CVPR 2025 Papers — Page 2
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2871 papers
Advancing Manga Analysis: Comprehensive Segmentation Annotations for the Manga109 Dataset
Minshan Xie (Centre for Perceptual and Interactive Intelligence), Tien-Tsin Wong (Monash University)
Object DetectionSegmentationDiffusion modelImage
🎯 What it does: This paper adds pixel-level segmentation annotations to the Manga109 dataset, creating MangaSeg.
Advancing Multiple Instance Learning with Continual Learning for Whole Slide Imaging
Xianrui Li (City University of Hong Kong), Antoni B. Chan (City University of Hong Kong)
Computational EfficiencyKnowledge DistillationTransformerImageBiomedical Data
🎯 What it does: This paper analyzes the catastrophic forgetting of multi-instance learning (MIL) in continual learning and proposes two solutions: Attention Knowledge Distillation (AKD) and Pseudo-Package Memory Pool (PMP).
Advancing Myopia To Holism: Fully Contrastive Language-Image Pre-training
Haicheng Wang (Taobao and Tmall Group of Alibaba), Yanfeng Wang (Taobao and Tmall Group of Alibaba)
RetrievalRepresentation LearningTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: To address the 'nearsightedness' problem caused by CLIP's one-to-one alignment, this paper proposes Holistic CLIP, which aligns images with multiple texts (multi-perspective, multi-level, multi-granularity) and introduces a multi-branch structure in the image encoder to achieve multi-dimensional representation of images. Subsequently, multi-to-multi alignment contrastive learning is employed for pre-training.
Advancing Semantic Future Prediction through Multimodal Visual Sequence Transformers
Efstathios Karypidis (Archimedes Athena Research Center), Nikos Komodakis
SegmentationDepth EstimationAutonomous DrivingTransformerSupervised Fine-TuningImageVideoMultimodality
🎯 What it does: Proposes the FUTURIST framework, which uses a unified visual sequence Transformer for future frame prediction in multimodal (semantic segmentation, depth maps).
Adventurer: Optimizing Vision Mamba Architecture Designs for Efficiency
Feng Wang (Johns Hopkins University), Cihang Xie (University of California Santa Cruz)
ClassificationObject DetectionSegmentationOptimizationComputational EfficiencyRepresentation LearningTransformerLarge Language ModelImage
🎯 What it does: Designed and implemented the Adventurer series models, utilizing one-dimensional image block sequences and causal language models to achieve linear complexity visual representation learning.
Adversarial Diffusion Compression for Real-World Image Super-Resolution
Bin Chen (Peking University), Lei Zhang (Hong Kong Polytechnic University)
RestorationSuper ResolutionCompressionKnowledge DistillationDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes AdcSR, a real-time single-step super-resolution model for real scene images that compresses OSEDiff using the ADC framework.
Adversarial Domain Prompt Tuning and Generation for Single Domain Generalization
Zhipeng Xu (Xidian University), Xinbo Gao (Chongqing University of Posts and Telecommunications)
ClassificationGenerationDomain AdaptationPrompt EngineeringDiffusion modelImage
🎯 What it does: This paper proposes a single-domain generalization framework (PAPT) based on a pre-trained text-image (T2I) diffusion model, which generates diverse domain-style images by learning two sets of abstract phrases: category prompts and domain prompts, thereby enhancing the model's performance on unseen domains.
AerialMegaDepth: Learning Aerial-Ground Reconstruction and View Synthesis
Khiem Vuong (Carnegie Mellon University), Shubham Tulsiani (Carnegie Mellon University)
Data SynthesisDepth EstimationImage
🎯 What it does: AerialMegaDepth mixed dataset was constructed, combining 3D city grid pseudo-synthetic views with real ground images, and existing multi-view geometry and view synthesis models were fine-tuned on this data.
AeroGen: Enhancing Remote Sensing Object Detection with Diffusion-Driven Data Generation
Datao Tang (Xi'an Jiaotong University), Deyu Meng (Xi'an Jiaotong University)
Object DetectionGenerationData SynthesisDiffusion modelImage
🎯 What it does: AeroGen, a layout-controllable generation framework based on diffusion models, is proposed for synthesizing high-quality synthetic images that meet horizontal or rotated bounding box constraints in remote sensing image object detection tasks, and an end-to-end data augmentation pipeline is constructed.
AeSPa : Attention-guided Self-supervised Parallel Imaging for MRI Reconstruction
Jinho Joo (Yonsei University), Dosik Hwang (Yonsei University)
RestorationConvolutional Neural NetworkBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This paper proposes a zero-shot self-supervised parallel imaging MRI reconstruction method called AeSPa, aimed at achieving fast and high-quality imaging without the need for fully sampled reference data.
Aesthetic Post-Training Diffusion Models from Generic Preferences with Step-by-step Preference Optimization
Zhanhao Liang (Australian National University), Liang Zheng (Microsoft)
GenerationOptimizationDiffusion modelImageText
🎯 What it does: A Stepwise Preference Optimization (SPO) method is proposed to enhance the aesthetic quality of images in text-to-image diffusion models.
AesthetiQ: Enhancing Graphic Layout Design via Aesthetic-Aware Preference Alignment of Multi-modal Large Language Models
Sohan Patnaik (Adobe), Mausoom Sarkar (Adobe)
GenerationOptimizationTransformerLarge Language ModelMultimodality
🎯 What it does: Utilizing multimodal large language models and aesthetic preference alignment techniques to generate graphic layouts that conform to human aesthetics.
