CVPR 2024 Papers — Page 3
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2716 papers
BadCLIP: Trigger-Aware Prompt Learning for Backdoor Attacks on CLIP
Jiawang Bai (Tsinghua University), Wei Liu (Tencent Data Platform)
ClassificationAdversarial AttackTransformerPrompt EngineeringContrastive LearningImageMultimodality
🎯 What it does: This paper introduces triggers during the prompt learning phase of CLIP, constructing a trigger-aware context generator to perform backdoor attacks on the model;
Balancing Act: Distribution-Guided Debiasing in Diffusion Models
Rishubh Parihar (Indian Institute of Science), R. Venkatesh Babu (Indian Institute of Science)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: This paper proposes a method for achieving fair generation without retraining the pre-trained diffusion model, by using Distribution Guidance to make the attribute distribution of generated images approximate the user-specified reference distribution.
BANF: Band-Limited Neural Fields for Levels of Detail Reconstruction
Akhmedkhan Shabanov (Simon Fraser University), Andrea Tagliasacchi (Google DeepMind)
RestorationSuper ResolutionNeural Radiance FieldImagePoint CloudMesh
🎯 What it does: This paper proposes a neural field with low-pass filtering achieved through interpolation during the training process (BANF), which can perform frequency decomposition of the neural field and achieve multi-detail (LOD) reconstruction.
Batch Normalization Alleviates the Spectral Bias in Coordinate Networks
Zhicheng Cai (Nanjing University), Xun Cao (Nanjing University)
RestorationSuper ResolutionCompressionNeural Radiance FieldImageBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: This paper addresses the inherent spectral bias problem in Coordinate Networks by incorporating Batch Normalization, enhancing the network's ability to learn high-frequency signals.
Bayes' Rays: Uncertainty Quantification for Neural Radiance Fields
Lily Goli (University of Toronto), Andrea Tagliasacchi (Simon Fraser University)
Depth EstimationOptimizationExplainability and InterpretabilityNeural Radiance FieldPoint Cloud
🎯 What it does: Proposes Bayes' Rays, a post-processing algorithm to estimate spatial uncertainty and eliminate artifacts caused by occlusion or missing data without retraining any pre-trained NeRF.
Bayesian Differentiable Physics for Cloth Digitalization
Deshan Gong (University of Leeds), He Wang (University College London)
ImagePhysics Related
🎯 What it does: A digital fabric modeling method based on a Bayesian differentiable physical model is proposed, using standard two-dimensional contour images obtained from the Cusick folding test to infer the physical parameters of the fabric.
Bayesian Diffusion Models for 3D Shape Reconstruction
Haiyang Xu (University of Science and Technology of China), Zhuowen Tu (University of California San Diego)
GenerationData SynthesisDiffusion modelPoint Cloud
🎯 What it does: A framework for 3D shape reconstruction through Bayesian inference using prior information and data-driven processes via joint diffusion is proposed (Bayesian Diffusion Models, BDM).
Bayesian Exploration of Pre-trained Models for Low-shot Image Classification
Yibo Miao (Shanghai Jiao Tong University), Zhijie Deng (Shanghai Jiao Tong University)
ClassificationImage
🎯 What it does: Modeling low-shot image classification using Gaussian Processes (GP), integrating pre-trained knowledge by using zero-shot CLIP as the mean function and a weighted sum of deep kernels from various pre-trained models as the kernel function.
BEHAVIOR Vision Suite: Customizable Dataset Generation via Simulation
Yunhao Ge (Stanford University), Jiajun Wu (Meta)
Data SynthesisRobotic IntelligenceImage
🎯 What it does: BVS provides a customizable synthetic data generation tool for system evaluation of visual models.
Behind the Veil: Enhanced Indoor 3D Scene Reconstruction with Occluded Surfaces Completion
Su Sun (Purdue University), Liu Ren (Bosch Research North America)
RestorationGenerationContrastive LearningPoint CloudMesh
🎯 What it does: A method for 3D reconstruction of indoor scenes is proposed using a series of depth images, capable of simultaneously recovering visible surfaces and occluded (invisible) surfaces, achieving the generation of a complete 3D mesh.
BEM: Balanced and Entropy-based Mix for Long-Tailed Semi-Supervised Learning
Hongwei Zheng (Meituan), Xiaoming Xu (Meituan)
ClassificationSupervised Fine-TuningImage
🎯 What it does: A long-tail semi-supervised learning framework based on data mixing, BEM (Balanced and Entropy-based Mix), is proposed to enhance model performance by rebalancing sample quantity and class uncertainty.
Benchmarking Audio Visual Segmentation for Long-Untrimmed Videos
Chen Liu (University of Queensland), Xin Yu (University of Queensland)
RecognitionSegmentationTransformerVideoMultimodalityBenchmarkAudio
🎯 What it does: A large-scale long untrimmed audio-visual segmentation dataset LU-AVS is proposed, along with a benchmark evaluation framework.
Benchmarking Implicit Neural Representation and Geometric Rendering in Real-Time RGB-D SLAM
Tongyan Hua (Hong Kong University of Science and Technology), Lin Wang (Hong Kong University of Science and Technology)
Pose EstimationOptimizationComputational EfficiencySimultaneous Localization and MappingPoint CloudBenchmark
🎯 What it does: A unified RGB-D SLAM benchmark has been established to evaluate the performance of different implicit neural representations (MLP, dense/hash grid, tri-plane, factorization) combined with geometric rendering functions.
