CVPR 2025 Papers — Page 3
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2871 papers
Audio-Visual Instance Segmentation
Ruohao Guo (Peking University), Buwen Liang
Object DetectionObject TrackingSegmentationTransformerVideoMultimodalityBenchmarkAudio
🎯 What it does: This paper proposes the Audio-Visual Instance Segmentation (AVIS) task and implements a strong baseline model (AVISM) for the classification, segmentation, and tracking of sound-emitting objects.
Audio-Visual Semantic Graph Network for Audio-Visual Event Localization
Liang Liu (Chongqing University of Posts and Telecommunications), Yongqiang Zhu (Chongqing University of Posts and Telecommunications)
ClassificationRecognitionGraph Neural NetworkContrastive LearningVideoMultimodalityAudio
🎯 What it does: An Audio-Visual Semantic Graph Network (AVSGN) is proposed to achieve audio-visual event localization through cross-modal semantic alignment and graph interaction.
Augmented Deep Contexts for Spatially Embedded Video Coding
Yifan Bian (University of Science and Technology of China), Dong Liu (University of Science and Technology of China)
CompressionTransformerVideo
🎯 What it does: This paper proposes a Spatially Embedded Video Codec (SEVC), which first compresses low-resolution videos to obtain spatial references, and then uses these spatial references along with temporal references to enhance motion vectors and features, ultimately generating a mixed spatial-temporal context and improved latent priors.
Augmenting Multimodal LLMs with Self-Reflective Tokens for Knowledge-based Visual Question Answering
Federico Cocchi (University of Modena and Reggio Emilia), Rita Cucchiara (University of Modena and Reggio Emilia)
RetrievalTransformerLarge Language ModelVision Language ModelTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: A retrieval-enhanced multimodal large language model, ReflectiVA, has been developed, which uses reflective tokens to automatically determine whether external knowledge needs to be retrieved and to filter relevant segments for knowledge-based visual question answering.
Augmenting Perceptual Super-Resolution via Image Quality Predictors
Fengjia Zhang (Samsung Electronics), Alex Levinshtein (Samsung Electronics)
RestorationSuper ResolutionGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes the use of a No-Reference Image Quality Assessment (NR-IQA) model to guide Super Resolution (SR) training, which can either sample multiple target images weighted by NR-IQA or directly use NR-IQA as a differentiable optimization objective.
AuraFusion360: Augmented Unseen Region Alignment for Reference-based 360deg Unbounded Scene Inpainting
Chung-Ho Wu (National Yang Ming Chiao Tung University), Yu-Lun Liu (National Yang Ming Chiao Tung University)
RestorationDiffusion modelGaussian SplattingImageStochastic Differential Equation
🎯 What it does: A method called AuraFusion360 based on reference images is proposed, which implements object removal and unseen area reconstruction in 360° unbounded scenes using 3D Gaussian point clouds.
Auto Cherry-Picker: Learning from High-quality Generative Data Driven by Language
Yicheng Chen (Shanghai AI Laboratory), Kai Chen (Shanghai AI Laboratory)
Object DetectionSegmentationGenerationData SynthesisTransformerLarge Language ModelDiffusion modelImageMultimodality
🎯 What it does: Developed the Auto Cherry-Picker (ACP) framework, which utilizes large language models (LLM) to generate scene graphs and layouts based on object combinations, then employs controllable diffusion models to generate images, and filters high-quality synthetic training samples through the Composite Layout and Image Score (CLIS);
Auto-Encoded Supervision for Perceptual Image Super-Resolution
MinKyu Lee (Sungkyunkwan University), Jae-Pil Heo (Sungkyunkwan University)
RestorationSuper ResolutionAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: A supervised loss based on autoencoders (AESOP) is proposed to address the blurriness issue caused by pixel-level reconstruction loss in perceptual super-resolution (SR).
AutoLUT: LUT-Based Image Super-Resolution with Automatic Sampling and Adaptive Residual Learning
Yuheng Xu (Nanjing University), Gangshan Wu (Nanjing University)
RestorationSuper ResolutionConvolutional Neural NetworkImage
🎯 What it does: This paper proposes an image super-resolution framework based on lookup tables (LUT) called Auto LUT, which integrates two pluggable modules: Auto Sample for automatic sampling and Ada R L for adaptive residual learning. It significantly reduces memory usage and inference latency while maintaining performance.
Automated Generation of Challenging Multiple-Choice Questions for Vision Language Model Evaluation
Yuhui Zhang (Stanford University), Serena Yeung-Levy (Stanford University)
GenerationData SynthesisTransformerLarge Language ModelAgentic AIVision Language ModelTextMultimodalityBenchmark
🎯 What it does: Proposes AutoConverter, a multi-agent framework that automatically transforms open-ended visual question answering (VQA) problems into multiple-choice questions with challenging distractor options; based on this, a unified VMCBench evaluation benchmark is constructed;
Automated Proof of Polynomial Inequalities via Reinforcement Learning
Banglong Liu (East China Normal University), Zhengfeng Yang (East China Normal University)
OptimizationReinforcement Learning
🎯 What it does: This paper proposes a method for automated proof of polynomial inequalities based on reinforcement learning, transforming the inequality proof into a linear programming problem using the Krivine basis representation, constructing an RL environment, and training an agent with DQN to gradually select basis elements, ultimately obtaining a non-negative representation.
Automatic Joint Structured Pruning and Quantization for Efficient Neural Network Training and Compression
Xiaoyi Qu (Lehigh University), Tianyi Chen (Microsoft)
CompressionOptimizationComputational EfficiencyConvolutional Neural NetworkTransformerImageText
🎯 What it does: The GETA framework is proposed to achieve joint automated training of structured pruning and quantization, supporting any deep network model.