AffordDP: Generalizable Diffusion Policy with Transferable Affordance
Shijie Wu (ShanghaiTech University), Jingya Wang (ShanghaiTech University)
Robotic IntelligenceTransformerReinforcement LearningDiffusion modelPoint Cloud
🎯 What it does: This paper proposes AffordDP, a diffusion-based imitation learning method that leverages transferable 3D affordances (static contact points and dynamic trajectories) to achieve generalization to unseen objects and unseen categories in tasks such as robot grasping, pulling doors, and placing items.
AFL: A Single-Round Analytic Approach for Federated Learning with Pre-trained Models
Run He (South China University of Technology), Huiping Zhuang (South China University of Technology)
Federated LearningConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes Analytic Federated Learning (AFL), a gradient-free federated learning framework that achieves single-round aggregation on pre-trained models.
AG-VPReID: A Challenging Large-Scale Benchmark for Aerial-Ground Video-based Person Re-Identification
Huy Nguyen (Queensland University of Technology), Clinton Fookes (Queensland University of Technology)
RecognitionRetrievalTransformerContrastive LearningVideoBenchmark
🎯 What it does: A large-scale aerial-ground video person re-identification benchmark AG-VPReID and a three-stream end-to-end model AG-VPReID-Net are proposed.
AI-Face: A Million-Scale Demographically Annotated AI-Generated Face Dataset and Fairness Benchmark
Li Lin (Purdue University), Shu Hu (Purdue University)
RecognitionGenerationData SynthesisDiffusion modelGenerative Adversarial NetworkImageBenchmark
🎯 What it does: An AI-Face dataset with a scale of one million, containing real, synthetic, and deepfake faces, annotated for skin color, gender, and age, has been constructed, and a fairness benchmark has been established based on this dataset.
AIGV-Assessor: Benchmarking and Evaluating the Perceptual Quality of Text-to-Video Generation with LMM
Jiarui Wang (Shanghai Jiao Tong University), Xiongkuo Min (Shanghai Jiao Tong University)
GenerationData SynthesisTransformerLarge Language ModelContrastive LearningVideoTextMultimodalityBenchmark
🎯 What it does: A database and evaluation model for assessing the quality of text-to-video generation (AIGV) has been proposed;
AIM-Fair: Advancing Algorithmic Fairness via Selectively Fine-Tuning Biased Models with Contextual Synthetic Data
Zengqun Zhao (Queen Mary University of London), Ioannis Patras (Queen Mary University of London)
ClassificationGenerationData SynthesisConvolutional Neural NetworkLarge Language ModelSupervised Fine-TuningDiffusion modelImage
🎯 What it does: The AIM-Fair method is proposed, which utilizes context prompt-driven diffusion models generated by large language models to synthesize data and selectively fine-tune pre-trained biased models, thereby significantly improving fairness while maintaining overall accuracy.
AIpparel: A Multimodal Foundation Model for Digital Garments
Kiyohiro Nakayama (Stanford University), Gordon Wetzstein (Stanford University)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: Developed the AIpparel multimodal foundation model, capable of generating and editing complex sewing patterns based on text, images, or editing instructions, and directly used for 3D garment simulation.
AirRoom: Objects Matter in Room Reidentification
Runmao Yao (University at Buffalo), Chen Wang (University at Buffalo)
RecognitionRetrievalConvolutional Neural NetworkTransformerContrastive LearningImage
🎯 What it does: The AirRoom framework is proposed, utilizing multi-layer object information such as global semantics, object patches, instance segmentation, and key points to achieve indoor room re-identification.
AKiRa: Augmentation Kit on Rays for Optical Video Generation
Xi Wang (LIX, École Polytechnique, IP Paris), Vicky Kalogeiton (University of Rennes, IRISA, Inria, CNRS)
GenerationData SynthesisDiffusion modelVideoText
🎯 What it does: The AKiRa framework is proposed, utilizing optical Ray augmentation to train a camera adapter, enabling text-to-video diffusion models to precisely control optical parameters such as camera motion, focal length, distortion, and depth of field, generating realistic zoom, fisheye, and Bokeh effects.
Alias-Free Latent Diffusion Models: Improving Fractional Shift Equivariance of Diffusion Latent Space
Yifan Zhou, Xingang Pan
GenerationData SynthesisDiffusion modelAuto EncoderImage
🎯 What it does: This paper proposes Alias-Free Latent Diffusion Models (AF-LDM), which enhance the shift equivariance of latent diffusion models by improving the design of VAE and U-Net's anti-aliasing, equivariant loss, and equivariant attention, resulting in more consistent and stable generation results.
ALIEN: Implicit Neural Representations for Human Motion Prediction under Arbitrary Latency
Dong Wei (Nanjing University of Science and Technology), Huaijiang Sun (Nanjing University of Science and Technology)
Pose EstimationVideo
🎯 What it does: The ALiEN model is proposed, which implements human motion prediction under arbitrary delays using implicit neural representations.
Align-A-Video: Deterministic Reward Tuning of Image Diffusion Models for Consistent Video Editing
Shengzhi Wang (Shanghai Research Institute for Autonomous Intelligent Systems), Qingwen Liu (Shanghai Research Institute for Autonomous Intelligent Systems)
GenerationData SynthesisReinforcement LearningDiffusion modelVideoBenchmark
🎯 What it does: Proposes Align-A-Video, a text-driven video editing framework achieved through reward fine-tuning and anchor frame feature propagation.
Align-KD: Distilling Cross-Modal Alignment Knowledge for Mobile Vision-Language Large Model Enhancement
Qianhan Feng (Peking University), Xinghao Chen (Huawei)
Computational EfficiencyKnowledge DistillationTransformerVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes Align-KD, which uses the teacher model 7B MobileVLM V2 to perform cross-modal alignment knowledge distillation for the 1.7B version, significantly improving the reasoning performance of mobile visual-language models.