Benchmarking Segmentation Models with Mask-Preserved Attribute Editing
Zijin Yin (Beijing University of Posts and Telecommunications), Jun Guo (Beijing University of Posts and Telecommunications)
SegmentationGenerationTransformerLarge Language ModelDiffusion modelImageTextBenchmark
🎯 What it does: A mask-preserved attribute editing pipeline based on diffusion models has been constructed, which accurately edits local and global attributes of images while keeping the original segmentation labels unchanged, generating editable images that can be used to evaluate the robustness of semantic segmentation.
Benchmarking the Robustness of Temporal Action Detection Models Against Temporal Corruptions
Runhao Zeng (Shenzhen MSU-BIT University), Yong Guo (South China University of Technology)
RecognitionObject DetectionConvolutional Neural NetworkTransformerVideoBenchmark
🎯 What it does: This paper addresses the vulnerability of temporal information in videos by constructing two sets of time corruption robustness benchmarks based on THUMOS14 and ActivityNet v1.3. It systematically evaluates the performance of seven mainstream temporal action detection models under varying degrees of time corruption and proposes a robust training strategy through FrameDrop data augmentation and Temporal-Robust Consistency (TRC) loss.
BerfScene: Bev-conditioned Equivariant Radiance Fields for Infinite 3D Scene Generation
Qihang Zhang (Chinese University of Hong Kong), Ceyuan Yang (Shanghai Artificial Intelligence Laboratory)
GenerationData SynthesisAutonomous DrivingNeural Radiance FieldGenerative Adversarial NetworkPoint Cloud
🎯 What it does: A conformal radiative field based on bird's-eye view conditions is proposed for the generation and editing of infinite-scale 3D scenes.
BEVNeXt: Reviving Dense BEV Frameworks for 3D Object Detection
Zhenxin Li (Fudan University), Zuxuan Wu (Fudan University)
Object DetectionAutonomous DrivingImagePoint CloudBenchmark
🎯 What it does: This paper proposes a novel dense BEV (Bird’s Eye View) framework called BEVNeXt for 3D object detection from multi-view images, significantly improving detection accuracy and localization precision through three major enhancement components.
BEVSpread: Spread Voxel Pooling for Bird's-Eye-View Representation in Vision-based Roadside 3D Object Detection
Wenjie Wang (Zhejiang University), Xi Li (Zhejiang University)
Object DetectionAutonomous DrivingComputational EfficiencyImage
🎯 What it does: A new voxel pooling strategy called BEVSpread is proposed, which significantly reduces positional approximation errors and improves the accuracy of roadside 3D object detection by diffusing single-point image features to surrounding BEV grids and using adaptive weights.
Beyond Average: Individualized Visual Scanpath Prediction
Xianyu Chen (University of Minnesota), Qi Zhao (University of Minnesota)
Recurrent Neural NetworkTransformerSupervised Fine-TuningImage
🎯 What it does: This paper proposes a personalized scanpath prediction method aimed at capturing the unique attention movement patterns of different observers in various visual tasks.
Beyond First-Order Tweedie: Solving Inverse Problems using Latent Diffusion
Litu Rout (Google), Wen-Sheng Chu (Google)
Image TranslationRestorationSuper ResolutionDiffusion modelImage
🎯 What it does: This paper proposes STSL (a posterior sampler based on second-order Tweedie approximation) to efficiently solve linear inverse problems and achieve image editing.
Beyond Image Super-Resolution for Image Recognition with Task-Driven Perceptual Loss
Jaeha Kim (Seoul National University), Kyoung Mu Lee (Seoul National University)
ClassificationObject DetectionSegmentationSuper ResolutionConvolutional Neural NetworkTransformerContrastive LearningImage
🎯 What it does: A super-resolution framework SR4IR is proposed, which utilizes task-driven perceptual loss (TDP) and cross-quality patch mixing (CQMix) to enhance the performance of low-resolution images in high-level vision tasks such as semantic segmentation, object detection, and image classification.
Beyond Seen Primitive Concepts and Attribute-Object Compositional Learning
Nirat Saini (University of Maryland), Abhinav Shrivastava (University of Maryland)
ClassificationObject DetectionConvolutional Neural NetworkVision Language ModelImageMultimodalityBenchmark
🎯 What it does: This paper proposes the Open Vocabulary Combinatorial Zero-Shot Learning (OV-CZSL) task, establishes three new benchmarks, and designs a neighborhood expansion loss for semantic transfer and combinatorial reasoning of attributes and objects in this task.
Beyond Text: Frozen Large Language Models in Visual Signal Comprehension
Lei Zhu (Peking University), Yanye Lu (Peking University)
Image TranslationRestorationGenerationTransformerLarge Language ModelVision Language ModelGenerative Adversarial NetworkContrastive LearningImageTextMultimodality
🎯 What it does: Proposes the Vision-to-Language Tokenizer (V2L), which maps images to the vocabulary of LLM to generate discrete tokens, allowing the frozen LLM to directly perform visual understanding and image denoising.
Beyond Textual Constraints: Learning Novel Diffusion Conditions with Fewer Examples
Yuyang Yu, Shengfeng He
GenerationData SynthesisDiffusion modelImage
🎯 What it does: The study investigates how to enable a text-to-image diffusion model based on Stable Diffusion to learn new control conditions under the constraint of only 100 samples, achieving high-quality and structurally consistent generation.
Bezier Everywhere All at Once: Learning Drivable Lanes as Bezier Graphs
Hugh Blayney (dRISK.ai), Panagiotis Angeloudis (Imperial College London)
Autonomous DrivingTransformerImageGraph
🎯 What it does: A lane graph structure based on shared parameterized cubic Bézier curves (Bézier Graph) is proposed, which generates drivable lane networks directly from aerial images using an end-to-end Transformer.
