ICCV 2023 Papers — Page 2
IEEE/CVF International Conference on Computer Vision · 2156 papers
Aggregating Feature Point Cloud for Depth Completion
Zhu Yu (Zhejiang University), Hui-Liang Shen (Zhejiang University)
RestorationDepth EstimationAutonomous DrivingTransformerImagePoint Cloud
🎯 What it does: This paper proposes a feature point cloud aggregation framework called PointDC, which achieves depth completion by aggregating sparse 3D point clouds with 2D features from RGB images.
Agile Modeling: From Concept to Classifier in Minutes
Otilia Stretcu (Google Research), Ariel Fuxman
ClassificationRetrievalContrastive LearningImage
🎯 What it does: An Agile Modeling framework is proposed, allowing non-machine learning experts to build classifiers for subjective visual concepts in minutes through an interactive, active learning approach.
AIDE: A Vision-Driven Multi-View, Multi-Modal, Multi-Tasking Dataset for Assistive Driving Perception
Dingkang Yang (Fudan University), Lihua Zhang (Fudan University)
Autonomous DrivingConvolutional Neural NetworkTransformerVideoMultimodality
🎯 What it does: The AIDE driving perception dataset is proposed, and various baseline frameworks are constructed on this dataset, covering multi-view, multi-modal, and multi-task driving monitoring tasks.
Algebraically Rigorous Quaternion Framework for the Neural Network Pose Estimation Problem
Chen Lin (Flatiron Institute), Sonya M. Hanson (Flatiron Institute)
Pose EstimationPoint Cloud
🎯 What it does: A training framework for attitude estimation based on the Adjugate Quaternion has been proposed to address the singularity and discontinuity issues encountered with traditional scalar quaternions during neural network training.
AlignDet: Aligning Pre-training and Fine-tuning in Object Detection
Ming Li (ByteDance Inc), Xin Pan (ByteDance Inc)
Object DetectionContrastive LearningImage
🎯 What it does: This paper proposes AlignDet, a unified unsupervised pre-training framework that is divided into image domain pre-training (training the backbone) and box domain pre-training (training all modules of the detector) to address the data, model, and task mismatch issues between pre-training and fine-tuning, significantly improving the performance of various detectors.
Alignment Before Aggregation: Trajectory Memory Retrieval Network for Video Object Segmentation
Rui Sun (University of Science and Technology of China), Feng Wu (University of Science and Technology of China)
Object TrackingSegmentationConvolutional Neural NetworkAgentic AIVideo
🎯 What it does: A Trajectory Memory Retrieval Network (TMRN) is proposed for video object segmentation, which includes a spatial alignment module and a temporal aggregation module.
Alignment-free HDR Deghosting with Semantics Consistent Transformer
Steven Tel, Dominique Ginhac (University of Burgundy)
RestorationTransformerImage
🎯 What it does: A non-aligned HDR ghosting removal method (SCTNet) is proposed, utilizing a semantically consistent Transformer to simultaneously model foreground motion and background semantics, generating high-quality HDR images.
ALIP: Adaptive Language-Image Pre-Training with Synthetic Caption
Kaicheng Yang (DeepGlint), Tongliang Liu (University of Sydney)
ClassificationRetrievalTransformerVision Language ModelContrastive LearningImageText
🎯 What it does: This paper proposes a dual-path adaptive contrastive pre-training framework called ALIP, which enhances image-text alignment using synthetic titles generated by OFA.
All in Tokens: Unifying Output Space of Visual Tasks via Soft Token
Jia Ning (Huazhong University of Science and Technology), Han Hu (Huazhong University of Science and Technology)
SegmentationPose EstimationDepth EstimationTransformerAuto EncoderImage
🎯 What it does: A unified framework for visual task output space, AiT, is proposed, which discretizes the high-dimensional outputs of different visual tasks into tokens and uses autoregressive Transformers for prediction, supporting various tasks such as depth estimation, instance segmentation, and keypoint detection.
All-to-Key Attention for Arbitrary Style Transfer
Mingrui Zhu (Xidian University), Xinbo Gao (Chongqing University of Post and Telecommunications)
Image TranslationConvolutional Neural NetworkGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: A novel all-key attention (A2K) mechanism is proposed for efficient and stable arbitrary style transfer.
All4One: Symbiotic Neighbour Contrastive Learning via Self-Attention and Redundancy Reduction
Imanol G. Estepa (Universitat de Barcelona), Petia Radeva (Universitat de Barcelona)
Representation LearningTransformerContrastive LearningImage
🎯 What it does: The All4One framework is proposed, which generates centroid representations through self-attention, merging multi-neighbor information into a single centroid and comparing it with the original image, further combining redundancy reduction objectives to achieve self-supervised feature learning.
Alleviating Catastrophic Forgetting of Incremental Object Detection via Within-Class and Between-Class Knowledge Distillation
Mengxue Kang (Intelligent Science and Technology Academy of China Aerospace Science and Industry Corporation), Xuhui Huang (Institute of Automation Chinese Academy of Sciences)
Object DetectionKnowledge DistillationTransformerImage
🎯 What it does: This study addresses the problem of catastrophic forgetting in Incremental Object Detection (IOD) and proposes to perform both between-class and within-class knowledge distillation in Transformer detectors to maintain the structure of the semantic feature space.
ALWOD: Active Learning for Weakly-Supervised Object Detection
Yuting Wang (Rutgers University), Vladimir Pavlovic (Rutgers University)
Object DetectionTransformerImage
🎯 What it does: The ALWOD framework is proposed, which combines active learning with weak/semi-supervised object detection, achieving efficient labeling and training through an auxiliary image generator and a student-teacher network.