Align3R: Aligned Monocular Depth Estimation for Dynamic Videos
Jiahao Lu (Hong Kong University of Science and Technology), Yuan Liu (Hong Kong University of Science and Technology)
Pose EstimationDepth EstimationTransformerVideoPoint Cloud
🎯 What it does: This paper proposes the Align3R method, which combines single-frame depth estimation with the DUSt3R model, utilizing a Transformer to extract depth features and inject them into the DUSt3R decoder. It then obtains temporally consistent depth sequences and camera poses through global alignment optimization.
AlignMamba: Enhancing Multimodal Mamba with Local and Global Cross-modal Alignment
Yan Li (Pengcheng Laboratory), Dongmei Jiang (Pengcheng Laboratory)
ClassificationRecurrent Neural NetworkMultimodality
🎯 What it does: This paper proposes AlignMamba, which achieves efficient multimodal fusion by incorporating local Optimal Transport alignment and global Maximum Mean Discrepancy alignment at the tip of Mamba.
Alignment, Mining and Fusion: Representation Alignment with Hard Negative Mining and Selective Knowledge Fusion for Medical Visual Question Answering
Yuanhao Zou (University of Michigan), Zhaozheng Yin (Stony Brook University)
Representation LearningTransformerVision Language ModelContrastive LearningMultimodalityBiomedical DataElectronic Health Records
🎯 What it does: A unified multimodal alignment framework AMiF is proposed, combining hard negative sample mining and selective knowledge fusion, specifically designed for pre-training and fine-tuning in the medical visual question answering (Med-VQA) task.
All Languages Matter: Evaluating LMMs on Culturally Diverse 100 Languages
Ashmal Vayani (Mohamed bin Zayed University of AI), Fahad Shahbaz Khan (Linköping University)
Large Language ModelTextMultimodalityBenchmark
🎯 What it does: A multimodal benchmark ALM-bench covering 100 languages has been proposed, and 16 LMMs have been evaluated.
All-Day Multi-Camera Multi-Target Tracking
Huijie Fan (Shenyang Institute of Automation, Chinese Academy of Sciences), Qiang Wang (Nanjing University of Posts and Telecommunications)
Object TrackingVideoMultimodality
🎯 What it does: The first multi-camera multi-target tracking dataset with infrared modality, M3Track, has been constructed, and an all-day all-weather tracking network, ADMCMT, has been proposed.
All-directional Disparity Estimation for Real-world QPD Images
Hongtao Yu (OMNIVISION), Chengming Liu (OMNIVISION)
Depth EstimationConvolutional Neural NetworkOptical FlowImage
🎯 What it does: Proposed differential estimation methods DPNet and QuadNet for dual-pixel and quad-pixel cameras, and created the first QPD disparity dataset.
All-Optical Nonlinear Diffractive Deep Network for Ultrafast Image Denoising
Xiaoling Zhou (Peking University), Shikun Zhang (Peking University)
RestorationOptimizationReinforcement LearningImage
🎯 What it does: A fully optical nonlinear diffraction deep network N3DNet is designed and implemented for ultrafast image denoising.
AlphaPre: Amplitude-Phase Disentanglement Model for Precipitation Nowcasting
Kenghong Lin (Harbin Institute of Technology), Yunming Ye (Harbin Institute of Technology)
GenerationOptimizationConvolutional Neural NetworkSupervised Fine-TuningImageTime Series
🎯 What it does: This paper proposes the AlphaPre model, which utilizes frequency domain amplitude-phase separation to predict changes in rainfall location through a phase network and changes in rainfall intensity through an amplitude network, and integrates them through AlphaMixer to achieve more refined rainfall forecasting.
AMO Sampler: Enhancing Text Rendering with Overshooting
Xixi Hu (Google), Hongliang Fei (Google)
GenerationDiffusion modelRectified FlowImageTextStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: This paper proposes a sampling method that is untrained and used only during the inference phase—Overshooting sampler and its improved version AMO sampler, specifically designed to enhance the rendering quality of the Rectified Flow (RF) model in text generation, addressing the text errors and distortions caused by traditional Euler sampling.
AMR-Transformer: Enabling Efficient Long-range Interaction for Complex Neural Fluid Simulation
Zeyi Xu (Shanghai Jiao Tong University), Bingbing Ni (Shanghai Jiao Tong University)
TransformerMeshPhysics Related
🎯 What it does: A neural CFD solving pipeline that combines Adaptive Mesh Refinement (AMR) with Transformers is proposed, which can efficiently capture long-range dependencies and fine-grained structures in fluid dynamics.
An End-to-End Robust Point Cloud Semantic Segmentation Network with Single-Step Conditional Diffusion Models
Wentao Qu (Nanjing University of Science and Technology), Liang Xiao (Nanjing University of Science and Technology)
Object DetectionSegmentationAutonomous DrivingTransformerDiffusion modelPoint Cloud
🎯 What it does: An end-to-end point cloud semantic segmentation network called CDSegNet is developed based on the Conditional-Noise Framework (CNF), utilizing the noise system of DDPM to achieve single-step inference and enhance robustness against noise and sparse data.
An Image-like Diffusion Method for Human-Object Interaction Detection
Xiaofei Hui (Lancaster University), Jun Liu (Lancaster University)
Object DetectionGenerationTransformerDiffusion modelImage
🎯 What it does: This paper transforms human-object interaction (HOI) detection into the problem of generating 'HOI images', using image diffusion models to generate high-quality HOI images, thereby obtaining human-object interaction relationships.
