Localization to Areas without Fine Ground Truth
Zimin Xia (École Polytechnique Fédérale de Lausanne), Julian F. P. Kooij (Delft University of Technology)
Domain AdaptationKnowledge DistillationImage
🎯 What it does: This paper addresses the scenario of cross-regional fine-grained view localization lacking precise ground truth, proposing a weakly supervised adaptation method based on knowledge self-distillation.
Adapting to Shifting Correlations with Unlabeled Data Calibration
Minh Nguyen (Cornell Tech), Mert Sabuncu
Domain AdaptationImage
🎯 What it does: Proposes a method called Generalized Propensity Adjustment (GPA) to address distribution shifts between different sites by using unlabeled data to calibrate model predictions.
Adaptive Annealing for Robust Averaging
Sidhartha Chitturi (Indian Institute of Science), Venu Madhav Govindu (Indian Institute of Science)
OptimizationGraphBenchmark
🎯 What it does: Propose an adaptive Annealing method based on the structure of the graph Laplacian matrix for robust average estimation (vector average and displacement average).
Adaptive Bounding Box Uncertainties via Two-Step Conformal Prediction
Alexander Timans (University of Amsterdam Bosch Delta Lab, University of Amsterdam), Eric Nalisnick (Bosch Center for AI, Robert Bosch GmbH)
Object DetectionConvolutional Neural NetworkImage
🎯 What it does: Propose a two-step adaptive bounding box uncertainty prediction framework, leveraging distribution-free conformal prediction to provide coverage guarantees for bounding boxes of multi-class detectors, and propagate classification uncertainty through class prediction sets.
Adaptive Compressed Sensing with Diffusion-Based Posterior Sampling
Noam Elata (Technion Israel Institute of Technology), Michael Elad (Technion Israel Institute of Technology)
CompressionDiffusion modelImageBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: This paper proposes an adaptive compressed sensing framework called AdaSense, which dynamically selects the most informative linear measurements during the measurement process by leveraging posterior sampling from pre-trained diffusion models.
Adaptive Correspondence Scoring for Unsupervised Medical Image Registration
Xiaoran Zhang (Yale University), James S. Duncan (Yale University)
Convolutional Neural NetworkBiomedical DataMagnetic Resonance ImagingUltrasound
🎯 What it does: This paper proposes a framework for unsupervised medical image registration using adaptive correspondence scoring, which dynamically reweights error residuals during training through a learnable correspondence scoring network, thereby suppressing the negative impact of noise, occlusions, and other non-corresponding regions on registration and improving registration accuracy.
Adaptive High-Frequency Transformer for Diverse Wildlife Re-Identification
Chenyue Li (Wuhan University), Mang Ye (Wuhan University)
RetrievalTransformerContrastive LearningImage
🎯 What it does: Proposed a unified multi-species wildlife re-identification framework called Adaptive High-Frequency Transformer, with the core idea of leveraging high-frequency information for feature learning.
Adaptive Human Trajectory Prediction via Latent Corridors
Neerja Thakkar (University Of California Berkeley), Jitendra Malik (University Of California Berkeley)
Object TrackingSegmentationDomain AdaptationPrompt EngineeringVideo
🎯 What it does: Adapt pre-trained human trajectory prediction models quickly to new scenarios using learnable image prompts (latent corridor), enabling the model to capture scenario-specific human behaviors with extremely limited data (e.g., 2 people for 30 seconds).
Adaptive Multi-head Contrastive Learning
Lei Wang (Australian National University), Liang Zheng (Australian National University)
Representation LearningTransformerContrastive LearningImage
🎯 What it does: Propose an Adaptive Multi-Head Contrastive Learning (AMCL) that improves representation learning by using multiple sets of projection heads and learning adaptive temperatures for each pair of samples.
Adaptive Multi-modal Fusion of Spatially Variant Kernel Refinement with Diffusion Model for Blind Image Super-Resolution
Junxiong Lin, Wenqiang Zhang (Shanghai Ocean University)
Depth EstimationSuper ResolutionConvolutional Neural NetworkDiffusion modelImageMultimodality
🎯 What it does: This paper proposes a new blind image super-resolution framework called SSR, achieving high-quality super-resolution by combining pre-trained diffusion models with precise estimation of spatially variant convolution kernels (SVKR) and depth information fusion.
Adaptive Multi-task Learning for Few-shot Object Detection
Yan Ren (Nanyang Technological University), Adams Wai-Kin Kong (Nanyang Technological University)
Object DetectionKnowledge DistillationMeta LearningConvolutional Neural NetworkVision Language ModelImageMultimodality
🎯 What it does: Propose an adaptive multi-task learning framework to address the conflicts between classification and localization tasks in few-shot object detection;
Adaptive Parametric Activation
Konstantinos P Alexandridis (Huawei Noah's Ark Lab), Shan Luo (King's College London)
ClassificationObject DetectionImage
🎯 What it does: Proposed and implemented an adaptive parameterized activation function APA, which unifies various activation forms and can be used across different layers.