Among Us: Adversarially Robust Collaborative Perception by Consensus
Yiming Li (New York University), Chen Feng (New York University)
Object DetectionAutonomous DrivingAdversarial AttackPoint Cloud
🎯 What it does: The ROBOSAC framework is proposed, utilizing the Random Sample Consensus (RANSAC) idea to achieve robustness against collaborative perception adversarial attacks;
An Adaptive Model Ensemble Adversarial Attack for Boosting Adversarial Transferability
Bin Chen (Fuzhou University), Ximeng Liu (Fuzhou University)
Adversarial AttackConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes AdaEA, which utilizes adaptive gradient modulation and a difference reduction filter to optimize multi-model ensemble attacks, enhancing the transferability of adversarial attacks between CNNs and ViTs.
An Embarrassingly Simple Backdoor Attack on Self-supervised Learning
Changjiang Li (Pennsylvania State University), Ting Wang (Pennsylvania State University)
Representation LearningAdversarial AttackContrastive LearningImage
🎯 What it does: This paper proposes a self-supervised learning backdoor attack method named CTRL, which can implant a backdoor by contaminating a small amount of training data.
Anatomical Invariance Modeling and Semantic Alignment for Self-supervised Learning in 3D Medical Image Analysis
Yankai Jiang (Alibaba Group), Minfeng Xu (Alibaba Group)
SegmentationRepresentation LearningTransformerContrastive LearningImageBiomedical DataComputed Tomography
🎯 What it does: This paper proposes a self-supervised learning framework called Alice, which utilizes cross-volume sampling to obtain positive samples of the same anatomical structure and performs semantic alignment within the same volume, thereby explicitly modeling anatomically invariant features.
Anchor Structure Regularization Induced Multi-view Subspace Clustering via Enhanced Tensor Rank Minimization
Jintian Ji (Beijing Jiaotong University), Songhe Feng (Beijing Jiaotong University)
OptimizationImage
🎯 What it does: This paper proposes a multi-view subspace clustering method based on anchor structure regularization and enhanced tensor rank minimization (ASR-ETR).
Anchor-Intermediate Detector: Decoupling and Coupling Bounding Boxes for Accurate Object Detection
Yilong Lv (Xi'an Institute of High Technology), Aitao Yang (Xi'an Institute of High Technology)
Object DetectionConvolutional Neural NetworkImage
🎯 What it does: By combining anchor-based and anchor-free detection frameworks, and introducing Box Decouple-Couple (BDC) and Corner-Aware heads, we propose the Anchor-Intermediate Detector (AID) to improve object detection accuracy.
Animal3D: A Comprehensive Dataset of 3D Animal Pose and Shape
Jiacong Xu, Adam Kortylewski
Pose EstimationImageBenchmark
🎯 What it does: The Animal3D dataset is proposed, achieving 3D pose and shape annotations for multiple species (40 species), including 26 key points and SMAL model parameters.
Anomaly Detection Under Distribution Shift
Tri Cao (Singapore Management University), Guansong Pang (Singapore Management University)
Anomaly DetectionKnowledge DistillationImageBenchmark
🎯 What it does: An unsupervised anomaly detection method called GNL is proposed for scenarios with distribution shifts, and benchmarks are established on four commonly used datasets.
Anomaly Detection using Score-based Perturbation Resilience
Woosang Shin (Korea Institute of Industrial Technology), Jong Pil Yun (University of Science and Technology)
Anomaly DetectionDiffusion modelScore-based ModelImageBenchmark
🎯 What it does: This paper proposes a method for unsupervised anomaly detection using a score-based diffusion model, with the core idea being to use the 'restoration error' from the score recovery of samples after perturbation as an anomaly metric.
Anti-DreamBooth: Protecting Users from Personalized Text-to-image Synthesis
Thanh Van Le (VinAI Research), Anh Tran (VinAI Research)
GenerationSafty and PrivacyAdversarial AttackDiffusion modelImage
🎯 What it does: In response to the potential misuse of the DreamBooth personalized text-image model, this paper proposes Anti-DreamBooth, which adds nearly invisible perturbations before users upload images, resulting in poor quality or distorted personalized images generated by any DreamBooth fine-tuning model trained on these images, thereby protecting user privacy.
Aperture Diffraction for Compact Snapshot Spectral Imaging
Tao Lv (Nanjing University), Xun Cao (Nanjing University)
RestorationTransformerImage
🎯 What it does: A compact snapshot spectral imaging system ADIS, composed solely of an ultra-thin orthogonal aperture mask and a standard imaging lens, has been designed, and a spectral reconstruction algorithm CSST based on this system has been proposed.
AREA: Adaptive Reweighting via Effective Area for Long-Tailed Classification
Xiaohua Chen (Institute of Information Engineering), Weiping Wang (Institute of Information Engineering)
ClassificationImage
🎯 What it does: The AREA (Adaptive Reweighting via Effective Area) method is proposed, which improves the class imbalance problem in long-tail classification tasks by calculating the effective area for each category to achieve adaptive reweighting.
Aria Digital Twin: A New Benchmark Dataset for Egocentric 3D Machine Perception
Xiaqing Pan (Meta Reality Labs), Yuheng (Carl) Ren (Meta Reality Labs)
Object DetectionSegmentationPose EstimationDomain AdaptationSimultaneous Localization and MappingImageMultimodalityBenchmark
🎯 What it does: The Aria Digital Twin (ADT) dataset is proposed and released, containing complete annotations of 200 sequences of egocentric 3D machine perception.