Analyzing the Synthetic-to-Real Domain Gap in 3D Hand Pose Estimation
Zhuoran Zhao (Hong Kong University of Science and Technology), Angela Yao (National University of Singapore)
Data SynthesisPose EstimationDomain AdaptationPoint Cloud
🎯 What it does: This study investigates the synthetic-to-real domain gap in 3D hand pose estimation and proposes a high-quality hand data synthesis pipeline.
Anatomical Consistency and Adaptive Prior-informed Transformation for Multi-contrast MR Image Synthesis via Diffusion Model
Yejee Shin (Yonsei University), Dosik Hwang (Korea Institute of Science and Technology)
GenerationData SynthesisVision Language ModelDiffusion modelImageBiomedical DataMagnetic Resonance ImagingAlzheimer's Disease
🎯 What it does: This paper proposes the APT model, which achieves the synthesis of multi-contrast MR images without modal loss through multi-contrast information fusion and anatomical consistency loss.
Anchor-Aware Similarity Cohesion in Target Frames Enables Predicting Temporal Moment Boundaries in 2D
Jiawei Tan (Chongqing University), Kang Dang (Xi'an Jiaotong-Liverpool University)
RetrievalConvolutional Neural NetworkVision Language ModelVideo
🎯 What it does: Two modules are proposed: Anchor-Aware Feature Alignment (AFAF) and Frame-Frame Similarity Guided Detection (F2SGD). AFAF achieves query-video feature alignment by inducing semantically compact segments through anchor frames, while F2SGD transforms moment boundary prediction into single-point detection within a 2D similarity matrix.
AniDoc: Animation Creation Made Easier
Yihao Meng (Hong Kong University of Science and Technology), Huamin Qu (Hong Kong University of Science and Technology)
GenerationData SynthesisDiffusion modelVideo
🎯 What it does: A full-process animation line art coloring and intermediate frame interpolation tool called AniDoc has been developed based on a video diffusion model. It can automatically color line art sequences according to character design references and generate coherent animations with only the start and end frame sketches provided.
AniGrad: Anisotropic Gradient-Adaptive Sampling for 3D Reconstruction From Monocular Video
Noah Stier (University of California Santa Barbara), Tobias Höllerer (University of California Santa Barbara)
Depth EstimationOptimizationComputational EfficiencyConvolutional Neural NetworkVideoPoint CloudMesh
🎯 What it does: In 3D reconstruction based on monocular video, an adaptive and anisotropic sampling strategy called AniGrad is introduced. It utilizes local basis functions to represent TSDF and quickly determines the sampling density of each voxel by combining gradient upper bounds, achieving high-quality and low-latency mesh extraction.
AniGS: Animatable Gaussian Avatar from a Single Image with Inconsistent Gaussian Reconstruction
Lingteng Qiu (Alibaba Group), Zilong Dong (Nanjing University)
GenerationData SynthesisPose EstimationTransformerDiffusion modelAuto EncoderGaussian SplattingImageVideo
🎯 What it does: Generate multi-view frontal images from a single portrait image and reconstruct an animatable 3D portrait model using 4D Gaussian projection.
Animate and Sound an Image
Xihua Wang (Renmin University of China), Yunfeng Wang (ZHI-TECH GROUP)
GenerationData SynthesisDiffusion modelImageVideoMultimodalityAudio
🎯 What it does: This paper studies the Image-to-Sounding-Video task and proposes the JointDiT model, which jointly pre-trains video and audio diffusion models to achieve synchronized audio-visual generation from images.
AnimateAnything: Consistent and Controllable Animation for Video Generation
Guojun Lei (Zhejiang University), Weiwei Xu (Zhejiang University)
GenerationData SynthesisDiffusion modelOptical FlowVideoText
🎯 What it does: A two-stage controllable video generation framework called AnimateAnything is proposed, which first unifies various control signals (such as user dragging, camera trajectories, reference videos, etc.) into optical flow, and then uses optical flow and text prompts to drive a video diffusion model to generate high-quality, coherent videos.
AniMer: Animal Pose and Shape Estimation Using Family Aware Transformer
Jin Lyu (Southern University of Science and Technology), Liang An (Tsinghua University)
Pose EstimationTransformerContrastive LearningImage
🎯 What it does: A Transformer-based animal pose and shape estimation framework called AniMer has been developed, which can directly regress SMAL model parameters from a single image, enabling pose and shape recovery across different animal species.
AniMo: Species-Aware Model for Text-Driven Animal Motion Generation
Xuan Wang (Zhejiang University), Gaoang Wang (Zhejiang University)
GenerationTransformerGenerative Adversarial NetworkVideoText
🎯 What it does: We propose AniMo, a two-stage model for text-driven animal motion generation, achieving diverse animal posture motion generation.
ANNEXE: Unified Analyzing, Answering, and Pixel Grounding for Egocentric Interaction
Yuejiao Su (Hong Kong Polytechnic University), Lap-Pui Chau (Hong Kong Polytechnic University)
RecognitionObject DetectionSegmentationTransformerLarge Language ModelImageTextMultimodalityBenchmark
🎯 What it does: The Ego-IRG (Egocentric Interaction Reasoning and pixel Grounding) task is proposed, which unifies the analysis of interactions, answering queries, and pixel-level localization, and based on this, the ANNEXE model is developed; simultaneously, a large-scale Ego-IRGBench dataset is constructed.
