CVPR 2024 Papers — Page 13
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2716 papers
Instance-aware Contrastive Learning for Occluded Human Mesh Reconstruction
Mi-Gyeong Gwon (Konkuk University), Wonjun Kim (Konkuk University)
Pose EstimationConvolutional Neural NetworkContrastive LearningImageMesh
🎯 What it does: A 3D human mesh reconstruction method based on instance-aware contrastive learning is proposed for scenes with multiple occlusions in a single image.
Instance-aware Exploration-Verification-Exploitation for Instance ImageGoal Navigation
Xiaohan Lei (University of Science and Technology of China), Houqiang Li (University of Science and Technology of China)
Object DetectionSegmentationRobotic IntelligenceReinforcement LearningSimultaneous Localization and MappingImage
🎯 What it does: This paper proposes an Instance-aware Exploration-Verification-Exploitation (IEVE) framework for navigation based on target instance images.
Instance-Aware Group Quantization for Vision Transformers
Jaehyeon Moon (Yonsei University), Bumsub Ham (Yonsei University)
ClassificationObject DetectionSegmentationTransformerImage
🎯 What it does: A method for instance-aware group quantization, IGQ-ViT, is proposed, which can dynamically partition the channels and softmax attention rows of ViT during the post-training phase and assign quantization parameters for each group, significantly reducing quantization error.
Instance-based Max-margin for Practical Few-shot Recognition
Minghao Fu (Nanjing University), Ke Zhu (Nanjing University)
ClassificationRecognitionContrastive LearningImage
🎯 What it does: A practice few-shot learning (pFSL) framework based on unsupervised pre-trained models is proposed, and within this framework, an instance-level maximum margin method (IbM2) is designed to achieve maximum margin by generating virtual samples around each training instance.
Instance-level Expert Knowledge and Aggregate Discriminative Attention for Radiology Report Generation
Shenshen Bu (Sun Yat-sen University), Zhiming Dai (Sun Yat-sen University)
GenerationKnowledge DistillationTransformerImageTextMultimodalityElectronic Health Records
🎯 What it does: This paper proposes a radiology report generation framework called EKAGen, which is based on instance-level expert knowledge and aggregated discriminative attention. It utilizes a unified embedding network to extract expert report knowledge, diagnose images, and retrieve corresponding instance-level knowledge, while focusing on key lesion areas through weakly supervised aggregated discriminative attention maps (ADM) and enhancing the model's global semantic consistency using a Global Information Self-Distillation (GID) strategy.
InstanceDiffusion: Instance-level Control for Image Generation
Xudong Wang (Meta), Ishan Misra (Meta)
GenerationData SynthesisDiffusion modelImageText
🎯 What it does: This study investigates an instance-level control text-to-image diffusion model that supports various positional formats and instance descriptions, generating images that conform to specified instance attributes and layouts.
Instantaneous Perception of Moving Objects in 3D
Di Liu (Rutgers University), Manmohan Chandraker (University of California San Diego)
Object DetectionObject TrackingAutonomous DrivingConvolutional Neural NetworkOptical FlowPoint Cloud
🎯 What it does: This study implements a detection and flow estimation method for instant perception of subtle vehicle movements (such as starting, reversing, and other slight actions) in autonomous driving scenarios using LiDAR point clouds.
InstantBooth: Personalized Text-to-Image Generation without Test-Time Finetuning
Jing Shi (Adobe Inc), Hyun Joon Jung (Adobe Inc)
GenerationDiffusion modelImageText
🎯 What it does: This paper proposes InstantBooth, which utilizes a pre-trained Stable Diffusion model to achieve personalized text-to-image generation without fine-tuning during testing. It supports multiple input images and can generate images with various poses, diverse backgrounds, language alignment, and identity preservation.
Instruct 4D-to-4D: Editing 4D Scenes as Pseudo-3D Scenes Using 2D Diffusion
Linzhan Mou (Zhejiang University), Yu-Xiong Wang (University of Illinois Urbana-Champaign)
GenerationData SynthesisDiffusion modelNeural Radiance FieldOptical FlowVideo
🎯 What it does: We propose Instruct 4D-to-4D, which can edit 4D dynamic scenes (represented by NeRF) based on natural language instructions while maintaining spatial and temporal consistency.
Instruct-Imagen: Image Generation with Multi-modal Instruction
Hexiang Hu (Google DeepMind), Xuhui Jia (Google Research)
GenerationDiffusion modelImageTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: The Instruct-Imagen model is proposed, which unifies the description of various image generation tasks through multimodal instructions and achieves zero-shot generalization.
Instruct-ReID: A Multi-purpose Person Re-identification Task with Instructions
Weizhen He (Zhejiang University), Yunfeng Yan (Zhejiang University)
RecognitionRetrievalTransformerContrastive LearningImageTextMultimodality
🎯 What it does: A unified instruction-driven person re-identification (Instruct-ReID) task is proposed, where the model accepts images and multimodal instructions as input, achieving a unified deployment for various re-identification scenarios including traditional, outfit change, template outfit change, infrared-visible, text retrieval, and language instructions.
InstructDiffusion: A Generalist Modeling Interface for Vision Tasks
Zigang Geng (University of Science and Technology of China), Baining Guo (Microsoft Research Asia)
Image TranslationSegmentationGenerationPose EstimationDiffusion modelImage
🎯 What it does: A unified instruction-driven diffusion framework has been constructed, capable of completing various visual tasks such as segmentation, keypoint detection, and image editing based on natural language instructions.