Adaptive Selection of Sampling-Reconstruction in Fourier Compressed Sensing
Seongmin Hong (Seoul National University), Se Young Chun (Seoul National University)
RestorationCompressionConvolutional Neural NetworkFlow-based ModelImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: Propose an adaptive selection sampling-reconstruction framework that selects the optimal sampling mask and corresponding reconstruction network for each input image in Fourier compressed sensing.
AdaShield: Safeguarding Multimodal Large Language Models from Structure-based Attack via Adaptive Shield Prompting
Yu Wang (University of Wisconsin-Madison), Chaowei Xiao (University of Wisconsin-Madison)
Adversarial AttackTransformerLarge Language ModelPrompt EngineeringVision Language ModelMultimodalityBenchmarkChain-of-Thought
🎯 What it does: Proposes AdaShield, which defends against structured jailbreak attacks on multimodal large language models by prepending adaptive safety prompts to inputs, without requiring model fine-tuning.
AddBiomechanics Dataset: Capturing the Physics of Human Motion at Scale
Keenon Werling (Stanford University), Jennifer Hicks (Stanford University)
OptimizationVideoBiomedical DataBenchmarkPhysics Related
🎯 What it does: Constructed and released the AddBiomechanics 1.0 dataset, containing optical marker and ground reaction force data from 273 subjects, over 70 hours, 24 million frames, with personalized musculoskeletal models generated for each subject, further proposing a benchmark for evaluating human motion dynamics.
AddMe: Zero-shot Group-photo Synthesis by Inserting People into Scenes
Dongxu Yue (International Digital Economy Academy), Yu Li (International Digital Economy Academy)
GenerationData SynthesisTransformerVision Language ModelDiffusion modelImageText
🎯 What it does: Propose AddMe, a zero-shot group photo generation framework that can insert personalized portraits into existing photos at any position using only one reference portrait and a user-provided mask, without requiring additional fine-tuning.
AddressCLIP: Empowering Vision-Language Models for City-wide Image Address Localization
Shixiong Xu (Chinese Academy of Sciences), Jieping Ye (Alibaba Cloud)
ClassificationTransformerVision Language ModelContrastive LearningImageText
🎯 What it does: Proposed an end-to-end vision-language model called AddressCLIP for predicting the readable text address of the location where the image was taken.
ADen: Adaptive Density Representations for Sparse-view Camera Pose Estimation
Hao Tang (Meta), Matt Feiszli (Meta)
Pose EstimationConvolutional Neural NetworkTransformerGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: Developed an adaptive density representation (ADen) for recovering camera poses from sparse view images, combining a generator-discriminator framework to output multiple candidate poses and select the optimal one.
ADMap: Anti-disturbance Framework for Vectorized HD Map Construction
Haotian Hu (Zhejiang Leapmotor Technology CO., LTD), Zhiwang Zhang (Ningbo Tech University)
Autonomous DrivingTransformerImageMultimodalityPoint Cloud
🎯 What it does: Proposed an end-to-end ADMap framework for online high-precision HD map vectorization construction, addressing point sequence jittering and sharpness phenomena.
AdvDiff: Generating Unrestricted Adversarial Examples using Diffusion Models
Xuelong Dai (Hong Kong Polytechnic University), Bin Xiao (Hong Kong Polytechnic University)
Adversarial AttackDiffusion modelImage
🎯 What it does: Proposed an unrestricted adversarial attack method AdvDiff based on diffusion models, used to generate realistic adversarial examples that can mislead target classifiers.
Adversarial Diffusion Distillation
Axel Sauer (Stability AI), Robin Rombach (Stability AI)
GenerationKnowledge DistillationConvolutional Neural NetworkTransformerVision Language ModelDiffusion modelScore-based ModelGenerative Adversarial NetworkImageMultimodality
🎯 What it does: Trained a low-step sampling image generation model ADD, achieving high-quality image generation within 1–4 steps through adversarial training and score distillation.
Adversarial Prompt Tuning for Vision-Language Models
Jiaming Zhang (Beijing Jiaotong Univisity), Jitao Sang (Fudan Univisity)
Adversarial AttackPrompt EngineeringVision Language ModelDiffusion modelContrastive LearningImageText
🎯 What it does: This paper proposes Adversarial Prompt Tuning (AdvPT), which enhances the robustness of the image encoder in CLIP against adversarial image attacks by fine-tuning the learnable prompt vectors in the CLIP text encoder.
Adversarial Robustification via Text-to-Image Diffusion Models
Daewon Choi (Korea Advanced Institute of Science and Technology), Jinwoo Shin (Korea Advanced Institute of Science and Technology)
ClassificationData SynthesisAdversarial AttackConvolutional Neural NetworkSupervised Fine-TuningVision Language ModelDiffusion modelImageText
🎯 What it does: Propose a 'denoising + classification' randomized smoothing framework based on text-to-image diffusion models without requiring training data, to achieve provable adversarial robustness for offline pre-trained classifiers.
AdversariaLeak: External Information Leakage Attack Using Adversarial Samples on Face Recognition Systems
Roye Katzav (Ben Gurion University Negev), Asaf Shabtai (Ben Gurion University Negev)
RecognitionAdversarial AttackImage
🎯 What it does: Studied an external information leakage attack (AdversariaLeak) based on adversarial examples, capable of inferring the attribute distribution of the training set of a face recognition model under an extremely small query budget.