Annotation Ambiguity Aware Semi-Supervised Medical Image Segmentation
Suruchi Kumari (Indian Institute of Technology Roorkee), Pravendra Singh (Indian Institute of Technology Roorkee)
SegmentationImageBiomedical DataComputed Tomography
🎯 What it does: A semi-supervised method called AmbiSSL is proposed, which combines a small amount of multi-label data with a large amount of unlabeled data to generate diverse and reliable medical image segmentation results.
AnomalyNCD: Towards Novel Anomaly Class Discovery in Industrial Scenarios
Ziming Huang (Huazhong University of Science and Technology), Yu Zhou (Huazhong University of Science and Technology)
Anomaly DetectionTransformerContrastive LearningImage
🎯 What it does: A multi-class industrial anomaly classification framework called AnomalyNCD is proposed, which is compatible with existing anomaly detection methods.
Anomize: Better Open Vocabulary Video Anomaly Detection
Fei Li (Wuhan University), Zheng Wang (Wuhan University)
Anomaly DetectionRecurrent Neural NetworkLarge Language ModelVision Language ModelVideoMultimodality
🎯 What it does: The Anomize framework is proposed, capable of simultaneously detecting and classifying both base and novel anomalies in open vocabulary video anomaly detection (OVVAD).
Antidote: A Unified Framework for Mitigating LVLM Hallucinations in Counterfactual Presupposition and Object Perception
Yuanchen Wu (Shanghai University), Xiaoqiang Li (Shanghai University)
GenerationData SynthesisOptimizationTransformerLarge Language ModelVision Language ModelDiffusion modelImageText
🎯 What it does: This paper proposes the Antidote framework, which trains a unified model to mitigate the erroneous generation of large visual language models in counterfactual premise questions and object perception hallucinations through synthesized data.
Any-Resolution AI-Generated Image Detection by Spectral Learning
Dimitrios Karageorgiou (Information Technologies Institute), Efstratios Gavves (University of Amsterdam)
Data SynthesisAnomaly DetectionTransformerDiffusion modelImage
🎯 What it does: A spectral distribution model of real images is constructed through self-supervised spectral learning, and spectral reconstruction similarity is utilized to detect AI-generated images.
Any3DIS: Class-Agnostic 3D Instance Segmentation by 2D Mask Tracking
Phuc Nguyen (MovianAI), Khoi Nguyen (Qualcomm AI Research)
Object DetectionSegmentationOptimizationComputational EfficiencyPoint Cloud
🎯 What it does: A class-agnostic 3D instance segmentation method based on 2D mask tracking, Any3DIS, is proposed, which can efficiently and accurately segment objects in point clouds.
Any6D: Model-free 6D Pose Estimation of Novel Objects
Taeyeop Lee (Korea Advanced Institute of Science and Technology), Kuk-Jin Yoon (Korea Advanced Institute of Science and Technology)
Pose EstimationImage
🎯 What it does: This paper proposes the Any6D framework, which can estimate the 6D pose and size of new objects in different scenes and viewpoints using only a single RGB-D anchor image, without the need for CAD models or multi-view references.
Anyattack: Towards Large-scale Self-supervised Adversarial Attacks on Vision-language Models
Jiaming Zhang (Hong Kong University of Science and Technology), Dit-Yan Yeung (Hong Kong University of Science and Technology)
RetrievalAdversarial AttackTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: A self-supervised AnyAttack framework is proposed, which pre-trains a noise generator on a large-scale unlabeled dataset and then fine-tunes it on downstream vision-language tasks to generate targeted adversarial images for any VLM.
AnyCam: Learning to Recover Camera Poses and Intrinsics from Casual Videos
Felix Wimbauer (Technical University of Munich), Daniel Cremers (Technical University of Munich)
Pose EstimationDepth EstimationOptimizationTransformerSimultaneous Localization and MappingOptical FlowVideo
🎯 What it does: We propose AnyCam, an end-to-end Transformer model that directly infers camera pose, intrinsic parameters, and uncertainty maps from everyday videos in a single forward pass, and allows for lightweight BA correction.
AnyDressing: Customizable Multi-Garment Virtual Dressing via Latent Diffusion Models
Xinghui Li (Bytedance Intelligent Creation), Qian He (Bytedance Intelligent Creation)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: A virtual try-on method for multiple garments based on a latent diffusion model, AnyDressing, is proposed, which can customize character images under arbitrary combinations of clothing and text prompts.
AnyEdit: Mastering Unified High-Quality Image Editing for Any Idea
Qifan Yu (Zhejiang University), Yueting Zhuang (Zhejiang University)
GenerationData SynthesisLarge Language ModelMixture of ExpertsDiffusion modelImage
🎯 What it does: A dataset of 2.5 million high-quality instruction-driven image editing pairs, called AnyEdit, has been constructed, and the AnySD model has been proposed to unify the handling of various editing tasks.
AnyMap: Learning a General Camera Model for Structure-from-Motion with Unknown Distortion in Dynamic Scenes
Andrea Porfiri Dal Cin (Qualcomm Technologies), Mohsen Ghafoorian (Qualcomm Technologies)
Pose EstimationDepth EstimationOptimizationOptical FlowVideo
🎯 What it does: This paper presents AnyMap, a differentiable structure from motion (SfM) framework that can simultaneously estimate dense 3D geometry, camera poses, and a general camera model implemented by a learnable inverse neural network (including radial and tangential distortion) while achieving motion regularization in dynamic scenes.
AnyMoLe: Any Character Motion In-betweening Leveraging Video Diffusion Models
Kwan Yun (KAIST), Junyong Noh (KAIST)
GenerationPose EstimationDomain AdaptationDiffusion modelVideo
🎯 What it does: The AnyMoLe method is proposed, which utilizes video diffusion models to achieve intermediate frame generation of arbitrary character movements without relying on external data.
