CVPR 2023 Papers — Page 11
IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2353 papers
Indiscernible Object Counting in Underwater Scenes
Guolei Sun (ETH Zurich), Luc Van Gool (KU Leuven)
Object DetectionTransformerImage
🎯 What it does: A new task called 'concealed object counting' is proposed, and a large underwater fish concealment counting dataset, IOCfish5K, is constructed. Based on this, a Transformer framework called IOCFormer is proposed, which integrates density and regression branches.
Inferring and Leveraging Parts From Object Shape for Improving Semantic Image Synthesis
Yuxiang Wei (Harbin Institute of Technology), Wangmeng Zuo (Harbin Institute of Technology)
SegmentationGenerationGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a PartNet based on a small number of supporting partial images to infer object part segmentation maps from semantic maps, and enhances the detail quality of semantic image synthesis through a part semantic modulation module.
Infinite Photorealistic Worlds Using Procedural Generation
Alexander Raistrick (Princeton University), Jia Deng (Princeton University)
GenerationData SynthesisImage
🎯 What it does: Infinigen has been developed, a fully procedural generator capable of infinitely diversifying and realistically rendering 3D scenes of the natural world (plants, animals, terrain, weather, etc.) and automatically rendering datasets with complete geometric ground truth.
Ingredient-Oriented Multi-Degradation Learning for Image Restoration
Jinghao Zhang (University of Science and Technology of China), Feng Zhao (University of Science and Technology of China)
RestorationTransformerImage
🎯 What it does: A multi-image denoising learning framework IDR based on 'components' rather than tasks is proposed, which achieves task knowledge collection and component knowledge integration in two stages, enhancing the scalability and generalization ability of various denoising tasks.
Initialization Noise in Image Gradients and Saliency Maps
Ann-Christin Woerl (Johannes Gutenberg University Mainz), Michael Wand (Johannes Gutenberg University Mainz)
Explainability and InterpretabilityConvolutional Neural NetworkImage
🎯 What it does: This paper reveals significant noise caused by the randomness of initialization and training by analyzing the gradients of image classification CNNs (including logit gradients and various saliency methods), and demonstrates that this noise leads to feature importance maps with substantial structural differences across different models.
Instance Relation Graph Guided Source-Free Domain Adaptive Object Detection
Vibashan VS (Johns Hopkins University), Vishal M. Patel (Johns Hopkins University)
Object DetectionDomain AdaptationGraph Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes a source-free domain adaptive object detection method that enhances target domain feature representation by constructing an Instance Relation Graph (IRG) and utilizing its guided contrastive learning.
Instance-Aware Domain Generalization for Face Anti-Spoofing
Qianyu Zhou (Shanghai Jiao Tong University), Lizhuang Ma (Shanghai Jiao Tong University)
Domain AdaptationAnomaly DetectionConvolutional Neural NetworkImage
🎯 What it does: Proposes a domain-agnostic instance-aware domain generalization framework for facial fraud detection.
Instance-Specific and Model-Adaptive Supervision for Semi-Supervised Semantic Segmentation
Zhen Zhao (University of Sydney), Luping Zhou (University of Sydney)
Object DetectionSegmentationImage
🎯 What it does: This paper proposes iMAS, an instance-specific and model-adaptive semi-supervised semantic segmentation method that utilizes a teacher-student model to evaluate the difficulty of each unlabeled sample based on IoU, and dynamically adjusts the weights of strong augmentation and consistency loss according to this difficulty.
Instant Domain Augmentation for LiDAR Semantic Segmentation
Kwonyoung Ryu (POSTECH), Jaesik Park (POSTECH)
SegmentationDomain AdaptationAutonomous DrivingSimultaneous Localization and MappingPoint Cloud
🎯 What it does: A real-time LiDAR data domain augmentation method called LiDomAug is proposed to address the performance degradation of perception algorithms caused by different LiDAR sensors.
Instant Multi-View Head Capture Through Learnable Registration
Timo Bolkart (Max Planck Institute for Intelligent Systems), Michael J. Black (University of Southern California)
SegmentationPose EstimationDepth EstimationConvolutional Neural NetworkImageMesh
🎯 What it does: Directly inferring a complete 3D head mesh from multi-view calibrated images, omitting traditional MVS reconstruction and non-rigid registration steps.
Instant Volumetric Head Avatars
Wojciech Zielonka (Max Planck Institute for Intelligent Systems), Justus Thies (Max Planck Institute for Intelligent Systems)
GenerationData SynthesisNeural Radiance FieldVideo
🎯 What it does: Based on monocular RGB video, a dynamic neural radiance field (NeRF) with geometric priors instantaneously constructs interactive 3D digital face avatars in less than 10 minutes.
Instant-NVR: Instant Neural Volumetric Rendering for Human-Object Interactions From Monocular RGBD Stream
Yuheng Jiang (ShanghaiTech University), Lan Xu (ShanghaiTech University)
Object DetectionObject TrackingPose EstimationDepth EstimationComputational EfficiencyNeural Radiance FieldSimultaneous Localization and MappingVideo
🎯 What it does: A system named Instant-NVR is proposed, which utilizes a single RGBD camera to achieve instant neural volume rendering of human-object interaction scenes, capable of generating high-quality rendering results from any viewpoint in real-time.