InstructVideo: Instructing Video Diffusion Models with Human Feedback
Hangjie Yuan (Zhejiang University), Dong Ni (Zhejiang University)
GenerationData SynthesisReinforcement Learning from Human FeedbackReinforcement LearningDiffusion modelVideoText
🎯 What it does: Using human feedback for reward fine-tuning of text-to-video diffusion models to make the generated videos more aligned with human preferences and improve visual quality.
Integrating Efficient Optimal Transport and Functional Maps For Unsupervised Shape Correspondence Learning
Tung Le (University of California), Xiaohui Xie (University of California)
OptimizationDiffusion modelMesh
🎯 What it does: An unsupervised shape correspondence learning framework is proposed by combining functional mapping with efficient piecewise Wasserstein distance, further enhanced by an adaptive refinement module to improve point-to-point mapping accuracy.
Intelligent Grimm - Open-ended Visual Storytelling via Latent Diffusion Models
Chang Liu (Shanghai Jiao Tong University), Weidi Xie (Shanghai Jiao Tong University)
GenerationData SynthesisDiffusion modelImageVideoText
🎯 What it does: A framework for automatic regression visual story generation called StoryGen is proposed, which can generate coherent image sequences based on textual stories.
Intensity-Robust Autofocus for Spike Camera
Changqing Su (Peking University), Tiejun Huang (Peking University)
Object DetectionOptimizationSpiking Neural NetworkImageVideo
🎯 What it does: An adaptive focusing framework based on Spike Dispersion (SD) is proposed, along with a real-time focus measurement and adaptive pyramid search algorithm SGFS for peak cameras, which does not require image reconstruction.
Inter-X: Towards Versatile Human-Human Interaction Analysis
Liang Xu (Shanghai Jiao Tong University), Xiaokang Yang (Shanghai Jiao Tong University)
RecognitionGenerationGraph Neural NetworkTransformerDiffusion modelVideoTextMultimodalityBenchmark
🎯 What it does: This paper constructs a large-scale human-human interaction MoCap dataset called Inter-X and proposes four types of downstream tasks (text-conditioned interaction generation, action-conditioned interaction generation, human response generation, and interaction recognition), providing a new benchmark for interaction analysis.
InteractDiffusion: Interaction Control in Text-to-Image Diffusion Models
Jiun Tian Hoe (Nanyang Technological University), Weipeng Hu (Universiti Malaya)
Object DetectionGenerationTransformerDiffusion modelImage
🎯 What it does: A pluggable interactive control module, InteractDiffusion, is proposed, which can add human-object interaction (HOI) information to existing text-to-image diffusion models (such as Stable Diffusion), enabling precise control over the interaction behaviors of objects in images.
Interactive Continual Learning: Fast and Slow Thinking
Biqing Qi (Harbin Institute of Technology), Bowen Zhou (Tsinghua University)
ClassificationRecognitionTransformerLarge Language ModelVision Language ModelImageMultimodality
🎯 What it does: An interactive continual learning framework ICL is proposed, using ViT as a fast system 1 and a cross-modal large language model as a slow system 2, achieving collaborative reasoning between the two.
Interactive3D: Create What You Want by Interactive 3D Generation
Shaocong Dong (Hong Kong University of Science and Technology), Dan Xu (Hong Kong University of Science and Technology)
GenerationData SynthesisDiffusion modelScore-based ModelGaussian SplattingTextPoint Cloud
🎯 What it does: A two-stage interactive 3D generation framework is proposed, utilizing efficient Gaussian scattering for direct user editing, and enhancing geometric quality through InstantNGP and hash refinement.
InterHandGen: Two-Hand Interaction Generation via Cascaded Reverse Diffusion
Jihyun Lee (KAIST), Tae-Kyun Kim (Imperial College London)
GenerationPose EstimationTransformerDiffusion modelPoint Cloud
🎯 What it does: Proposes the InterHandGen framework, which uses a hierarchical diffusion model to generate interactive hand poses (with or without objects) and can be integrated as a prior into other tasks;
InternVL: Scaling up Vision Foundation Models and Aligning for Generic Visual-Linguistic Tasks
Zhe Chen (OpenGVLab), Jifeng Dai (University of Science and Technology of China)
ClassificationRecognitionSegmentationGenerationRetrievalTransformerLarge Language ModelVision Language ModelContrastive LearningImageVideoTextMultimodality
🎯 What it does: We propose InternVL, a general vision-language foundation model that extends the visual encoder to 6 billion parameters and aligns with large language models, applicable to multimodal tasks such as image and video recognition, retrieval, captioning, and dialogue.
Interpretable Measures of Conceptual Similarity by Complexity-Constrained Descriptive Auto-Encoding
Alessandro Achille (AWS AI Labs), Stefano Soatto (AWS AI Labs)
RetrievalExplainability and InterpretabilityTransformerLarge Language ModelDiffusion modelAuto EncoderImageTextMultimodality
🎯 What it does: A concept similarity measurement based on the length of natural language descriptions is proposed, utilizing complexity-constrained description auto-encoding (CC:DAE) to measure the similarity of images, texts, and cross-modal data.
Intraoperative 2D/3D Image Registration via Differentiable X-ray Rendering
Vivek Gopalakrishnan, Polina Golland
Pose EstimationConvolutional Neural NetworkImageBiomedical DataComputed Tomography
🎯 What it does: A self-supervised differentiable 2D/3D X-ray and CT alignment framework, DiffPose, has been developed, achieving sub-millimeter registration without manual annotation.