Bi-Causal: Group Activity Recognition via Bidirectional Causality
Youliang Zhang (Wuhan University), Zheng Wang (Wuhan University)
RecognitionGraph Neural NetworkTransformerVideo
🎯 What it does: This paper proposes a framework for group action recognition called Bi-Causal, which combines human-human relationships (HR) and human-object interactions (HOI), and facilitates mutual promotion of the two through a bidirectional causal channel.
Bi-level Learning of Task-Specific Decoders for Joint Registration and One-Shot Medical Image Segmentation
Xin Fan (Dalian University of Technology), Risheng Liu (Dalian University of Technology)
SegmentationOptimizationImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A dual-layer optimized task-specific decoder framework (Bi-JROS) is proposed for one-shot medical image segmentation and registration.
Bi-SSC: Geometric-Semantic Bidirectional Fusion for Camera-based 3D Semantic Scene Completion
Yujie Xue (Hunan University), Mingxing Duan (Hunan University)
SegmentationAutonomous DrivingTransformerPoint Cloud
🎯 What it does: This paper proposes the Bi-SSC framework, which achieves 3D semantic scene completion based on a binocular camera through geometric-semantic bidirectional fusion.
Bidirectional Autoregessive Diffusion Model for Dance Generation
Canyu Zhang (University of South Carolina), Song Wang (University of South Carolina)
GenerationTransformerDiffusion modelVideoAudio
🎯 What it does: A bidirectional autoregressive diffusion model (BADM) is proposed for generating dance from music, capable of producing coherent and natural 3D dance sequences while maintaining beat synchronization.
Bidirectional Multi-Scale Implicit Neural Representations for Image Deraining
Xiang Chen (Nanjing University of Science and Technology), Jiangxin Dong (Nanjing University of Science and Technology)
RestorationTransformerImage
🎯 What it does: This paper proposes a single-image de-rain network called NeRD-Rain, which utilizes a multi-scale Transformer and implicit neural representation (INR) to achieve mutual communication of coarse and fine scale information through bidirectional feedback, aiming to obtain clearer de-rain results under various raindrop conditions.
BigGait: Learning Gait Representation You Want by Large Vision Models
Dingqiang Ye (Southern University of Science and Technology), Shiqi Yu (Michigan State University)
RecognitionRepresentation LearningContrastive LearningImageVideo
🎯 What it does: The BigGait framework is proposed, utilizing large-scale self-supervised visual models (such as DINOv2) to generate gait features under unsupervised conditions, replacing traditional task-specific pre-trained models.
Bilateral Adaptation for Human-Object Interaction Detection with Occlusion-Robustness
Guangzhi Wang (National University of Singapore), Mohan Kankanhalli (National University of Singapore)
RecognitionObject DetectionTransformerContrastive LearningImage
🎯 What it does: This paper proposes a Bidirectional Adaptation Network (BCOM), which extracts semantic features using the CLIP visual encoder after detection and combines them with the multi-scale spatial features of the detector. Furthermore, it enhances the robustness and accuracy of Human-Object Interaction (HOI) detection through Occlusion Part Extrapolation (OPE) and Conditional Context Mining (CCM) modules.
Bilateral Event Mining and Complementary for Event Stream Super-Resolution
Zhilin Huang (Shenzhen International Graduate School, Tsinghua University), Wenming Yang (Shenzhen International Graduate School, Tsinghua University)
RestorationSuper ResolutionConvolutional Neural NetworkVideo
🎯 What it does: A dual-stream network BMCNet is proposed, which processes positive and negative events separately and achieves complementarity through a bidirectional information exchange module, thereby completing the super-resolution of event streams.
Bilateral Propagation Network for Depth Completion
Jie Tang (National University of Defense Technology), Ping Tan (Hong Kong University of Science and Technology)
RestorationDepth EstimationConvolutional Neural NetworkImagePoint Cloud
🎯 What it does: This paper proposes a deep completion network based on bilateral propagation, BP-Net, which utilizes sparse depth measurements and synchronized color images for early-stage depth propagation, thereby avoiding the sensitivity of later convolution operations to sparse data.
BilevelPruning: Unified Dynamic and Static Channel Pruning for Convolutional Neural Networks
Shangqian Gao (University of Pittsburgh), Heng Huang (University of Maryland)
CompressionOptimizationConvolutional Neural NetworkImage
🎯 What it does: A unified dynamic and static channel pruning method is proposed, employing a dual-layer optimization framework to achieve both types of pruning simultaneously.
Binarized Low-light Raw Video Enhancement
Gengchen Zhang (Beijing Institute of Technology), Ying Fu (Beijing Institute of Technology)
RestorationConvolutional Neural NetworkVideo
🎯 What it does: A binary neural network for low-light raw video enhancement, BRVE, is proposed.
Binding Touch to Everything: Learning Unified Multimodal Tactile Representations
Fengyu Yang (Yale University), Alex Wong (Yale University)
ClassificationRecognitionRetrievalRepresentation LearningRobotic IntelligenceTransformerDiffusion modelContrastive LearningMultimodality
🎯 What it does: This paper presents UniTouch, a unified multimodal tactile representation model that can map tactile signals from different visual-tactile sensors to a common multimodal embedding space, enabling zero-shot tactile tasks.