InstantAvatar: Learning Avatars From Monocular Video in 60 Seconds
Tianjian Jiang (ETH Zurich), Otmar Hilliges (ETH Zurich)
GenerationPose EstimationComputational EfficiencyNeural Radiance FieldVideo
🎯 What it does: This paper presents InstantAvatar, a method that can learn high-fidelity animatable full-body avatars from monocular video in 60 seconds and render at 15 FPS.
InstMove: Instance Motion for Object-Centric Video Segmentation
Qihao Liu (Johns Hopkins University), Song Bai (ByteDance)
Object TrackingSegmentationRecurrent Neural NetworkOptical FlowVideo
🎯 What it does: Proposes the InstMove module, which improves video instance segmentation, video object segmentation, and multi-object tracking/segmentation through instance-level motion prediction.
InstructPix2Pix: Learning To Follow Image Editing Instructions
Tim Brooks (University of California), Alexei A. Efros (University of California)
Image TranslationGenerationLarge Language ModelPrompt EngineeringDiffusion modelImageText
🎯 What it does: Trained a conditional diffusion model InstructPix2Pix, enabling the model to perform image editing directly during the forward inference phase based solely on the input image and natural language editing instructions, without the need for additional inversion or fine-tuning.
Integral Neural Networks
Kirill Solodskikh (Huawei), Stamatios Lefkimmiatis (Huawei)
ClassificationSuper ResolutionOptimizationImage
🎯 What it does: This paper studies a deep network that represents the weights of network layers as continuous functions and replaces discrete convolution/fully connected operations with integration, referred to as Integral Neural Networks (INN).
Integrally Pre-Trained Transformer Pyramid Networks
Yunjie Tian (University of Chinese Academy of Sciences), Qixiang Ye (University of Chinese Academy of Sciences)
ClassificationObject DetectionSegmentationKnowledge DistillationTransformerImage
🎯 What it does: Overall pre-training of the visual Transformer, jointly optimizing the backbone network and the feature pyramid (neck), and proposing Masked Feature Modeling (MFM) for multi-stage supervision.
Interactive and Explainable Region-Guided Radiology Report Generation
Tim Tanida (Technical University of Munich), Daniel Rueckert (Imperial College London)
Object DetectionGenerationExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningImageTextMultimodality
🎯 What it does: A radiology report generation method based on anatomical region detection is proposed, which generates corresponding sentences for each key region using regional visual features and concatenates them into a complete report.
Interactive Cartoonization With Controllable Perceptual Factors
Namhyuk Ahn (NAVER WEBTOON AI), Seungkwon Kim (NAVER WEBTOON AI)
Image TranslationGenerationGenerative Adversarial NetworkImage
🎯 What it does: An interactive cartoonization model named CARTOONER has been developed, capable of real-time control over texture (stroke thickness and abstraction level) and color, achieving diverse and editable cartoon effects.
Interactive Segmentation As Gaussion Process Classification
Minghao Zhou (Xi'an Jiaotong University), Yefeng Zheng (Tencent)
SegmentationImage
🎯 What it does: Reformulate the interactive segmentation task as a pixel-level binary classification problem for each image, and use a Gaussian process classification model.
Interactive Segmentation of Radiance Fields
Rahul Goel (International Institute of Information Technology Hyderabad), P. J. Narayanan (International Institute of Information Technology Hyderabad)
SegmentationKnowledge DistillationTransformerNeural Radiance FieldImage
🎯 What it does: An interactive radiance field object segmentation method based on TensoRF, called ISRF, is proposed, allowing users to quickly obtain high-quality 3D segmentation results through positive and negative strokes.
InternImage: Exploring Large-Scale Vision Foundation Models With Deformable Convolutions
Wenhai Wang (Shanghai AI Laboratory), Yu Qiao (Shanghai AI Laboratory)
ClassificationObject DetectionSegmentationConvolutional Neural NetworkTransformerImage
🎯 What it does: A new large-scale convolutional foundation model, InternImage, is proposed, utilizing deformable convolution to achieve long-range dependencies and adaptive spatial aggregation.
Interventional Bag Multi-Instance Learning on Whole-Slide Pathological Images
Tiancheng Lin (Shanghai Jiao Tong University), Chang-Wen Chen
ClassificationConvolutional Neural NetworkTransformerImage
🎯 What it does: A multi-instance learning framework based on causal intervention, IBMIL, is proposed, which can suppress background correlation bias in whole slide image classification and improve bag-level prediction accuracy.
Intrinsic Physical Concepts Discovery With Object-Centric Predictive Models
Qu Tang (University of Chinese Academy of Sciences), Zhaoxiang Zhang (University of Chinese Academy of Sciences)
Auto EncoderVideoPhysics Related
🎯 What it does: The PHYCINE model is proposed, utilizing unsupervised object-centric predictive learning to connect low-level visual attributes to high-level physical concepts of quality and charge.
Introducing Competition To Boost the Transferability of Targeted Adversarial Examples Through Clean Feature Mixup
Junyoung Byun (Korea Advanced Institute of Science and Technology), Changick Kim (Korea Advanced Institute of Science and Technology)
ClassificationAdversarial AttackTransformerImage
🎯 What it does: This paper proposes a Clean Feature Mixup (CFM) method that mixes clean features in the feature space to introduce a competitive mechanism when generating targeted adversarial samples, thereby enhancing the cross-model transferability of adversarial samples.