Intriguing Properties of Diffusion Models: An Empirical Study of the Natural Attack Capability in Text-to-Image Generative Models
Takami Sato (University of California), Qi Alfred Chen (University of California)
Object DetectionGenerationAutonomous DrivingAdversarial AttackTransformerDiffusion modelImage
🎯 What it does: This study investigates the 'natural attack capability' of diffusion models in generating images, proposing the NDD attack (Natural Denoising Diffusion Attack) and constructing the NDDA dataset for systematic evaluation.
Intrinsic Image Diffusion for Indoor Single-view Material Estimation
Peter Kocsis (Technical University of Munich), Matthias Nießner (Technical University of Munich)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: A conditional diffusion model is proposed to generate multiple possible material interpretations (color, roughness, metallicity) from a single indoor image.
IntrinsicAvatar: Physically Based Inverse Rendering of Dynamic Humans from Monocular Videos via Explicit Ray Tracing
Shaofei Wang (ETH Zurich), Siyu Tang (ETH Zurich)
GenerationPose EstimationNeural Radiance FieldVideo
🎯 What it does: Using monocular video to achieve the inverse rendering of the physical properties of clothing on the human body, resulting in an animatable and relightable avatar model.
Inverse Rendering of Glossy Objects via the Neural Plenoptic Function and Radiance Fields
Haoyuan Wang (City University of Hong Kong), Rynson W.H. Lau (City University of Hong Kong)
Neural Radiance FieldImageMesh
🎯 What it does: A dual-stage inverse rendering framework is proposed: the first stage uses NeRF + NeuS + Mip-NeRF to reconstruct smooth geometry and ambient light; the second stage constructs a 5D Neural Plenoptic Function (NeP) and combines material-aware cone sampling for high-fidelity material estimation of the target object.
Inversion-Free Image Editing with Language-Guided Diffusion Models
Sihan Xu (University of Michigan), Joyce Chai (University of Michigan)
Image TranslationGenerationDiffusion modelImageBenchmark
🎯 What it does: We propose a text-guided image editing method called InfEdit that does not require traditional inversion processes and can complete high-quality image modifications in less than 3 seconds.
Investigating and Mitigating the Side Effects of Noisy Views for Self-Supervised Clustering Algorithms in Practical Multi-View Scenarios
Jie Xu (University of Electronic Science and Technology of China), Xiaofeng Zhu (University of Electronic Science and Technology of China)
Representation LearningAuto EncoderMultimodality
🎯 What it does: A deep self-supervised clustering framework MVCAN is proposed to address the noise view defect (NVD) in multi-view clustering, aiming to achieve efficient and robust clustering in practical multi-view scenarios.
Investigating Compositional Challenges in Vision-Language Models for Visual Grounding
Yunan Zeng (Center for Research on Intelligent Perception and Computing), Liang Wang (Center for Research on Intelligent Perception and Computing)
Object DetectionExplainability and InterpretabilityTransformerSupervised Fine-TuningVision Language ModelImageMultimodalityBenchmark
🎯 What it does: This paper studies the combinatorial reasoning ability of visual language models (VLM) in visual localization tasks and proposes the ARPGrounding three-dimensional combinatorial localization benchmark.
IPoD: Implicit Field Learning with Point Diffusion for Generalizable 3D Object Reconstruction from Single RGB-D Images
Yushuang Wu (Chinese University of Hong Kong Shenzhen), Xiaoguang Han (Chinese University of Hong Kong Shenzhen)
Object DetectionSegmentationGenerationTransformerDiffusion modelImagePoint Cloud
🎯 What it does: A framework called IPoD is proposed, which combines implicit field learning with point diffusion to achieve generalizable 3D object reconstruction from a single RGB-D image.
IQ-VFI: Implicit Quadratic Motion Estimation for Video Frame Interpolation
Mengshun Hu (Wuhan University), Yinqiang Zheng
Image TranslationRestorationKnowledge DistillationConvolutional Neural NetworkOptical FlowVideo
🎯 What it does: An implicit quadratic motion estimation framework IQ-VFI is proposed, which achieves high-precision motion modeling and video frame interpolation between two frames through an implicit acceleration estimation network and knowledge distillation.
IReNe: Instant Recoloring of Neural Radiance Fields
Alessio Mazzucchelli (Arquimea Research Center), Adrian Penate-Sanchez (Universidad de las Palmas de Gran Canaria)
Image TranslationGenerationNeural Radiance FieldImage
🎯 What it does: A method is proposed that allows for instant color editing of a pre-trained NeRF using only a single user-edited image, generating new scenes for panoramic visualization.
Is Ego Status All You Need for Open-Loop End-to-End Autonomous Driving?
Zhiqi Li (Nanjing University), Jose M. Alvarez (NVIDIA)
Autonomous DrivingPoint Cloud
🎯 What it does: A thorough analysis of open-loop end-to-end autonomous driving research is conducted, pointing out the dominance of ego state, dataset imbalance, and insufficient evaluation metrics, and proposing a new metric CCR as well as a lightweight baseline BEV-Planner that only uses BEV features and ego state.
Is Vanilla MLP in Neural Radiance Field Enough for Few-shot View Synthesis?
Hanxin Zhu (University of Science and Technology of China), Zhibo Chen (University of Science and Technology of China)
GenerationData SynthesisNeural Radiance FieldImage
🎯 What it does: This paper addresses the overfitting problem that often occurs in few-shot NeRF view synthesis by proposing a new network structure and auxiliary regularization method.
