CVPR 2023 Papers — Page 3
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2353 papers
BEV-LaneDet: An Efficient 3D Lane Detection Based on Virtual Camera via Key-Points
Ruihao Wang (HAOMO.AI Technology Co.), Jintao Xu (HAOMO.AI Technology Co.)
Autonomous DrivingOptimizationComputational EfficiencyConvolutional Neural NetworkImagePoint Cloud
🎯 What it does: This paper proposes an efficient 3D lane detection framework based on monocular images, called BEV‑LaneDet;
BEV-SAN: Accurate BEV 3D Object Detection via Slice Attention Networks
Xiaowei Chi (Peking University), Shanghang Zhang (Peking University)
Object DetectionAutonomous DrivingTransformerPoint Cloud
🎯 What it does: This paper proposes the BEV Slice Attention Network (BEV-SAN), which enhances BEV 3D object detection from multi-view cameras by segmenting the BEV space at different height dimensions, aggregating global and local slice features, and using a dual-stage attention fusion.
BEV@DC: Bird's-Eye View Assisted Training for Depth Completion
Wending Zhou (Chinese University of Hong Kong Shenzhen), Zhen Li (Chinese University of Hong Kong Shenzhen)
RestorationDepth EstimationAutonomous DrivingConvolutional Neural NetworkImagePoint Cloud
🎯 What it does: This paper proposes a BEV-assisted training framework that uses LiDAR information only during the training phase to enhance the performance of image-based depth completion.
BEVFormer v2: Adapting Modern Image Backbones to Bird's-Eye-View Recognition via Perspective Supervision
Chenyu Yang (Tsinghua University), Jifeng Dai (Tsinghua University)
Object DetectionAutonomous DrivingTransformerImage
🎯 What it does: This paper proposes a dual-stage Bird's Eye View (BEV) detection framework BEVFormer v2, which combines perspective supervision with existing BEV modules to achieve faster convergence and higher performance.
BEVHeight: A Robust Framework for Vision-Based Roadside 3D Object Detection
Lei Yang (Tsinghua University), Peng Chen (Alibaba Group)
Object DetectionAutonomous DrivingConvolutional Neural NetworkImage
🎯 What it does: Proposes the BEVHeight framework, which uses the pixel height distribution from roadside cameras to replace depth, constructing a 2D to 3D projection to achieve 3D object detection from a bird's-eye view.
Beyond Appearance: A Semantic Controllable Self-Supervised Learning Framework for Human-Centric Visual Tasks
Weihua Chen (Alibaba Group), Xiuyu Sun (Alibaba Group)
RecognitionRepresentation LearningTransformerContrastive LearningImage
🎯 What it does: A self-supervised learning framework called SOLIDER with adjustable semantic ratios was trained on a large-scale unlabeled human image dataset to learn general human visual representations.
Beyond Attentive Tokens: Incorporating Token Importance and Diversity for Efficient Vision Transformers
Sifan Long (Jilin University), Jingdong Wang (Baidu)
ClassificationComputational EfficiencyTransformerImage
🎯 What it does: A pruning method that simultaneously considers token importance and diversity in Vision Transformers is proposed—Token Decoupling & Merging.
Beyond mAP: Towards Better Evaluation of Instance Segmentation
Rohit Jena (University of Pennsylvania), Jianbo Shi (University of Pennsylvania)
Object DetectionSegmentationConvolutional Neural NetworkImage
🎯 What it does: This paper studies the limitations of instance segmentation evaluation metrics, proposes a new metric for measuring redundant predictions, and designs semantic ranking and semantic NMS modules to reduce redundant predictions.
Bi-Directional Distribution Alignment for Transductive Zero-Shot Learning
Zhicai Wang (University of Science and Technology of China), Xiangnan He (University of Science and Technology of China)
ClassificationGenerationAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a bidirectional distribution alignment generative model Bi-VAEGAN to enhance the performance of Transductive Zero-Shot Learning (TZSL).
Bi-Directional Feature Fusion Generative Adversarial Network for Ultra-High Resolution Pathological Image Virtual Re-Staining
Kexin Sun (Fudan University), Yu-Gang Jiang (Fudan University)
Image TranslationGenerationGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a Bidirectional Feature Fusion Generative Adversarial Network (BFF-GAN) to eliminate the block effect in the virtual re-staining process of ultra-high-resolution pathological images.
Bi-Level Meta-Learning for Few-Shot Domain Generalization
Xiaorong Qin (Chinese Academy of Sciences), Shuqiang Jiang (Chinese Academy of Sciences)
Domain AdaptationMeta LearningImage
🎯 What it does: This paper studies the problem of Few-Shot Domain Generalization (FSDG) and proposes a dual-layer meta-learning method to address the generalization ability from known domains to unknown domains by learning two levels of meta-knowledge.
Bi-LRFusion: Bi-Directional LiDAR-Radar Fusion for 3D Dynamic Object Detection
Yingjie Wang (University of Science and Technology of China), Yanyong Zhang (University of Science and Technology of China)
Object DetectionAutonomous DrivingPoint Cloud
🎯 What it does: A Bi-LRFusion bidirectional LiDAR-Radar fusion framework is proposed to enhance the performance of 3D dynamic object detection.
Bi3D: Bi-Domain Active Learning for Cross-Domain 3D Object Detection
Jiakang Yuan (Fudan University), Yu Qiao (Shanghai AI Laboratory)
Object DetectionDomain AdaptationAutonomous DrivingPoint Cloud
🎯 What it does: Bi3D is proposed—a dual-domain active learning framework for cross-domain 3D object detection under the condition of having very few labeled target domain samples.