BioCLIP: A Vision Foundation Model for the Tree of Life
Samuel Stevens (Ohio State University), Yu Su (Ohio State University)
ClassificationRecognitionTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: A visual foundation model called BIOCLIP aimed at the 'Tree of Life' has been developed, and the largest biological image dataset TREEOFLIFE-10M has been released;
BiPer: Binary Neural Networks using a Periodic Function
Edwin Vargas (Universidad Industrial de Santander), Henry Arguello (King Abdullah University of Science and Technology)
ClassificationOptimizationImage
🎯 What it does: A network binarization method based on binary periodic (square wave) functions is proposed (BiPer).
BiTT: Bi-directional Texture Reconstruction of Interacting Two Hands from a Single Image
Minje Kim (Korea Advanced Institute of Science and Technology), Tae-Kyun Kim (Imperial College London)
RestorationPose EstimationImage
🎯 What it does: The BiTT method is proposed to end-to-end reconstruct the re-lightable, poseable, and viewable texture of two hands from a single interactive RGB image.
BIVDiff: A Training-Free Framework for General-Purpose Video Synthesis via Bridging Image and Video Diffusion Models
Fengyuan Shi (Nanjing University), Limin Wang (Nanjing University)
GenerationData SynthesisDiffusion modelVideo
🎯 What it does: This paper presents BIVDiff, a training-free universal video synthesis framework that first generates videos frame by frame using a specific image diffusion model (IDM), then generates initial latent variables suitable for the video diffusion model (VDM) through Mixed Inversion, and finally performs temporal smoothing in the VDM to achieve high-quality, temporally consistent video generation.
Blind Image Quality Assessment Based on Geometric Order Learning
Nyeong-Ho Shin (Korea University), Chang-Su Kim (Korea University)
TransformerImage
🎯 What it does: This paper proposes a no-reference image quality assessment network QCN based on geometric order learning, which can rank image feature vectors by quality in the embedding space and predict quality scores through nearest neighbor projection.
BlockGCN: Redefine Topology Awareness for Skeleton-Based Action Recognition
Yuxuan Zhou (University of Mannheim), Xian-Sheng Hua (Terminus Group)
RecognitionPose EstimationGraph Neural NetworkVideoGraph
🎯 What it does: This paper proposes BlockGCN, an improved graph convolutional network designed to better capture spatial-temporal features in skeletal action sequences.
Blur-aware Spatio-temporal Sparse Transformer for Video Deblurring
Huicong Zhang (Harbin Institute of Technology), Hongxun Yao (Harbin Institute of Technology)
RestorationTransformerOptical FlowVideo
🎯 What it does: A spatiotemporal sparse transformer network guided by blurry images obtained from optical flow estimation and bidirectional feature propagation is proposed for video deblurring.
Blur2Blur: Blur Conversion for Unsupervised Image Deblurring on Unknown Domains
Bang-Dang Pham (VinAI Research), Minh Hoai (University of Adelaide)
Image TranslationRestorationGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes an unsupervised blur-to-blur translation framework called Blur2Blur, which first converts the unknown blur images captured by the target camera into a known blur domain, and then uses a pre-trained deblurring network to recover clear images.
BodyMAP - Jointly Predicting Body Mesh and 3D Applied Pressure Map for People in Bed
Abhishek Tandon (Carnegie Mellon University), Zackory Erickson (Carnegie Mellon University)
GenerationPose EstimationConvolutional Neural NetworkMultimodalityMesh
🎯 What it does: Using a depth camera and a pressure-sensitive mattress to obtain 2D pressure images, we jointly predict the 3D mesh of the human body (pose and shape) and the corresponding 3D applied pressure map.
Boosting Adversarial Training via Fisher-Rao Norm-based Regularization
Xiangyu Yin (University of Liverpool), Wenjie Ruan (University of Liverpool)
ClassificationAdversarial AttackConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: This paper proposes a Rademacher complexity analysis based on the Fisher-Rao norm, introducing a new logit variable Γ_ce, and further designs a stage-wise logit-oriented adversarial training framework (LOAT) to enhance the robustness of the model while maintaining or improving standard accuracy.
Boosting Adversarial Transferability by Block Shuffle and Rotation
Kunyu Wang (Chinese University of Hong Kong), Xiaosen Wang (Huawei Singularity Security Lab)
Adversarial AttackConvolutional Neural NetworkImage
🎯 What it does: A new input transformation method called Block Shuffle and Rotation (BSR) is proposed, which disrupts the attention heatmap by shuffling and rotating image blocks to enhance the transferability of adversarial samples.
Boosting Continual Learning of Vision-Language Models via Mixture-of-Experts Adapters
Jiazuo Yu (Dalian University of Technology), You He (Tsinghua University)
TransformerMixture of ExpertsVision Language ModelAuto EncoderImageMultimodality
🎯 What it does: A parameter-efficient continual learning framework is proposed, utilizing Mixture-of-Experts (MoE) adapters for dynamic expansion on a frozen CLIP model, and automatically distinguishing seen and unseen data through a Distribution-Discriminative Auto-Selector (DDAS), balancing memory retention and zero-shot transfer capability.
Boosting Diffusion Models with Moving Average Sampling in Frequency Domain
Yurui Qian (University of Science and Technology of China), Tao Mei (HiDream.ai Inc.)
GenerationData SynthesisOptimizationDiffusion modelImage
🎯 What it does: This paper proposes a frequency domain sampling method based on moving average (MASF), which stabilizes the denoising process of diffusion models by applying moving averages to different frequency sub-bands in the frequency domain.