ARNOLD: A Benchmark for Language-Grounded Task Learning with Continuous States in Realistic 3D Scenes
Ran Gong (University of California), Siyuan Huang (National Key Laboratory of General Artificial Intelligence)
Robotic IntelligenceReinforcement LearningImageVideoBenchmark
🎯 What it does: The ARNOLD benchmark has been constructed, which includes 8 language-driven robotic tasks aimed at continuous states, providing real 3D scenes, multi-camera observations, 10k expert demonstrations, and various generalization data splits; a systematic evaluation of language and state understanding has also been implemented.
ASAG: Building Strong One-Decoder-Layer Sparse Detectors via Adaptive Sparse Anchor Generation
Shenghao Fu (Sun Yat-sen University), Wei-Shi Zheng (Sun Yat-sen University)
Object DetectionTransformerImage
🎯 What it does: A single-decoder sparse detector based on the Adaptive Sparse Anchor Generator (ASAG) is proposed, which initializes queries by dynamically generating sparse anchors in the image.
ASIC: Aligning Sparse in-the-wild Image Collections
Kamal Gupta (Google), Abhishek Kar (Google)
RecognitionObject DetectionTransformerContrastive LearningImage
🎯 What it does: A self-supervised low-shot dense correspondence method named ASIC is proposed, which can achieve globally consistent dense registration of objects or categories using only 10 to 30 wild images.
ASM: Adaptive Skinning Model for High-Quality 3D Face Modeling
Kai Yang (Tencent AI Lab), Wei Yang (Tencent AI Lab)
GenerationOptimizationImage
🎯 What it does: A high-capacity 3D face parameter model ASM with adaptive skin binding is proposed for face reconstruction from multi-view uncalibrated images.
AssetField: Assets Mining and Reconfiguration in Ground Feature Plane Representation
Yuanbo Xiangli (Chinese University of Hong Kong), Dahua Lin (Chinese University of Hong Kong)
Data SynthesisRepresentation LearningNeural Radiance FieldPoint Cloud
🎯 What it does: The AssetField framework is proposed, which uses ground feature planes to represent scenes, enabling asset mining, grouping, and cross-scene asset library construction, while supporting object-level, category-level, and scene-level editing.
Atmospheric Transmission and Thermal Inertia Induced Blind Road Segmentation with a Large-Scale Dataset TBRSD
Junzhang Chen (Beihang University), Xiangzhi Bai (Beihang University)
Object DetectionSegmentationTransformerImage
🎯 What it does: This paper proposes a blind road semantic segmentation network based on thermal infrared images, and for the first time introduces two physical models, atmospheric transmission and thermal inertia effects, into the network, improving the accuracy of blind road detection in low-light environments.
ATT3D: Amortized Text-to-3D Object Synthesis
Jonathan Lorraine (NVIDIA Corporation), James Lucas (NVIDIA Corporation)
GenerationData SynthesisOptimizationDiffusion modelNeural Radiance FieldTextPoint CloudMesh
🎯 What it does: The ATT3D method is proposed, which generates high-quality 3D objects in just 1 second (1 GPU) during inference by training a unified model with multiple text prompts at once.
Attention Discriminant Sampling for Point Clouds
Cheng-Yao Hong (Academia Sinica), Tyng-Luh Liu (National Taiwan University)
ClassificationObject DetectionSegmentationPoint Cloud
🎯 What it does: An attention-based structure-aware sampling method ADS is proposed, which can simultaneously consider geometric and semantic information during point cloud sampling.
Attention Where It Matters: Rethinking Visual Document Understanding with Selective Region Concentration
Haoyu Cao (Tencent YouTu Lab), Xing Sun (Tencent YouTu Lab)
RecognitionGenerationComputational EfficiencyTransformerImageText
🎯 What it does: A novel end-to-end document understanding model named SeRum is proposed, which generates text directly from images, eliminating the OCR stage. It utilizes a query decoder to locate areas of interest and performs local decoding through content-aware token merging.
Attentive Mask CLIP
Yifan Yang (Microsoft Research), Yuqing Yang (Microsoft Research)
ClassificationRetrievalComputational EfficiencyTransformerContrastive LearningImage
🎯 What it does: An efficient CLIP pre-training framework A-CLIP is constructed through attention-driven sparse cropping of image tokens.
AttT2M: Text-Driven Human Motion Generation with Multi-Perspective Attention Mechanism
Chongyang Zhong (Institute of Computing Technology Chinese Academy of Sciences), Shihong Xia (Institute of Computing Technology Chinese Academy of Sciences)
GenerationData SynthesisTransformerAuto EncoderVideoTextMultimodality
🎯 What it does: This paper proposes AttT2M, a two-stage multi-view attention mechanism for text-driven human action generation. The first stage utilizes a spatiotemporal encoder with body part attention and VQ-VAE to learn a discrete latent space; the second stage captures the cross-modal correspondence between text and actions through global-local attention (sentence-level conditional self-attention + word-level cross-attention) and generates action sequences using a generative Transformer.
Audio-Enhanced Text-to-Video Retrieval using Text-Conditioned Feature Alignment
Sarah Ibrahimi (University of Amsterdam), Mohamed Omar (Amazon Prime Video)
RetrievalTransformerContrastive LearningVideoTextMultimodalityAudio
🎯 What it does: A text-conditioned audio-video feature alignment method (TEFAL) is proposed, which performs cross-attention alignment between audio and video features with text in text retrieval tasks, and then sums the two to obtain the final representation.