Automatic Spectral Calibration of Hyperspectral Images: Method, Dataset and Benchmark
Zhuoran Du (University of Amsterdam), Shikui Wei (Beijing Jiaotong University)
RestorationTransformerImageBenchmark
🎯 What it does: A Transformer-based automatic spectral correction method is proposed, which can achieve high-quality hyperspectral image correction without the need for a reference panel.
AutoPresent: Designing Structured Visuals from Scratch
Jiaxin Ge (Carnegie Mellon University), Trevor Darrell (University of California)
GenerationOptimizationTransformerLarge Language ModelVision Language ModelTextBenchmark
🎯 What it does: This paper studies a method for generating presentation slides from natural language instructions and proposes the SLIDESBENCH benchmark.
Autoregressive Distillation of Diffusion Transformers
Yeongmin Kim (Meta GenAI), Artsiom Sanakoyeu
GenerationKnowledge DistillationTransformerDiffusion modelImageTextOrdinary Differential Equation
🎯 What it does: This paper proposes a few-step distillation method for diffusion transformers (DiT) called Autoregressive Distillation (ARD), which significantly reduces exposure bias and improves generation quality by utilizing the complete historical information of the probability flow ODE trajectory to predict the next step.
Autoregressive Sequential Pretraining for Visual Tracking
Shiyi Liang (Xi'an Jiaotong University), Xing Wei (Xi'an Jiaotong University)
Object TrackingTransformerDiffusion modelVideo
🎯 What it does: This paper proposes Unsupervised Autoregressive Sequence Pretraining (ARP), which predicts the appearance of future frames through a conditional diffusion model and verifies it with a reverse autoregressive trajectory, achieving global pretraining for continuous videos and integrating it into ARTrackV2 to form ARPTrack.
AutoSSVH: Exploring Automated Frame Sampling for Efficient Self-Supervised Video Hashing
Niu Lian (Harbin Institute of Technology), Bin Chen (Harbin Institute of Technology)
RetrievalCompressionTransformerContrastive LearningVideo
🎯 What it does: AutoSSVH compresses unlabeled videos into high-quality hash codes for efficient video retrieval through automated adversarial frame sampling and contrastive learning.
AutoURDF: Unsupervised Robot Modeling from Point Cloud Frames Using Cluster Registration
Jiong Lin (Columbia University), Hod Lipson (Columbia University)
Robotic IntelligencePoint Cloud
🎯 What it does: This paper proposes AutoURDF, an unsupervised robot model generation method based on temporal point clouds, which can automatically generate URDF-compatible description files.
AvatarArtist: Open-Domain 4D Avatarization
Hongyu Liu (Hong Kong University of Science and Technology), Qifeng Chen (Ant Group)
GenerationData SynthesisDiffusion modelAuto EncoderGenerative Adversarial NetworkImageStochastic Differential Equation
🎯 What it does: Using GAN and diffusion models for joint training, the AvatarArtist model is constructed to transform single portraits in any style into animatable open-domain 4D avatars.
AVF-MAE++: Scaling Affective Video Facial Masked Autoencoders via Efficient Audio-Visual Self-Supervised Learning
Xuecheng Wu (Xi'an Jiaotong University), Liang He (Xi'an Jiaotong University)
RecognitionRepresentation LearningTransformerAuto EncoderVideoMultimodalityAudio
🎯 What it does: This paper proposes AVF-MAE++, a scalable audio-video masked autoencoder for emotional video facial analysis.
AVQACL: A Novel Benchmark for Audio-Visual Question Answering Continual Learning
Kaixuan Wu (Southeast University), Guoliang Wu (Southeast University)
Knowledge DistillationConvolutional Neural NetworkRecurrent Neural NetworkSupervised Fine-TuningVideoMultimodalityBenchmarkAudio
🎯 What it does: A continuous learning benchmark for audio-visual question answering, AVQACL, is proposed, along with the development of two datasets, Split-AVQA and Split-MUSIC-AVQA. A continuous learning method is introduced that combines question-guided cross-modal fusion (QCIF), task-specific knowledge distillation with spatiotemporal constraints (TKD-STFC), and question semantic consistency constraints (QSCC).
BACON: Improving Clarity of Image Captions via Bag-of-Concept Graphs
Zhantao Yang, Fan Cheng (Alibaba Group)
Object DetectionGenerationTransformerLarge Language ModelVision Language ModelImage
🎯 What it does: Designed and implemented the BACON method to generate structured and splittable image descriptions, and based on this, trained the LLAVA(BACON)-CAPTIONER to achieve high-quality annotations without GPT-4V.
BADGR: Bundle Adjustment Diffusion Conditioned by Gradients for Wide-Baseline Floor Plan Reconstruction
Yuguang Li (University of Washington), Alex Colburn
Pose EstimationOptimizationTransformerDiffusion modelImage
🎯 What it does: Developed BADGR, a diffusion-based bundle adjustment model that can jointly optimize camera pose and 2D layout from sparse 360° panoramic images.
BadToken: Token-level Backdoor Attacks to Multi-modal Large Language Models
Zenghui Yuan (Huazhong University of Science and Technology), Lichao Sun (Lehigh University)
GenerationAdversarial AttackData-Centric LearningTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes a token-level backdoor attack method called BadToken for multimodal large language models (MLLMs), which can embed triggers in input images to achieve fine-grained word replacement or word addition in the output text, thereby completing the attack while maintaining the overall performance of the model.
Balanced Direction from Multifarious Choices: Arithmetic Meta-Learning for Domain Generalization
Xiran Wang (Nanjing University), Yinghuan Shi (Nanjing University)
Domain AdaptationMeta LearningImageBenchmark
🎯 What it does: An arithmetic meta-learning (Arith) framework is proposed, which achieves gradient matching between source domains through a gradient weighting approach and obtains more balanced model parameters by approximating the centroid of the optimal parameters of the source domains.