Adversarially Robust Distillation by Reducing the Student-Teacher Variance Gap
Junhao Dong (Nanyang Technological University), Yew-Soon Ong (Nanyang Technological University)
Knowledge DistillationAdversarial AttackImage
🎯 What it does: The study enhances adversarial robust knowledge distillation effectiveness by reducing the variance gap between student and teacher features.
AEDNet: Adaptive Embedding and Multiview-Aware Disentanglement for Point Cloud Completion
Zhiheng Fu (University of Western Australia), Mohammed Bennamoun (University of Western Australia)
GenerationTransformerPoint Cloud
🎯 What it does: Proposed the AEDNet model, designing adaptive embedding and multi-view decoupling modules for point cloud completion tasks.
AFF-ttention! Affordances and Attention models for Short-Term Object Interaction Anticipation
Lorenzo Mur-Labadia (University of Zaragoza), Antonino Furnari (University of Catania)
Object DetectionTransformerVision-Language-Action ModelVideoText
🎯 What it does: Proposes the STAformer architecture for first-person short-term object interaction prediction, enhancing prediction accuracy through two affordance modules: an environmental affordance database and interaction hotspot prediction.
Affective Visual Dialog: A Large-Scale Benchmark for Emotional Reasoning Based on Visually Grounded Conversations
Kilichbek Haydarov, Mohamed Elhoseiny (King Abdullah University of Science and Technology)
ClassificationRetrievalTransformerLarge Language ModelSupervised Fine-TuningMultimodalityBenchmarkRetrieval-Augmented Generation
🎯 What it does: This paper proposes the Affective Visual Dialog task and constructs a large-scale dataset, AffectVisDial, for studying emotion generation and interpretation in visual dialogues.
Affine steerers for structured keypoint description
Georg Bökman (Chalmers University of Technology), Fredrik Kahl (Chalmers University of Technology)
Pose EstimationRetrievalRepresentation LearningConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: Proposed a framework called Affine Steerer that trains deep learning keypoint descriptors to be equivariant to local affine transformations.
AFreeCA: Annotation-Free Counting for All
Adriano D'Alessandro (Simon Fraser University), Ghassan Hamarneh (Simon Fraser University)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: Proposed an unsupervised and annotation-free object counting framework, AFreeCA, which leverages text-to-image diffusion models to generate ranking data and counting data. It trains a ranking network to learn object counting features, aligns the count values by training only a linear layer, and finally combines a density classification guided partitioning strategy to achieve multi-resolution counting in high-density regions.
Agent Attention: On the Integration of Softmax and Linear Attention
Dongchen Han (Tsinghua University), Gao Huang (Tsinghua University)
ClassificationObject DetectionSegmentationGenerationTransformerDiffusion modelImage
🎯 What it does: This paper proposes the Agent Attention (Agent Attention) mechanism, integrating traditional Softmax attention with linear attention to form an efficient quadruple attention (Q, A, K, V) capable of global semantic modeling.
Agent3D-Zero: An Agent for Zero-shot 3D Understanding
sha zhang, Yanyong Zhang (University of Science and Technology of China)
RecognitionTransformerLarge Language ModelAgentic AIPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: Propose the Agent3D-Zero framework, leveraging large vision-language models to achieve 3D scene understanding under zero-shot conditions through active perspective selection and visual prompts.
Agglomerative Token Clustering
Joakim Bruslund Haurum (Aalborg University), Thomas B. Moeslund (Aalborg University)
ClassificationObject DetectionSegmentationGenerationCompressionTransformerImage
🎯 What it does: Propose the Agglomerative Token Clustering (ATC) method, which uses parameter-free hierarchical agglomerative clustering to progressively merge the most similar tokens in Vision Transformers, thereby reducing the number of tokens;
AID-AppEAL: Automatic Image Dataset and Algorithm for Content Appeal Enhancement and Assessment Labeling
Sherry X. Chen (University of California, Santa Barbara), Pradeep Sen (University of California, Santa Barbara)
GenerationData SynthesisSuper ResolutionPrompt EngineeringVision Language ModelDiffusion modelImage
🎯 What it does: Designed and implemented an automated pipeline for generating the Image Content Attractiveness (ICAA) dataset, named AID-AppEAL. Proposed content attractiveness estimators based on relative and absolute evaluation, and developed enhanced methods combining text inversion with Stable Diffusion. Ultimately, generated datasets with over 70K images each in the food and interior design domains.
Align before Collaborate: Mitigating Feature Misalignment for Robust Multi-Agent Perception
Dingkang Yang (Fudan University), Liang Song (Duke Kunshan University)
Object DetectionAutonomous DrivingPoint Cloud
🎯 What it does: Proposed a lightweight plugin called NEAT to dynamically align features in multi-agent collaborative perception, eliminating feature alignment errors caused by transmission delays and localization errors, thereby enhancing perception robustness.