AnySat: One Earth Observation Model for Many Resolutions, Scales, and Modalities
Guillaume Astruc (Univ Gustave Eiffel), Loic Landrieu (Univ Gustave Eiffel)
ClassificationSegmentationContrastive LearningImageMultimodality
🎯 What it does: This paper presents AnySat, a self-supervised learning model for Earth observation that can simultaneously handle various resolutions, scales, and sensors.
APHQ-ViT: Post-Training Quantization with Average Perturbation Hessian Based Reconstruction for Vision Transformers
Zhuguanyu Wu (Beihang University), Yunhong Wang (Beihang University)
ClassificationObject DetectionSegmentationTransformerImage
🎯 What it does: For post-training quantization of Vision Transformers, the APHQ-ViT method is proposed.
Apollo: An Exploration of Video Understanding in Large Multimodal Models
Orr Zohar (Meta GenAI), Xide Xia (Stanford University)
RecognitionGenerationComputational EfficiencyTransformerLarge Language ModelVision Language ModelVideoTextMultimodalityBenchmark
🎯 What it does: This paper systematically explores the design space of large video models, proposing the 'Scaling Consistency' theory and constructing the Apollo series of multimodal models based on this theory, which can efficiently understand long videos.
Apply Hierarchical-Chain-of-Generation to Complex Attributes Text-to-3D Generation
Yiming Qin (Peking University), Yang Liu (Peking University)
GenerationData SynthesisLarge Language ModelGaussian SplattingText
🎯 What it does: An automated 3D Gaussian Splatting generation framework called HCoG is proposed, which can automatically chunk and generate and refine 3D assets according to complex attributes and occlusion relationships in the text in an internal-to-external order.
APT: Adaptive Personalized Training for Diffusion Models with Limited Data
JungWoo Chae (LGCNS AI Research), Sangheum Hwang (Seoul National University of Science and Technology)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: Fine-tuning pre-trained diffusion models with a small amount of data, proposing the APT framework.
AR-Diffusion: Asynchronous Video Generation with Auto-Regressive Diffusion
Mingzhen Sun (Institute of Automation), Jing Liu (Institute of Automation)
GenerationData SynthesisTransformerDiffusion modelVideo
🎯 What it does: A self-regressive diffusion model AR-Diffusion is proposed for asynchronous video generation;
Arbitrary-steps Image Super-resolution via Diffusion Inversion
Zongsheng Yue (Xi'an Jiaotong University), Chen Change Loy (Nanyang Technological University)
RestorationSuper ResolutionDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: A diffusion inversion-based image super-resolution method (InvSR) is proposed, which constructs the intermediate state of the diffusion model by predicting noise mapping, thereby generating high-resolution images with arbitrary sampling steps.
Arc2Avatar: Generating Expressive 3D Avatars from a Single Image via ID Guidance
Dimitrios Gerogiannis (Imperial College London), Stefanos Zafeiriou (Imperial College London)
GenerationData SynthesisDiffusion modelScore-based ModelGaussian SplattingImage
🎯 What it does: Generate high-fidelity, expressive 3D head avatars from a single facial image.
ArcPro: Architectural Programs for Structured 3D Abstraction of Sparse Points
Qirui Huang (Shenzhen University), Hui Huang (Shenzhen University)
GenerationData SynthesisTransformerPoint CloudMesh
🎯 What it does: This paper proposes the ArcPro method, which uses a program-based DSL to map sparse point clouds to structured 3D architectural abstractions.
Are Images Indistinguishable to Humans Also Indistinguishable to Classifiers?
Zebin You (Renmin University of China), Chongxuan Li (Baidu)
ClassificationGenerationConvolutional Neural NetworkTransformerDiffusion modelImage
🎯 What it does: This paper proposes a distribution classification task to evaluate the distribution gap between images generated by diffusion models and real images. The results show that even with a low FID, the classifier can still distinguish between the two with an accuracy of over 98%.
Are Spatial-Temporal Graph Convolution Networks for Human Action Recognition Over-Parameterized?
Jianyang Xie (University of Liverpool), Yalin Zheng (University of Liverpool)
RecognitionGraph Neural NetworkSupervised Fine-TuningVideo
🎯 What it does: Proves that ST-GCN is over-parameterized and proposes a sparse generator, which further improves action recognition accuracy through multi-layer sparse fusion while maintaining performance comparable to dense networks.
Argus: A Compact and Versatile Foundation Model for Vision
Weiming Zhuang (Sony AI), Lingjuan Lyu (Sony AI)
Object DetectionSegmentationPose EstimationAnomaly DetectionTransformerImage
🎯 What it does: A compact multi-task vision foundation model named Argus is proposed, supporting 12 visual tasks, using a two-stage training approach: first, multi-task pre-training on core tasks (only training lightweight adapters), then freezing the backbone and fine-tuning the decoder for new tasks.
Argus: Vision-Centric Reasoning with Grounded Chain-of-Thought
Yunze Man (University of Illinois Urbana-Champaign), Zhiding Yu (NVIDIA)
RecognitionObject DetectionTransformerLarge Language ModelMixture of ExpertsVision Language ModelImageTextMultimodalityChain-of-Thought
🎯 What it does: A multimodal large model named Argus is proposed, which achieves precise localization and reasoning of images by combining goal-oriented visual attention with chain-of-thought (CoT).
