ECCV 2024 Papers — Page 12
European Conference on Computer Vision · 2387 papers
IntrinsicAnything: Learning Diffusion Priors for Inverse Rendering Under Unknown Illumination
Xi Chen (Zhejiang University), Xiaowei Zhou (Alibaba Group)
RestorationTransformerVision Language ModelDiffusion modelImageMesh
🎯 What it does: This paper proposes an inverse rendering method based on diffusion model priors, which recovers object material properties from multi-view images under unknown static lighting.
Introducing Routing Functions to Vision-Language Parameter-Efficient Fine-Tuning with Low-Rank Bottlenecks
Tingyu Qu (KU Leuven), Marie-Francine Moens (KU Leuven)
Computational EfficiencyRepresentation LearningTransformerVision Language ModelImageTextMultimodality
🎯 What it does: Propose a routing function used in vision-language parameter-efficient fine-tuning (VL-PEFT) to better align visual and language features in low-rank bottlenecks;
Invertible Neural Warp for NeRF
Shin-Fang Chng (Adelaide University), Simon Lucey (Adelaide University)
GenerationPose EstimationOptimizationFlow-based ModelNeural Radiance FieldImage
🎯 What it does: This paper jointly optimizes the camera pose and scene reconstruction in NeRF, proposing to use invertible neural networks (INNs) to over-parameterize the camera pose, thereby improving optimization convergence and pose accuracy.
Investigating Style Similarity in Diffusion Models
Gowthami Somepalli (University of Maryland, College Park), Tom Goldstein (Columbia University)
GenerationRetrievalRepresentation LearningTransformerDiffusion modelContrastive LearningImage
🎯 What it does: This paper proposes a style descriptor CSD based on multi-label contrastive learning and constructs a large-scale real-style label dataset ContraStyles for image style retrieval and style attribution of Diffusion generative models.
IRGen: Generative Modeling for Image Retrieval
Yidan Zhang (Beijing Normal University), Baining Guo (Tsinghua University)
RetrievalTransformerImage
🎯 What it does: IRGen treats the image retrieval task as a generation task, using a sequence-to-sequence model to directly output the image identifier of the nearest neighbor of the query image, achieving end-to-end differentiable search.
IRSAM: Advancing Segment Anything Model for Infrared Small Target Detection
Mingjin Zhang (Xidian University), Jing Zhang (University of Sydney)
Object DetectionTransformerImage
🎯 What it does: Proposed the IRSAM model, modifying the original Segment Anything Model (SAM) to specialize in infrared small target detection.
Is Retain Set All You Need in Machine Unlearning? Restoring Performance of Unlearned Models with Out-Of-Distribution Images
Jacopo Bonato (Leonardo Labs), Luigi Sabetta (Leonardo Labs)
Safty and PrivacyKnowledge DistillationImage
🎯 What it does: Propose the SCAR algorithm that achieves machine model forgetting (unlearning) without requiring a retention set, by migrating the feature vectors of samples to be forgotten to the distribution of the nearest incorrect class to remove information, and using knowledge distillation techniques to maintain model performance on OOD data.
Is user feedback always informative? Retrieval Latent Defending for Semi-Supervised Domain Adaptation without Source Data
Junha Song (KAIST), Jaegul Choo (KAIST)
Domain AdaptationConvolutional Neural NetworkTransformerImageBiomedical Data
🎯 What it does: Under the source-free semi-supervised domain adaptation (SemiSDA) framework where source data is inaccessible, this paper investigates the potential negative bias in user feedback (Negatively Biased Feedback, NBF) and proposes a pluggable retrieval-based latent defending (RLD) method to mitigate the adverse effects of NBF on model adaptation performance.
Isomorphic Pruning for Vision Models
Gongfan Fang (National University of Singapore), Xinchao Wang (National University of Singapore)
Computational EfficiencyConvolutional Neural NetworkTransformerImage
🎯 What it does: Proposed a structured pruning method based on graph isomorphism called Isomorphic Pruning, which first splits the network into groups of isomorphic substructures and then performs importance ranking and pruning within each group;
Iterative Ensemble Training with Anti-Gradient Control for Mitigating Memorization in Diffusion Models
Xiao Liu (Wuhan University), Jiaxu Miao (Sun Yat-sen University)
GenerationDiffusion modelImage
🎯 What it does: Propose an iterative ensemble training with anti-gradient control (IET-AGC) framework, which weakens memorization in diffusion models by sharding data to train multiple models, periodically aggregating parameters, and dynamically removing low-loss samples during training.