AlignDiff: Aligning Diffusion Models for General Few-Shot Segmentation
Ri-Zhao Qiu (University of Illinois Urbana-Champaign), Kris Hauser (University of Illinois Urbana-Champaign)
SegmentationGenerationData SynthesisPrompt EngineeringDiffusion modelImageText
🎯 What it does: This paper proposes the AlignDiff framework, which utilizes a small number of real samples to guide text-to-image diffusion models in generating synthetic training data that is both aligned with target category instances and equipped with pixel-level annotations, thereby enhancing few-shot semantic segmentation performance.
Aligning Neuronal Coding of Dynamic Visual Scenes with Foundation Vision Models
Rining Wu (University of Leeds), Jian Liu (University of Leeds)
Representation LearningConvolutional Neural NetworkTransformerContrastive LearningVideoTime Series
🎯 What it does: This study proposes the Vi-ST model, which combines pre-trained ViT (DINOv2) with spatiotemporal convolutional networks to directly extract features from dynamic natural videos. It predicts the electrical activity of retinal ganglion cells (RGC) through a multi-scale spatiotemporal module and conditional fusion with retinal receptor field information.
Alignist: CAD-Informed Orientation Distribution Estimation by Fusing Shape and Correspondences
Shishir Reddy Vutukur (Technical University of Munich), Tolga Birdal (Imperial College London)
Pose EstimationMixture of ExpertsMesh
🎯 What it does: Propose the Alignist method, which utilizes geometric and symmetry correspondence information from CAD models to estimate the multi-modal probability distribution of 6D pose, and employs a dual-branch MLP to learn rotation during training.
AlignZeg: Mitigating Objective Misalignment for Zero-shot Semantic Segmentation
Jiannan Ge (University of Science and Technology of China), Qi Tian (University of Science and Technology of China)
SegmentationTransformerContrastive LearningImage
🎯 What it does: Propose the AlignZeg framework to address the target error alignment problem in zero-shot semantic segmentation.
All You Need is Your Voice: Emotional Face Representation with Audio Perspective for Emotional Talking Face Generation
Seongho Kim (Inha University), Byung Cheol Song (Inha University)
GenerationRecurrent Neural NetworkGenerative Adversarial NetworkVideoAudio
🎯 What it does: This paper proposes the Emotional Face Representation with Audio Perspective (EAP) method, addressing the issues of input bias and emotional intensity saturation in audio-driven emotional speaker video generation. It first neutralizes arbitrary emotional faces through Audio-to-Neutral Translation (ANT), then extracts emotional context and frame-level emotional intensity from audio via Emotional Representation from Audio (ERA), ultimately achieving natural and emotionally accurate emotional speaker video generation.
Alternate Diverse Teaching for Semi-supervised Medical Image Segmentation
Zhen Zhao (Shanghai AI Lab), Luping Zhou (University of Sydney)
SegmentationConvolutional Neural NetworkBiomedical Data
🎯 What it does: Propose the AD-MT framework, achieving semi-supervised medical image segmentation through a single student network and two sets of alternately updated teacher networks;
AMD: Automatic Multi-step Distillation of Large-scale Vision Models
Cheng Han (University of Missouri - Kansas City), Dongfang Liu (Virginia Tech)
CompressionKnowledge DistillationTransformerImage
🎯 What it does: Proposed and implemented a compression method called Automatic Multi-step Distillation (AMD), which can compress Transformer visual models to only 10% of their original size while maintaining high performance, suitable for knowledge distillation of large-scale visual models.
AMEGO: Active Memory from long EGOcentric videos
Gabriele Goletto (Politecnico di Torino), Dima Damen
Object DetectionObject TrackingRepresentation LearningTransformerVideoBenchmark
🎯 What it does: This paper proposes AMEGO, a structured representation based on active memory, for capturing object interactions and key positions from long-term first-person videos, enabling multiple queries without semantic labels.
AMES: Asymmetric and Memory-Efficient Similarity Estimation for Instance-level Retrieval
Pavel Suma (Czech Technical University in Prague), Giorgos Tolias (Czech Technical University in Prague)
RetrievalComputational EfficiencyKnowledge DistillationTransformerImage
🎯 What it does: Proposed and implemented a Transformer-based heterogeneous and low-memory similarity estimation framework, AMES, for instance-level image retrieval, achieving approximately 1KB memory usage per image through techniques such as heterogeneous vectors, binary descriptors, and knowledge distillation.
An accurate detection is not all you need to combat label noise in web-noisy datasets
Paul Albert (University of Adelaide), Jack Valmadre (University of Adelaide)
ClassificationData-Centric LearningContrastive LearningImage
🎯 What it does: On a web-crawled dataset, the paper proposes a Linear Separation Alternating (LSA) strategy by combining the linear separation from unsupervised contrastive learning with existing noise detection methods, and integrates it with the PLS algorithm to form PLS-LSA, significantly improving accuracy in image classification tasks under label noise.
An Adaptive Screen-Space Meshing Approach for Normal Integration
Moritz Heep (University of Bonn), Eduard Zell (University of Bonn)
GenerationComputational EfficiencyImageMesh
🎯 What it does: To address the normal maps obtained via photometric stereo, this paper proposes constructing an adaptive triangular mesh in screen space first, then performing normal integration on this mesh to complete 3D surface reconstruction.