Inverse Rendering of Translucent Objects Using Physical and Neural Renderers
Chenhao Li (Osaka University), Hajime Nagahara (Osaka University)
GenerationData SynthesisNeural Radiance FieldImage
🎯 What it does: This study investigates the inverse rendering of the shape, surface reflection, subsurface scattering (SSS), and ambient lighting of translucent objects from flash and non-flash images.
Inversion-Based Style Transfer With Diffusion Models
Yuxin Zhang (Institute of Automation, Chinese Academy of Sciences), Changsheng Xu (Institute of Automation, Chinese Academy of Sciences)
Image TranslationGenerationDiffusion modelImage
🎯 What it does: This paper proposes an artistic style transfer framework called InST, which is based on reverse learning text descriptions from a single artwork. It utilizes attention-based text reverse and random reverse methods to use the learned style information as a conditional driver for a diffusion model to generate new artistic images, achieving precise transfer of attributes such as semantics, texture, brushstrokes, and color.
Invertible Neural Skinning
Yash Kant (University of Toronto), Igor Gilitschenski (Snap Research)
GenerationPose EstimationMesh
🎯 What it does: An end-to-end differentiable Inverse Neural Skinning (INS) pipeline is proposed, utilizing a Pose-Conditioned Inverse Network (PIN) to capture the nonlinear deformations of clothing and muscles on both sides of differential LBS, achieving surface correspondence during pose re-targeting with only one mesh extraction.
Inverting the Imaging Process by Learning an Implicit Camera Model
Xin Huang (Northwestern Polytechnical University), Qing Wang (Tencent AI Lab)
RestorationGenerationGenerative Adversarial NetworkOptical FlowImage
🎯 What it does: This paper proposes an implicit neural camera model (including a blur generator and a tone mapper) that can reverse the camera imaging process and generate panoramic focused HDR images through self-supervised learning under the supervision of multi-focus/multi-exposure images.
IPCC-TP: Utilizing Incremental Pearson Correlation Coefficient for Joint Multi-Agent Trajectory Prediction
Dekai Zhu (Technical University of Munich), Benjamin Busam (Technical University of Munich)
Autonomous DrivingTransformerPoint Cloud
🎯 What it does: The IPCC-TP plugin module is proposed, which extends multi-agent trajectory prediction from marginal probability distributions to a joint Gaussian distribution, capturing interactions between agents through the Pearson correlation coefficient of motion increments.
iQuery: Instruments As Queries for Audio-Visual Sound Separation
Jiaben Chen (University of California San Diego), Jianbo Shi (University of Pennsylvania)
TransformerPrompt EngineeringVideoMultimodalityAudio
🎯 What it does: This paper proposes the iQuery framework, which achieves audio-visual sound source separation in Transformers through learnable audio queries, allowing for precise separation of sounds from different instruments or events in corresponding videos within mixed audio.
Is BERT Blind? Exploring the Effect of Vision-and-Language Pretraining on Visual Language Understanding
Morris Alper (Tel Aviv University), Hadar Averbuch-Elor (Tel Aviv University)
TransformerVision Language ModelTextMultimodality
🎯 What it does: This paper studies the impact of visual-language pre-training on text encoders in visual reasoning tasks (visual language understanding) and proposes a set of visual language understanding task collections and a zero-shot Stroop detection method.
IS-GGT: Iterative Scene Graph Generation With Generative Transformers
Sanjoy Kundu (Oklahoma State University), Sathyanarayanan N. Aakur (Oklahoma State University)
Object DetectionGenerationTransformerImageGraph
🎯 What it does: This paper proposes a two-stage generative transformer framework IS-GGT for scene graph generation: first, it samples the interaction graph between entities using a generative transformer, and then classifies the predicates of the sampled edges;
ISBNet: A 3D Point Cloud Instance Segmentation Network With Instance-Aware Sampling and Box-Aware Dynamic Convolution
Tuan Duc Ngo (VinAI Research), Khoi Nguyen (VinAI Research)
Object DetectionSegmentationConvolutional Neural NetworkPoint Cloud
🎯 What it does: This paper proposes ISBNet, a non-clustering 3D point cloud instance segmentation framework based on dynamic convolution;
Iterative Geometry Encoding Volume for Stereo Matching
Gangwei Xu (Huazhong University of Science and Technology), Xin Yang (Huazhong University of Science and Technology)
Depth EstimationComputational EfficiencyConvolutional Neural NetworkRecurrent Neural NetworkImage
🎯 What it does: This paper proposes an Iterative Geometric Encoding Volume (IGEV) framework that generates Geometric Encoding Volumes (GEV) using a lightweight 3D CNN and fuses them with All-Pairs Correlation (APC) to form a Combined Geometric Encoding Volume (CGEV). It then uses ConvGRU to iteratively update the disparity and accelerates convergence by regressing the initial disparity through soft-argmin, extending it to multi-view stereo (MVS) tasks.
Iterative Next Boundary Detection for Instance Segmentation of Tree Rings in Microscopy Images of Shrub Cross Sections
Alexander Gillert (Fraunhofer Institute for Computer Graphics Research), Uwe Freiherr von Lukas (Fraunhofer Institute for Computer Graphics Research)
Object DetectionSegmentationConvolutional Neural NetworkImage
🎯 What it does: An iterative next ring detection (INBD) method is proposed for tree ring instance segmentation in cell-level high-resolution microscopic images of tree cross-sections.