Boosting Flow-based Generative Super-Resolution Models via Learned Prior
Li-Yuan Tsao (National Tsing Hua University), Chun-Yi Lee (National Tsing Hua University)
RestorationGenerationSuper ResolutionFlow-based ModelGenerative Adversarial NetworkImage
🎯 What it does: A conditional learning prior (latent module) is proposed, which directly predicts latent codes during the inference phase of the flow-based super-resolution model, thereby avoiding the limitations of grid artifacts, exponential inverse problems, and fixed sampling temperatures.
Boosting Image Quality Assessment through Efficient Transformer Adaptation with Local Feature Enhancement
Kangmin Xu (Wuhan University), Weisi Lin (Nanyang Technological University)
Convolutional Neural NetworkTransformerImage
🎯 What it does: The LoDa method is proposed, which combines multi-scale distortion features extracted from a pre-trained ViT and CNN for no-reference image quality assessment, achieving parameter-efficient adaptation.
Boosting Image Restoration via Priors from Pre-trained Models
Xiaogang Xu (Zhejiang Lab), Hujun Bao (Zhejiang University)
RestorationDiffusion modelImage
🎯 What it does: This paper proposes a lightweight pre-trained guidance refinement module (PTG-RM) that enhances the performance of various image restoration tasks by utilizing features from pre-trained models such as CLIP and Stable Diffusion.
Boosting Neural Representations for Videos with a Conditional Decoder
Xinjie Zhang (Hong Kong University of Science and Technology), Jun Zhang (Hong Kong University of Science and Technology)
RestorationCompressionNeural Radiance FieldVideo
🎯 What it does: This paper proposes a general enhancement framework that improves the reconstruction, compression, repair, and interpolation capabilities of implicit video representation (INR) models by introducing a temporal-aware conditional decoder (TAT) and a sine activation NeRV-like module (SNeRV).
Boosting Object Detection with Zero-Shot Day-Night Domain Adaptation
Zhipeng Du (King's College London), Jiankang Deng (Imperial College London)
Object DetectionDomain AdaptationImage
🎯 What it does: A zero-shot day-night domain adaptation framework DAI-Net is designed, utilizing reflectance representation learning to enhance object detection performance in low-light scenarios.
Boosting Order-Preserving and Transferability for Neural Architecture Search: a Joint Architecture Refined Search and Fine-tuning Approach
Beichen Zhang (Shanghai Jiao Tong University), Junchi Yan (Shanghai Jiao Tong University)
Neural Architecture SearchImage
🎯 What it does: Proposes Supernet Shifting in NAS, combining architecture search and supernet fine-tuning to enhance global and local ranking consistency.
Boosting Self-Supervision for Single-View Scene Completion via Knowledge Distillation
Keonhee Han, Daniel Cremers
Autonomous DrivingKnowledge DistillationImage
🎯 What it does: A multi-view density field reconstruction model MVBTS based on self-supervised learning is proposed, and its knowledge is transferred to a single-view model KDBTS through knowledge distillation, significantly improving the performance of single-view scene reconstruction and occupancy prediction.
Boosting Spike Camera Image Reconstruction from a Perspective of Dealing with Spike Fluctuations
Rui Zhao (Peking University), Tiejun Huang (Peking University)
RestorationConvolutional Neural NetworkImageVideo
🎯 What it does: A deep network based on multi-order differential spike timing fusion (DSFT) and multi-granularity alignment is proposed to reconstruct high-quality images from high-frequency, noise, and quantization error-prone pulse sequences captured by spike cameras.
Bootstrapping Autonomous Driving Radars with Self-Supervised Learning
Yiduo Hao (University of Cambridge), Haitham Hassanieh (École Polytechnique Fédérale de Lausanne)
Object DetectionAutonomous DrivingContrastive LearningMultimodality
🎯 What it does: Pre-train a radar perception model through self-supervised learning to learn features from a large amount of unlabeled radar data, improving the performance of radar single target detection.
Bootstrapping Chest CT Image Understanding by Distilling Knowledge from X-ray Expert Models
Weiwei Cao (Alibaba Group), Ling Zhang (Alibaba Group)
GenerationRetrievalKnowledge DistillationTransformerContrastive LearningImageTextBiomedical DataComputed Tomography
🎯 What it does: This paper proposes the use of knowledge distillation from a 2D X-ray expert model and language-guided retrieval to train a 3D CT image understanding model, while employing robust contrastive learning to achieve semantic alignment between CT images and reports, supporting zero-shot diagnosis, report generation, and fine-tuning.
Bootstrapping SparseFormers from Vision Foundation Models
Ziteng Gao (National University of Singapore), Mike Zheng Shou (National University of Singapore)
SegmentationRetrievalComputational EfficiencyRepresentation LearningTransformerContrastive LearningImageMultimodality
🎯 What it does: By inheriting the transformer parameters of large pre-trained visual backbone models (such as AugReg and CLIP) and only training a small number of dedicated focus transformers and intermediate layer weights, a rapid bootstrapping of SparseFormer is achieved.
BoQ: A Place is Worth a Bag of Learnable Queries
Amar Ali-bey (Laval University), Philippe Giguère (Laval University)
RecognitionRetrievalTransformerImage
🎯 What it does: This paper proposes a Transformer-based aggregation method called Bag-of-Queries (BoQ), which aggregates local features extracted by CNN/ViT into a robust global descriptor using learnable global queries and a cross-attention mechanism for visual place recognition.