IS-Fusion: Instance-Scene Collaborative Fusion for Multimodal 3D Object Detection
Junbo Yin (Beijing Institute of Technology), Wenguan Wang (Ecole Polytechnique Federale de Lausanne)
Object DetectionAutonomous DrivingTransformerMultimodalityPoint Cloud
🎯 What it does: A multi-modal 3D object detection framework named IS-FUSION is proposed, which enhances BEV features through instance-level and scene-level collaborative fusion.
It's All About Your Sketch: Democratising Sketch Control in Diffusion Models
Subhadeep Koley (University of Surrey), Yi-Zhe Song (University of Surrey)
GenerationRetrievalDiffusion modelImage
🎯 What it does: This paper proposes a diffusion model control framework aimed at beginner hand-drawn sketches, allowing sketches to directly generate high-fidelity images that align with the intended concept.
Iterated Learning Improves Compositionality in Large Vision-Language Models
Chenhao Zheng (University of Michigan), Ranjay Krishna (University of Washington)
Knowledge DistillationRepresentation LearningTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: A training framework for visual-language models based on iterative learning has been developed to enhance the model's combinatorial reasoning ability.
iToF-flow-based High Frame Rate Depth Imaging
Yu Meng (Nanjing University), Tao Yue (Nanjing University)
Depth EstimationConvolutional Neural NetworkFlow-based ModelOptical FlowImage
🎯 What it does: This paper addresses the errors caused by indirect time-of-flight (iToF) multi-modal measurements in dynamic scenes, proposing the iToF-flow model. Based on this, it designs two main components: cross-mode flow and uni-mode photometric residual flow, constructing an end-to-end high frame rate depth reconstruction network that includes LCTM (Local Linear Transformation Module) and UPCM (Uni-mode Photometric Correction Module).
Jack of All Tasks Master of Many: Designing General-Purpose Coarse-to-Fine Vision-Language Model
Shraman Pramanick (Johns Hopkins University), Amjad Almahairi (Meta)
ClassificationSegmentationGenerationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: We propose VistaLLM, a unified vision-language model capable of simultaneously handling both coarse-grained (e.g., classification, question answering, caption generation) and fine-grained (e.g., segmentation, localization) tasks under single and multiple image inputs.
JDEC: JPEG Decoding via Enhanced Continuous Cosine Coefficients
Woo Kyoung Han (Korea University), Kyong Hwan Jin (Korea University)
RestorationCompressionTransformerImage
🎯 What it does: This paper presents JDEC, a JPEG decoder that utilizes local implicit neural representations and continuous cosine coefficients, capable of directly recovering high-quality images from quantized DCT spectra without the need for traditional decoders, and can process JPEG files with different quality factors in one go.
JeDi: Joint-Image Diffusion Models for Finetuning-Free Personalized Text-to-Image Generation
Yu Zeng (Johns Hopkins University), Yogesh Balaji (NVIDIA Research)
GenerationData SynthesisLarge Language ModelDiffusion modelImage
🎯 What it does: The JeDi model is proposed, achieving personalized text-to-image generation without fine-tuning, utilizing joint image diffusion and directly using reference images for personalized generation during sampling.
JoAPR: Cleaning the Lens of Prompt Learning for Vision-Language Models
Yuncheng Guo (Fudan University), Xiaodong Gu (Fudan University)
ClassificationRecognitionTransformerPrompt EngineeringVision Language ModelImageMultimodality
🎯 What it does: This paper proposes the JoAPR framework to address the issue of insufficient robustness in visual language pre-training models (VL-PTM) caused by label noise during the prompt learning process.
Joint Reconstruction of 3D Human and Object via Contact-Based Refinement Transformer
Hyeongjin Nam (Seoul National University), Kyoung Mu Lee (Seoul National University)
Object DetectionPose EstimationTransformerImage
🎯 What it does: The paper proposes a model called CONTHO that simultaneously reconstructs 3D human and object models from a single image, utilizing contact information between the human and the object for 3D refinement.
Joint-Task Regularization for Partially Labeled Multi-Task Learning
Kento Nishi (Harvard University), Hanspeter Pfister (Harvard University)
SegmentationDepth EstimationAuto EncoderImage
🎯 What it does: This paper proposes the Joint-Task Regularization (JTR) method, which aims to achieve cross-task joint regularization in partially labeled multi-task learning scenarios by imposing distance constraints on the predictions and labels of all tasks in a joint task latent space.
Joint2Human: High-Quality 3D Human Generation via Compact Spherical Embedding of 3D Joints
Muxin Zhang (Tianjin University), Kun Li (Tianjin University)
GenerationData SynthesisPose EstimationDiffusion modelAuto EncoderMesh
🎯 What it does: This paper presents Joint2Human, a method for directly generating high-quality 3D human meshes using a 2D Diffusion model through Fourier Occupancy Field (FOF).
Jointly Training and Pruning CNNs via Learnable Agent Guidance and Alignment
Alireza Ganjdanesh (University of Maryland), Heng Huang (University of Maryland)
OptimizationConvolutional Neural NetworkReinforcement LearningAgentic AIImage
🎯 What it does: A framework for joint training and structured pruning is proposed, using a reinforcement learning agent to decide the pruning ratio of CNN layers in real-time, and aligning model weights with the pruning structure through soft regularization.
JointSQ: Joint Sparsification-Quantization for Distributed Learning
Weiying Xie (Xidian University), Leyuan Fang (Hunan University)
OptimizationFederated LearningComputational EfficiencyConvolutional Neural NetworkRecurrent Neural NetworkTransformerImageText
🎯 What it does: A joint sparsification-quantization framework called JointSQ is proposed, which treats sparsification as 0-bit quantization for gradient compression in distributed learning.