Iterative Proposal Refinement for Weakly-Supervised Video Grounding
Meng Cao (Peking University), Daxin Jiang (Microsoft)
Knowledge DistillationRepresentation LearningTransformerVision Language ModelVideo
🎯 What it does: An Iterative Proposal Refinement Network (IRON) is proposed for weakly supervised video grounding, utilizing semantic and conceptual knowledge from a pre-trained VL model for dual distillation, and iteratively updating proposal confidence through label propagation.
Iterative Vision-and-Language Navigation
Jacob Krantz (Oregon State University), Jesse Thomason (University of Southern California)
TransformerReinforcement LearningTextMultimodalityBenchmark
🎯 What it does: This paper proposes the Iterative Visual and Language Navigation (IVLN) paradigm and constructs two navigation benchmarks, discrete and continuous (IR2R and IR2R-CE), for evaluating agents' learning and memory capabilities in multi-segment instruction sequences.
IterativePFN: True Iterative Point Cloud Filtering
Dasith de Silva Edirimuni (Deakin University), Ying He (Nanyang Technological University)
RestorationGraph Neural NetworkPoint Cloud
🎯 What it does: A multi-iteration module point cloud denoising network called IterativePFN is proposed, which can achieve a complete iterative filtering process within the network.
itKD: Interchange Transfer-Based Knowledge Distillation for 3D Object Detection
Hyeon Cho (Ajou University), Wonjun Hwang (Ajou University)
Object DetectionAutonomous DrivingKnowledge DistillationAuto EncoderPoint Cloud
🎯 What it does: This paper proposes a knowledge distillation method called itKD for 3D point cloud object detection, aimed at training lightweight detectors.
JacobiNeRF: NeRF Shaping With Mutual Information Gradients
Xiaomeng Xu (Tsinghua University), Leonidas Guibas (Google Research)
Object DetectionSegmentationNeural Radiance FieldContrastive LearningImage
🎯 What it does: This paper proposes aligning the mutual information (MI) in the gradient space of Neural Radiance Fields (NeRF) so that when the network weights are perturbed, semantically related scene points or regions can produce resonant mutual responses, thereby achieving tasks such as sparse label propagation, instance selection, and editing.
JAWS: Just a Wild Shot for Cinematic Transfer in Neural Radiance Fields
Xi Wang (Inria), Marc Christie (Inria)
Image TranslationPose EstimationOptimizationNeural Radiance FieldOptical FlowVideo
🎯 What it does: A differentiable pipeline based on NeRF is proposed, achieving cinematic style transfer of reference videos through inverse optimization of camera pose, focal length, and time.
Jedi: Entropy-Based Localization and Removal of Adversarial Patches
Bilel Tarchoun (University of Sousse), Ihsen Alouani (Queen's University Belfast)
RestorationObject DetectionAdversarial AttackAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a defense method against adversarial patches called Jedi, which can accurately locate patches in images and repair them using an autoencoder without modifying the target model, followed by patch removal and inpainting recovery.
Joint Appearance and Motion Learning for Efficient Rolling Shutter Correction
Bin Fan (Northwestern Polytechnical University), Qi Liu (Northwestern Polytechnical University)
RestorationComputational EfficiencyTransformerImageVideo
🎯 What it does: This paper proposes and implements a single-stage JAMNet network for recovering high-quality global shutter images from two frames of rolling shutter images.
Joint HDR Denoising and Fusion: A Real-World Mobile HDR Image Dataset
Shuaizheng Liu (Hong Kong Polytechnic University), Lei Zhang (Hong Kong Polytechnic University)
RestorationTransformerImage
🎯 What it does: A real-world mobile phone HDR image dataset (Mobile-HDR) has been constructed, and a joint denoising and fusion Transformer network called Joint-HDRDN has been proposed to generate noise-free HDR images from three differently exposed LDR raw images.
Joint Token Pruning and Squeezing Towards More Aggressive Compression of Vision Transformers
Siyuan Wei (MEGVII Technology), Jiajun Liang (MEGVII Technology)
CompressionTransformerImage
🎯 What it does: This paper proposes a Token Pruning and Compression (TPS) module to more aggressively compress visual Transformers while retaining the information of pruned tokens.
Joint Video Multi-Frame Interpolation and Deblurring Under Unknown Exposure Time
Wei Shang (Harbin Institute of Technology), Wangmeng Zuo (Harbin Institute of Technology)
Image TranslationRestorationContrastive LearningVideo
🎯 What it does: An end-to-end method is proposed to simultaneously perform multi-frame interpolation and deblurring for videos with unknown exposure times.
Joint Visual Grounding and Tracking With Natural Language Specification
Li Zhou (Harbin Institute of Technology), Zhenyu He (Harbin Institute of Technology)
Object DetectionObject TrackingTransformerImageVideoText
🎯 What it does: A joint visual localization and tracking framework is proposed, utilizing natural language descriptions to simultaneously achieve target localization and tracking.
JRDB-Pose: A Large-Scale Dataset for Multi-Person Pose Estimation and Tracking
Edward Vendrow (Stanford University), Hamid Rezatofighi (Monash University)
Object TrackingPose EstimationVideoMultimodalityBenchmark
🎯 What it does: The JRDB-Pose dataset and benchmark are proposed for multi-person pose estimation and tracking.