BOTH2Hands: Inferring 3D Hands from Both Text Prompts and Body Dynamics
Wenqian Zhang (ShanghaiTech University), Lan Xu (ShanghaiTech University)
GenerationData SynthesisPose EstimationTransformerDiffusion modelVideoTextMultimodality
🎯 What it does: This paper presents a large-scale dual-hand action dataset called BOTH57M and develops a model named BOTH2Hands, which can generate realistic 3D dual-hand actions by simultaneously utilizing text prompts and body dynamics.
Brain Decodes Deep Nets
Huzheng Yang (University of Pennsylvania), Jianbo Shi (University of Pennsylvania)
TransformerSupervised Fine-TuningDiffusion modelImageMagnetic Resonance Imaging
🎯 What it does: A visualization tool based on brain encoding, FactorTopy, has been developed to map the four-dimensional features (spatial, layer, scale, channel) of pre-trained visual models onto the human brain's visual cortex, revealing the internal computational processes of the network.
BrainWash: A Poisoning Attack to Forget in Continual Learning
Ali Abbasi (Vanderbilt University), Soheil Kolouri (Vanderbilt University)
OptimizationAdversarial AttackImage
🎯 What it does: A backdoor data poisoning attack named BrainWash is proposed, which allows continual learning models to forget previously learned knowledge after learning new tasks.
Breathing Life Into Sketches Using Text-to-Video Priors
Rinon Gal, Gal Chechik
GenerationDiffusion modelScore-based ModelVideoText
🎯 What it does: A pre-trained text-to-video diffusion model is optimized through score distillation to automatically animate a single subject's vector sketch using text prompts, outputting short videos that retain vector form.
Bridging Remote Sensors with Multisensor Geospatial Foundation Models
Boran Han (Amazon Web Services), Markus Reichstein (Max-Planck-Institute for Biogeochemistry)
ClassificationRestorationSegmentationTransformerMixture of ExpertsImage
🎯 What it does: A multi-sensor geospatial foundation model named msGFM is proposed, capable of jointly learning image features from four types of sensors: RGB, Sentinel-2, SAR, and DSM, and achieving unified modeling of paired and unpaired data through cross-sensor pre-training.
Bridging the Gap Between End-to-End and Two-Step Text Spotting
Mingxin Huang (South China University of Technology), Lianwen Jin (South China University of Technology)
RecognitionObject DetectionOptimizationTransformerImageText
🎯 What it does: Proposes the Bridging Text Spotting (BTS) framework, which optimizes a pre-trained detector and recognizer in an end-to-end manner by freezing them and using a zero-initialized bridging network and Adapter, enhancing text recognition accuracy while maintaining modularity.
Bridging the Gap: A Unified Video Comprehension Framework for Moment Retrieval and Highlight Detection
Yicheng Xiao (Tsinghua University), Xiu Li (Tsinghua University)
RetrievalTransformerContrastive LearningVideoMultimodalityAudio
🎯 What it does: A unified video understanding framework UVCOM is proposed, jointly addressing the tasks of moment retrieval (MR) and highlight detection (HD).
Bridging the Synthetic-to-Authentic Gap: Distortion-Guided Unsupervised Domain Adaptation for Blind Image Quality Assessment
Aobo Li (Xidian University), Leida Li (Xidian University)
Domain AdaptationConvolutional Neural NetworkImage
🎯 What it does: A distortion-guided unsupervised multi-source domain adaptation framework DGQA is proposed to address the cross-domain generalization problem of assessing the quality of synthetic distorted images to real distorted images.
Bring Event into RGB and LiDAR: Hierarchical Visual-Motion Fusion for Scene Flow
Hanyu Zhou (Huazhong University of Science and Technology), Zhiwei Shi (Huazhong University of Science and Technology)
Autonomous DrivingRecurrent Neural NetworkOptical FlowMultimodalityPoint Cloud
🎯 What it does: This paper proposes a hierarchical visual-motion fusion framework, VisMoFlow, which uses event cameras as a bridge between RGB and LiDAR for all-day (day and night) scene flow estimation.
Brush2Prompt: Contextual Prompt Generator for Object Inpainting
Mang Tik Chiu (University of Illinois Urbana-Champaign), Humphrey Shi (Adobe)
RestorationGenerationTransformerPrompt EngineeringVision Language ModelDiffusion modelImageText
🎯 What it does: Designed and implemented Brush2Prompt, a diffusion prior model based on the CLIP space, to automatically generate diverse and contextually relevant text prompts for text-driven object filling after the user provides an image and a mask.
BSNet: Box-Supervised Simulation-assisted Mean Teacher for 3D Instance Segmentation
Jiahao Lu (University of Science and Technology of China), Tianzhu Zhang (University of Science and Technology of China)
Object DetectionSegmentationTransformerSupervised Fine-TuningPoint Cloud
🎯 What it does: This paper proposes a weakly supervised 3D instance segmentation framework called BSNet, which utilizes SAFormer to generate pseudo-labels and train the instance segmentation network.
BT-Adapter: Video Conversation is Feasible Without Video Instruction Tuning
Ruyang Liu (Peking University), Ge Li (Peking University)
RecognitionRetrievalComputational EfficiencyTransformerVision Language ModelContrastive LearningVideoMultimodality
🎯 What it does: This paper proposes BT-Adapter, a method for efficiently transferring from image-language models to video-dialogue models by inserting a branch spatio-temporal attention module while keeping the CLIP visual encoder frozen.