Bias in Pruned Vision Models: In-Depth Analysis and Countermeasures
Eugenia Iofinova (IST Austria), Dan Alistarh (Neural Magic)
ClassificationCompressionAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: This paper systematically evaluates the impact of neural network pruning on visual model bias and proposes a bias prediction and mitigation method based on dense models.
Bias Mimicking: A Simple Sampling Approach for Bias Mitigation
Maan Qraitem (Boston University), Bryan A. Plummer (Boston University)
ClassificationRecognitionImage
🎯 What it does: This paper proposes a sampling-based bias mitigation method called Bias Mimicking, aimed at eliminating spurious correlations in visual recognition datasets caused by sensitive attributes such as gender and race.
Bias-Eliminating Augmentation Learning for Debiased Federated Learning
Yuan-Yi Xu (National Taiwan University), Yu-Chiang Frank Wang (NVIDIA)
ClassificationFederated LearningImage
🎯 What it does: The research addresses the issue of eliminating local data bias in federated learning and proposes the FedBEAL framework, which learns a Bias-Eliminating Augmenter (BEA) at each client to generate bias-adversarial samples, thereby achieving debiased local updates and obtaining a global model through aggregation.
BiasAdv: Bias-Adversarial Augmentation for Model Debiasing
Jongin Lim (Samsung Advanced Institute of Technology), Seungju Han (Samsung Advanced Institute of Technology)
Adversarial AttackData-Centric LearningGenerative Adversarial NetworkImage
🎯 What it does: Proposes BiasAdv, a debiasing data augmentation method that generates bias-conflicted samples using adversarial attacks.
BiasBed - Rigorous Texture Bias Evaluation
Nikolai Kalischek (ETH Zurich), Konrad Schindler (ETH Zurich)
ClassificationOptimizationConvolutional Neural NetworkAuto EncoderImageBenchmark
🎯 What it does: A BiasBed evaluation platform has been constructed for the systematic comparison and assessment of various algorithms aimed at reducing texture bias in convolutional networks, providing a unified process for experiments, model selection, and statistical testing.
BiCro: Noisy Correspondence Rectification for Multi-Modality Data via Bi-Directional Cross-Modal Similarity Consistency
Shuo Yang (University of Technology Sydney), Min Xu (University of Technology Sydney)
RetrievalContrastive LearningImageTextMultimodality
🎯 What it does: This paper proposes a framework called BiCro based on bidirectional cross-modal similarity consistency to correct noisy correspondences in multimodal matching.
Bidirectional Copy-Paste for Semi-Supervised Medical Image Segmentation
Yunhao Bai (East China Normal University), Yan Wang (East China Normal University)
SegmentationSupervised Fine-TuningImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A Bidirectional Copy-Paste method is proposed, which combines the Mean Teacher framework to achieve semi-supervised medical image segmentation.
Bidirectional Cross-Modal Knowledge Exploration for Video Recognition With Pre-Trained Vision-Language Models
Wenhao Wu (University of Sydney), Wanli Ouyang (Shanghai AI Laboratory)
RecognitionTransformerVision Language ModelContrastive LearningVideoText
🎯 What it does: This paper proposes the BIKE framework, which utilizes the bidirectional cross-modal knowledge of the pre-trained vision-language model (CLIP) to enhance video recognition performance, mainly consisting of an attribute branch and a video branch.
BiFormer: Learning Bilateral Motion Estimation via Bilateral Transformer for 4K Video Frame Interpolation
Junheum Park, Chang-Su Kim
RestorationGenerationTransformerOptical FlowVideo
🎯 What it does: In 4K video frame interpolation, BiFormer is proposed, which first roughly estimates the global motion field through a bilateral Transformer, and then refines it to synthesize the intermediate frames.
BiFormer: Vision Transformer With Bi-Level Routing Attention
Lei Zhu (City University of Hong Kong), Rynson W.H. Lau (City University of Hong Kong)
ClassificationObject DetectionSegmentationTransformerImage
🎯 What it does: This paper proposes a Bi-Level Routing Attention (BRA) mechanism and constructs a new visual Transformer backbone network called BiFormer, which is used for tasks such as image classification, object detection, instance segmentation, and semantic segmentation.
Bilateral Memory Consolidation for Continual Learning
Xing Nie (Chinese Academy of Sciences), Shiming Xiang (Chinese Academy of Sciences)
ClassificationKnowledge DistillationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A Bilateral Memory Consolidation (BiMeCo) framework is proposed, which divides model parameters into two channels: short-term memory and long-term memory. It utilizes knowledge distillation and momentum updates to achieve dynamic interaction between the two, significantly reducing forgetting during the continual learning process.
Binarizing Sparse Convolutional Networks for Efficient Point Cloud Analysis
Xiuwei Xu (Tsinghua University), Jiwen Lu (Tsinghua University)
SegmentationComputational EfficiencyNeural Architecture SearchConvolutional Neural NetworkPoint Cloud
🎯 What it does: This study investigates how to binarize sparse convolutional networks and proposes BSC-Net to achieve efficient point cloud semantic segmentation.