Audio-Visual Class-Incremental Learning
Weiguo Pian (University of Texas at Dallas), Yapeng Tian (University of Texas at Dallas)
ClassificationKnowledge DistillationConvolutional Neural NetworkContrastive LearningVideoMultimodalityBenchmarkAudio
🎯 What it does: Proposes an audio-visual joint incremental learning framework AV-CIL, focusing on the issues of forgetting and association loss between audio and visual modalities in incremental learning.
Audio-Visual Deception Detection: DOLOS Dataset and Parameter-Efficient Crossmodal Learning
Xiaobao Guo (Nanyang Technological University), Alex Kot (Nanyang Technological University)
ClassificationRecognitionAnomaly DetectionTransformerSupervised Fine-TuningVideoMultimodalityAudio
🎯 What it does: This study investigates audio-visual deception detection in video dialogues based on game shows, constructs the largest DOLOS dataset, and proposes a parameter-efficient cross-modal learning framework called PECL.
Audio-Visual Glance Network for Efficient Video Recognition
Muhammad Adi Nugroho (Korea Advanced Institute of Science and Technology), Changick Kim (Korea Advanced Institute of Science and Technology)
RecognitionComputational EfficiencyTransformerReinforcement LearningVideoMultimodalityAudio
🎯 What it does: Designed the Audio-Visual Glance Network (AVGN), which utilizes lightweight audio and visual encoders to first extract global features, then filters key frames through the Audio-Visual Temporal Saliency Transformer (AV-TeST), and processes only spatiotemporally important areas using Audio-Enhanced Patch Attention (AESPA) and a patch extraction network, achieving efficient video recognition.
Audiovisual Masked Autoencoders
Mariana-Iuliana Georgescu (Google Research), Anurag Arnab (Google Research)
ClassificationDomain AdaptationRepresentation LearningTransformerAuto EncoderVideoMultimodalityAudio
🎯 What it does: This paper proposes an audiovisual masked autoencoder (Audiovisual MAE) that learns audio and video features through joint encoding and cross-modal reconstruction.
Augmented Box Replay: Overcoming Foreground Shift for Incremental Object Detection
Yuyang Liu (Chinese Academy of Sciences), Joost van de Weijer (University Autonoma de Barcelona)
Object DetectionKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: This paper addresses the foreground drift problem in incremental object detection and proposes solutions including Augmented Box Replay (ABR) and Attentive RoI Distillation.
Augmenting and Aligning Snippets for Few-Shot Video Domain Adaptation
Yuecong Xu (Institute for Infocomm Research), Xiaoli Li (Institute for Infocomm Research)
RecognitionDomain AdaptationTransformerVideo
🎯 What it does: A new method called SSA2lign is proposed to address the problem of video domain adaptation with only a small number of target video samples.
AutoAD II: The Sequel - Who, When, and What in Movie Audio Description
Tengda Han (Visual Geometry Group, University of Oxford), Andrew Zisserman (Visual Geometry Group, University of Oxford)
RecognitionGenerationTransformerLarge Language ModelVision Language ModelVideoMultimodalityAudio
🎯 What it does: This study proposes a complete movie audio description generation system, which includes three sub-tasks: time segment prediction, character recognition, and description generation, capable of automatically generating contextually appropriate descriptive text during dialogue pauses.
AutoDiffusion: Training-Free Optimization of Time Steps and Architectures for Automated Diffusion Model Acceleration
Lijiang Li (Xiamen University), Rongrong Ji (Xiamen University)
GenerationOptimizationDiffusion modelImage
🎯 What it does: This paper proposes AutoDiffusion, which utilizes a training-free evolutionary search to simultaneously find the optimal time step sequence and a compressed noise prediction network structure to accelerate the generation of diffusion models.
Automated Knowledge Distillation via Monte Carlo Tree Search
Lujun Li (Hong Kong University of Science and Technology), Ya Yang (City University of Hong Kong)
ClassificationObject DetectionSegmentationKnowledge DistillationTransformerImage
🎯 What it does: Proposes Auto-KD, which automates the design of knowledge distillation for the first time, constructing a unified tree search space and using Monte Carlo Tree Search for efficient searching;
Automatic Animation of Hair Blowing in Still Portrait Photos
Wenpeng Xiao (ByteDance Intelligent Creation Lab), Bing Li (King Abdullah University of Science and Technology)
SegmentationGenerationTransformerImageVideo
🎯 What it does: This paper proposes a user intervention-free algorithm that can automatically blow hair in static portrait photos, generating high-quality cinemagraph videos.
Automatic Network Pruning via Hilbert-Schmidt Independence Criterion Lasso under Information Bottleneck Principle
Song Guo (Xiamen University), Rongrong Ji (Xiamen University)
CompressionOptimizationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes an automatic channel pruning method called APIB based on the Information Bottleneck (IB) principle, which uses HSIC Lasso to solve the IB approximation and automatically determines the pruning ratio for each layer.
AutoReP: Automatic ReLU Replacement for Fast Private Network Inference
Hongwu Peng (University of Connecticut), Caiwen Ding (University of Connecticut)
Safty and PrivacyComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: We propose AutoReP, an automatic ReLU replacement framework for private inference that significantly reduces ReLU operations while maintaining high accuracy.
AutoSynth: Learning to Generate 3D Training Data for Object Point Cloud Registration
Zheng Dang (École Polytechnique Fédérale de Lausanne), Mathieu Salzmann (École Polytechnique Fédérale de Lausanne)
Data SynthesisDomain AdaptationOptimizationAuto EncoderPoint Cloud
🎯 What it does: Automatically generate and select a synthetic training dataset suitable for 3D point cloud registration (AutoSynth).