Balanced Rate-Distortion Optimization in Learned Image Compression
Yichi Zhang (Purdue University), Fengqing Zhu (Purdue University)
CompressionOptimizationImage
🎯 What it does: This paper views rate-distortion optimization in image compression as a multi-objective optimization problem and proposes two strategies to balance rate and distortion updates.
Balancing Two Classifiers via A Simplex ETF Structure for Model Calibration
Jiani Ni (Jilin University), Hongyuan Zha (Chinese University of Hong Kong)
ClassificationOptimizationConvolutional Neural NetworkImage
🎯 What it does: The BalCAL method is proposed to enhance the confidence calibration of deep networks by adding a fixed classifier based on Simplex ETF after a standard learnable classifier and using an adapter to align the features of both.
BARD-GS: Blur-Aware Reconstruction of Dynamic Scenes via Gaussian Splatting
Yiren Lu (Case Western Reserve University), Yu Yin (Case Western Reserve University)
RestorationGenerationGaussian SplattingVideo
🎯 What it does: A BARD-GS method based on 3D Gaussian splatting is proposed, which can directly recover high-quality dynamic scene reconstruction and new view synthesis from monocular videos containing camera motion blur and object motion blur.
BASKET: A Large-Scale Video Dataset for Fine-Grained Skill Estimation
Yulu Pan (University of North Carolina at Chapel Hill), Gedas Bertasius (University of North Carolina at Chapel Hill)
ClassificationRecognitionSupervised Fine-TuningVideoBenchmark
🎯 What it does: This work constructs a large-scale basketball video dataset called BASKET, aimed at evaluating players' levels in 20 fine-grained basketball skills (such as three-point shooting, rebounding, passing, etc.) and assessing existing long video recognition models based on this task.
Bayesian Prompt Flow Learning for Zero-Shot Anomaly Detection
Zhen Qu (Institute of Automation Chinese Academy of Sciences), Guiguang Ding (Tsinghua University)
Anomaly DetectionPrompt EngineeringVision Language ModelImageBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: A zero-shot anomaly detection method based on Bayesian Prompt Flow Learning, called Bayes-PFL, is proposed.
Bayesian Test-Time Adaptation for Vision-Language Models
Lihua Zhou (Chinese Academy of Sciences), Zhen Lei (Chinese Academy of Sciences)
ClassificationDomain AdaptationTransformerVision Language ModelImageMultimodality
🎯 What it does: This paper proposes a test-time adaptive method based on Bayesian inference—Bayesian Class Adaptation (BCA)—to achieve real-time online adaptation for zero-shot image classification on the pre-trained vision-language model CLIP.
Be More Specific: Evaluating Object-centric Realism in Synthetic Images
Anqi Liang, Aleix Martinez
Object DetectionData SynthesisVision Language ModelImage
🎯 What it does: This study investigates the assessment of realism in synthetic images from an object-oriented perspective, constructing an object-level realism (OcR) dataset and proposing a corresponding evaluation framework and model.
Believing is Seeing: Unobserved Object Detection using Generative Models
Subhransu S. Bhattacharjee (Australian National University), Rahul Shome (Australian National University)
Object DetectionTransformerVision Language ModelDiffusion modelImage
🎯 What it does: Proposed and implemented the task of 'unobserved object detection', predicting the location distribution of objects outside the image field of view or occluded in 2D, 2.5D, and 3D spaces.
Benchmarking Large Vision-Language Models via Directed Scene Graph for Comprehensive Image Captioning
Fan Lu (University of Science and Technology of China), Zheng-Jun Zha (University of Science and Technology of China)
Object DetectionGenerationTransformerLarge Language ModelVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: A fine-grained image description evaluation benchmark named CompreCap has been constructed, with manually annotated segmentation, attributes, and directional relationships, generating a complete scene graph structure, and an evaluation process has been proposed to measure the quality of detailed descriptions generated by large visual language models.
Benchmarking Object Detectors under Real-World Distribution Shifts in Satellite Imagery
Sara A. Al-Emadi (Qatar Computing Research Institute), Ferda Ofli (Qatar Computing Research Institute)
Object DetectionDomain AdaptationImageBenchmark
🎯 What it does: This paper proposes three benchmark datasets specifically designed to evaluate the generalization performance of satellite image object detection under real distribution shifts (especially spatial distribution shifts) (RWDS-CZ, RWDS-FR, RWDS-HE), and systematically assesses the performance of various mainstream object detection models under single-source and multi-source training settings.
beta-FFT: Nonlinear Interpolation and Differentiated Training Strategies for Semi-Supervised Medical Image Segmentation
Ming Hu (Xi'an Institute of Optics and Precision Mechanics, Chinese Academy of Sciences), Quan Wang (Xi'an Institute of Optics and Precision Mechanics, Chinese Academy of Sciences)
SegmentationImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper proposes a FFT-based nonlinear interpolation and differentiation training strategy (β-FFT) to address the homogenization problem caused by co-training in semi-supervised medical image segmentation.
BEVDiffuser: Plug-and-Play Diffusion Model for BEV Denoising with Ground-Truth Guidance
Xin Ye (Bosch Research North America), Liu Ren (Bosch Research North America)
Object DetectionAutonomous DrivingDiffusion modelPoint Cloud
🎯 What it does: A BEV feature denoising module called BEVDiffuser based on a conditional diffusion model has been designed and implemented, which can serve as a plug-and-play module to enhance the performance of existing BEV models during the training phase.