Binary Latent Diffusion
Ze Wang (Purdue University), Qiang Qiu (Purdue University)
GenerationData SynthesisTransformerDiffusion modelAuto EncoderImage
🎯 What it does: This paper proposes a binary latent space image representation and generation framework, using a binary autoencoder to learn a bidirectional mapping between images and binary latent codes, and training a binary diffusion model in this space for efficient image generation.
Biomechanics-Guided Facial Action Unit Detection Through Force Modeling
Zijun Cui (Rensselaer Polytechnic Institute), Qiang Ji (Rensselaer Polytechnic Institute)
ClassificationRecognitionConvolutional Neural NetworkImageOrdinary Differential Equation
🎯 What it does: A method for detecting facial action units (AUs) based on facial biomechanics is proposed, utilizing a muscle activation dynamics model and a differentiable ODE solver to estimate muscle forces and jointly predict AUs with image features.
BioNet: A Biologically-Inspired Network for Face Recognition
Pengyu Li (Terminus Group)
RecognitionConvolutional Neural NetworkImage
🎯 What it does: A facial recognition network called BioNet is proposed, which is based on the feature partitioning of the lower temporal cortex of the human brain and integrates attribute knowledge to enhance recognition performance.
Bit-Shrinking: Limiting Instantaneous Sharpness for Improving Post-Training Quantization
Chen Lin (Hikvision Research Institute), Shiliang Pu (Hikvision Research Institute)
Object DetectionCompressionOptimizationConvolutional Neural NetworkTransformerImage
🎯 What it does: This study investigates the impact of the loss landscape in low-precision post-training quantization (PTQ) on performance and proposes an improvement to PTQ through a Bit-shrinking method that limits 'instant' sharpness.
BITE: Beyond Priors for Improved Three-D Dog Pose Estimation
Nadine Rüegg (ETH Zurich), Silvia Zuffi (Max Planck Institute for Intelligent Systems)
Pose EstimationConvolutional Neural NetworkImage
🎯 What it does: Reconstructing the 3D shape and pose of a dog from a single image, the BITE framework combines a dog-specific D-SMAL model with ground contact constraints.
Bitstream-Corrupted JPEG Images Are Restorable: Two-Stage Compensation and Alignment Framework for Image Restoration
Wenyang Liu (Nanyang Technological University), Lap-Pui Chau (Hong Kong Polytechnic University)
RestorationGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a method for recovering JPEG images affected by bit errors in encrypted bitstreams. It first designs a robust decoder to complete the full decoding, and then achieves high-quality recovery within a two-stage compensation and alignment framework (SCA and GCA).
BKinD-3D: Self-Supervised 3D Keypoint Discovery From Multi-View Videos
Jennifer J. Sun (California Institute of Technology), Pietro Perona (California Institute of Technology)
Pose EstimationConvolutional Neural NetworkContrastive LearningVideo
🎯 What it does: A self-supervised 3D keypoint discovery framework BKinD-3D is proposed, which can automatically learn 3D poses from unlabeled data in multi-view action videos.
Black-Box Sparse Adversarial Attack via Multi-Objective Optimisation
Phoenix Neale Williams (University of Exeter), Ke Li (University of Exeter)
OptimizationAdversarial AttackImage
🎯 What it does: A black-box sparse adversarial attack method based on multi-objective evolutionary optimization (SA-MOO) is proposed, which minimizes the L0 and L2 norms of perturbations while maintaining a high attack success rate.
BlackVIP: Black-Box Visual Prompting for Robust Transfer Learning
Changdae Oh (University of Seoul), Kyungwoo Song (Yonsei University)
ClassificationDomain AdaptationOptimizationTransformerPrompt EngineeringAuto EncoderImage
🎯 What it does: A black-box visual prompt method called BlackVIP is designed and implemented, utilizing an input-adaptive visual prompt generation network (Coordinator) and a zero-order optimization algorithm (SPSA-GC) to achieve transfer learning without accessing pre-trained model parameters and with extremely low memory requirements.
Blemish-Aware and Progressive Face Retouching With Limited Paired Data
Lianxin Xie (South China University of Technology), Hau San Wong (City University of Hong Kong)
Image TranslationRestorationGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a two-stage facial defect repair framework BPFRe, which utilizes an encoder-decoder for rough repair and a StyleGAN generator for refinement, employing a defect-aware attention module to explicitly suppress defect areas.
BlendFields: Few-Shot Example-Driven Facial Modeling
Kacper Kania (Warsaw University of Technology), Marek Kowalski (Microsoft)
GenerationData SynthesisNeural Radiance FieldMesh
🎯 What it does: A controllable expression rendering model called BlendFields is constructed based on sparse expression training, which can refine high-frequency expression details while maintaining low-frequency geometric deformations.
Blind Image Quality Assessment via Vision-Language Correspondence: A Multitask Learning Perspective
Weixia Zhang (Shanghai Jiao Tong University), Kede Ma (City University of Hong Kong)
ClassificationRecognitionTransformerVision Language ModelContrastive LearningImage
🎯 What it does: A multi-task learning framework based on visual-language correspondence is proposed, using the CLIP visual and text encoders to jointly predict blind image quality, scene classification, and distortion types, with the three tasks sharing the same visual representation and automatically adjusting loss weights.
Blind Video Deflickering by Neural Filtering With a Flawed Atlas
Chenyang Lei (Chinese Academy of Sciences), Qifeng Chen (Hong Kong University of Science and Technology)
RestorationConvolutional Neural NetworkOptical FlowVideo
🎯 What it does: A novel unsupervised blind deflickering framework is proposed, utilizing a neural atlas and neural filtering strategy to automatically remove flicker from a single input video while maintaining temporal consistency.