ARKit LabelMaker: A New Scale for Indoor 3D Scene Understanding
Guangda Ji (ETH Zurich), Hermann Blum (University of Bonn)
Object DetectionSegmentationTransformerNeural Radiance FieldPoint Cloud
🎯 What it does: This paper presents ARKit LabelMaker, which uses an improved LabelMakerV2 to automatically generate dense semantic labels for the 3D RGB-D data of ARKitScenes, constructing the largest real-world indoor 3D semantic annotation dataset.
ARM: Appearance Reconstruction Model for Relightable 3D Generation
Xiang Feng (University of Utah), Yin Yang (University of Utah)
GenerationData SynthesisMesh
🎯 What it does: Single-view to 3D reconstruction, generating high-quality re-lightable 3D meshes and PBR textures.
Around the World in 80 Timesteps: A Generative Approach to Global Visual Geolocation
Nicolas Dufour (Ecole Polytechnique), Loic Landrieu (Ecole des Ponts)
GenerationRetrievalTransformerDiffusion modelImage
🎯 What it does: This paper proposes a generative global visual localization method based on diffusion and Riemannian flow matching, achieving the estimation of the image capture location by directly denoising coordinates on the Earth's surface.
ART: Anonymous Region Transformer for Variable Multi-Layer Transparent Image Generation
Yifan Pu (Microsoft Research), Baining Guo (Microsoft Research)
GenerationData SynthesisTransformerDiffusion modelAuto EncoderImage
🎯 What it does: This paper proposes the Anonymous Region Transformer (ART), which can directly generate transparent images with variable layers based solely on global text prompts and anonymous region layouts.
ArtFormer: Controllable Generation of Diverse 3D Articulated Objects
Jiayi Su (Xiamen University Malaysia), Botian Xu (Tsinghua University)
GenerationTransformerDiffusion modelTextMultimodality
🎯 What it does: A Transformer-based framework called Articulation Transformer is proposed, introducing a controllable Shape Prior to generate multi-part 3D articulated objects conditioned on text or images.
Articulated Kinematics Distillation from Video Diffusion Models
Xuan Li (University of California Los Angeles), Donglai Xiang (NVIDIA)
GenerationData SynthesisPose EstimationKnowledge DistillationDiffusion modelScore-based ModelVideoMesh
🎯 What it does: A framework called Articulated Kinematics Distillation (AKD) is proposed, which generates high-fidelity character animations by combining the advantages of skeletal-based animation and modern generative models.
ArticulatedGS: Self-supervised Digital Twin Modeling of Articulated Objects using 3D Gaussian Splatting
Junfu Guo (University of Science and Technology of China), Ruizhen Hu (Shenzhen University)
SegmentationPose EstimationGaussian SplattingImage
🎯 What it does: This paper proposes ArticulatedGS, a self-supervised 3D Gaussian Splatting method that constructs digital twin models of movable parts using two sets of multi-view RGB images with different poses.
ArtiFade: Learning to Generate High-quality Subject from Blemished Images
Shuya Yang (University of Hong Kong), Kwan-Yee K. Wong (University of Hong Kong)
RestorationGenerationDiffusion modelImageBenchmark
🎯 What it does: To address the issues of watermarking, advertising noise, and other defects in the main image, the ArtiFade method is proposed to fine-tune the iconic embedding and attention parameters in the diffusion model, achieving defect removal and generating high-quality main images.
ArtiScene: Language-Driven Artistic 3D Scene Generation Through Image Intermediary
Zeqi Gu (NVIDIA), Yifan Ding (NVIDIA)
Object DetectionGenerationPose EstimationDepth EstimationTransformerLarge Language ModelDiffusion modelImageText
🎯 What it does: This paper presents ArtiScene, a language-based 3D scene generation pipeline that requires no training. It uses 2D generated images as an intermediary to automatically extract layouts and styles, generate corresponding 3D assets, and ultimately stitch together an editable complete scene.
ASAP: Advancing Semantic Alignment Promotes Multi-Modal Manipulation Detecting and Grounding
Zhenxing Zhang (Hefei University of Technology), Meng Wang (Hefei Comprehensive National Science Center)
ClassificationRecognitionObject DetectionTransformerLarge Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: The ASAP framework is proposed for detecting and locating multimodal media forgery, utilizing subtitles and explanatory texts generated by multimodal large language models to enhance the semantic alignment between images and text.
ASHiTA: Automatic Scene-grounded HIerarchical Task Analysis
Yun Chang (Massachusetts Institute of Technology), Jiuguang Wang (Robotics and AI Institute)
Object DetectionSegmentationRobotic IntelligenceTransformerLarge Language ModelVision Language ModelSimultaneous Localization and MappingImagePoint Cloud
🎯 What it does: A framework named ASHiTA has been designed and implemented, which can automatically decompose high-level natural language tasks into fine-grained sub-tasks corresponding to 3D scene graphs, and semantically align the sub-tasks with objects and locations in the scene.
ASIGN: An Anatomy-aware Spatial Imputation Graphic Network for 3D Spatial Transcriptomics
Junchao Zhu (Vanderbilt University), Yuankai Huo (Vanderbilt University)
Graph Neural NetworkTransformerBiomedical Data
🎯 What it does: A method for inferring 3D spatial transcriptomic data using 3D WSI and a single 2D ST sample, called ASIGN, is proposed.
Assessing and Learning Alignment of Unimodal Vision and Language Models
Le Zhang (Mila - Quebec AI Institute), Aishwarya Agrawal (Mila - Quebec AI Institute)
ClassificationSegmentationRetrievalTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Research and evaluate the cross-modal alignment capability of unimodal vision and language models, and propose an efficient alignment framework SAIL;
Associative Transformer
Yuwei Sun (Araya Research), Ryota Kanai (Microsoft Research)
ClassificationComputational EfficiencyAdversarial AttackTransformerImage
🎯 What it does: This paper proposes the Associative Transformer (AiT), a visual transformer that achieves sparse attention and associative memory through bottleneck attention, low-rank explicit memory, and continuous Hopfield networks.