Beyond Background Shift: Rethinking Instance Replay in Continual Semantic Segmentation
Hongmei Yin (Tianjin University), Liang Wan (Tianjin University)
SegmentationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: An Enhanced Instance Replay (EIR) framework is proposed to address the issues of background shift and catastrophic forgetting in continual semantic segmentation through instance-level storage and fusion.
Beyond Clean Training Data: A Versatile and Model-Agnostic Framework for Out-of-Distribution Detection with Contaminated Training Data
Yuchuan Li (Queen's University), Il-Min Kim (Queen's University)
Anomaly DetectionImage
🎯 What it does: This study investigates OOD detection in scenarios where the training data is mixed with OOD samples and proposes a model-agnostic iterative framework that allows existing OOD methods to train on contaminated data and accurately estimate the contamination ratio.
Beyond Generation: A Diffusion-based Low-level Feature Extractor for Detecting AI-generated Images
Nan Zhong (City University of Hong Kong), Xinpeng Zhang (Fudan University)
Anomaly DetectionTransformerDiffusion modelImage
🎯 What it does: Construct a low-level feature extractor that generates image pairs differing only at the pixel level through a diffusion model denoising process. Subsequently, model the low-level features of real photos as a class distribution and consider all images deviating from this distribution as AI-generated images.
Beyond Human Perception: Understanding Multi-Object World from Monocular View
Keyu Guo (Changan University), Ajmal Saeed Mian (Changan University)
Object DetectionSegmentationTransformerLarge Language ModelPrompt EngineeringImageTextMultimodality
🎯 What it does: The MonoMulti-3DVG task is proposed, and a large-scale multi-target 3D visual alignment (3DVG) dataset called MonoMulti3D-ROPE is constructed from monocular RGB images. The CyclopsNet network is designed to achieve multi-modal semantic alignment and fusion.
Beyond Image Classification: A Video Benchmark and Dual-Branch Hybrid Discrimination Framework for Compositional Zero-Shot Learning
Dongyao Jiang (Xi'an Jiaotong University), Nanning Zheng (Xi'an Jiaotong University)
ClassificationContrastive LearningImageVideoBenchmark
🎯 What it does: This paper proposes a dual-branch hybrid discriminative framework capable of zero-shot classification for unseen combinations in video and image data.
Beyond Local Sharpness: Communication-Efficient Global Sharpness-aware Minimization for Federated Learning
Debora Caldarola (Politecnico di Torino), Marco Ciccone (Vector Institute)
OptimizationFederated LearningImage
🎯 What it does: The FedGLOSS algorithm is proposed in federated learning, utilizing Sharpness-aware Minimization (SAM) on the server side to achieve the flattening of the global model, and avoiding additional communication by using the pseudo-gradient from the previous round to approximate the global sharpness;
Beyond Sight: Towards Cognitive Alignment in LVLM via Enriched Visual Knowledge
Yaqi Zhao (Peking University), Wentao Zhang (Peking University)
RecognitionObject DetectionTransformerLarge Language ModelVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: This paper studies the cognitive mismatch problem between visual encoders (VE) and large language models (LLM), and proposes an Entity Enhanced Cognitive Alignment (EECA) method based on multi-granularity supervision, significantly improving the performance of multimodal visual language models in landmark recognition tasks.
Beyond Single-Modal Boundary: Cross-Modal Anomaly Detection through Visual Prototype and Harmonization
Kai Mao (Xi'an Jiaotong University), Nanning Zheng (Xi'an Jiaotong University)
Anomaly DetectionContrastive LearningImageMultimodalityBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper proposes a cross-modal anomaly detection method that can generalize to unknown modalities after training on known modalities.
Beyond Words: Augmenting Discriminative Richness via Diffusions in Unsupervised Prompt Learning
Hairui Ren (Jilin University), Yi Chang (Jilin University)
ClassificationGenerationData SynthesisDomain AdaptationPrompt EngineeringVision Language ModelDiffusion modelImage
🎯 What it does: Using diffusion models to generate high-quality synthetic images, constructing auxiliary classifiers, and integrating with text classifiers to enhance the quality of pseudo-labels for unlabeled data, thereby improving unsupervised prompt learning;
BF-STVSR: B-Splines and Fourier---Best Friends for High Fidelity Spatial-Temporal Video Super-Resolution
Eunjin Kim (Ulsan National Institute of Science and Technology), Jaejun Yoo (Ulsan National Institute of Science and Technology)
RestorationSuper ResolutionOptical FlowVideo
🎯 What it does: A continuous space-time video super-resolution framework BF-STVSR based on B-Spline and Fourier mapping is proposed, capable of generating high-quality video frames at any spatial scale and temporal interpolation points.
BFANet: Revisiting 3D Semantic Segmentation with Boundary Feature Analysis
Weiguang Zhao (University of Liverpool), Kaizhu Huang (Duke Kunshan University)
SegmentationTransformerPoint Cloud
🎯 What it does: This paper addresses four types of errors in 3D point cloud semantic segmentation (false positives, merging errors, displacement errors, and region classification errors), proposes corresponding evaluation metrics, and designs the BFANet network to enhance segmentation performance through boundary feature analysis.
BG-Triangle: Bezier Gaussian Triangle for 3D Vectorization and Rendering
Minye Wu (KU Leuven), Jingyi Yu (ShanghaiTech University)
GenerationData SynthesisOptimizationNeural Radiance FieldGaussian SplattingImage
🎯 What it does: Designed and implemented a B’ezier Gaussian Triangle mixed representation for differentiable rendering of 3D vectorization and view synthesis.