JRDB-PanoTrack: An Open-world Panoptic Segmentation and Tracking Robotic Dataset in Crowded Human Environments
Duy Tho Le (Monash University), Hamid Rezatofighi (Monash University)
Object TrackingSegmentationRobotic IntelligenceTransformerImageMultimodalityPoint CloudBenchmark
🎯 What it does: Created and released JRDB-PanoTrack, an open-world panoramic instance-segmentation and tracking dataset for robotic vision, along with benchmark evaluations.
JRDB-Social: A Multifaceted Robotic Dataset for Understanding of Context and Dynamics of Human Interactions Within Social Groups
Simindokht Jahangard (Monash University), Hamid Rezatofighi (Monash University)
Robotic IntelligenceTransformerLarge Language ModelVideoMultimodalityBenchmarkAudio
🎯 What it does: This paper constructs the JRDB-Social dataset, which provides gender, age, and race labels for each individual based on JRDB, records multi-label interactions between every pair of members within groups at the frame level, and gives spatial locations, scene associations, and purpose descriptions for the groups.
Just Add ?! Pose Induced Video Transformers for Understanding Activities of Daily Living
Dominick Reilly (University of North Carolina Charlotte), Srijan Das (University of North Carolina Charlotte)
ClassificationRecognitionPose EstimationTransformerVideo
🎯 What it does: A Pose-Induced Video Transformer (π-ViT) is proposed, which injects 2D/3D keypoint information into the RGB representation of the video transformer by inserting two auxiliary modules, 2D-SIM and 3D-SIM, during the training phase, and does not require pose information during inference.
Kandinsky Conformal Prediction: Efficient Calibration of Image Segmentation Algorithms
Joren Brunekreef (Netherlands Cancer Institute), Jonas Teuwen (Netherlands Cancer Institute)
SegmentationConvolutional Neural NetworkImageBiomedical DataComputed Tomography
🎯 What it does: An adaptive calibration method of the Kandinsky type used in image segmentation is proposed, which clusters the non-conformity scores of the segmenter based on spatial similarity between pixels, thereby achieving efficient calibration of pixel-level classifiers.
KD-DETR: Knowledge Distillation for Detection Transformer with Consistent Distillation Points Sampling
Yu Wang (Baidu), Errui Ding (Baidu)
Object DetectionKnowledge DistillationTransformerImage
🎯 What it does: This paper proposes KD-DETR, a knowledge distillation framework for DETR, which achieves consistent distillation points by introducing shared dedicated query points.
Kernel Adaptive Convolution for Scene Text Detection via Distance Map Prediction
Jinzhi Zheng (Institute of Software Chinese Academy of Sciences), Libo Zhang (Institute of Software Chinese Academy of Sciences)
Object DetectionSegmentationConvolutional Neural NetworkImage
🎯 What it does: Proposed the Kernel Adaptive Convolution (KAC) network, which utilizes adaptive convolution kernels and distance maps to achieve text detection in scenes of arbitrary shapes;
KeyPoint Relative Position Encoding for Face Recognition
Minchul Kim (Michigan State University), Xiaoming Liu (Michigan State University)
RecognitionTransformerImage
🎯 What it does: Proposes Keypoint Relative Position Encoding (KP-RPE), which adjusts the relative position encoding of ViT self-attention through keypoints, thereby enhancing robustness to unseen affine transformations.
KITRO: Refining Human Mesh by 2D Clues and Kinematic-tree Rotation
Fengyuan Yang (National University of Singapore), Angela Yao (National University of Singapore)
Pose EstimationMesh
🎯 What it does: Proposes the KITRO method, which refines 3D human meshes from 2D keypoints by solving skeletal orientations in a closed form and utilizing a motion chain-based decision tree;
Know Your Neighbors: Improving Single-View Reconstruction via Spatial Vision-Language Reasoning
Rui Li (ETH Zurich), Federico Tombari (Technical University of Munich)
Depth EstimationAutonomous DrivingTransformerVision Language ModelNeural Radiance FieldImageMultimodality
🎯 What it does: The KYN method is proposed, which infers the density of each 3D point using a single view image through semantic and spatial context, achieving complete 3D scene reconstruction.
Knowledge-Enhanced Dual-stream Zero-shot Composed Image Retrieval
Yucheng Suo (Zhejiang University), Yi Yang (Zhejiang University)
RetrievalContrastive LearningImageTextMultimodality
🎯 What it does: A knowledge-enhanced dual-stream framework (KEDs) is proposed for zero-shot compositional image retrieval (ZSCIR) tasks without the need for triplet data.
Koala: Key Frame-Conditioned Long Video-LLM
Reuben Tan (Boston University), Kate Saenko (Boston University)
TransformerLarge Language ModelPrompt EngineeringVideoTextBenchmark
🎯 What it does: By introducing key frames, conditional segments (CS), and conditional video (CV) tokenizers into a pre-trained short video LLM, understanding and question answering for minute-long videos have been achieved;
KP-RED: Exploiting Semantic Keypoints for Joint 3D Shape Retrieval and Deformation
Ruida Zhang (Tsinghua University), Xiangyang Ji (Tsinghua University)
RetrievalPoint Cloud
🎯 What it does: A unified semantic keypoint-based 3D shape retrieval and deformation framework, KP-RED, is proposed, which can retrieve the most similar CAD models from global or partial scanned point clouds and guide fine-grained deformation through keypoints to achieve high-quality reconstruction results.