ItTakesTwo: Leveraging Peer Representations for Semi-supervised LiDAR Semantic Segmentation
Yuyuan Liu (University of Adelaide), Gustavo Carneiro (University of Surrey)
SegmentationAutonomous DrivingConvolutional Neural NetworkContrastive LearningPoint Cloud
🎯 What it does: Propose the ITTakesTwo framework, which leverages different LiDAR representations from the same scene for consistency learning and contrastive learning, thereby significantly improving the performance of semi-supervised semantic segmentation.
IVTP: Instruction-guided Visual Token Pruning for Large Vision-Language Models
Kai Huang (Alibaba Group), Liang Yu (Alibaba Group)
Computational EfficiencyTransformerLarge Language ModelPrompt EngineeringVision Language ModelContrastive LearningMultimodalityBenchmark
🎯 What it does: Propose an instruction-guided visual token pruning method (IVTP), which performs two-phase pruning in large vision-language models. It reduces redundant visual patches in the visual encoder through group-wise token pruning (GTP), and further eliminates visual tokens irrelevant to the current text instruction by generating pseudo CLS using the CLIP text encoder in the early layers of the LLM.
JDT3D: Addressing the Gaps in LiDAR-Based Tracking-by-Attention
Brian Cheong (University of Toronto), Steven L Waslander (University of Toronto)
Object DetectionObject TrackingAutonomous DrivingConvolutional Neural NetworkTransformerMultimodalityPoint Cloud
🎯 What it does: This paper proposes and implements a LiDAR-based attention joint detection and tracking framework called JDT3D, and verifies its performance on nuScenes.
Joint RGB-Spectral Decomposition Model Guided Image Enhancement in Mobile Photography
Kailai Zhou (Nanjing University), Xun Cao (Nanjing University)
RestorationConvolutional Neural NetworkTransformerMixture of ExpertsImageMultimodality
🎯 What it does: This paper proposes a Joint RGB-Spectral Decomposition Model Guided Image Enhancement Framework (JDM-HDRNet), which decomposes low-resolution multispectral images (Lr-MSI) and RGB images into shadow, reflection, and material priors jointly, and introduces these priors into HDRNet for tone mapping, local brightness adaptation, and semantic mesh expert learning;
JointDreamer: Ensuring Geometry Consistency and Text Congruence in Text-to-3D Generation via Joint Score Distillation
ChenHan Jiang, Dit-Yan Yeung (Huawei Noah's Ark Lab)
GenerationDiffusion modelScore-based ModelNeural Radiance FieldTextMesh
🎯 What it does: Propose a framework called JointDreamer that generates geometry-consistent and high-fidelity 3D assets from text inputs, and introduce Joint Score Distillation (JSD) to achieve multi-view consistency.
Just a Hint: Point-Supervised Camouflaged Object Detection
Huafeng Chen (Unmanned System Research Institute, Northwestern Polytechnical University), shan gao
Object DetectionSegmentationTransformerPrompt EngineeringContrastive LearningImage
🎯 What it does: This paper proposes a camouflaged object detection method using only point annotations, achieving the transfer from single-point hints to complete target segmentation.
Kalman-Inspired Feature Propagation for Video Face Super-Resolution
Ruicheng Feng (Nanyang Technological University), Chen Change Loy (Nanyang Technological University)
Super ResolutionConvolutional Neural NetworkOptical FlowVideo
🎯 What it does: Propose a feature propagation framework called KEEP based on the Kalman filtering idea for video face super-resolution, which can maintain stable face priors in the time dimension and achieve consistent recovery of consecutive frames.
KDProR: A Knowledge-Decoupling Probabilistic Framework for Video-Text Retrieval
Xianwei Zhuang (Peking University), Yuexian Zou (Peking University)
RetrievalTransformerVision Language ModelContrastive LearningVideoTextMultimodality
🎯 What it does: This paper proposes KDProR, a video-text retrieval method achieved through knowledge decoupling and a probabilistic framework.
Kernel Diffusion: An Alternate Approach to Blind Deconvolution
Yash Sanghvi (Purdue University), Stanley Chan
RestorationConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: Propose the Kernel-Diffusion method, which samples the blur kernel using a kernel-priority diffusion model and reconstructs the sharp image by combining a differentiable non-blind solver, thus achieving blind deconvolution.
Keypoint Promptable Re-Identification
Vladimir Somers (Ecole Polytechnique Fédérale de Lausanne), Christophe De Vleeschouwer (Université Catholique de Louvain)
Pose EstimationRetrievalTransformerPrompt EngineeringImageVideoBenchmark
🎯 What it does: Propose a promptable person re-identification model named KPR that leverages semantic keypoint hints to address multi-person occlusion-induced ambiguity problems.