An Economic Framework for 6-DoF Grasp Detection
Xiao-Ming Wu (Sun Yat-sen University), Wei-Shi Zheng (Sun Yat-sen University)
Pose EstimationRobotic IntelligenceConvolutional Neural NetworkPoint Cloud
🎯 What it does: Proposes an economical 6-DoF grasp detection framework called EconomicGrasp, aiming to maintain high grasp performance while reducing training resources.
An Efficient and Effective Transformer Decoder-Based Framework for Multi-Task Visual Grounding
Wei Chen (Wuhan University), Yu Wu (Wuhan University)
SegmentationTransformerVision Language ModelImageText
🎯 What it does: Proposes a multi-task visual localization framework (EEVG) based on Transformer Decoder, which utilizes the Decoder for vision-language fusion and significantly improves computational efficiency and localization accuracy through parameter-agnostic visual token elimination strategies and lightweight mask heads.
An Empirical Study and Analysis of Text-to-Image Generation Using Large Language Model-Powered Textual Representation
Zhiyu Tan (Fudan University), Hao Li (Fudan University)
GenerationTransformerLarge Language ModelDiffusion modelMultimodality
🎯 What it does: Propose a three-stage training pipeline called OmniDiffusion, leveraging large language models (LLM) as text encoders to enable text-to-image generation with multilingual and long-text capabilities.
An Explainable Vision Question Answer Model via Diffusion Chain-of-Thought
Chunhao LU, Jake Luo (University of Wisconsin Milwaukee)
Explainability and InterpretabilityTransformerVision Language ModelDiffusion modelMultimodalityChain-of-Thought
🎯 What it does: Proposed a diffusion-based chain-of-thought interpretable visual question answering model that can generate reasoning processes and final answers step by step.
An Image is Worth 1/2 Tokens After Layer 2: Plug-and-Play Inference Acceleration for Large Vision-Language Models
Liang Chen (Peking University), Baobao Chang (Alibaba Group)
Computational EfficiencyTransformerVision Language ModelMultimodality
🎯 What it does: Identify the phenomenon where visual tokens in deep layers of large vision-language models are inefficiently attended to, and propose the FastV method to dynamically prune visual tokens after the second layer, significantly reducing inference FLOPs without sacrificing multimodal task performance.
An Incremental Unified Framework for Small Defect Inspection
Jiaqi Tang (Hong Kong University of Science and Technology (Guangzhou)), Fugee Tsung (Hong Kong University of Science and Technology (Guangzhou))
Anomaly DetectionTransformerAuto EncoderImage
🎯 What it does: This paper proposes an Incremental Unified Framework (IUF) that can progressively learn micro-defect detection for various industrial objects without storing feature memory, achieving image-level and pixel-level localization.
An Information Theoretical View for Out-Of-Distribution Detection
Hu Jinjing, Xilin Chen (Chinese Academy of Sciences)
Anomaly DetectionConvolutional Neural NetworkImage
🎯 What it does: Propose the OOD Entropy Regularization (OER) method, enhancing OOD detection performance through temperature scaling and information entropy regularization.
An Optimal Control View of LoRA and Binary Controller Design for Vision Transformers
Chi Zhang (National University of Singapore), Qianxiao Li (National University of Singapore)
ClassificationObject DetectionTransformerSupervised Fine-TuningImage
🎯 What it does: This paper proposes to treat LoRA as a control variable in a controlled system, derives the optimal control form of LoRA using Pontryagin's maximum principle, and further designs a Bang-Bang controller (BiC) that uses only ±1 binary control, achieving low-precision and efficient fine-tuning of ViT models.
An Optimization Framework to Enforce Multi-View Consistency for Texturing 3D Meshes
Zhengyi Zhao (Alibaba Group), Qixing Huang (University of Texas at Austin)
GenerationOptimizationDiffusion modelTextMesh
🎯 What it does: Designed a four-phase optimization framework to achieve multi-view consistent texture generation from text prompts to 3D meshes;
Analysis-by-Synthesis Transformer for Single-View 3D Reconstruction
Dian Jia (University of Illinois Chicago), Wei Tang (University of Illinois Chicago)
GenerationTransformerImageMesh
🎯 What it does: Propose a unified analysis-synthesis Transformer framework for 3D reconstruction and texture generation from single-view images without 3D annotations.
Analytic-Splatting: Anti-Aliased 3D Gaussian Splatting via Analytic Integration
Zhihao Liang, Kui Jia
GenerationGaussian SplattingImage
🎯 What it does: Propose a 3D Gaussian rendering method that approximates Gaussian distribution responses within pixel windows via analytical integration, significantly enhancing anti-aliasing effects in view synthesis.
AnatoMask: Enhancing Medical Image Segmentation with Reconstruction-guided Self-masking
Yuheng Li (Emory University), Xiaofeng Yang (Emory University)
SegmentationKnowledge DistillationConvolutional Neural NetworkAuto EncoderImageBiomedical DataComputed TomographyPositron Emission Tomography
🎯 What it does: Proposes AnatoMask, a reconstruction-guided self-supervised masking method that improves the pre-training process for 3D medical image segmentation.