Block Selection Method for Using Feature Norm in Out-of-Distribution Detection
Yeonguk Yu (Gwangju Institute of Science and Technology), Kyoobin Lee (Gwangju Institute of Science and Technology)
Anomaly DetectionConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a post-hoc discrete distribution detection framework based on feature map norms, utilizing a block selection method to identify the most suitable network blocks for detection without the need for external OOD samples during training, and using the feature norm of these blocks for OOV discrimination.
Blowing in the Wind: CycleNet for Human Cinemagraphs From Still Images
Hugo Bertiche (Universitat de Barcelona), Duygu Ceylan (Adobe Research)
GenerationData SynthesisRecurrent Neural NetworkOptical FlowImageVideo
🎯 What it does: This paper proposes a CycleNet method based on recurrent neural networks, which can automatically generate realistic human clothing animations (cinemagraphs) from a single RGB image, simulating the delicate motion of garments swaying in the wind.
Blur Interpolation Transformer for Real-World Motion From Blur
Zhihang Zhong (University of Tokyo), Imari Sato (National Institute of Informatics)
RestorationTransformerImageVideo
🎯 What it does: A new Transformer-based model called BiT is proposed to recover clear video frames at any moment from motion-blurred images.
Boost Vision Transformer With GPU-Friendly Sparsity and Quantization
Chong Yu (NVIDIA Corporation), Jiayuan Fan (Fudan University)
Object DetectionSegmentationCompressionKnowledge DistillationTransformerImage
🎯 What it does: This paper proposes a GPU-oriented ViT compression scheme called GPUSQ-ViT, which combines 2:4 sparse pruning and quantization for efficient deployment.
Boosting Accuracy and Robustness of Student Models via Adaptive Adversarial Distillation
Bo Huang (Hong Kong University of Science and Technology), Wei Wang (Hong Kong University of Science and Technology)
Knowledge DistillationAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: An Adaptive Adversarial Distillation (AdaAD) method is proposed, which involves the teacher model in the inner optimization phase to search for points of maximum prediction difference, thereby enhancing both the natural accuracy and adversarial robustness of the student model during the distillation process.
Boosting Detection in Crowd Analysis via Underutilized Output Features
Shaokai Wu (Jilin University), Fengyu Yang (University of Michigan)
Object DetectionConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: Proposes the Crowd Hat module, which enhances dense crowd analysis (counting, localization, detection) using the area and confidence features from detection box outputs.
Boosting Low-Data Instance Segmentation by Unsupervised Pre-Training With Saliency Prompt
Hao Li (Northwestern Polytechnical University), Junwei Han (Huazhong University of Science and Technology)
Object DetectionSegmentationConvolutional Neural NetworkTransformerPrompt EngineeringContrastive LearningImage
🎯 What it does: A method for unsupervised pre-training based on visual prompts is proposed to enhance the performance of query-based end-to-end instance segmentation models (such as K-Net) in low-data environments.
Boosting Semi-Supervised Learning by Exploiting All Unlabeled Data
Yuhao Chen (MEGVII Technology), Xuequan Lu (Deakin University)
ClassificationSegmentationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes a FullMatch framework based on FixMatch, which fully utilizes all unlabeled data to enhance semi-supervised learning performance through Entropy Meaning Loss (EML) and Adaptive Negative Learning (ANL).
Boosting Transductive Few-Shot Fine-Tuning With Margin-Based Uncertainty Weighting and Probability Regularization
Ran Tao (Carnegie Mellon University), Marios Savvides (Carnegie Mellon University)
ClassificationMeta LearningConvolutional Neural NetworkSupervised Fine-TuningImageBenchmark
🎯 What it does: A transductive fine-tuning method for few-shot learning, TF-MP, is proposed, which utilizes test set information to enhance model robustness.
Boosting Verified Training for Robust Image Classifications via Abstraction
Zhaodi Zhang (East China Normal University), Min Zhang (East China Normal University)
ClassificationConvolutional Neural NetworkImage
🎯 What it does: Proposes an abstract-based interval training method for robust training of image classification models, accompanied by a black-box verifier.
Boosting Video Object Segmentation via Space-Time Correspondence Learning
Yurong Zhang (Shanghai Jiao Tong University), Wenjun Zhang (Zhejiang University)
Object TrackingSegmentationContrastive LearningVideo
🎯 What it does: An unsupervised spatio-temporal correspondence learning framework is designed to enhance the correspondence of matching-based video object segmentation models, thereby improving mask tracking performance.
Boosting Weakly-Supervised Temporal Action Localization With Text Information
Guozhang Li (Xidian University), Xinbo Gao (Chongqing University of Posts and Telecommunications)
RecognitionObject DetectionTransformerVideoText
🎯 What it does: This paper proposes a weakly supervised temporal action localization framework that utilizes action label text information to achieve two objectives through text-segment mining and video-text language completion, significantly improving localization accuracy.