Auxiliary Tasks Benefit 3D Skeleton-based Human Motion Prediction
Chenxin Xu (Shanghai Jiao Tong University), Yanfeng Wang (Shanghai Jiao Tong University)
Pose EstimationTransformerTime Series
🎯 What it does: The AuxFormer framework is proposed, which simultaneously learns the main prediction task and two auxiliary tasks (denoising and occlusion recovery) in the 3D skeleton motion prediction task, enhancing the model's ability to capture spatial-temporal dependencies through auxiliary learning.
AvatarCraft: Transforming Text into Neural Human Avatars with Parameterized Shape and Pose Control
Ruixiang Jiang (Hong Kong Polytechnic University), Jing Liao (City University of Hong Kong)
GenerationData SynthesisPose EstimationDiffusion modelScore-based ModelNeural Radiance FieldMesh
🎯 What it does: This paper proposes the AvatarCraft method, which utilizes diffusion models to guide neural implicit fields in generating high-quality, animatable human avatars from text prompts, supporting pose and shape control through SMPL parameters.
Backpropagation Path Search On Adversarial Transferability
Zhuoer Xu (Ant Group), Weiqiang Wang (Ant Group)
Adversarial AttackConvolutional Neural NetworkImage
🎯 What it does: A framework for backpropagation path search (PAS) based on structural search is designed and implemented, which enhances the transferability of adversarial samples on black-box models by reparameterizing convolutional layers with SkipConv and constructing a DAG search space.
BallGAN: 3D-aware Image Synthesis with a Spherical Background
Minjung Shin (Yonsei University), Youngjung Uh (Yonsei University)
GenerationData SynthesisNeural Radiance FieldGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a 3D-aware GAN framework called BallGAN, which utilizes a spherical surface to approximate the background, achieving clear separation of foreground and background in an unsupervised manner from single-view data, and supports rendering the foreground from any viewpoint.
BANSAC: A Dynamic BAyesian Network for Adaptive SAmple Consensus
Valter Piedade (Instituto Superior Tecnico), Pedro Miraldo (Mitsubishi Electric Research Labs)
Pose EstimationOptimizationComputational EfficiencyReinforcement LearningImage
🎯 What it does: This paper proposes BANSAC, a variant of RANSAC that utilizes dynamic Bayesian networks to dynamically update the inlier probabilities of each matching point and performs probability-weighted sampling; it also provides a probability-based termination criterion.
BaRe-ESA: A Riemannian Framework for Unregistered Human Body Shapes
Emmanuel Hartman (Florida State University), Mohamed Daoudi (Institut Mines-Télécom)
GenerationOptimizationMesh
🎯 What it does: The study proposes the BaRe-ESA framework, which can directly perform shape representation, interpolation, extrapolation, and random generation on unregistered human meshes without the need for point correspondence or consistent meshes.
Batch-based Model Registration for Fast 3D Sherd Reconstruction
Jiepeng Wang (University of Hong Kong), Wenping Wang (Texas A&M University)
Object DetectionSegmentationOptimizationPoint CloudMesh
🎯 What it does: After scanning multiple ceramic shards in a single session, the partial 3D reconstruction results on the front and back sides are automatically matched and precisely registered using Bidirectional Boundary ICP (BBICP), ultimately resulting in a high-precision 3D model of the complete ceramic shard.
Bayesian Optimization Meets Self-Distillation
HyunJae Lee (Lunit Inc), Donggeun Yoo (Lunit Inc)
ClassificationOptimizationKnowledge DistillationConvolutional Neural NetworkImageBiomedical Data
🎯 What it does: By combining Bayesian optimization with self-distillation, the BOSS framework is proposed to utilize the parameters and performance knowledge of previously trained models in each BO iteration to enhance the final model performance.
Bayesian Prompt Learning for Image-Language Model Generalization
Mohammad Mahdi Derakhshani (University of Amsterdam), Brais Martinez (Samsung AI Cambridge)
ClassificationDomain AdaptationTransformerPrompt EngineeringVision Language ModelImageText
🎯 What it does: By viewing prompt learning as Bayesian variational inference, a Bayesian prompt learning method is proposed to regularize the prompt space and enhance the generalization ability of unseen prompts.
Be Everywhere - Hear Everything (BEE): Audio Scene Reconstruction by Sparse Audio-Visual Samples
Mingfei Chen (University of Washington), Eli Shlizerman (University of Washington)
GenerationData SynthesisMultimodalityAudio
🎯 What it does: A BEE framework is proposed, utilizing sparse audio and video sensors to reconstruct spatial audio in real-time for any listener position in dynamic scenes.
Beating Backdoor Attack at Its Own Game
Min Liu (Carnegie Mellon University), Xiangyu Yue (Chinese University of Hong Kong)
Anomaly DetectionAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a Non-Adversarial Backdoor (NAB) framework that defends against backdoor attacks by injecting non-adversarial backdoors into a small number of suspected samples to suppress original backdoor attacks.
BeLFusion: Latent Diffusion for Behavior-Driven Human Motion Prediction
German Barquero (Universitat de Barcelona), Cristina Palmero (Universitat de Barcelona)
GenerationData SynthesisPose EstimationRecurrent Neural NetworkDiffusion modelAuto EncoderVideoTime Series
🎯 What it does: A latent diffusion model based on behavior decoupled latent space (BeLFusion) is designed to generate coherent and diverse random human motion prediction sequences that align with observed historical movements.