Animal Avatars: Reconstructing Animatable 3D Animals from Casual Videos
Remy Sabathier (Meta), Niloy Mitra
GenerationPose EstimationNeural Radiance FieldSimultaneous Localization and MappingVideoMesh
🎯 What it does: This paper proposes a method for constructing an animatable 3D dog model from monocular video, combining joint optimization of shape, pose, and texture;
AnimatableDreamer: Text-Guided Non-rigid 3D Model Generation and Reconstruction with Canonical Score Distillation
Xinzhou Wang (Tongji University), Bin He (Didi)
GenerationData SynthesisPose EstimationDiffusion modelScore-based ModelNeural Radiance FieldVideoTextMesh
🎯 What it does: The study proposes AnimatableDreamer, which extracts skeletons from monocular videos and achieves text-based non-rigid 4D generation and reconstruction through Canonical Score Distillation (CSD).
Animate Your Motion: Turning Still Images into Dynamic Videos
Mingxiao Li (KU Leuven), Tinne Tuytelaars (KU Leuven)
GenerationConvolutional Neural NetworkVision Language ModelDiffusion modelImageVideoText
🎯 What it does: Generate customized videos based on a given static image, object motion trajectory (bounding box sequence), and text description.
AnimateMe: 4D Facial Expressions via Diffusion Models
Dimitrios Gerogiannis (Imperial College London), Stefanos Zafeiriou (Imperial College London)
GenerationData SynthesisGraph Neural NetworkDiffusion modelVideoMesh
🎯 What it does: This paper proposes a 4D facial expression generation method based on diffusion models, which directly trains a graph neural network on the mesh space for denoising. It can generate coherent facial animations according to expression intensity and time steps, and achieves temporal consistency through consistent noise sampling.
Any Target Can be Offense: Adversarial Example Generation via Generalized Latent Infection
Youheng Sun (University of Electronic Science and Technology of China), Jingkuan Song (University of Electronic Science and Technology of China)
Adversarial AttackConvolutional Neural NetworkVision Language ModelImageMultimodality
🎯 What it does: Propose a generator-based adversarial attack method called GAKer, which can generate adversarial examples from any target object (regardless of whether it appears in the training set) by implanting source images with potential features of the target image, achieving attacks on both known and unknown categories.
Any2Point: Empowering Any-modality Transformers for Efficient 3D Understanding
Yiwen Tang (Shanghai AI Laboratory), Xuelong Li (TeleAI)
ClassificationTransformerSupervised Fine-TuningPoint Cloud
🎯 What it does: Propose the Any2Point framework, which leverages any modality's pre-trained large models (e.g., language, vision, audio) to achieve 3D point cloud understanding through parameter-efficient fine-tuning.
AnyControl: Create Your Artwork with Versatile Control on Text-to-Image Generation
Yanan Sun, Kai Chen (Shanghai AI Laboratory)
GenerationConvolutional Neural NetworkTransformerVision Language ModelDiffusion modelImageTextMultimodality
🎯 What it does: This paper proposes AnyControl, a multi-control image synthesis framework capable of generating high-quality, semantically consistent images under flexible combinations of text prompts and arbitrary numbers of multi-modal spatial conditions (such as depth maps, segmentation maps, edge maps, pose maps, etc.).
AnyHome: Open-Vocabulary Large-Scale Indoor Scene Generation with First-Person View Exploration
Rao Fu (Brown University), Srinath Sridhar (Brown University)
GenerationData SynthesisGraph Neural NetworkTransformerLarge Language ModelPrompt EngineeringDiffusion modelScore-based ModelTextMeshGraphRetrieval-Augmented Generation
🎯 What it does: Propose the AnyHome framework, which can convert arbitrary text descriptions into structured, textured home 3D scenes and support multi-level editing.
Anytime Continual Learning for Open Vocabulary Classification
Zhen Zhu (University of Illinois at Urbana-Champaign), Derek Hoiem (University of Illinois at Urbana-Champaign)
ClassificationComputational EfficiencyTransformerSupervised Fine-TuningVision Language ModelImageBenchmark
🎯 What it does: Propose an open-vocabulary image classification framework for continuous learning at any time, which can instantly update and continuously improve performance when receiving new samples at arbitrary time points, while maintaining the ability to predict for any label set.
APL: Anchor-based Prompt Learning for One-stage Weakly Supervised Referring Expression Comprehension
Yaxin Luo, Gen Luo (International Digital Economy Academy)
RecognitionObject DetectionConvolutional Neural NetworkPrompt EngineeringVision Language ModelContrastive LearningMultimodality
🎯 What it does: Propose an Anchor-based Prompt Learning framework (APL), which improves the weakly supervised Referring Expression Comprehension (REC) task by dynamically generating multimodal prompts (including position, color, and category) and fusing them into Anchor features.
Appearance-based Refinement for Object-Centric Motion Segmentation
Junyu Xie (University of Oxford), Andrew Zisserman (University of Oxford)
SegmentationTransformerSupervised Fine-TuningContrastive LearningOptical FlowVideo
🎯 What it does: First, objects are discovered in videos using optical flow, and then predicted masks are selected and corrected by leveraging temporally consistent appearance information, achieving unsupervised motion segmentation.