Bootstrap Your Own Prior: Towards Distribution-Agnostic Novel Class Discovery
Muli Yang (Xidian University), Hanwang Zhang (Nanyang Technological University)
ClassificationImage
🎯 What it does: This paper proposes a distribution-agnostic novel class discovery method based on self-estimated priors—BYOP;
Bootstrapping Objectness From Videos by Relaxed Common Fate and Visual Grouping
Long Lian (University of California Berkeley), Stella X. Yu (University of Michigan)
Object DetectionSegmentationOptical FlowVideo
🎯 What it does: This paper proposes an unsupervised video object segmentation method based on Relaxed Common Fate (RCF) and visual grouping. In a two-stage training process, it first learns object segmentation using motion signals, and then corrects the segmentation using low-level/high-level appearance constraints to achieve unsupervised hyperparameter tuning.
Both Style and Distortion Matter: Dual-Path Unsupervised Domain Adaptation for Panoramic Semantic Segmentation
Xu Zheng (Northeastern University), Lin Wang (Hong Kong University of Science and Technology)
SegmentationDomain AdaptationTransformerContrastive LearningImage
🎯 What it does: This paper proposes a dual-projection unsupervised domain adaptation framework called DPPASS, which utilizes ERP and tangent projection to address the inherent gap and format distortion issues in semantic segmentation of panoramic images through cross-projection and internal projection training.
Boundary Unlearning: Rapid Forgetting of Deep Networks via Shifting the Decision Boundary
Min Chen (Huazhong University of Science and Technology), Chen Wang (Huazhong University of Science and Technology)
ClassificationRecognitionSafty and PrivacyConvolutional Neural NetworkImage
🎯 What it does: A Boundary Unlearning method is proposed, which achieves the forgetting of entire classes of samples by quickly moving the decision boundary on a trained deep network.
Boundary-Aware Backward-Compatible Representation via Adversarial Learning in Image Retrieval
Tan Pan (Ant Group), Wei Chu (Ant Group)
RetrievalRepresentation LearningAdversarial AttackGenerative Adversarial NetworkImage
🎯 What it does: A new backward-compatible training method (AdvBCT) is proposed, allowing image retrieval systems to upgrade models without recalculating database embeddings.
Boundary-Enhanced Co-Training for Weakly Supervised Semantic Segmentation
Shenghai Rong (University of Science and Technology of China), Junjie Li (University of Science and Technology of China)
SegmentationConvolutional Neural NetworkTransformerImage
🎯 What it does: In the second stage of weakly supervised semantic segmentation, a boundary-enhanced co-training (BECO) framework is proposed to handle noisy pseudo-labels in a robust learning manner, improving segmentation performance.
Box-Level Active Detection
Mengyao Lyu (Tsinghua University), Guiguang Ding (Tsinghua University)
Object DetectionConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper proposes a box-level active detection framework and designs the Complementary Pseudo Active Strategy (ComPAS) method, which selects the most informative target boxes through an input-end committee and completes sparse annotations using pseudo-labels, thereby reducing annotation redundancy and improving detection performance.
BoxTeacher: Exploring High-Quality Pseudo Labels for Weakly Supervised Instance Segmentation
Tianheng Cheng (Huazhong University of Science and Technology), Wenyu Liu (Horizon Robotics)
Object DetectionSegmentationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: This paper studies weakly supervised instance segmentation and proposes the BoxTeacher framework, which utilizes box annotations to generate high-quality pseudo labels and enhances performance through teacher-student learning.
Breaching FedMD: Image Recovery via Paired-Logits Inversion Attack
Hideaki Takahashi (University of Tokyo), Yang Liu (Institute for AI Industry Research Tsinghua University)
RestorationFederated LearningAdversarial AttackGenerative Adversarial NetworkImage
🎯 What it does: This paper studies and demonstrates the privacy vulnerabilities of Federated Learning with Model Distillation (FedMD) when only sharing public data output logits, and proposes a gradient-free, knowledge-decoupled image inversion attack (PLI) based on paired logits, which can recover clients' private images from public logits without accessing gradients or model parameters.
Breaking the "Object" in Video Object Segmentation
Pavel Tokmakov (Toyota Research Institute), Adrien Gaidon (Toyota Research Institute)
SegmentationTransformerVideo
🎯 What it does: This paper first collects and annotates a video object segmentation dataset VOST focused on object deformation and transformation, and systematically evaluates existing VOS methods based on this dataset; subsequently, an enhanced spatiotemporal Transformer variant AOT+ is proposed for the best-performing AOT model to improve segmentation performance in deformation scenarios.
Bridging Precision and Confidence: A Train-Time Loss for Calibrating Object Detection
Muhammad Akhtar Munir (Muhammad bin Zayed University of Artificial Intelligence), Fahad Shahbaz Khan (Muhammad bin Zayed University of Artificial Intelligence)
Object DetectionAutonomous DrivingTransformerImage
🎯 What it does: This paper proposes a training-time auxiliary loss for object detection models (BPC), which achieves model calibration by aligning the confidence of predicted boxes with detection accuracy (TP/FN);
Bridging Search Region Interaction With Template for RGB-T Tracking
Tianrui Hui (Institute of Information Engineering, Chinese Academy of Sciences), Si Liu (Institute of Artificial Intelligence, Beihang University)
Object TrackingTransformerMultimodality
🎯 What it does: This paper proposes a Template Bridging-based Cross-Modal Search Region Interaction (TBSI) module for RGB-T tracking;
Bridging the Gap Between Model Explanations in Partially Annotated Multi-Label Classification
Youngwook Kim (Seoul National University), Jungwoo Lee (Seoul National University)
ClassificationExplainability and InterpretabilityConvolutional Neural NetworkImage
🎯 What it does: This paper studies the impact of incorrect negative labels on model interpretation (CAM) in partially labeled multi-label classification, assuming that unobserved labels are negative. It proposes a lightweight enhancement function, BoostLU, which improves the model's attribution scores for positive labels during inference or training, thereby enhancing classification performance.