Benchmarking Algorithmic Bias in Face Recognition: An Experimental Approach Using Synthetic Faces and Human Evaluation
Hao Liang, Guha Balakrishnan
RecognitionData SynthesisGenerative Adversarial NetworkImageBenchmark
🎯 What it does: This paper proposes an experimental method based on synthetic faces and human judgment to quantify the biases of facial recognition systems regarding protected attributes such as race and gender.
Benchmarking and Analyzing Robust Point Cloud Recognition: Bag of Tricks for Defending Adversarial Examples
Qiufan Ji (New Jersey Institute of Technology), Lichao Sun (Rutgers University)
ClassificationRecognitionAdversarial AttackPoint CloudBenchmark
🎯 What it does: This paper first establishes a unified and rigorous 3D point cloud adversarial robustness benchmark to systematically evaluate various attack and defense methods; subsequently, it proposes a Hybrid Training and a 'Bag-of-Tricks' defense framework based on preprocessing, reconstruction, and augmentation, and validates its effectiveness on this benchmark.
Benchmarking Low-Shot Robustness to Natural Distribution Shifts
Aaditya Singh (Georgia Institute of Technology), Judy Hoffman (Georgia Institute of Technology)
ClassificationDomain AdaptationData-Centric LearningConvolutional Neural NetworkTransformerSupervised Fine-TuningContrastive LearningImageBiomedical DataBenchmark
🎯 What it does: Under limited labeled samples (low sample), the system evaluates the impact of different pre-training strategies, model architectures, and robustness interventions on out-of-distribution (OOD) robustness, providing an in-depth comparison of three datasets: ImageNet, iWildCam, and Camelyon for the first time.
Betrayed by Captions: Joint Caption Grounding and Generation for Open Vocabulary Instance Segmentation
Jianzong Wu (Peking University), Chen Change Loy (Nanyang Technological University)
Object DetectionSegmentationGenerationTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: This paper proposes a framework for joint image annotation alignment and generation (CGG) for open vocabulary instance segmentation tasks.
Better May Not Be Fairer: A Study on Subgroup Discrepancy in Image Classification
Ming-Chang Chiu (University of Southern California), Xuezhe Ma (University of Southern California)
ClassificationData-Centric LearningConvolutional Neural NetworkFlow-based ModelImage
🎯 What it does: Manually annotated the CIFAR-10/100 test set based on background color to generate the CIFAR-B dataset, studied subgroup performance differences, and proposed the FlowAug semantic data augmentation method to reduce subgroup differences.
BEV-DG: Cross-Modal Learning under Bird's-Eye View for Domain Generalization of 3D Semantic Segmentation
Miaoyu Li (Xiamen University), Yun Fu (Northeastern University)
SegmentationDomain AdaptationAutonomous DrivingContrastive LearningMultimodalityPoint Cloud
🎯 What it does: A cross-modal learning framework BEV-DG based on a bird's-eye view is proposed for domain generalization in 3D semantic segmentation.
BEVPlace: Learning LiDAR-based Place Recognition using Bird's Eye View Images
Lun Luo (Zhejiang University), Hui-Liang Shen (Zhejiang University)
RecognitionPose EstimationAutonomous DrivingConvolutional Neural NetworkSimultaneous Localization and MappingPoint CloudBenchmark
🎯 What it does: The study uses Bird's Eye View (BEV) as a representation of LiDAR point clouds and proposes a rotation-invariant network called BEVPlace based on group convolution and NetVLAD for place recognition and location estimation of point clouds.
Beyond Image Borders: Learning Feature Extrapolation for Unbounded Image Composition
Xiaoyu Liu (Harbin Institute of Technology), Wangmeng Zuo (Harbin Institute of Technology)
Image TranslationRecommendation SystemTransformerImage
🎯 What it does: A joint model UNIC is proposed, providing boundaryless recommendations for camera angles and image cropping, achieving automatic adjustment of perspective and composition during shooting.
Beyond Object Recognition: A New Benchmark towards Object Concept Learning
Yong-Lu Li (Shanghai Jiao Tong University), Cewu Lu (Shanghai Jiao Tong University)
RecognitionObject DetectionConvolutional Neural NetworkTransformerImageBenchmark
🎯 What it does: This paper proposes the Object Concept Learning (OCL) task, constructs a large-scale knowledge base with three layers of causal annotations: categories, attributes, and usability, and introduces the Object Concept Reasoning Network (OCRN) based on causal intervention.
Beyond One-to-One: Rethinking the Referring Image Segmentation
Yutao Hu (University of Hong Kong), Ping Luo (University of Hong Kong)
SegmentationTransformerContrastive LearningImageTextMultimodality
🎯 What it does: A Dual Multi-Modal Interaction (DMMI) network is proposed to address the segmentation difficulties when natural language descriptions point to multiple targets or no targets.
Beyond Single Path Integrated Gradients for Reliable Input Attribution via Randomized Path Sampling
Giyoung Jeon (LG AI Research), Jaesik Choi (KAIST)
Explainability and InterpretabilityConvolutional Neural NetworkImage
🎯 What it does: A method for input attribution based on random path sampling is proposed—Stick-breaking Path Integration (SPI), which reduces the noise and unreliability of traditional Integrated Gradients by averaging gradient integrals over different paths.
Beyond Skin Tone: A Multidimensional Measure of Apparent Skin Color
William Thong (Sony AI), Alice Xiang (Sony AI)
RecognitionSegmentationGenerationGenerative Adversarial NetworkImage
🎯 What it does: Proposes and evaluates a multidimensional skin color measurement method based on the CIELAB color space (L* represents skin color depth, h* represents skin color hue), and uses this method to detect skin color bias in image datasets and computer vision models.