K-Planes: Explicit Radiance Fields in Space, Time, and Appearance
Sara Fridovich-Keil, Angjoo Kanazawa
Data SynthesisCompressionNeural Radiance FieldImageVideo
🎯 What it does: The k-planes model is proposed, which achieves an explicit representation of radiance fields for 3D static and 4D dynamic scenes by splitting the d-dimensional space into (d choose 2) two-dimensional planes.
K3DN: Disparity-Aware Kernel Estimation for Dual-Pixel Defocus Deblurring
Yan Yang (Australian National University), Miaomiao Liu (Australian National University)
RestorationImage
🎯 What it does: This paper proposes a deblurring framework K3DN based on dual-pixel (DP) image pairs, utilizing DP disparity information to achieve adaptive estimation of spatially varying blur kernels, thereby restoring panoramic focused images.
KD-DLGAN: Data Limited Image Generation via Knowledge Distillation
Kaiwen Cui (Nanyang Technological University), Eric P. Xing (Mohamed bin Zayed University of Artificial Intelligence)
GenerationKnowledge DistillationGenerative Adversarial NetworkImage
🎯 What it does: Proposes KD-DLGAN, a GAN trained with knowledge distillation for data-constrained scenarios.
KERM: Knowledge Enhanced Reasoning for Vision-and-Language Navigation
Xiangyang Li (Chinese Academy of Sciences), Shuqiang Jiang (Chinese Academy of Sciences)
RetrievalRobotic IntelligenceTransformerVision Language ModelMultimodalityRetrieval-Augmented Generation
🎯 What it does: A knowledge-enhanced reasoning model KERM is proposed, which retrieves knowledge facts corresponding to views and integrates visual, historical, and instruction features to improve visual-language navigation.
Kernel Aware Resampler
Michael Bernasconi (ETH Zurich), Christopher Schroers (Disney Research Studios)
RestorationSuper ResolutionConvolutional Neural NetworkImage
🎯 What it does: A 'Kernel Aware Resampler' based on deep learning is proposed, which can resample any image (enlargement, rotation, distortion correction, etc.) and can automatically estimate the blur kernel during the downsampling process.
KiUT: Knowledge-Injected U-Transformer for Radiology Report Generation
Zhongzhen Huang (Shanghai Jiao Tong University), Shaoting Zhang (SenseTime Research)
GenerationKnowledge DistillationTransformerImageTextElectronic Health Records
🎯 What it does: The KiUT framework is proposed, which uses a U-Connection Transformer and a knowledge-injected decoder to generate radiology reports.
Knowledge Combination To Learn Rotated Detection Without Rotated Annotation
Tianyu Zhu (Amazon), Anton van den Hengel (Monash University)
Object DetectionDomain AdaptationImage
🎯 What it does: Proposes the KCR framework, which utilizes the joint training of source data's rotation annotations and target data's axis-aligned annotations to achieve rotation object detection without the need for rotation annotations.
Knowledge Distillation for 6D Pose Estimation by Aligning Distributions of Local Predictions
Shuxuan Guo (École Polytechnique Fédérale de Lausanne), Mathieu Salzmann (École Polytechnique Fédérale de Lausanne)
Pose EstimationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: A knowledge distillation method for 6D pose estimation is designed, which distills a lightweight network by aligning the local prediction distributions (keypoints or dense representations) of the teacher and student, combined with segmentation scores.
L-CoIns: Language-Based Colorization With Instance Awareness
Zheng Chang (Beijing University of Posts and Telecommunications), Boxin Shi (Peking University)
Image TranslationGenerationTransformerImageText
🎯 What it does: A language-based instance-aware colorization framework L-CoIns is proposed, which can automatically color grayscale images based on user-provided descriptions and has the ability to distinguish different instances corresponding to the same object word.
Label Information Bottleneck for Label Enhancement
Qinghai Zheng (Fuzhou University), Haoyu Tang (Shandong University)
ClassificationAuto EncoderTabular
🎯 What it does: The Label Information Bottleneck (LIB) method is proposed for label enhancement, restoring the complete label distribution based on logical labels.
Label-Free Liver Tumor Segmentation
Qixin Hu (Huazhong University of Science and Technology), Zongwei Zhou (Johns Hopkins University)
SegmentationData SynthesisConvolutional Neural NetworkImageComputed TomographyBenchmark
🎯 What it does: By synthesizing realistic liver tumors in healthy liver CT scans, AI models can be trained to perform liver tumor segmentation without the need for manual voxel annotation.
LANA: A Language-Capable Navigator for Instruction Following and Generation
Xiaohan Wang (Zhejiang University), Yi Yang (Zhejiang University)
GenerationExplainability and InterpretabilityRobotic IntelligenceTransformerVision Language ModelMultimodality
🎯 What it does: A single model named LANA has been developed, capable of simultaneously executing visual language navigation instructions and generating path descriptions, achieving bidirectional human-machine language interaction.
Language Adaptive Weight Generation for Multi-Task Visual Grounding
Wei Su (Zhejiang University), Xi Li (Shanghai Institute for Advanced Study of Zhejiang University)
Object DetectionSegmentationTransformerVision Language ModelImage
🎯 What it does: A language-adaptive weight-based active perception visual localization framework VG-LAW is proposed, and a lightweight multi-task head is designed to simultaneously perform reference-based localization (REC) and segmentation (RES) tasks.