Bringing Inputs to Shared Domains for 3D Interacting Hands Recovery in the Wild
Gyeongsik Moon (Meta Reality Labs)
Pose EstimationConvolutional Neural NetworkImageMesh
🎯 What it does: This paper proposes the InterWild framework, which recovers the 3D mesh of two interacting hands from a single RGB image in the wild.
BUFFER: Balancing Accuracy, Efficiency, and Generalizability in Point Cloud Registration
Sheng Ao (Sun Yat-sen University), Yulan Guo (Sun Yat-sen University)
RecognitionComputational EfficiencyConvolutional Neural NetworkPoint Cloud
🎯 What it does: A point cloud registration framework named BUFFER is proposed, combining point-level and block-level feature learning, while achieving high accuracy, high efficiency, and strong generalization.
Building Rearticulable Models for Arbitrary 3D Objects From 4D Point Clouds
Shaowei Liu (University of Illinois Urbana-Champaign), Shenlong Wang (University of Illinois Urbana-Champaign)
GenerationOptimizationRobotic IntelligenceNeural Radiance FieldOptical FlowPoint Cloud
🎯 What it does: Learn and generate 1-DOF joint re-animatable 3D models with arbitrary numbers of components and arbitrary topological structures from 4D point cloud sequences.
BundleSDF: Neural 6-DoF Tracking and 3D Reconstruction of Unknown Objects
Bowen Wen (NVIDIA), Stan Birchfield (NVIDIA)
Object TrackingPose EstimationTransformerSimultaneous Localization and MappingVideoPoint Cloud
🎯 What it does: This paper proposes a method for real-time 6-DoF tracking and neural implicit 3D reconstruction of unknown rigid objects using monocular RGB-D video, given only the object mask of the first frame.
BUOL: A Bottom-Up Framework With Occupancy-Aware Lifting for Panoptic 3D Scene Reconstruction From a Single Image
Tao Chu (South China University of Technology), Jiaqi Wang (Shanghai AI Laboratory)
SegmentationDepth EstimationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a bottom-up framework for single image processing, combining occupancy-aware lifting to achieve panoramic 3D scene reconstruction and semantic/instance segmentation.
Burstormer: Burst Image Restoration and Enhancement Transformer
Akshay Dudhane, Ming-Hsuan Yang
RecognitionRestorationConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes a new computer vision method aimed at improving the accuracy of image recognition.
C-SFDA: A Curriculum Learning Aided Self-Training Framework for Efficient Source Free Domain Adaptation
Nazmul Karim (University of Central Florida), Nazanin Rahnavard (SRI International)
ClassificationSegmentationDomain AdaptationContrastive LearningImage
🎯 What it does: This paper proposes a curriculum learning-based self-training framework C-SFDA, which avoids noise propagation by reliable pseudo-label filtering to achieve unsupervised domain adaptation.
CABM: Content-Aware Bit Mapping for Single Image Super-Resolution Network With Large Input
Senmao Tian (Beijing Jiaotong University), Shunli Zhang (Beijing Jiaotong University)
RestorationSuper ResolutionSupervised Fine-TuningImage
🎯 What it does: A content-aware bit mapping (CABM) method is proposed for large-size input in single image super-resolution (SISR), which establishes a lookup table from edge scores to quantized bit widths, replacing the traditional MLP bit selector, thereby reducing computational and storage overhead.
CafeBoost: Causal Feature Boost To Eliminate Task-Induced Bias for Class Incremental Learning
Benliu Qiu (University of Electronic Science and Technology of China), Lili Pan (University of Electronic Science and Technology of China)
ClassificationImage
🎯 What it does: This paper proposes a causal intervention-based feature debiasing module (CafeBoost) to eliminate task-induced bias in class-incremental learning and integrates it as a plug-in into existing class-incremental learning frameworks.
Camouflaged Instance Segmentation via Explicit De-Camouflaging
Naisong Luo (University of Science and Technology of China), Feng Wu (University of Science and Technology of China)
Object DetectionSegmentationConvolutional Neural NetworkImage
🎯 What it does: An end-to-end Camouflaged Instance Segmentation network DCNet is proposed, which includes two main modules: pixel-level de-camouflaging and instance-level suppression, significantly improving the segmentation accuracy of camouflaged instances.
Camouflaged Object Detection With Feature Decomposition and Edge Reconstruction
Chunming He (Shenzhen International Graduate School Tsinghua University), Xiu Li (Shenzhen International Graduate School Tsinghua University)
Object DetectionSegmentationImageOrdinary Differential Equation
🎯 What it does: This paper proposes a FEDER model for detecting concealed objects, achieving precise segmentation through feature frequency domain decomposition and edge reconstruction.
CAMS: CAnonicalized Manipulation Spaces for Category-Level Functional Hand-Object Manipulation Synthesis
Juntian Zheng (Tsinghua University), Li Yi (Tsinghua University)
GenerationOptimizationRobotic IntelligenceAuto EncoderVideo
🎯 What it does: This paper proposes a category-level functional hand-object interaction motion generation method, designing a two-level Canonicalized Manipulation Spaces (CAMS) and implementing a two-stage generation framework based on a CVAE planner and optimized hand pose fitting;
Can't Steal? Cont-Steal! Contrastive Stealing Attacks Against Image Encoders
Zeyang Sha (CISPA Helmholtz Center for Information Security), Yang Zhang (CISPA Helmholtz Center for Information Security)
Knowledge DistillationRepresentation LearningAdversarial AttackConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper studies model stealing attacks on self-supervised image encoders and proposes a stealing method based on contrastive learning called Cont-Steal.