Approaching Outside: Scaling Unsupervised 3D Object Detection from 2D Scene
Ruiyang Zhang (University of Macau), Zhedong Zheng (University of Macau)
Object DetectionAutonomous DrivingImagePoint Cloud
🎯 What it does: Propose a framework called LiSe that utilizes 2D images to assist LiDAR in unsupervised 3D object detection;
Arbitrary-Scale Video Super-Resolution with Structural and Textural Priors
Wei Shang (Harbin Institute of Technology), Kede Ma (Jiangxi University of Finance and Economics)
Super ResolutionConvolutional Neural NetworkOptical FlowVideo
🎯 What it does: This paper proposes a strong baseline B-AVSR based on optical flow guided recursive units, optical flow corrected cross-attention units, and hyper-decoupled hyper-sampling units. By incorporating multi-scale structural and texture priors, it forms ST-AVSR to achieve video super-resolution at arbitrary scales.
Arc2Face: A Foundation Model for ID-Consistent Human Faces
Foivos Paraperas Papantoniou (Imperial College London), Stefanos Zafeiriou (Imperial College London)
GenerationSupervised Fine-TuningDiffusion modelContrastive LearningImage
🎯 What it does: Propose Arc2Face, a foundational model that uses ArcFace ID embedding as the sole condition to generate high-quality, identity-consistent facial images under any pose, expression, or scene.
Are Synthetic Data Useful for Egocentric Hand-Object Interaction Detection?
Rosario Leonardi (University of Catania), Giovanni Maria Farinella
Object DetectionData SynthesisDomain AdaptationSupervised Fine-TuningImageBenchmark
🎯 What it does: This paper investigates the effectiveness of synthetic data in first-person hand-object interaction detection, and constructs the HOI-Synth benchmark based on a new data generation pipeline and simulator.
ARoFace: Alignment Robustness to Improve Low-quality Face Recognition
Mohammad Saeed Ebrahimi Saadabadi (West Virginia University), Nasser Nasrabadi (West Virginia University)
RecognitionImage
🎯 What it does: This paper proposes the ARoFace method, which enhances the robustness of low-quality face recognition (LQ FR) by incorporating adversarial data augmentation with alignment error (FAE) during training.
ArtVLM: Attribute Recognition Through Vision-Based Prefix Language Modeling
William Yicheng Zhu (Google Research), Feng Yang (Google Research)
RecognitionRetrievalLarge Language ModelSupervised Fine-TuningVision Language ModelContrastive LearningImageTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: Propose a generative retrieval method for attribute recognition through visual prefix language modeling, restructuring the attribute recognition framework as an image-object-attribute conditional probability model.
Assessing Sample Quality via the Latent Space of Generative Models
Jingyi Xu (Stony Brook University), Dimitris Samaras (Stony Brook University)
GenerationData SynthesisDiffusion modelAuto EncoderGenerative Adversarial NetworkImagePoint CloudBiomedical Data
🎯 What it does: This paper proposes a latent density score based on the inherent latent space of a generative model to evaluate the quality of individual generated samples, directly measuring quality by calculating the Gaussian kernel density between the latent code of the sample and the latent codes of the training set;
Asymmetric Mask Scheme for Self-Supervised Real Image Denoising
Xiangyu Liao (Sichuan University), Chao Ren (Sichuan University)
RestorationImage
🎯 What it does: Propose AMSNet, a self-supervised denoising framework that employs single-mask training and multi-mask inference, compatible with various existing denoisers and avoiding limitations imposed by BSN.
Asynchronous Bioplausible Neuron for Spiking Neural Networks for Event-Based Vision
Hussain Sajwani (Khalifa University of Science and Technology), Sanket Mr Kachole (Kingston University)
ClassificationSegmentationSpiking Neural NetworkTime Series
🎯 What it does: Proposed Asynchronous Bioplastic Neuron (ABN) for spiking neural networks in event vision, achieving network stability and energy efficiency optimization through dynamic threshold regulation.
Asynchronous Large Language Model Enhanced Planner for Autonomous Driving
Yuan Chen (Beihang University), Si Liu (Beihang University)
Autonomous DrivingTransformerLarge Language ModelSupervised Fine-Tuning
🎯 What it does: Propose an asynchronous LLM-enhanced closed-loop planning framework called AsyncDriver, which uses LLM to extract scene-related instruction features to guide the real-time planner in generating more accurate and controllable trajectories.
Attention Beats Linear for Fast Implicit Neural Representation Generation
Shuyi Zhang (Zhejiang University), Haishuai Wang (Zhejiang University)
RestorationGenerationTransformerMultimodality
🎯 What it does: Proposed a local attention-based implicit neural representation (ANR) model that generates instance feature vectors using a transformation network and fuses them with coordinate features through a local attention layer, achieving efficient and fast continuous signal reconstruction.