BHViT: Binarized Hybrid Vision Transformer
Tian Gao (Nanjing University of Science and Technology), Hui Kong (University of Macau)
SegmentationOptimizationComputational EfficiencyKnowledge DistillationTransformerImage
🎯 What it does: This paper proposes a binarizable hybrid visual Transformer (BHViT) aimed at significantly reducing computational load and energy consumption while maintaining high accuracy, making it suitable for deployment on edge devices.
Bias for Action: Video Implicit Neural Representations with Bias Modulation
Alper Kayabasi (University of California Riverside), Vishwanath Saragadam (University of California Riverside)
RestorationSuper ResolutionVideo
🎯 What it does: The ActINR model is proposed, which achieves continuous modeling of videos through implicit neural representations (INR) with bias modulation, supporting tasks such as slow motion, super-resolution, denoising, and restoration.
BIGS: Bimanual Category-agnostic Interaction Reconstruction from Monocular Videos via 3D Gaussian Splatting
Jeongwan On (UNIST), Seungryul Baek (UNIST)
Object DetectionGenerationPose EstimationDiffusion modelGaussian SplattingVideo
🎯 What it does: Designed and implemented the BIGS pipeline, which utilizes 3D Gaussian splatting to reconstruct interactive scenes between two hands and unknown objects in monocular videos, and supports high-quality rendering from new viewpoints.
BiLoRA: Almost-Orthogonal Parameter Spaces for Continual Learning
Hao Zhu (Data61 CSIRO), Piotr Koniusz (Data61 CSIRO)
ClassificationDomain AdaptationImage
🎯 What it does: Proposes BiLoRA - a parameter-efficient continual learning method achieved through bilinear rewriting and fixed orthogonal bases (Fourier).
BiM-VFI: Bidirectional Motion Field-Guided Frame Interpolation for Video with Non-uniform Motions
Wonyong Seo (Korea Advanced Institute of Science and Technology), Munchurl Kim (Korea Advanced Institute of Science and Technology)
Image TranslationRestorationKnowledge DistillationConvolutional Neural NetworkOptical FlowVideo
🎯 What it does: This paper studies and implements a high-quality video frame interpolation method based on Bidirectional Motion Field (BiM) called BiM-VFI.
BimArt: A Unified Approach for the Synthesis of 3D Bimanual Interaction with Articulated Objects
Wanyue Zhang (Max Planck Institute for Informatics), Christian Theobalt (Max Planck Institute for Informatics)
GenerationRobotic IntelligenceTransformerDiffusion modelMultimodality
🎯 What it does: The BimArt framework based on diffusion models generates diverse and contactable two-handed 3D interaction animations using the trajectories of articulated objects.
BIMBA: Selective-Scan Compression for Long-Range Video Question Answering
Md Mohaiminul Islam (University of North Carolina), Lorenzo Torresani (Meta AI)
RetrievalCompressionTransformerLarge Language ModelSupervised Fine-TuningVideoMultimodality
🎯 What it does: A long video question-answering model named BIMBA is proposed, which compresses massive video visual tokens into a small number of high-information tokens using a Mamba-based selective scanning state space model, and then inputs them into a LLM to generate answers.
Binarized Mamba-Transformer for Lightweight Quad Bayer HybridEVS Demosaicing
Shiyang Zhou (Harbin Institute of Technology), Jingyong Su (Harbin Institute of Technology)
RestorationTransformerImageVideo
🎯 What it does: A lightweight binary Mamba-Transformer network BMTNet is proposed for image demosaicing in Quad Bayer HybridEVS cameras.
Binarized Neural Network for Multi-spectral Image Fusion
Junming Hou (Southeast University), Liang-Jian Deng (University of Electronic Science and Technology of China)
Image TranslationRestorationGenerationConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: A binary neural network for multispectral image fusion (BNNPan) is proposed, which generates high-resolution multispectral images by fusing high-resolution multispectral images with low-resolution multispectral images and high-resolution panchromatic images.
BiomedCoOp: Learning to Prompt for Biomedical Vision-Language Models
Taha Koleilat (Concordia University), Yiming Xiao (Concordia University)
ClassificationAnomaly DetectionKnowledge DistillationTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageBiomedical DataMagnetic Resonance ImagingComputed TomographyUltrasound
🎯 What it does: For the task of medical image classification with limited annotations, we propose BiomedCoOp, a prompt learning-based framework that utilizes the BiomedCLIP vision-language model and diverse medical prompts generated by GPT-4. By combining semantic consistency and knowledge distillation mechanisms, it achieves efficient and transferable few-shot learning.
BIOMEDICA: An Open Biomedical Image-Caption Archive, Dataset, and Vision-Language Models Derived from Scientific Literature
Alejandro Lozano (Stanford University), Serena Yeung-Levy (Stanford University)
RetrievalTransformerVision Language ModelContrastive LearningImageTextBiomedical DataBenchmark
🎯 What it does: The BIOMEDICA framework was constructed, collecting and organizing 24 million image-text pairs from 6 million PubMed Central Open Access papers, and providing 27 metadata fields; subsequently, a variant of CLIP (BMC-CLIP) was continuously pre-trained on this large-scale dataset.
BioX-CPath: Biologically-driven Explainable Diagnostics for Multistain IHC Computational Pathology
Amaya Gallagher-Syed (Queen Mary University of London), Gregory Slabaugh
ClassificationExplainability and InterpretabilityGraph Neural NetworkImageBiomedical Data
🎯 What it does: This paper proposes an interpretable multi-staining immunohistochemistry graph neural network capable of multi-label classification on whole slide images.
BIP3D: Bridging 2D Images and 3D Perception for Embodied Intelligence
Xuewu Lin (Horizon Robotics), Zhizhong Su (Horizon Robotics)
Object DetectionRobotic IntelligenceTransformerVision Language ModelImageTextBenchmark
🎯 What it does: We propose BIP3D, an image-centric 3D perception model capable of simultaneously performing 3D object detection and 3D visual localization. It takes multi-view images, text, and optional depth maps as input and outputs 3D bounding boxes.