Language in a Bottle: Language Model Guided Concept Bottlenecks for Interpretable Image Classification
Yue Yang (University of Pennsylvania), Mark Yatskar (University of Pennsylvania)
ClassificationExplainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringVision Language ModelImage
🎯 What it does: This paper proposes an interpretable image classification framework called LaBo, which does not require manually defined concepts. It utilizes large language models to generate candidate concepts and aligns them visually through CLIP, forming a concept bottleneck model.
Language-Guided Audio-Visual Source Separation via Trimodal Consistency
Reuben Tan (Boston University), Kate Saenko (Boston University)
RestorationTransformerVision Language ModelContrastive LearningVideoTextMultimodalityAudio
🎯 What it does: A self-supervised audio source separation method called VAST is proposed, which can separate corresponding audio sources from mixed audio based on natural language queries or video inputs.
Language-Guided Music Recommendation for Video via Prompt Analogies
Daniel McKee (University of Illinois at Urbana-Champaign), Bryan Russell (Adobe Research)
RetrievalRecommendation SystemTransformerLarge Language ModelPrompt EngineeringContrastive LearningVideoTextMultimodality
🎯 What it does: This paper proposes a natural language-guided music recommendation framework that can select suitable background music through free text descriptions while given a video.
LANIT: Language-Driven Image-to-Image Translation for Unlabeled Data
Jihye Park (Korea University), Seungryong Kim (Korea University)
Image TranslationData SynthesisVision Language ModelGenerative Adversarial NetworkImage
🎯 What it does: A language-driven unsupervised image translation framework called LANIT is proposed, which generates multi-hot domain labels using dataset-level textual domain descriptions to drive image-to-image translation.
Large-Capacity and Flexible Video Steganography via Invertible Neural Network
Chong Mou (Peking University), Jian Zhang (Peking University)
Data SynthesisSafty and PrivacyFlow-based ModelVideo
🎯 What it does: A large-capacity, reversible video steganography network is proposed, capable of hiding up to 7 secret videos within a cover video and achieving complete recovery of these secret videos through a single reversible neural network.
Large-Scale Training Data Search for Object Re-Identification
Yue Yao (Australian National University), Liang Zheng (Australian National University)
RecognitionRetrievalImage
🎯 What it does: The paper proposes a Search and Prune (SnP) framework to quickly construct a training set that is close to the target domain distribution and has a controllable size from a large-scale data pool, in order to enhance the performance of object re-identification in the target domain.
LargeKernel3D: Scaling Up Kernels in 3D Sparse CNNs
Yukang Chen (Chinese University of Hong Kong), Jiaya Jia (Chinese University of Hong Kong)
Object DetectionSegmentationAutonomous DrivingConvolutional Neural NetworkMultimodalityPoint Cloud
🎯 What it does: Designed and implemented a spatially partitioned convolution that can use ultra-large convolution kernels in 3D sparse convolution networks, and based on this, constructed the LargeKernel3D backbone network for indoor 3D semantic segmentation and outdoor 3D object detection.
LaserMix for Semi-Supervised LiDAR Semantic Segmentation
Lingdong Kong (Nanyang Technological University), Ziwei Liu (Nanyang Technological University)
SegmentationAutonomous DrivingPoint Cloud
🎯 What it does: This paper presents LaserMix, a semi-supervised LiDAR semantic segmentation method that utilizes spatial priors of laser beams.
LASP: Text-to-Text Optimization for Language-Aware Soft Prompting of Vision & Language Models
Adrian Bulat (Samsung AI Cambridge), Georgios Tzimiropoulos (Samsung AI Cambridge)
Domain AdaptationOptimizationTransformerPrompt EngineeringVision Language ModelImageTextMultimodality
🎯 What it does: Proposes the LASP method, which learns soft prompts using cross-entropy loss between texts to improve the few-shot adaptation of V&L models.
Latency Matters: Real-Time Action Forecasting Transformer
Harshayu Girase (Honda Research Institute USA), Karttikeya Mangalam (Honda Research Institute USA)
RecognitionObject DetectionAutonomous DrivingComputational EfficiencyTransformerVideo
🎯 What it does: Proposed and implemented RAFTformer, a two-stage Transformer network for low-latency real-time action prediction.
Latent-NeRF for Shape-Guided Generation of 3D Shapes and Textures
Gal Metzer, Daniel Cohen-Or
GenerationData SynthesisDiffusion modelScore-based ModelNeural Radiance FieldImageMesh
🎯 What it does: Proposes the Latent-NeRF framework, which utilizes Score Distillation from the Latent Diffusion model for text-guided 3D shape and texture generation, and achieves finer control through two shape constraints: Sketch-Shape and Latent-Paint.
LAVENDER: Unifying Video-Language Understanding As Masked Language Modeling
Linjie Li (Microsoft), Lijuan Wang (Microsoft)
GenerationRetrievalTransformerVision Language ModelVideoText
🎯 What it does: LAVENDER is proposed, a unified video-language model that uses Masked Language Modeling (MLM) as a unified interface for all pre-training and downstream tasks, without the need for task-specific heads.