KeypointDETR: An End-to-End 3D Keypoint Detector
Hairong Jin (Zhejiang University), Youyi Zheng (Zhejiang University)
Pose EstimationTransformerPoint Cloud
🎯 What it does: Proposed an end-to-end 3D keypoint detection framework called KeypointDETR, which directly outputs multi-heatmap and corresponding probabilities without requiring post-processing.
KFD-NeRF: Rethinking Dynamic NeRF with Kalman Filter
Yifan Zhan (University of Tokyo), Yinqiang Zheng (Kyoto University)
GenerationNeural Radiance FieldVideo
🎯 What it does: Propose KFD-NeRF, a dynamic NeRF integrated with a Kalman filter, achieving high-quality time-consistent rendering through smooth tri-plane encoding and shallow MLP;
Kinetic Typography Diffusion Model
Seonmi Park (GIST), Hae-Gon Jeon (GIST)
GenerationTransformerDiffusion modelVideoText
🎯 What it does: Based on video diffusion models, we propose KineTy, which can generate readable and dynamic kinetic typography videos according to user text descriptions;
KMTalk: Speech-Driven 3D Facial Animation with Key Motion Embedding
Zhihao Xu (South China University of Technology), Shuangping Huang (South China University of Technology)
GenerationTransformerMeshAudio
🎯 What it does: This paper proposes a speech-driven 3D facial animation generation method based on key action embedding.
Knowledge Transfer with Simulated Inter-Image Erasing for Weakly Supervised Semantic Segmentation
Tao Chen (Nanjing University of Science and Technology), Yazhou Yao (Nanjing University of Science and Technology)
SegmentationKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a Knowledge Transfer-based Simulated Cross-image Erasing (KTSE) framework for weakly supervised semantic segmentation, addressing the under-activation and over-expansion issues in traditional adversarial erasing methods.
Knowledge-enhanced Visual-Language Pretraining for Computational Pathology
Xiao Zhou (Shanghai Artificial Intelligence Laboratory), Yan-Feng Wang (Shanghai Artificial Intelligence Laboratory)
ClassificationRetrievalKnowledge DistillationRepresentation LearningTransformerVision Language ModelContrastive LearningImageTextMultimodalityBiomedical Data
🎯 What it does: This paper investigates leveraging structured pathological knowledge to enhance visual-language pre-training, aiming to improve performance in retrieval, zero-shot classification, and whole-slide tumor subtyping tasks in computational pathology.
L-DiffER: Single Image Reflection Removal with Language-based Diffusion Model
Yuchen Hong (Peking University), Boxin Shi (Peking University)
RestorationPrompt EngineeringVision Language ModelDiffusion modelImageText
🎯 What it does: Propose a language-guided diffusion model called L-DiffER for single-image reflection removal.
Label-anticipated Event Disentanglement for Audio-Visual Video Parsing
Jinxing Zhou (Hefei University of Technology), Meng Wang (Hefei University of Technology)
SegmentationConvolutional Neural NetworkTransformerVision Language ModelVideoTextMultimodalityAudio
🎯 What it does: This paper proposes a Label Semantic Projection (LEAP) decoding strategy, which utilizes event category text embeddings to perform multi-step projections on audio-visual features, achieving decoupling and localization of multiple events.
Label-free Neural Semantic Image Synthesis
Jiayi Wang (Bosch Center for Artificial Intelligence), Anna Khoreva (Bosch Center for Artificial Intelligence)
GenerationData SynthesisRepresentation LearningTransformerVision Language ModelDiffusion modelContrastive LearningImage
🎯 What it does: Propose an unlabeled neural semantic image synthesis method that uses neural layouts extracted from pre-trained base models as conditions for image generation within the ControlNet framework.
LabelDistill: Label-guided Cross-modal Knowledge Distillation for Camera-based 3D Object Detection
Sanmin Kim (Korea Advanced Institue of Science and Technology), Dongsuk Kum (Korea Advanced Institue of Science and Technology)
Autonomous DrivingKnowledge DistillationConvolutional Neural NetworkAuto EncoderImagePoint Cloud
🎯 What it does: Proposes a label-guided cross-modal knowledge distillation method that combines LiDAR teacher models and label features to enhance camera-based 3D object detection performance, while preserving semantic information of the student model through feature partitioning.
Labeled Data Selection for Category Discovery
Bingchen Zhao (University Of Edinburgh), Oisin Mac Aodha (University Of Copenhagen)
RecognitionData-Centric LearningTransformerContrastive LearningImage
🎯 What it does: This paper studies how to select appropriate labeled data to improve the performance of category discovery on unlabeled data in the Generalized Category Discovery (GCD) task.