Asynchronous Collaborative Graph Representation for Frames and Events
Dianze Li (Peking University), Yonghong Tian (Peking University)
Object DetectionDepth EstimationAdversarial AttackGraph Neural NetworkMultimodality
🎯 What it does: An Asynchronous Collaborative Graph Representation (ACGR) is proposed, which unifies the modeling of frames and events to achieve high-performance, low-latency visual task inference.
ATA: Adaptive Transformation Agent for Text-Guided Subject-Position Variable Background Inpainting
Yizhe Tang (Shanghai Jiao Tong University), Fangyuan Zou (Tencent)
Image TranslationRestorationGenerationTransformerDiffusion modelImageText
🎯 What it does: This paper proposes a text-guided variable subject position background filling task and achieves dynamic adjustment of subject position during background filling through the Adaptive Transformation Agent (A TA).
AToM: Aligning Text-to-Motion Model at Event-Level with GPT-4Vision Reward
Haonan Han, Xiu Li
GenerationTransformerLarge Language ModelReinforcement LearningVideoTextMultimodality
🎯 What it does: Designed and implemented the ATOM framework, utilizing GPT-4Vision for event-level alignment evaluation of generated motion and text prompts, constructed the MotionPrefer dataset, and fine-tuned MotionGPT through LoRA+IPO reinforcement learning, significantly improving the quality of text-motion alignment.
ATP-LLaVA: Adaptive Token Pruning for Large Vision Language Models
Xubing Ye (Tsinghua University), Yansong Tang (Tsinghua University)
OptimizationComputational EfficiencyRepresentation LearningTransformerVision Language ModelImageMultimodality
🎯 What it does: Designed and implemented ATP-LLaVA, which utilizes an Adaptive Token Pruning module to perform hierarchical and instance-adaptive pruning of visual tokens in large-scale vision-language models, significantly reducing inference costs.
ATP: Adaptive Threshold Pruning for Efficient Data Encoding in Quantum Neural Networks
Mohamed Afane (Fordham University), Junaid Farooq (University of Michigan-Dearborn)
ClassificationOptimizationComputational EfficiencyImage
🎯 What it does: The Adaptive Threshold Pruning (ATP) method is proposed, which performs adaptive threshold pruning on input images to reduce the use of qubits and entanglement entropy.
Attend to Not Attended: Structure-then-Detail Token Merging for Post-training DiT Acceleration
Haipeng Fang (Institute of Computing Technology), Tong-Yee Lee (National Cheng Kung University)
CompressionComputational EfficiencyTransformerDiffusion modelImage
🎯 What it does: A post-training acceleration method called SDTM is proposed, which dynamically merges useless visual tokens in the diffusion transformer (DiT) using a structure-detail denoising prior.
Attention Distillation: A Unified Approach to Visual Characteristics Transfer
Yang Zhou (Shenzhen University), Hui Huang (Shenzhen University)
Image TranslationGenerationOptimizationKnowledge DistillationDiffusion modelImage
🎯 What it does: By introducing attention distillation loss into the self-attention mechanism of the diffusion model, the visual features of example images are transferred by directly optimizing the generated images in the latent space.
Attention IoU: Examining Biases in CelebA using Attention Maps
Aaron Serianni (Princeton University), Vikram V. Ramaswamy (Princeton University)
ClassificationSegmentationConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a Bias-IoU metric based on attention maps for quantitatively assessing the internal bias of image classification models across different attributes. Its effectiveness is validated on synthetic datasets Waterbirds and CelebA, further revealing latent confounding variables in CelebA that are not directly captured by labels.
Attraction Diminishing and Distributing for Few-Shot Class-Incremental Learning
Li-Jun Zhao (Shandong University), Xin-Shun Xu (Shandong University)
ClassificationDomain AdaptationContrastive LearningImage
🎯 What it does: This paper addresses the 'hubness' problem in few-shot incremental learning, providing theoretical analysis and improvement strategies.
Attribute-formed Class-specific Concept Space: Endowing Language Bottleneck Model with Better Interpretability and Scalability
Jianyang Zhang (University of Electronic Science and Technology of China), Fengmao Lv (Southwest Jiaotong University)
ClassificationExplainability and InterpretabilityTransformerLarge Language ModelVision Language ModelImage
🎯 What it does: This paper proposes the Attribute-formed Language Bottleneck Model (ALBM), which achieves interpretable image classification by constructing an attribute-based class-specific concept space and combining Visual Attribute Prompt Learning (VAPL) with LLM for automatic concept set generation (DSS).
Attribute-Missing Multi-view Graph Clustering
Bowen Zhao (Xidian University), Quanxue Gao (Tulane University)
Graph Neural NetworkGraph
🎯 What it does: A framework for Attribute Missing Multi-View Graph Clustering (AMMGC) is proposed, which first iteratively fills in missing node attributes using neighborhood information, then aligns the multi-view graph structure through a dual-structure consistency module, and finally enhances clustering reliability using a high-confidence guidance module.
AudCast: Audio-Driven Human Video Generation by Cascaded Diffusion Transformers
Jiazhi Guan (Tsinghua University), Ziwei Liu (Nanyang Technological University)
GenerationData SynthesisTransformerDiffusion modelVideoMultimodalityMeshAudio
🎯 What it does: Designed the AudCast framework to generate full-body videos driven by a single-frame reference image and arbitrary audio;