Layout-Based Causal Inference for Object Navigation
Sixian Zhang (Chinese Academy of Sciences), Shuqiang Jiang (Chinese Academy of Sciences)
Robotic IntelligenceReinforcement Learning
🎯 What it does: A layout-based soft total direct effect (L-sTDE) framework is proposed, which adjusts the positive and negative influences of experiences by estimating environmental layout differences in target navigation, enhancing generalization ability in unknown environments.
LayoutDiffusion: Controllable Diffusion Model for Layout-to-Image Generation
Guangcong Zheng (Zhejiang University), Xi Li (Tencent)
GenerationData SynthesisTransformerDiffusion modelImage
🎯 What it does: This paper presents LayoutDiffusion, an end-to-end diffusion model for layout-to-image generation that allows for fine control over the positions, sizes, and categories of multiple objects while maintaining high quality.
LayoutDM: Discrete Diffusion Model for Controllable Layout Generation
Naoto Inoue (CyberAgent), Kota Yamaguchi (CyberAgent)
GenerationTransformerDiffusion modelMultimodality
🎯 What it does: A controllable layout generation framework called LayoutDM based on discrete diffusion models is proposed, capable of generating structured layouts under both unconditional and various conditional settings (such as category, size, position, relationship, etc.).
LayoutDM: Transformer-Based Diffusion Model for Layout Generation
Shang Chai (University of Science and Technology of China), Fengying Yan (Tianjin University)
GenerationData SynthesisTransformerDiffusion modelImage
🎯 What it does: A conditional layout generation model, LayoutDM, is proposed, which utilizes diffusion models and a Transformer architecture to progressively generate high-quality layouts that meet attribute constraints from noise.
LayoutFormer++: Conditional Graphic Layout Generation via Constraint Serialization and Decoding Space Restriction
Zhaoyun Jiang (Xi'an Jiaotong University), Dongmei Zhang (Microsoft Research Asia)
GenerationTransformerImage
🎯 What it does: A unified conditional layout generation model, LayoutFormer++, is proposed, which can automatically generate high-quality layouts that satisfy various constraints provided by users (element types, sizes, relationships, local modifications, etc.).
Leapfrog Diffusion Model for Stochastic Trajectory Prediction
Weibo Mao (Shanghai Jiao Tong University), Yanfeng Wang (Shanghai Jiao Tong University)
GenerationData SynthesisAutonomous DrivingComputational EfficiencyTransformerDiffusion modelTime SeriesSequential
🎯 What it does: A Leapfrog Diffusion Model (LED) is proposed, which achieves fast multimodal inference for trajectory prediction through a trainable jump pre-initializer.
Learnable Skeleton-Aware 3D Point Cloud Sampling
Cheng Wen (University of Sydney), Dacheng Tao (University of Sydney)
ClassificationRetrievalGraph Neural NetworkPoint Cloud
🎯 What it does: A learnable 3D point cloud sampling method based on object skeletons is proposed, which achieves efficient sampling while preserving geometric topology information.
Learned Image Compression With Mixed Transformer-CNN Architectures
Jinming Liu (Waseda University), Jiro Katto (Waseda University)
CompressionConvolutional Neural NetworkTransformerAuto EncoderImage
🎯 What it does: This paper proposes an end-to-end learning-based image compression framework, which includes a parallel Transformer-CNN mixed block (TCM) and a channel-compressed Swintower attention module (SWAtten). Based on this, a channel autoregressive entropy model is constructed to improve rate-distortion performance.
Learned Two-Plane Perspective Prior Based Image Resampling for Efficient Object Detection
Anurag Ghosh (Carnegie Mellon University), Srinivasa G. Narasimhan (Carnegie Mellon University)
Object DetectionConvolutional Neural NetworkImage
🎯 What it does: A learnable image resampling method based on two-plane perspective priors is proposed, which uses geometric constraints to adaptively enlarge small targets that are far away and above, significantly improving detection performance while maintaining low latency and low memory.
Learning 3D Representations From 2D Pre-Trained Models via Image-to-Point Masked Autoencoders
Renrui Zhang (Shanghai Artificial Intelligence Laboratory), Hongsheng Li (Chinese University of Hong Kong)
ClassificationSegmentationRepresentation LearningTransformerAuto EncoderPoint Cloud
🎯 What it does: By transferring the knowledge of existing 2D pre-trained models to 3D point clouds, I2P-MAE (an image-point Masked Autoencoder) is proposed for self-supervised pre-training, which can learn high-quality 3D representations without the need for a large amount of 3D data.
Learning 3D Scene Priors With 2D Supervision
Yinyu Nie (Technical University of Munich), Matthias Nießner (Technical University of Munich)
SegmentationGenerationData SynthesisTransformerPoint CloudMesh
🎯 What it does: Without relying on any 3D supervision, we learn 3D scene priors through multi-view 2D instance masks and can generate complete 3D scene layouts and object geometries.
Learning 3D-Aware Image Synthesis With Unknown Pose Distribution
Zifan Shi (Hong Kong University of Science and Technology), Dit-Yan Yeung (Hong Kong University of Science and Technology)
GenerationData SynthesisPose EstimationNeural Radiance FieldGenerative Adversarial NetworkImage
🎯 What it does: A 3D-aware image synthesis method called PoF3D is proposed, which can learn high-quality 3D generative models without using any pose priors.