Beyond the Limitation of Monocular 3D Detector via Knowledge Distillation
Yiran Yang (Chinese Academy of Sciences), Xinming Li (Chinese Academy of Sciences)
Object DetectionAutonomous DrivingKnowledge DistillationPoint CloudBenchmark
🎯 What it does: A 3D monocular detection knowledge distillation framework is proposed, utilizing implicit deep distribution for perspective-induced feature imitation and depth-guided prediction distillation.
Beyond the Pixel: a Photometrically Calibrated HDR Dataset for Luminance and Color Prediction
Christophe Bolduc (Laval University), Jean-François Lalonde (Laval University)
RestorationSegmentationConvolutional Neural NetworkImage
🎯 What it does: This study conducted photometric calibration of HDR panoramic images captured by the Canon 5D Mark III in conjunction with a colorimeter, establishing the first large-scale indoor photometric calibration HDR dataset. Based on this, three new tasks were defined: pixel-level brightness prediction, pixel-level chromaticity prediction, and plane illuminance prediction, utilizing the U-Net network for training and evaluation on this dataset.
Bidirectional Alignment for Domain Adaptive Detection with Transformers
Liqiang He (Oregon State University), Sinisa Todorovic (Oregon State University)
Object DetectionDomain AdaptationTransformerImage
🎯 What it does: A cross-domain object detection method named BiADT is proposed, which separates domain-invariant and domain-specific features for each token in the encoder and decoder of the Transformer, and achieves alignment of domain-invariant features and distinction of domain-specific features through bidirectional alignment.
Bidirectionally Deformable Motion Modulation For Video-based Human Pose Transfer
Wing-Yin Yu (City University of Hong Kong), Kun Li (City University of Hong Kong)
Image TranslationGenerationPose EstimationRecurrent Neural NetworkGenerative Adversarial NetworkVideo
🎯 What it does: An end-to-end video human pose transfer framework is designed, utilizing deformable motion modulation and bidirectional recursive propagation to generate continuous frames.
BiFF: Bi-level Future Fusion with Polyline-based Coordinate for Interactive Trajectory Prediction
Yiyao Zhu (Hong Kong University of Science and Technology), Shaojie Shen (Hong Kong University of Science and Technology)
Autonomous DrivingTransformerPoint Cloud
🎯 What it does: The BiFF model is proposed, which generates scene-consistent interaction trajectory predictions through High-level Future Intention Fusion (HFIF) and Low-level Future Behavior Fusion (LFBF).
Bird's-Eye-View Scene Graph for Vision-Language Navigation
Rui Liu (Zhejiang University), Yi Yang (Zhejiang University)
Object DetectionRobotic IntelligenceTransformerVision Language ModelMultimodalityPoint Cloud
🎯 What it does: By constructing a scene graph based on Bird's Eye View (BEV), this paper achieves three-dimensional perception of indoor environments and utilizes the scene graph for visual-language navigation decision-making.
BiViT: Extremely Compressed Binary Vision Transformers
Yefei He (Zhejiang University), Bohan Zhuang (Monash University)
ClassificationObject DetectionSegmentationCompressionKnowledge DistillationTransformerImage
🎯 What it does: A method for binarizing the Vision Transformer (ViT) model is proposed, mainly addressing the softmax distribution of the self-attention module and the issue of retaining pre-training information. Three techniques are constructed: Softmax-aware Binarization, Cross-layer Binarization, and Parameterized Weight Scales.
Black Box Few-Shot Adaptation for Vision-Language Models
Yassine Ouali (Samsung AI), Georgios Tzimiropoulos (Queen Mary University of London)
Domain AdaptationPrompt EngineeringVision Language ModelImage
🎯 What it does: This paper proposes a novel black-box few-shot visual-language model adaptation method called LFA, which achieves cross-domain adaptation using only pre-computed image and text features without accessing model weights.
Black-Box Unsupervised Domain Adaptation with Bi-Directional Atkinson-Shiffrin Memory
Jingyi Zhang (Nanyang Technological University), Shijian Lu (Nanyang Technological University)
ClassificationObject DetectionSegmentationDomain AdaptationImage
🎯 What it does: This study focuses on black-box unsupervised domain adaptation and proposes the BiMem bidirectional memory mechanism to correct pseudo-labels and alleviate the 'forgetting' problem during the training process.
BlendFace: Re-designing Identity Encoders for Face-Swapping
Kaede Shiohara (University of Tokyo), Takafumi Taketomi (CyberAgent AI Lab)
RecognitionImage TranslationGenerationGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes BlendFace, a novel identity encoder for the face swapping task, which eliminates the attribute bias of traditional recognition models (such as ArcFace) by using synthetic attribute-mixed images during the pre-training phase, achieving better separation of identity and attributes.
Blending-NeRF: Text-Driven Localized Editing in Neural Radiance Fields
Hyeonseop Song (LG Electronics), Taehyeong Kim (Seoul National University)
GenerationData SynthesisNeural Radiance FieldContrastive LearningImage
🎯 What it does: This paper proposes Blending-NeRF, which achieves text-based local 3D object editing, including color changes and density adjustments, by freezing a pre-trained NeRF and training an editable NeRF.
BlindHarmony: "Blind" Harmonization for MR Images via Flow Model
Hwihun Jeong (Seoul National University), Jongho Lee (Seoul National University)
Image HarmonizationSegmentationFlow-based ModelImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: We propose BlindHarmony, a MR image harmonization framework that trains solely on target domain data, capable of effectively harmonizing unseen source domain images.