Canonical Fields: Self-Supervised Learning of Pose-Canonicalized Neural Fields
Rohith Agaram (International Institute of Information Technology Hyderabad), Srinath Sridhar (Brown University)
Pose EstimationRepresentation LearningNeural Radiance FieldContrastive LearningPoint Cloud
🎯 What it does: A self-supervised Canonical Field Network (CaFi-Net) is proposed, which directly normalizes the density field of Neural Radiance Fields (NeRF) at the category level to generate a unified normalized field.
CAP-VSTNet: Content Affinity Preserved Versatile Style Transfer
Linfeng Wen (Sun Yat-Sen University), Changqing Zou (Zhejiang Lab)
Image TranslationImageVideo
🎯 What it does: The CAP-VSTNet framework is proposed, utilizing reversible residual networks and unbiased linear transformations to achieve content affinity-preserving multi-scene style transfer.
CAP: Robust Point Cloud Classification via Semantic and Structural Modeling
Daizong Ding (Fudan University), Min Yang (Fudan University)
ClassificationAdversarial AttackContrastive LearningPoint Cloud
🎯 What it does: This paper proposes a defense framework named CAP, which enhances the robustness of point cloud classification models against adversarial attacks by utilizing attention pooling and dynamic contrastive learning.
Cap4Video: What Can Auxiliary Captions Do for Text-Video Retrieval?
Wenhao Wu (University of Sydney), Wanli Ouyang (Baidu Inc)
RetrievalTransformerContrastive LearningVideoText
🎯 What it does: The Cap4Video framework is proposed, which utilizes zero-shot video subtitle generation as auxiliary subtitles to enhance text-video retrieval performance in three ways (data augmentation, feature interaction, score fusion).
CapDet: Unifying Dense Captioning and Open-World Detection Pretraining
Yanxin Long (Shenzhen Campus of Sun Yat-sen University), Xiaodan Liang (MBZUAI)
Object DetectionGenerationTransformerVision Language ModelContrastive LearningImageText
🎯 What it does: This paper proposes CapDet, which unifies the tasks of open-world detection and dense description. The model can identify targets based on a given list of categories and can also generate natural language descriptions for unknown targets.
CAPE: Camera View Position Embedding for Multi-View 3D Object Detection
Kaixin Xiong (Huazhong University of Science and Technology), Xiang Bai (Huazhong University of Science and Technology)
Object DetectionAutonomous DrivingTransformerPoint Cloud
🎯 What it does: A multi-view 3D object detection method based on Camera Perspective Position Embedding (CAPE) is proposed, and it is further extended to temporal modeling.
CaPriDe Learning: Confidential and Private Decentralized Learning Based on Encryption-Friendly Distillation Loss
Nurbek Tastan (Mohamed bin Zayed University of Artificial Intelligence), Karthik Nandakumar (Mohamed bin Zayed University of Artificial Intelligence)
Federated LearningSafty and PrivacyKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: A CaPriDe learning framework is proposed, utilizing fully homomorphic encryption (FHE) to achieve decentralized collaborative learning with data confidentiality, and transferring model knowledge through knowledge distillation.
CARTO: Category and Joint Agnostic Reconstruction of ARTiculated Objects
Nick Heppert (University of Freiburg), Thomas Kollar (Toyota Research Institute)
Object DetectionPose EstimationImage
🎯 What it does: A single forward-pass CARTO framework is designed to simultaneously detect multiple instances, estimate 6D poses, sizes, and joint types and states from a single stereo RGB image, and recover their complete 3D shapes in a single code space.
Cascade Evidential Learning for Open-World Weakly-Supervised Temporal Action Localization
Mengyuan Chen (Chinese Academy of Sciences), Changsheng Xu (Chinese Academy of Sciences)
RecognitionObject DetectionVideo
🎯 What it does: This paper proposes a Cascade Evidence Learning Framework (CELL) for open-world weakly supervised temporal action localization.
Cascaded Local Implicit Transformer for Arbitrary-Scale Super-Resolution
Hao-Wei Chen (National Tsing Hua University), Chun-Yi Lee (National Tsing Hua University)
RestorationSuper ResolutionTransformerImage
🎯 What it does: Proposed a Local Implicit Transformer (LIT) and a Cascaded LIT (CLIT) to achieve super-resolution at arbitrary scales.
CASP-Net: Rethinking Video Saliency Prediction From an Audio-Visual Consistency Perceptual Perspective
Junwen Xiong (Northwestern Polytechnical University), Guangtao Zhai (Shanghai Jiao Tong University)
RecognitionSegmentationComputational EfficiencyConvolutional Neural NetworkVideoMultimodalityAudio
🎯 What it does: A consistency-aware audio-video attention prediction network (CASP-Net) is proposed, which achieves video saliency prediction through a dual-stream encoder, an audio-video interaction module, consistency-aware prediction coding, and a multi-scale decoder.