KPConvX: Modernizing Kernel Point Convolution with Kernel Attention
Hugues Thomas (Apple), Jian Zhang (Apple)
Object DetectionSegmentationConvolutional Neural NetworkPoint Cloud
🎯 What it does: Two new types of point cloud convolution operations, KPConvD and KPConvX, have been introduced to improve the efficiency and performance of KPConv.
KTPFormer: Kinematics and Trajectory Prior Knowledge-Enhanced Transformer for 3D Human Pose Estimation
Jihua Peng (Hong Kong Polytechnic University), P. Y. Mok (Hong Kong Polytechnic University)
Pose EstimationTransformerVideo
🎯 What it does: This paper proposes a Transformer network named KTPFormer, which enhances 3D pose estimation by incorporating two types of prior attention modules—Kinematics Prior Attention (KPA) and Trajectory Prior Attention (TPA)—utilizing prior information about human structure and motion trajectories.
KVQ: Kwai Video Quality Assessment for Short-form Videos
Yiting Lu (University of Science and Technology of China), Zhibo Chen (Kuaishou Technology)
ClassificationRecognitionCompressionTransformerContrastive LearningVideo
🎯 What it does: The first large-scale short video quality assessment database KVQ has been constructed, and a quality assessment model specifically for short videos, KSVQE, has been proposed.
L-MAGIC: Language Model Assisted Generation of Images with Coherence
Zhipeng Cai (Intel Labs), Michael Paulitsch (Intel Labs)
GenerationData SynthesisTransformerLarge Language ModelDiffusion modelImageVideoTextMultimodalityPoint Cloud
🎯 What it does: A diffusion framework L-MAGIC assisted by large language models has been developed, which can automatically generate coherent 360° panoramic images from single-view images or multimodal inputs, and supports subsequent applications such as generating viewpoint videos and 3D point clouds.
L0-Sampler: An L0 Model Guided Volume Sampling for NeRF
Liangchen Li (University of Science and Technology of China), Juyong Zhang (University of Science and Technology of China)
GenerationData SynthesisOptimizationNeural Radiance FieldPoint Cloud
🎯 What it does: Proposes L0-Sampler, which improves the hierarchical volume sampling strategy of NeRF, making the sampling more concentrated near the surface.
L2B: Learning to Bootstrap Robust Models for Combating Label Noise
Yuyin Zhou (University of California), Lei Xing (Stanford University)
ClassificationSegmentationMeta LearningConvolutional Neural NetworkSupervised Fine-TuningImageMagnetic Resonance Imaging
🎯 What it does: This paper proposes a meta-learning-based 'self-guided' method (L2B) that dynamically adjusts the training loss by simultaneously learning the weights α and β of each sample's true labels and pseudo-labels, implicitly re-labeling and suppressing the impact of noisy labels.
L4D-Track: Language-to-4D Modeling Towards 6-DoF Tracking and Shape Reconstruction in 3D Point Cloud Stream
Jingtao Sun (Hunan University), Mike Zheng Shou (National University of Singapore)
Object TrackingSegmentationPose EstimationTransformerLarge Language ModelContrastive LearningVideoPoint Cloud
🎯 What it does: A language-driven 4D model L4D-Track is proposed, capable of achieving zero-shot 6-DoF trajectory tracking and 3D shape reconstruction in 3D point cloud streams.
LAA-Net: Localized Artifact Attention Network for Quality-Agnostic and Generalizable Deepfake Detection
Dat Nguyen, Djamila Aouada
ClassificationAnomaly DetectionConvolutional Neural NetworkSupervised Fine-TuningImageVideo
🎯 What it does: A deep fake detection framework LAA-Net is proposed, utilizing an explicit attention mechanism (heatmap and self-consistency branch) and an enhanced feature pyramid network (E-FPN) through multi-task learning;
Label Propagation for Zero-shot Classification with Vision-Language Models
Vladan Stojni?, Giorgos Tolias (NAVER LABS Europe)
ClassificationRecognitionVision Language ModelImageMultimodality
🎯 What it does: This paper proposes a new method for zero-shot classification using unlabeled data—ZLaP, which enhances the zero-shot recognition performance of VLM by performing label propagation on a bimodal graph composed of text class vectors and unlabeled image vectors.
Label-Efficient Group Robustness via Out-of-Distribution Concept Curation
Yiwei Yang (University of Washington), Bill Howe (University of Washington)
ClassificationOptimizationDiffusion modelImage
🎯 What it does: Proposes the Concept Correction framework, which infers group labels using a set of concepts outside of discrete tasks and implements group robust training for unlabeled samples in conjunction with Group DRO.
LAENeRF: Local Appearance Editing for Neural Radiance Fields
Lukas Radl (Graz University of Technology), Markus Steinberger (Huawei Technologies)
Image TranslationGenerationNeural Radiance FieldImage
🎯 What it does: This paper proposes LAENeRF, a unified framework for local appearance editing (color re-mapping and style transfer) of NeRF scenes.
LAFS: Landmark-based Facial Self-supervised Learning for Face Recognition
Zhonglin Sun (Queen Mary University of London), Georgios Tzimiropoulos (Queen Mary University of London)
RecognitionRepresentation LearningTransformerContrastive LearningImage
🎯 What it does: A self-supervised learning framework LAFS based on facial key points is proposed, which pre-trains transferable facial representations using unlabeled facial images.
LAKE-RED: Camouflaged Images Generation by Latent Background Knowledge Retrieval-Augmented Diffusion
Pancheng Zhao (Nankai University), Jufeng Yang (Nankai University)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: This study proposes a latent background knowledge retrieval enhanced diffusion model (LAKE-RED) that automatically generates camouflage images without background input, capable of naturally integrating specified foreground objects into the generated background.