Building a Strong Pre-Training Baseline for Universal 3D Large-Scale Perception
Haoming Chen (East China Normal University), Yuan Xie (East China Normal University)
Object DetectionSegmentationAutonomous DrivingContrastive LearningMultimodalityPoint Cloud
🎯 What it does: Proposes the CSC (Coherent Semantic Consistency) framework, which utilizes cross-scene semantic consistency for unsupervised pre-training of 3D representations.
Building Bridges across Spatial and Temporal Resolutions: Reference-Based Super-Resolution via Change Priors and Conditional Diffusion Model
Runmin Dong (Tsinghua University), Haohuan Fu (Tsinghua University)
RestorationSuper ResolutionDiffusion modelImage
🎯 What it does: A reference image super-resolution method based on a conditional diffusion model, Ref-Diff, is proposed, utilizing land cover change priors to guide denoising, enhancing the content authenticity and texture transfer of large-scale remote sensing images.
Building Optimal Neural Architectures using Interpretable Knowledge
Keith G. Mills (University of Alberta), Di Niu (University of Alberta)
SegmentationNeural Architecture SearchGraph Neural NetworkReinforcement LearningImage
🎯 What it does: AutoBuild learns to align subgraph embeddings in the network structure with performance labels to obtain interpretable importance scores, using these scores to directly construct or significantly reduce the search space, achieving high-quality networks without complete search.
Building Vision-Language Models on Solid Foundations with Masked Distillation
Sepehr Sameni (University of Bern), Simon Jenni (Adobe Research)
SegmentationRetrievalKnowledge DistillationRepresentation LearningTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: A visual language model SF-CLIP based on CLIP has been constructed, combining mask knowledge distillation and contrastive learning to enhance spatial and linguistic understanding.
Byzantine-robust Decentralized Federated Learning via Dual-domain Clustering and Trust Bootstrapping
Peng Sun (Hunan University), Bo Liu (Chinese University of Hong Kong)
Federated LearningConvolutional Neural NetworkImage
🎯 What it does: A decentralized federated learning framework named DFL-Dual is proposed, which identifies and filters Byzantine attackers through dual-domain clustering and trust guidance.
C^2RV: Cross-Regional and Cross-View Learning for Sparse-View CBCT Reconstruction
Yiqun Lin (Hong Kong University of Science and Technology), Xiaomeng Li (Hong Kong University of Science and Technology)
RestorationConvolutional Neural NetworkSupervised Fine-TuningImageComputed Tomography
🎯 What it does: A cross-region and cross-view learning-based sparse-view CBCT reconstruction framework C2RV is proposed.
C2KD: Bridging the Modality Gap for Cross-Modal Knowledge Distillation
Fushuo Huo (Hong Kong Polytechnic University), Song Guo (Hong Kong University of Science and Technology)
Knowledge DistillationImageTextMultimodalityAudio
🎯 What it does: To address the failures caused by modality imbalance and soft label mismatch in Cross-Modal Knowledge Distillation (CMKD), the authors analyze the reasons from both theoretical and empirical perspectives and propose a new framework called Customized Cross-Modal Knowledge Distillation (C2KD).
C3: High-Performance and Low-Complexity Neural Compression from a Single Image or Video
Hyunjik Kim (Google DeepMind), Emilien Dupont (Google DeepMind)
CompressionImageVideo
🎯 What it does: This paper presents C3, a low-complexity neural compression method for single images/videos, trained frame by frame to achieve high compression performance.
C3Net: Compound Conditioned ControlNet for Multimodal Content Generation
Juntao Zhang (Hong Kong University of Science and Technology), Chi-Keung Tang (Hong Kong University of Science and Technology)
GenerationData SynthesisDiffusion modelContrastive LearningImageTextMultimodalityAudio
🎯 What it does: C3Net is proposed, a generative model that can simultaneously receive audio, text, and images as conditions and generate multimodal content (images, text, audio).
CA-Jaccard: Camera-aware Jaccard Distance for Person Re-identification
Yiyu Chen (Beijing Institute of Technology), Yixuan Zhu (Beijing Institute of Technology)
RecognitionRetrievalImage
🎯 What it does: A camera-aware Jaccard distance CA-Jaccard is proposed to enhance the reliability of neighbors and the quality of pseudo-labels in person retrieval.
Cache Me if You Can: Accelerating Diffusion Models through Block Caching
Felix Wimbauer (Meta GenAI), Jialiang Wang (Meta GenAI)
GenerationComputational EfficiencyDiffusion modelImage
🎯 What it does: Cache the outputs of intermediate blocks in the U-Net of the diffusion model to reduce redundant computations during the multi-step denoising process, thereby accelerating image generation.
CAD-SIGNet: CAD Language Inference from Point Clouds using Layer-wise Sketch Instance Guided Attention
Mohammad Sadil Khan (University of Luxembourg), Djamila Aouada (University of Luxembourg)
GenerationData SynthesisTransformerPoint Cloud
🎯 What it does: An end-to-end autoregressive network CAD‑SIGNet has been developed to recover CAD design history (i.e., a series of sketch-and-extrusion token sequences) from input point clouds and can generate multiple feasible design solutions.
CAD: Photorealistic 3D Generation via Adversarial Distillation
Ziyu Wan (City University of Hong Kong), Leonidas Guibas (Stanford University)
GenerationData SynthesisDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a 3D generation framework based on a pre-trained diffusion model called CAD (Consistent Adversarial Distillation), which learns the corresponding 3D distribution on a 3D GAN (StyleGAN2 + triplane) using a single input image and text prompts, thereby achieving high-quality, realistic, and diverse 3D object generation.