Castling-ViT: Compressing Self-Attention via Switching Towards Linear-Angular Attention at Vision Transformer Inference
Haoran You (Georgia Institute of Technology), Yingyan (Celine) Lin (Georgia Institute of Technology)
ClassificationObject DetectionSegmentationComputational EfficiencyKnowledge DistillationTransformerImage
🎯 What it does: Proposes the Castling-ViT framework, which simultaneously uses linear angular attention and auxiliary softmax attention during training, and retains only linear angular attention during inference to reduce computational costs.
CAT: LoCalization and IdentificAtion Cascade Detection Transformer for Open-World Object Detection
Shuailei Ma (Northeast University), Fanbing Lv (Changsha Hisense Intelligent System Research Institute Co., Ltd)
Object DetectionTransformerImage
🎯 What it does: A Cascaded Detection Transformer (CAT) for open-world object detection is proposed.
Catch Missing Details: Image Reconstruction With Frequency Augmented Variational Autoencoder
Xinmiao Lin (Rochester Institute of Technology), Yu Kong (Michigan State University)
RestorationGenerationTransformerAuto EncoderImageText
🎯 What it does: The Frequency Augmented VAE (FA-VAE) model is proposed, which enhances image reconstruction quality by embedding a Frequency Completion Module (FCM) and Dynamic Spectrum Loss (DSL) in the VQ-VAE decoder, and extends to text-to-image generation tasks with the introduction of a Cross-Attention Autoregressive Transformer (CAT).
Category Query Learning for Human-Object Interaction Classification
Chi Xie (Tongji University), Yichen Wei (Shanghai Jiao Tong University)
ClassificationTransformerImage
🎯 What it does: Proposes a HOI classification method based on category query learning, mapping category queries to images and enhancing interaction classification performance by combining a transformer decoder.
Causally-Aware Intraoperative Imputation for Overall Survival Time Prediction
Xiang Li (Fudan University), Yanwei Fu (Fudan University)
ClassificationConvolutional Neural NetworkImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper proposes a causal inference system using intraoperative indicators—CAWIM—to predict overall survival time for patients with primary liver cancer.
CCuantuMM: Cycle-Consistent Quantum-Hybrid Matching of Multiple Shapes
Harshil Bhatia (Indian Institute of Technology), Vladislav Golyanik (MPI for Informatics)
OptimizationMesh
🎯 What it does: A periodic consistent quantum-hybrid multi-shape matching method called CCuantuMM is proposed, which simplifies the multi-shape matching problem into a continuous three-shape matching problem, and uses quantum annealing (QA) to solve the quadratic unconstrained binary optimization (QUBO) at each step to update the correspondences.
CDDFuse: Correlation-Driven Dual-Branch Feature Decomposition for Multi-Modality Image Fusion
Zixiang Zhao (Xi'an Jiaotong University), Luc Van Gool (Computer Vision Lab, ETH Zürich)
Image TranslationObject DetectionSegmentationConvolutional Neural NetworkTransformerImageMultimodalityMagnetic Resonance ImagingComputed TomographyPositron Emission Tomography
🎯 What it does: Proposes the CDDFuse dual-branch Transformer-CNN network, which decomposes low-frequency shared features and high-frequency exclusive features of multimodal images, and generates fused images.
CelebV-Text: A Large-Scale Facial Text-Video Dataset
Jianhui Yu (University of Sydney), Wayne Wu (Shanghai AI Laboratory)
GenerationRetrievalGenerative Adversarial NetworkVideoTextBenchmark
🎯 What it does: We propose CelebV-Text, a novel dataset containing 70,000 high-quality facial videos (with a resolution of at least 512×512) and 1,400,000 text descriptions, generated through a semi-automated template method to produce naturally diverse text.
Center Focusing Network for Real-Time LiDAR Panoptic Segmentation
Xiaoyan Li (Beijing University of Technology), Baocai Yin (Beijing University of Technology)
SegmentationAutonomous DrivingPoint CloudBenchmark
🎯 What it does: This paper proposes a real-time proposal-free Center Focus Network (CFNet) for LiDAR panoramic segmentation.
CF-Font: Content Fusion for Few-Shot Font Generation
Chi Wang (Zhejiang University), Weiwei Xu (Zhejiang University)
GenerationGenerative Adversarial NetworkImage
🎯 What it does: Proposes a content fusion module and projection character loss to improve few-shot font generation.
CFA: Class-Wise Calibrated Fair Adversarial Training
Zeming Wei (Peking University), Yisen Wang (Peking University)
ClassificationAdversarial AttackGenerative Adversarial NetworkImage
🎯 What it does: This study investigates the sensitivity of different categories to adversarial training configurations and proposes an adaptive category-level adversarial training framework called CFA.
Change-Aware Sampling and Contrastive Learning for Satellite Images
Utkarsh Mall (Cornell University), Kavita Bala (Cornell University)
ClassificationSegmentationRetrievalContrastive LearningImage
🎯 What it does: A self-supervised learning framework utilizing long-term and short-term temporal differences in satellite images is proposed, featuring a new Change-Aware Contrastive Loss (CACo) and an improved geographic sampling strategy.
Chat2Map: Efficient Scene Mapping From Multi-Ego Conversations
Sagnik Majumder (University of Texas at Austin), Vamsi Krishna Ithapu (Reality Labs Research)
GenerationComputational EfficiencyTransformerReinforcement LearningVideoMultimodalityAudio
🎯 What it does: Proposes the Chat2Map task, which utilizes first-person audio-visual dialogue data from multiple participants to actively select camera sampling frames and generate scene occupancy maps.