Learning a 3D Morphable Face Reflectance Model From Low-Cost Data
Yuxuan Han (Tsinghua University), Feng Xu (SenseTime Research)
RestorationGenerationNeural Radiance FieldImage
🎯 What it does: Developed and trained the first 3D deformable facial reflection model with spatially varying BRDF that can learn from low-cost public data.
Learning a Deep Color Difference Metric for Photographic Images
Haoyu Chen (City University of Hong Kong), Kede Ma (City University of Hong Kong)
Image TranslationRestorationFlow-based ModelImage
🎯 What it does: This paper proposes a color difference (CD) metric method based on multi-scale autoregressive normalizing flows—CD-Flow—for evaluating color differences in real photographic images.
Learning a Depth Covariance Function
Eric Dexheimer (Imperial College London), Andrew J. Davison (Imperial College London)
Depth EstimationOptimizationConvolutional Neural NetworkSimultaneous Localization and MappingImage
🎯 What it does: Learn and apply deep covariance functions for depth completion, bundle adjustment, and monocular dense visual odometry.
Learning a Practical SDR-to-HDRTV Up-Conversion Using New Dataset and Degradation Models
Cheng Guo (Communication University of China), Xiuhua Jiang (Communication University of China)
Image TranslationRestorationTransformerVideo
🎯 What it does: This paper studies the upsampling from SDR video to HDR-WCG television, proposing a new dataset HDRTV4K and an improved HDR→SDR degradation model.
Learning a Simple Low-Light Image Enhancer From Paired Low-Light Instances
Zhenqi Fu (Xiamen University), Kai-Kuang Ma (Nanyang Technological University)
RestorationAuto EncoderImage
🎯 What it does: A no-reference low-light image enhancement method called PairLIE is proposed, which utilizes low-light image pairs to learn adaptive priors and removes inappropriate features through a projection network, ultimately achieving enhancement with a simple network.
Learning a Sparse Transformer Network for Effective Image Deraining
Xiang Chen (Nanjing University of Science and Technology), Jinshan Pan (Nanjing University of Science and Technology)
RestorationTransformerMixture of ExpertsImage
🎯 What it does: A sparse Transformer network called DRSformer is proposed for single image rain removal, achieving finer raindrop removal through top-k selection of self-attention and a multi-scale feedforward network.
Learning Accurate 3D Shape Based on Stereo Polarimetric Imaging
Tianyu Huang (Chinese University of Hong Kong), Yun-Hui Liu (Chinese University of Hong Kong)
RestorationDepth EstimationConvolutional Neural NetworkTransformerImage
🎯 What it does: A framework that combines stereo polarization imaging and deep learning is proposed, capable of simultaneously recovering the surface normals and disparity of images.
Learning Action Changes by Measuring Verb-Adverb Textual Relationships
Davide Moltisanti (University of Edinburgh), Laura Sevilla-Lara (University of Edinburgh)
RecognitionTransformerVideoText
🎯 What it does: This paper proposes a method to learn action changes in videos by measuring the relationship between verbs and adverbs in text, and designs a regression target based on this idea; at the same time, a new dataset called Adverbs in Recipes (AIR) is constructed for training and evaluating the task of action adverb recognition.
Learning Adaptive Dense Event Stereo From the Image Domain
Hoonhee Cho (Korea Advanced Institute of Science and Technology), Kuk-Jin Yoon (Korea Advanced Institute of Science and Technology)
Depth EstimationDomain AdaptationGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes the ADES framework, which transfers image-based dense disparity networks to the unlabeled event domain, utilizing unsupervised domain adaptation techniques to achieve accurate disparity estimation in the target event domain.
Learning Analytical Posterior Probability for Human Mesh Recovery
Qi Fang (NetEase Games), Weidong Zhang (NetEase Games)
Pose EstimationConvolutional Neural NetworkMesh
🎯 What it does: A posterior-based 3D human mesh recovery framework is proposed by learning and utilizing the posterior probability of the matrix Fisher distribution based on SO(3).
Learning Anchor Transformations for 3D Garment Animation
Fang Zhao (Tencent AI Lab), Ying Shan (Tencent AI Lab)
Recurrent Neural NetworkMesh
🎯 What it does: Proposes the AnchorDEF model, which predicts 3D clothing animation using anchor point transformations and vertex displacements within a unified framework.
Learning and Aggregating Lane Graphs for Urban Automated Driving
Martin Büchner (University of Freiburg), Wolfram Burgard (University of Technology Nuremberg)
Autonomous DrivingGraph Neural NetworkGraph
🎯 What it does: Using a lane-based graph neural network (LaneGNN) to predict the reachable successor lane graph from the vehicle's perspective, and generating a globally consistent lane graph through iterative aggregation.
Learning Articulated Shape With Keypoint Pseudo-Labels From Web Images
Anastasis Stathopoulos (Rutgers University), Dimitris N. Metaxas (Rutgers University)
Object DetectionPose EstimationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a method for training a monocular 3D shape reconstruction model using unlabeled images based on a small number of 2D keypoint annotations and web scraping, through pseudo-labeling and data filtering methods.
Learning Attention As Disentangler for Compositional Zero-Shot Learning
Shaozhe Hao (Hong Kong University), Kwan-Yee K. Wong (Hong Kong University)
ClassificationRecognitionTransformerImage
🎯 What it does: Proposes the ADE (Attention as Disentangler) method, which utilizes cross-attention to decouple image pairs with shared attributes or objects, and conducts zero-shot learning inference based on this.