Birth and Death of a Rose
Chen Geng, Jiajun Wu
GenerationData SynthesisDiffusion modelScore-based ModelVideoText
🎯 What it does: A full-process method based on a two-dimensional diffusion model is proposed to generate sequences of three-dimensional object geometry and material properties (such as color, roughness, and metallicity) over time, without the need for any three-dimensional data.
BizGen: Advancing Article-level Visual Text Rendering for Infographics Generation
Yuyang Peng (Tsinghua University), Yuhui Yuan (Tsinghua University)
GenerationTransformerDiffusion modelImageTextRetrieval-Augmented Generation
🎯 What it does: This paper presents BizGen, specifically designed to address the task of business content generation (infographics, slides) at the article level for visual text rendering.
Black Hole-Driven Identity Absorbing in Diffusion Models
Muhammad Shaheryar (Kyungpook National University), Soon Ki Jung (Kyungpook National University)
GenerationSafty and PrivacyDiffusion modelImage
🎯 What it does: A Black Hole-Driven Identity Absorption (BIA) framework is designed for the semantic latent space (h-space) in diffusion models, achieving identity erasure for specified identities, thereby enhancing the model's privacy protection and content deletion capabilities when generating images.
Black Swan: Abductive and Defeasible Video Reasoning in Unpredictable Events
Aditya Chinchure (University of British Columbia), Leonid Sigal (University of British Columbia)
Anomaly DetectionTransformerLarge Language ModelVision Language ModelVideoMultimodalityBenchmark
🎯 What it does: A reasoning benchmark for rare events in videos, called BlackSwanSuite, is proposed, which includes three types of tasks: forecasting, detective work, and reporting.
Black-Box Forgery Attacks on Semantic Watermarks for Diffusion Models
Andreas Müller, Erwin Quiring (Ruhr University Bochum)
GenerationAdversarial AttackDiffusion modelImage
🎯 What it does: A black-box forgery and removal attack against semantic watermarks is proposed, demonstrating that a high success rate of watermark transfer can be achieved between different models with just one reference watermark image.
BLADE: Single-view Body Mesh Estimation through Accurate Depth Estimation
Shengze Wang (UNC Chapel Hill), Michael Stengel (NVIDIA)
Pose EstimationDepth EstimationOptimizationTransformerImageMesh
🎯 What it does: This paper proposes the BLADE method, which accurately estimates the 3D mesh of the human body and the camera perspective from a single image, focusing on solving the perspective distortion problem in close-range scenes.
BlenderGym: Benchmarking Foundational Model Systems for Graphics Editing
Yunqi Gu (Stanford University), Leonidas Guibas (Stanford University)
Large Language ModelVision Language ModelMeshBenchmark
🎯 What it does: Designed and implemented the BlenderGym benchmark for evaluating the code editing capabilities of visual language models in 3D graphic editing;
Blind Bitstream-corrupted Video Recovery via Metadata-guided Diffusion Model
Shuyun Wang (University of Queensland), Xin Yu (University of Queensland)
RestorationTransformerDiffusion modelVideo
🎯 What it does: A recovery method for damaged videos with blind spot flow is proposed, which completely relies on video metadata to guide the diffusion model for recovery without the need for manually annotated masks.
BlobGEN-Vid: Compositional Text-to-Video Generation with Blob Video Representations
Weixi Feng (University of California Santa Barbara), Weili Nie (NVIDIA)
GenerationData SynthesisLarge Language ModelDiffusion modelVideoText
🎯 What it does: This paper proposes a blob video representation and the BlobGEN-Vid framework, which achieves fine-grained control of motion, appearance, and camera through blob conditions in text-to-video generation.
BlockDance: Reuse Structurally Similar Spatio-Temporal Features to Accelerate Diffusion Transformers
Hui Zhang (Fudan University), Zuxuan Wu (Fudan University)
RestorationGenerationComputational EfficiencyTransformerReinforcement LearningDiffusion modelImageVideoText
🎯 What it does: A training-independent BlockDance method is proposed, which accelerates the iterative denoising process of the Diffusion Transformer by caching and reusing structurally similar spatio-temporal features. Based on this, an instance-adaptive acceleration network called BlockDance-Ada is designed.
Blood Flow Speed Estimation with Optical Coherence Tomography Angiography Images
Wensheng Cheng (Stony Brook University), Haibin Ling (Stony Brook University)
Depth EstimationTransformerImageBiomedical Data
🎯 What it does: This paper proposes an end-to-end deep learning framework called OCTA-Flow, which can directly estimate blood flow velocity from Optical Coherence Tomography Angiography (OCTA) images, replacing the traditional Optical Doppler Tomography (ODT) that requires expensive hardware and complex signal processing.
BlueLM-V-3B: Algorithm and System Co-Design for Multimodal Large Language Models on Mobile Devices
Xudong Lu (vivo AI Lab), Hongsheng Li (CUHK MMLab)
OptimizationComputational EfficiencyTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: Introducing BlueLM-V-3B, a 3B parameter multimodal large language model suitable for mobile devices, achieving collaborative optimization of algorithms and systems.
Blurred LiDAR for Sharper 3D: Robust Handheld 3D Scanning with Diffuse LiDAR and RGB
Nikhil Behari (Massachusetts Institute of Technology), Ramesh Raskar (Massachusetts Institute of Technology)
SegmentationDepth EstimationGaussian SplattingSimultaneous Localization and MappingImagePoint Cloud
🎯 What it does: This study investigates how to achieve robust 3D reconstruction in low-texture, low-light, and low-reflectivity scenes using a scattering LiDAR and RGB fusion on handheld devices.