LAMP: Learn A Motion Pattern for Few-Shot Video Generation
Ruiqi Wu (Nankai University), Xiangyu Zhang (MEGVII Technology)
GenerationData SynthesisDiffusion modelVideoText
🎯 What it does: A text-to-video generation framework called LAMP is proposed, which trains on a small number of videos. It first generates the initial frame using a high-quality text-to-image model, and then uses a modified diffusion model to predict subsequent frames, achieving the learning of motion patterns.
LaMPilot: An Open Benchmark Dataset for Autonomous Driving with Language Model Programs
Yunsheng Ma (Purdue University), Ziran Wang (Purdue University)
Autonomous DrivingTransformerLarge Language ModelPrompt EngineeringTextBenchmarkChain-of-Thought
🎯 What it does: This paper proposes the LaMPilot framework, which utilizes large language models to generate executable driving strategy code and creates the LaMPilot-Bench benchmark dataset to evaluate its instruction-following capability.
LAN: Learning to Adapt Noise for Image Denoising
Changjin Kim (Hanyang University), Sungyong Baik (Hanyang University)
RestorationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper proposes an algorithm called Learning-to-Adapt-Noise (LAN), which learns adjustable offsets on input noise to transfer the image noise distribution of unseen noise to the noise distribution familiar to the pre-trained denoising network, thereby achieving adaptive denoising for unknown noise.
Lane2Seq: Towards Unified Lane Detection via Sequence Generation
Kunyang Zhou (Southeast University)
Object DetectionAutonomous DrivingTransformerReinforcement LearningImage
🎯 What it does: This paper proposes Lane2Seq, which transforms lane detection into a sequence generation task using a Transformer encoder-decoder architecture, eliminating traditional dedicated detection heads and multi-task losses.
LaneCPP: Continuous 3D Lane Detection using Physical Priors
Maximilian Pittner (Bosch Mobility Solutions), Alexandru P. Condurache (Bosch Mobility Solutions)
Autonomous DrivingConvolutional Neural NetworkImage
🎯 What it does: This paper proposes LaneCPP, a continuous 3D lane detection framework based on monocular images.
LangSplat: 3D Language Gaussian Splatting
Minghan Qin (Tsinghua University), Hanspeter Pfister (Harvard University)
SegmentationRetrievalOptimizationComputational EfficiencyAuto EncoderGaussian SplattingPoint Cloud
🎯 What it does: A language field (LangSplat) based on 3D Gaussian distribution has been constructed to achieve efficient and accurate 3D open-language queries.
Language Embedded 3D Gaussians for Open-Vocabulary Scene Understanding
Jin-Chuan Shi (Beihang University), Shao-Hua Guan (Beihang University)
SegmentationGenerationData SynthesisContrastive LearningGaussian SplattingPoint Cloud
🎯 What it does: A new scene representation method is proposed that embeds language into 3D Gaussian point clouds for open vocabulary queries and high-quality new view synthesis.
Language Model Guided Interpretable Video Action Reasoning
Ning Wang (Xidian University), Mohammed Bennamoun (University of Western Australia)
RecognitionExplainability and InterpretabilityTransformerLarge Language ModelVideo
🎯 What it does: This paper proposes a video action reasoning framework called LaIAR, which utilizes language models to guide the process, achieving interpretable action recognition by aligning the logical reasoning knowledge of the language model with the video model.
Language Models as Black-Box Optimizers for Vision-Language Models
Shihong Liu (Carnegie Mellon University), Deva Ramanan (Carnegie Mellon University)
ClassificationGenerationOptimizationTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: This paper proposes using dialogue-based large language models (ChatGPT/GPT-4-V) as black-box optimizers to iteratively improve the text prompts of visual-language models (VLM) through interactive feedback, completing classification tasks from zero to one and optimizing text-to-image generation.
Language-aware Visual Semantic Distillation for Video Question Answering
Bo Zou (Tsinghua University), Youjian Zhao (Tsinghua University)
Knowledge DistillationRepresentation LearningTransformerContrastive LearningVideoText
🎯 What it does: This paper proposes the VideoDistill framework, which utilizes a Language-Aware Gate (LA-Gate) to achieve goal-driven visual semantic distillation, including a differentiable sparse sampling module and a visual refinement module, specifically designed for VideoQA.
Language-conditioned Detection Transformer
Jang Hyun Cho (University of Texas Austin), Philipp Krähenbühl (University of Texas Austin)
Object DetectionTransformerVision Language ModelImageText
🎯 What it does: This paper proposes a language-conditioned detection Transformer (DECOLA) that can dynamically adjust the internal mechanisms of the detector based on the given text category, generating high-quality pseudo-labels using image-level labels, and then training an open-vocabulary detector with these pseudo-labels.
Language-driven All-in-one Adverse Weather Removal
Hao Yang (Beijing Institute of Technology), Wei Liang (Beijing Institute of Technology)
RestorationConvolutional Neural NetworkMixture of ExpertsVision Language ModelImage
🎯 What it does: This paper proposes a language-driven all-weather degradation removal framework (LDR) that can adaptively restore images affected by various severe weather conditions such as rain, snow, raindrops, and fog within a single network, without the need for explicit weather labels or severity annotations.