Body Knowledge and Uncertainty Modeling for Monocular 3D Human Body Reconstruction
Yufei Zhang (Rensselaer Polytechnic Institute), Qiang Ji (Rensselaer Polytechnic Institute)
Pose EstimationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A two-stage probabilistic network (KNOWN) based on human knowledge and uncertainty modeling is proposed, enabling the reconstruction of 3D human posture and shape from monocular RGB images without the need for 3D annotations.
Bold but Cautious: Unlocking the Potential of Personalized Federated Learning through Cautiously Aggressive Collaboration
Xinghao Wu (Beihang University), Shaojie Tang (University of Texas at Dallas)
Federated LearningImage
🎯 What it does: This paper proposes a personalized federated learning framework called FedCAC, which is based on parameter sensitivity and client data distribution similarity, enabling better collaboration and personalization among clients in non-IID scenarios.
BoMD: Bag of Multi-label Descriptors for Noisy Chest X-ray Classification
Yuanhong Chen (Australian Institute for Machine Learning, University of Adelaide), Gustavo Carneiro (Centre for Vision, Speech and Signal Processing, University of Surrey)
ClassificationConvolutional Neural NetworkLarge Language ModelImageBiomedical DataBenchmark
🎯 What it does: A two-stage method for handling noisy multi-label chest X-ray image classification is proposed—Bag of Multi-Label Descriptors (BoMD).
Boosting 3-DoF Ground-to-Satellite Camera Localization Accuracy via Geometry-Guided Cross-View Transformer
Yujiao Shi (Australian National University), Hongdong Li (Australian National University)
Pose EstimationAutonomous DrivingOptimizationTransformerImage
🎯 What it does: This paper proposes a geometry-guided cross-view Transformer and a neural pose optimizer to achieve refined 3-DoF (position and orientation) localization after matching ground camera and satellite images.
Boosting Adversarial Transferability via Gradient Relevance Attack
Hegui Zhu (Northeastern University), Wuming Jiang (Beijing EyeCool Technology)
Adversarial AttackImage
🎯 What it does: This paper proposes the Gradient Relevance Attack (GRA), which enhances the transferability of adversarial samples through a gradient relevance framework and decay indicators.
Boosting Few-shot Action Recognition with Graph-guided Hybrid Matching
Jiazheng Xing (Zhejiang University), Yong Liu (Zhejiang University)
RecognitionGraph Neural NetworkVideo
🎯 What it does: A graph-guided hybrid matching framework (GgHM) is proposed for few-shot action recognition.
Boosting Long-tailed Object Detection via Step-wise Learning on Smooth-tail Data
Na Dong (Harbin Institute of Technology), Gim Hee Lee (National University of Singapore)
Object DetectionKnowledge DistillationConvolutional Neural NetworkTransformerSupervised Fine-TuningImage
🎯 What it does: A stepwise learning framework is proposed, which significantly improves long-tail object detection performance by pre-training on the full long-tail data, fine-tuning on the head class dominant subset of the smooth tail data, and then performing knowledge transfer on the tail class dominant subset.
Boosting Multi-modal Model Performance with Adaptive Gradient Modulation
Hong Li (ShanghaiTech University), Yi Zhou (University of Science and Technology of China)
ClassificationRecognitionOptimizationTransformerMultimodalityAudio
🎯 What it does: The paper proposes an Adaptive Gradient Modulation (AGM) method to enhance the performance of multimodal models.
Boosting Novel Category Discovery Over Domains with Soft Contrastive Learning and All in One Classifier
Zelin Zang (Westlake University), Stan Z. Li (Westlake University)
Domain AdaptationContrastive LearningImage
🎯 What it does: This paper proposes a framework called Soft-contrastive All-in-one Network (SAN) to address the issues of viewpoint noise and overconfidence in new categories in cross-domain unsupervised domain adaptation.
Boosting Positive Segments for Weakly-Supervised Audio-Visual Video Parsing
Kranthi Kumar Rachavarapu (Indian Institute of Technology Madras), Rajagopalan A. N. (Indian Institute of Technology Madras)
ClassificationRecognitionSegmentationVideoMultimodalityAudio
🎯 What it does: This paper proposes a weakly supervised audio-video event parsing framework based on the Poisson-Binomial distribution, which improves the detection rate of positive sample segments through EM iteration, significantly enhancing event localization performance.
Boosting Semantic Segmentation from the Perspective of Explicit Class Embeddings
Yuhe Liu (Beihang University), Zengchang Qin (Beihang University)
SegmentationTransformerImage
🎯 What it does: ECENet is proposed, utilizing explicit category embeddings extracted from predicted masks and enabling interaction between multi-stage features to achieve high accuracy in semantic segmentation.
Boosting Single Image Super-Resolution via Partial Channel Shifting
Xiaoming Zhang (Southwest Jiaotong University), Xiaole Zhao (Southwest Jiaotong University)
RestorationSuper ResolutionConvolutional Neural NetworkImage
🎯 What it does: This paper studies a feature enhancement method without additional parameters or computational overhead—Partial Channel Shifting (PCS)—to improve the performance of single-image super-resolution models.
Boosting Whole Slide Image Classification from the Perspectives of Distribution, Correlation and Magnification
Linhao Qu (Fudan University), Zhijian Song (Fudan University)
ClassificationTransformerContrastive LearningImage
🎯 What it does: A multi-scale multi-stage MIL framework called MILBooster is proposed for WSI classification.