Blurry-Edges: Photon-Limited Depth Estimation from Defocused Boundaries
Wei Xu (Purdue University), Qi Guo (Purdue University)
Depth EstimationConvolutional Neural NetworkTransformerImage
🎯 What it does: A deep estimation framework utilizing blurry edge information (Blurry-Edges) is proposed, capable of recovering object depth from boundary blurriness in photon-limited and severely noisy image pairs.
BOE-ViT: Boosting Orientation Estimation with Equivariance in Self-Supervised 3D Subtomogram Alignment
Runmin Jiang (Carnegie Mellon University), Min Xu (Carnegie Mellon University)
Pose EstimationTransformerImage
🎯 What it does: This paper presents BOE-ViT, a self-supervised 3D subtomogram alignment method based on Vision Transformer.
BOLT: Boost Large Vision-Language Model Without Training for Long-form Video Understanding
Shuming Liu (King Abdullah University of Science and Technology), Bernard Ghanem (King Abdullah University of Science and Technology)
RetrievalTransformerVision Language ModelVideoMultimodality
🎯 What it does: The BOLT method is proposed, which enhances the performance of large-scale visual language models in long video understanding tasks through frame selection strategies such as inverse transformation sampling, without the need for additional training.
Boltzmann Attention Sampling for Image Analysis with Small Objects
Theodore Zhao (Microsoft), Mu Wei (Microsoft)
Object DetectionSegmentationTransformerImageBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: This paper proposes a Transformer architecture named BoltzFormer, which utilizes Boltzmann distribution for dynamic sampling of sparse attention regions to achieve small object detection and segmentation based on text prompts.
Boost the Inference with Co-training: A Depth-guided Mutual Learning Framework for Semi-supervised Medical Polyp Segmentation
Yuxin Li (Ocean University of China), Zhibin Yu (Ocean University of China)
SegmentationConvolutional Neural NetworkContrastive LearningImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper proposes a semi-supervised polyp segmentation framework RD-Net based on Mean Teacher, which utilizes an auxiliary student network to train depth images and achieves high-precision segmentation during inference using only RGB through depth-guided cross-modal mutual learning.
Boost Your Human Image Generation Model via Direct Preference Optimization
Sanghyeon Na (Kakao), Hyunjoon Lee (Kakao)
GenerationOptimizationReinforcement LearningDiffusion modelImageText
🎯 What it does: A method called HG-DPO based on Direct Preference Optimization (DPO) is proposed, which uses real human portraits as 'winning' samples during training and combines three-stage curriculum learning to generate high-quality and highly realistic human images.
Boosting Adversarial Transferability through Augmentation in Hypothesis Space
Yu Guo (Xiamen University), Cheng Wang (Xiamen University)
Adversarial AttackConvolutional Neural NetworkTransformerImagePoint Cloud
🎯 What it does: A model space augmentation-based adversarial attack method called OPS is proposed, which constructs neighborhood models in the hypothesis space using random input transformation operators and perturbations, and solves random optimization, significantly enhancing the transferability of adversarial samples.
Boosting Domain Incremental Learning: Selecting the Optimal Parameters is All You Need
Qiang Wang (Xi'an Jiaotong University), Yihong Gong (Xi'an Jiaotong University)
ClassificationObject DetectionDomain AdaptationTransformerSupervised Fine-TuningImageAudio
🎯 What it does: The SOYO framework is proposed to improve domain label prediction and parameter selection in Parameter-Isolated Domain Incremental Learning (PIDIL), thereby enhancing the model's continuous learning performance in multi-task scenarios.
Boosting Point-Supervised Temporal Action Localization through Integrating Query Reformation and Optimal Transport
Mengnan Liu (Xi'an Jiaotong University), Gang Hua (Dolby Laboratories)
RecognitionObject DetectionTransformerVideo
🎯 What it does: An end-to-end point annotation temporal action localization framework QROT is proposed, which integrates query reconstruction and optimal transport to generate high-quality pseudo labels and perform decoder matching.
Boosting the Dual-Stream Architecture in Ultra-High Resolution Segmentation with Resolution-Biased Uncertainty Estimation
Rong Qin (Nankai University), Jufeng Yang (Nankai University)
SegmentationConvolutional Neural NetworkImage
🎯 What it does: A dual-stream ultra-high resolution (UHR) semantic segmentation framework is proposed, which simultaneously achieves three main objectives: feature fusion, important region amplification, and detail supplementation by estimating resolution deviation uncertainty.
BOOTPLACE: Bootstrapped Object Placement with Detection Transformers
Hang Zhou (University of Alberta), Li Cheng (University of Alberta)
Object DetectionData SynthesisTransformerImage
🎯 What it does: A 'placement-by-detection' framework based on detection transformers is proposed, which first detects potential placement areas in the image and then matches target objects to the best areas through an object-region association network, achieving precise copy-paste object placement.
Bootstrap Your Own Views: Masked Ego-Exo Modeling for Fine-grained View-invariant Video Representations
Jungin Park (Yonsei University), Kwanghoon Sohn (Yonsei University)
RecognitionRetrievalRepresentation LearningTransformerAuto EncoderContrastive LearningVideo
🎯 What it does: A framework is proposed that utilizes a self-supervised masking strategy (self-view mask and cross-view mask) for representation learning on unpaired and asynchronous first-person and third-person videos.
BooW-VTON: Boosting In-the-Wild Virtual Try-On via Mask-Free Pseudo Data Training
Xuanpu Zhang (Tianjin University), An-An Liu (Tianjin University)
Image TranslationData SynthesisDiffusion modelImage
🎯 What it does: A Mask-free virtual try-on model called BooW-VTON is proposed, which achieves high-quality clothing try-on in wild scenes without the need for a portrait parser by utilizing pseudo-triplet data, wild scene data augmentation, and try-on localization loss.