Attention Decomposition for Cross-Domain Semantic Segmentation
Liqiang He (Oregon State University), Sinisa Todorovic (Oregon State University)
SegmentationDomain AdaptationTransformerGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: This paper proposes a new Transformer network called ADFormer for cross-domain semantic segmentation tasks;
Attention Prompting on Image for Large Vision-Language Models
Runpeng Yu (National University of Singapore), Xinchao Wang (National University of Singapore)
Explainability and InterpretabilityComputational EfficiencyRepresentation LearningTransformerPrompt EngineeringVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: Propose Attention Prompting on Image (API), which enhances the visual perception and answering performance of large visual language models (LVLM) by superimposing heatmaps related to text queries onto the original image as visual prompts.
Attention-Challenging Multiple Instance Learning for Whole Slide Image Classification
Yunlong Zhang (Zhejiang University), Lin Yang (Westlake University)
ClassificationTransformerBiomedical Data
🎯 What it does: Proposed a multi-instance learning (MIL) framework called ACMIL for whole slide image (WSI) classification, aiming to alleviate overfitting caused by excessive attention concentration.
AttentionHand: Text-driven Controllable Hand Image Generation for 3D Hand Reconstruction in the Wild
Junho Park (Sogang University), Suk-Ju Kang (Sogang University)
GenerationData SynthesisTransformerPrompt EngineeringDiffusion modelAuto EncoderImageTextMultimodality
🎯 What it does: Propose a text-driven controllable hand image generation framework called AttentionHand
AttnZero: Efficient Attention Discovery for Vision Transformers
Lujun Li, Yike Guo (Hong Kong University of Science and Technology)
ClassificationObject DetectionSegmentationComputational EfficiencyNeural Architecture SearchTransformerImageBenchmark
🎯 What it does: Propose the AttnZero framework, which leverages evolutionary search to automatically discover efficient linear attention modules, enhancing the performance and computational efficiency of Vision Transformers.
Audio-driven Talking Face Generation with Stabilized Synchronization Loss
Dogucan Yaman (Karlsruhe Institute of Technology), Alexander Waibel (Istanbul Technical University)
GenerationData SynthesisConvolutional Neural NetworkGenerative Adversarial NetworkVideoMultimodalityAudio
🎯 What it does: Propose an audio-driven speaker facial generation model to address the conflict between audio-lip synchronization and visual quality.
Audio-Synchronized Visual Animation
Lin Zhang (University of Wisconsin Madison), Pedro Morgado (University of Wisconsin Madison)
GenerationData SynthesisDiffusion modelVideoMultimodalityAudio
🎯 What it does: Proposed the Audio-Synchronized Image Animation (ASVA) task and constructed a high-quality AVSync15 dataset and AVSyncD diffusion model.
Audio-visual Generalized Zero-shot Learning the Easy Way
Shentong Mo (Carnegie Mellon University), Pedro Morgado (University of Wisconsin-Madison)
ClassificationRepresentation LearningConvolutional Neural NetworkTransformerContrastive LearningVideoTextMultimodalityAudio
🎯 What it does: Propose a concise framework named EZ-AVGZL that achieves cross-modal zero-shot learning through a supervised text-audio-visual contrastive loss.
AUFormer: Vision Transformers are Parameter-Efficient Facial Action Unit Detectors
Kaishen Yuan (Tianjin University), Jingyu Yang (Tianjin University)
RecognitionTransformerImage
🎯 What it does: Proposed AUFormer, a parameter-efficient facial action unit (AU) detection framework based on Vision Transformer, along with the MoKE collaboration mechanism and MDWA loss.
AugDETR: Improving Multi-scale Learning for Detection Transformer
Jinpeng Dong (Xi'an Jiaotong University), Nanning Zheng (Xi'an Jiaotong University)
Object DetectionTransformerImage
🎯 What it does: Under the existing DETR framework, the multi-scale learning capability is enhanced by introducing two modules: Hybrid Attention Encoder and Encoder-Mixing Cross-Attention, thereby improving object detection performance.
Augmented Neural Fine-tuning for Efficient Backdoor Purification
Nazmul Karim (University of Central Florida), Nazanin Rahnavard (University of Central Florida)
Computational EfficiencyAdversarial AttackSupervised Fine-TuningImageVideoTextPoint Cloud
🎯 What it does: This paper proposes a post-backdoor removal framework called NFT, which removes backdoors in deep networks by utilizing MixUp data augmentation and neural mask fine-tuning, avoiding the costly trigger reverse search.
AugUndo: Scaling Up Augmentations for Monocular Depth Completion and Estimation
Yangchao Wu (UCLA Vision Lab), Alex Wong (Yale Vision Lab)
Depth EstimationImage
🎯 What it does: Propose the AugUndo framework, enabling unsupervised depth completion and estimation to utilize richer geometric and photometric augmentations. The method reverses geometric transformations on the output depth, allowing the loss to be computed on the original input, thereby avoiding artifacts introduced by augmentations.
Auto-DAS: Automated Proxy Discovery for Training-free Distillation-aware Architecture Search
Haosen Sun (Hong Kong University of Science and Technology), Shitong Shao (Hong Kong University of Science and Technology Guangzhou)
OptimizationKnowledge DistillationNeural Architecture SearchImageBenchmark
🎯 What it does: This paper proposes an automated proxy discovery framework, Auto-DAS, for rapidly evaluating the distillation performance of candidate student models in training-free distillation-aware architecture search.