Language-Driven Anchors for Zero-Shot Adversarial Robustness
Xiao Li (Tsinghua University), Xiaolin Hu (Tsinghua University)
Representation LearningAdversarial AttackContrastive LearningImage
🎯 What it does: This paper proposes a language-driven anchor point adversarial training (LAAT) that achieves zero-shot adversarial robustness by utilizing text anchors generated by the CLIP text encoder.
Language-driven Grasp Detection
An Dinh Vuong (FPT Software AI Center), Anh Nguyen (FPT Software AI Center)
Object DetectionSegmentationGenerationRobotic IntelligenceVision Language ModelDiffusion modelContrastive LearningImageTextMultimodality
🎯 What it does: This paper proposes a language-driven grasp detection task and a large-scale dataset Grasp-Anything++, and designs the LGD method based on diffusion models and contrastive learning, completing robotic grasping experiments.
Language-driven Object Fusion into Neural Radiance Fields with Pose-Conditioned Dataset Updates
Ka Chun Shum (Hong Kong University of Science and Technology), Sai-Kit Yeung (Hong Kong University of Science and Technology)
GenerationData SynthesisDiffusion modelNeural Radiance FieldImage
🎯 What it does: Utilizing a text-driven diffusion model to synthesize target objects in multi-view background images and achieve object insertion and removal in NeRF through pose-conditioned dataset updates.
Language-guided Image Reflection Separation
Haofeng Zhong (Peking University), Boxin Shi (Peking University)
Image TranslationConvolutional Neural NetworkContrastive LearningImageTextMultimodality
🎯 What it does: A framework for image reflection separation guided by natural language descriptions is proposed.
Language-only Training of Zero-shot Composed Image Retrieval
Geonmo Gu (NAVER Vision), Sangdoo Yun (NAVER AI Lab)
RetrievalTransformerVision Language ModelContrastive LearningText
🎯 What it does: A zero-shot combined image retrieval framework LinCIR is proposed, which is trained solely using text.
Laplacian-guided Entropy Model in Neural Codec with Blur-dissipated Synthesis
Atefeh Khoshkhahtinat (West Virginia University), Nasser M. Nasrabadi (West Virginia University)
CompressionTransformerDiffusion modelImage
🎯 What it does: A neural image compression decoder based on a conditional non-homogeneous variance diffusion model is proposed, incorporating an explicit frequency-induced bias;
LaRE^2: Latent Reconstruction Error Based Method for Diffusion-Generated Image Detection
Yunpeng Luo (Tencent YouTu Lab), Shouhong Ding (Tencent YouTu Lab)
GenerationTransformerDiffusion modelImage
🎯 What it does: A diffusion-based image detection method LaRE2 is proposed, which utilizes latent space single-step reconstruction error and error-guided feature refinement to quickly extract discriminative features and improve detection performance.
Large Language Models are Good Prompt Learners for Low-Shot Image Classification
Zhaoheng Zheng (University of Southern California), Ram Nevatia (University of Southern California)
ClassificationKnowledge DistillationTransformerLarge Language ModelPrompt EngineeringImage
🎯 What it does: This paper proposes the LLaMP framework, which utilizes the knowledge of large language models for low-shot image classification, generating category-specific prompts that are compatible with the CLIP text encoder.
LASA: Instance Reconstruction from Real Scans using A Large-scale Aligned Shape Annotation Dataset
Haolin Liu (Chinese University of Hong Kong, Shenzhen), Xiaoguang Han (Chinese University of Hong Kong, Shenzhen)
Object DetectionSegmentationTransformerDiffusion modelMultimodalityPoint Cloud
🎯 What it does: A large-scale high-quality aligned CAD annotation dataset LASA is proposed, and based on this dataset, a cross-modal diffusion reconstruction model DisCo and an occupancy-guided 3D object detection method OccGOD are designed.
LASIL: Learner-Aware Supervised Imitation Learning For Long-term Microscopic Traffic Simulation
Ke Guo (Alibaba Group), Jia Pan (University of Hong Kong)
Autonomous DrivingGraph Neural NetworkAuto EncoderTime Series
🎯 What it does: The paper proposes a learner-aware supervised imitation learning framework (LASIL) for generating long-term stable micro-traffic simulations.
LASO: Language-guided Affordance Segmentation on 3D Object
Yicong Li (National University of Singapore), Tat-seng Chua (National University of Singapore)
Object DetectionSegmentationTransformerLarge Language ModelTextPoint Cloud
🎯 What it does: Proposed the language-guided 3D object functional segmentation task LASO and constructed the corresponding dataset.
Latency Correction for Event-guided Deblurring and Frame Interpolation
Yixin Yang (Peking University), Boxin Shi (Peking University)
RestorationVideo
🎯 What it does: This paper addresses the delay issue of event cameras and proposes a delay correction method based on a parameterized delay-brightness curve. It improves deblurring and frame interpolation tasks through an event-driven double integral model and a self-supervised event delay availability metric.
Latent Modulated Function for Computational Optimal Continuous Image Representation
Zongyao He (Sun Yat-sen University), Zhi Jin (Sun Yat-sen University)
Super ResolutionOptimizationComputational EfficiencyImage
🎯 What it does: A Latent Modulated Function (LMF) framework is proposed, which splits high-resolution (HR) high-dimensional decoding into shared decoding in low-resolution (LR) high-dimensional space and rendering in HR low-dimensional space, thus achieving efficient continuous image representation for arbitrary scale super-resolution (ASSR).
Layout-Agnostic Scene Text Image Synthesis with Diffusion Models
Qilong Zhangli (Rutgers University), Praveen Krishnan (Meta AI)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: This paper presents SceneTextGen, a layout-free scene text image synthesis method based on diffusion models.