Brain-Inspired Spiking Neural Networks for Energy-Efficient Object Detection
Ziqi Li (Chang'an University), Qianxi Zhang (Chang'an University)
Object DetectionSpiking Neural NetworkImage
🎯 What it does: A multi-scale object detection framework MSD based on spiking neural networks is proposed, which can be directly trained, is low-energy, and achieves object detection on static images and event data.
Breaking the Low-Rank Dilemma of Linear Attention
Qihang Fan (Chinese Academy of Sciences), Ran He (Chinese Academy of Sciences)
ClassificationObject DetectionSegmentationTransformerImage
🎯 What it does: This study investigates the low-rank bottleneck of linear attention, proposing Rank-Augmented Linear Attention (RALA) and constructing the Rank-Augmented Vision Linear Transformer (RAVLT) based on it to achieve efficient visual modeling.
Breaking the Memory Barrier of Contrastive Loss via Tile-Based Strategy
Zesen Cheng (Alibaba Group), Lidong Bing (Alibaba Group)
RetrievalOptimizationComputational EfficiencyContrastive LearningImage
🎯 What it does: A contrastive loss strategy based on tile-wise computation is proposed to reduce GPU memory consumption during large batch training.
BrepGiff: Lightweight Generation of Complex B-rep with 3D GAT Diffusion
Hao Guo (Northwest Polytechnical University), Yilei Shi (Northwest Polytechnical University)
GenerationData SynthesisComputational EfficiencyGraph Neural NetworkDiffusion modelMesh
🎯 What it does: A lightweight B-rep direct generation method based on 3D graphs is proposed, utilizing a GAT diffusion model to construct CAD models from nodes (face centroids) and edges (inter-face connectivity).
Bridge Frame and Event: Common Spatiotemporal Fusion for High-Dynamic Scene Optical Flow
Hanyu Zhou (National University of Singapore), Luxin Yan (Huazhong University of Science and Technology)
Autonomous DrivingRecurrent Neural NetworkTransformerOptical FlowVideoMultimodality
🎯 What it does: A framework for optical flow estimation (ComST-Flow) that utilizes the simultaneous spatiotemporal fusion of frame cameras and event cameras is proposed, achieving dense and continuous optical flow in high dynamic scenes by constructing a simultaneous spatiotemporal gradient space.
Bridge the Gap: From Weak to Full Supervision for Temporal Action Localization with PseudoFormer
Ziyi Liu (University of Science and Technology Beijing), Yangcen Liu (Georgia Institute of Technology)
RecognitionObject DetectionTransformerVideo
🎯 What it does: This paper proposes a two-branch framework called PseudoFormer, aimed at bridging the gap between weakly supervised and fully supervised temporal action localization through pseudo-labels.
Bridging Gait Recognition and Large Language Models Sequence Modeling
Shaopeng Yang (Beijing Normal University), Yongzhen Huang (Beijing Normal University)
RecognitionTransformerLarge Language ModelVideoText
🎯 What it does: Utilizing large language models for sequence modeling of gait sequences, the GaitLLM framework is proposed to convert gait features into text format (G2L module) for input into a frozen LLM, and then map the LLM output back to the gait feature space (L2G module), thereby enhancing gait recognition performance.
Bridging Modalities: Improving Universal Multimodal Retrieval by Multimodal Large Language Models
Xin Zhang (Hong Kong Polytechnic University), Min Zhang (Soochow University)
RetrievalTransformerLarge Language ModelContrastive LearningImageTextMultimodalityBenchmark
🎯 What it does: A unified multimodal retrieval framework GME is proposed, and a large-scale fused modality training set and UMRB benchmark are constructed.
Bridging Past and Future: End-to-End Autonomous Driving with Historical Prediction and Planning
Bozhou Zhang (Fudan University), Li Zhang (Fudan University)
Autonomous DrivingTransformerPoint Cloud
🎯 What it does: Proposes the BridgeAD framework, which integrates historical prediction and planning information into an end-to-end autonomous driving system.
Bridging the Gap between Gaussian Diffusion Models and Universal Quantization for Image Compression
Lucas Relic (ETH Zurich), Christopher Schroers (ETH Zurich)
CompressionDiffusion modelImage
🎯 What it does: A unified quantization-based diffusion model image compression framework is proposed, improving the forward diffusion process and employing a uniform noise diffusion model to achieve more realistic image reconstruction at low bit rates.
Bridging the Vision-Brain Gap with an Uncertainty-Aware Blur Prior
Haitao Wu (Tianjin University), Xiaomin Ying (Beijing Institute of Basic Medical Sciences)
RetrievalContrastive LearningImageMultimodalityMagnetic Resonance Imaging
🎯 What it does: This paper proposes an Uncertainty-aware Blur Prior (UBP) based on uncertainty perception to address the systematic gap (System GAP) and random gap (Random GAP) in visual-brain decoding, achieving zero-shot retrieval from EEG/MRI signals to natural images.
Bridging Viewpoint Gaps: Geometric Reasoning Boosts Semantic Correspondence
Qiyang Qian (University of California), Chenfeng Xu (Stanford University)
RecognitionObject DetectionDomain AdaptationComputational EfficiencyKnowledge DistillationTransformerDiffusion modelImagePoint Cloud
🎯 What it does: This paper proposes a semantic correspondence method that combines 3D geometric alignment with deformation and sparse semantic matching, utilizing DUSt3R trained on synthetic cross-instance perspective data to achieve efficient and robust correspondences.