CaDeT: a Causal Disentanglement Approach for Robust Trajectory Prediction in Autonomous Driving
Mozhgan Pourkeshavarz (Huawei), Amir Rasouli (Huawei)
Autonomous DrivingGraph Neural NetworkTransformerTime Series
🎯 What it does: A trajectory prediction method based on causal decomposition, CaDeT, is proposed, which enhances the model's robustness under distribution shifts by separating causal features from non-causal features and performing uncertainty-driven interventions in the latent space.
CADTalk: An Algorithm and Benchmark for Semantic Commenting of CAD Programs
Haocheng Yuan (University of Edinburgh), Changjian Li (University of Edinburgh)
Object DetectionSegmentationLarge Language ModelImageBenchmark
🎯 What it does: This paper proposes the task of semantic annotation for CAD programs and develops the CADTalker algorithm, which can automatically parse OpenSCAD programs, render multi-view images after execution, and generate corresponding shape component annotations for each code block using image segmentation and voting.
CAGE: Controllable Articulation GEneration
Jiayi Liu, Manolis Savva
GenerationGraph Neural NetworkDiffusion modelMesh
🎯 What it does: A controllable 3D joint object generation method called CAGE is proposed, which can generate high-quality joint models that meet constraints based on categories, graph structures, and user-defined component shapes, joint types, and axes.
CaKDP: Category-aware Knowledge Distillation and Pruning Framework for Lightweight 3D Object Detection
Haonan Zhang, Bihan Wen
Object DetectionKnowledge DistillationConvolutional Neural NetworkPoint Cloud
🎯 What it does: The paper explores a specific problem in the field of computer vision and proposes a new solution.
Calibrating Multi-modal Representations: A Pursuit of Group Robustness without Annotations
Chenyu You (Yale University), James S. Duncan (Yale University)
ClassificationRepresentation LearningConvolutional Neural NetworkTransformerVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: A contrastive feature calibration (CFR) method based on CLIP is proposed without using any group labels, aimed at alleviating the short-sighted features (spurious correlation) of visual-language models and enhancing group robustness.
CAM Back Again: Large Kernel CNNs from a Weakly Supervised Object Localization Perspective
Shunsuke Yasuki, Masato Taki
ClassificationObject DetectionConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: A lightweight convolutional network RepLKNet is proposed, utilizing large convolutional kernels and achieving efficient training and inference through reparameterization.
Cam4DOcc: Benchmark for Camera-Only 4D Occupancy Forecasting in Autonomous Driving Applications
Junyi Ma (Shanghai Jiao Tong University), Hesheng Wang (Shanghai Jiao Tong University)
Autonomous DrivingConvolutional Neural NetworkPoint CloudBenchmark
🎯 What it does: Proposed the Cam4DOcc benchmark, established a unified 4D occupancy prediction data format and standard evaluation protocol, and implemented four baselines (static world, point cloud projection, BEV instance replication, end-to-end OCCNet).
CAMEL: CAusal Motion Enhancement Tailored for Lifting Text-driven Video Editing
Guiwei Zhang (Beihang University), Qing Yang (Du Xiaoman Financial)
GenerationData SynthesisDiffusion modelVideoText
🎯 What it does: This paper proposes CAMEL, which utilizes learnable motion prompts and causal motion to enhance attention, achieving motion coherence and visual consistency in text-driven video editing.
CAMixerSR: Only Details Need More "Attention"
Yan Wang (Bytedance Inc.), Li Zhang (Bytedance Inc.)
RestorationSuper ResolutionImage
🎯 What it does: This paper proposes a Content-Aware Mixer (CAMixer) and a super-resolution model based on it, CAMixerSR, which can dynamically switch between convolution and self-attention operations in different regions, thereby reducing computational load while maintaining high quality.
Can Biases in ImageNet Models Explain Generalization?
Paul Gavrikov (Offenburg University), Janis Keuper (Offenburg University)
ClassificationRecognitionConvolutional Neural NetworkImage
🎯 What it does: For 48 ImageNet models trained using ResNet-50, we systematically measured their texture/shape bias, spectral bias, and critical band characteristics, and correlated these biases with various generalization benchmarks (ID, robustness, concept transfer, adversarial robustness).
Can I Trust Your Answer? Visually Grounded Video Question Answering
Junbin Xiao (National University of Singapore), Tat-Seng Chua (National University of Singapore)
RecognitionOptimizationTransformerVision Language ModelContrastive LearningVideoMultimodalityBenchmark
🎯 What it does: A weakly supervised visual-oriented video question answering (NExT-GQA) dataset was constructed, and existing visual-language models (VLMs) were evaluated on it. The NG+ method was proposed, which combines differentiable Gaussian masking optimization with cross-modal self-supervision to enhance the visual credibility of answers.
Can Language Beat Numerical Regression? Language-Based Multimodal Trajectory Prediction
Inhwan Bae (Gwangju Institute of Science and Technology), Hae-Gon Jeon (Gwangju Institute of Science and Technology)
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringImageTextMultimodality
🎯 What it does: Proposes transforming the trajectory prediction task into a question-answer format of a language model, using text prompts to predict future trajectories.
Can Protective Perturbation Safeguard Personal Data from Being Exploited by Stable Diffusion?
Zhengyue Zhao (University of Chinese Academy of Sciences), Xing Hu (Institute of Computing Technology, Chinese Academy of Sciences)
GenerationData SynthesisSafty and PrivacyDiffusion modelImage
🎯 What it does: This study investigates the feasibility of using adversarial perturbations to protect personal images from being misappropriated by Stable Diffusion and proposes a GrIDPure method for removing perturbations.