Lagrangian Hashing for Compressed Neural Field Representations
Shrisudhan Govindarajan, Andrea Tagliasacchi
GenerationCompressionNeural Radiance FieldAuto EncoderGaussian SplattingImagePoint Cloud
🎯 What it does: Proposes a hybrid neural field representation combining Eulerian grids and Lagrangian point clouds, called Lagrangian Hashing, for efficient compression and high-quality view synthesis
LaMI-DETR: Open-Vocabulary Detection with Language Model Instruction
Penghui Du (Beihang University), Si Liu (Baidu)
Object DetectionFederated LearningRepresentation LearningConvolutional Neural NetworkTransformerLarge Language ModelVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Propose a LaMI-DETR framework based on DETR, which generates visual descriptions using language model instructions (GPT-3.5 + T5), constructs visual concepts and cross-category relationships, thereby improving concept representation and reducing overfitting to base classes.
Lane Graph as Path: Continuity-preserving Path-wise Modeling for Online Lane Graph Construction
Bencheng Liao (Huazhong University of Science & Technology), Xinggang Wang (Huazhong University of Science & Technology)
Autonomous DrivingGraph Neural NetworkTransformerImageGraph
🎯 What it does: Propose the end-to-end LaneGAP method, which directly detects continuous road paths from camera inputs and converts paths into complete lane graphs through Path2Graph.
Language-Assisted Skeleton Action Understanding for Skeleton-Based Temporal Action Segmentation
Haoyu Ji (Harbin Institute of Technology), Honghai Liu (Harbin Institute of Technology)
SegmentationGraph Neural NetworkTransformerLarge Language ModelVision-Language-Action ModelContrastive LearningTextGraph
🎯 What it does: Propose a language-assisted skeleton action segmentation method called LaSA, which models the semantic relationships between skeleton joints and actions by leveraging language priors, and enhances action classification and boundary prediction performance through contrastive learning with aligned text descriptions.
Language-Driven 6-DoF Grasp Detection Using Negative Prompt Guidance
Toan Nguyen (FPT Software AI Center), Anh Nguyen (University of Liverpool)
Pose EstimationRobotic IntelligenceConvolutional Neural NetworkTransformerPrompt EngineeringVision Language ModelDiffusion modelTextPoint Cloud
🎯 What it does: Propose a 6-DoF grasp detection framework based on natural language prompts, which can locate and grasp specified objects in cluttered point cloud scenes according to user instructions.
Language-Driven Physics-Based Scene Synthesis and Editing via Feature Splatting
Ri-Zhao Qiu (University of California San Diego), Xiaolong Wang (University of California San Diego)
GenerationData SynthesisVision Language ModelGaussian SplattingImageTextPhysics Related
🎯 What it does: Propose Feature Splatting, which converts static 3D Gaussian scenes into editable and physically simulatable dynamic scenes using natural language queries, and achieves semi-automated scene decomposition and material assignment through semantic features.
Language-Image Pre-training with Long Captions
Kecheng Zheng (Zhejiang University), Yujun Shen (Northeastern University)
ClassificationSegmentationRetrievalTransformerLarge Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: DreamLIP significantly enhances cross-modal representation ability by regenerating long text descriptions for 30 million images and dynamically sampling subtitles from images in a contrastive learning framework for multi-positive sample training.
LaPose: Laplacian Mixture Shape Modeling for RGB-Based Category-Level Object Pose Estimation
Ruida Zhang (Tsinghua University), Xiangyang Ji (California Institute of Technology)
Pose EstimationConvolutional Neural NetworkImage
🎯 What it does: Proposed a RGB-based category-level object pose estimation framework named LaPose, which models object shape uncertainty through a Laplacian Mixture Model and introduces a scale-agnostic pose representation to achieve precise estimation of 9DoF poses.
LAPT: Label-driven Automated Prompt Tuning for OOD Detection with Vision-Language Models
Yabin Zhang (Hong Kong Polytechnic University), Lei Zhang (Hong Kong Polytechnic University)
Anomaly DetectionTransformerPrompt EngineeringContrastive LearningImageTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: Automatically learn distribution-aware prompts using only the ID category names to enhance the performance of vision-language models in OOD detection.
LaRa: Efficient Large-Baseline Radiance Fields
Anpei Chen (University of Tübingen), Andreas Geiger (University of Tübingen)
GenerationConvolutional Neural NetworkTransformerNeural Radiance FieldGaussian SplattingImageMesh
🎯 What it does: Propose LaRa, a Feed-Forward model based on 2D Gaussian projection, which can reconstruct high-quality 360° illumination fields and meshes using only a small number (e.g., 4) of large-baseline views.
Large Motion Model for Unified Multi-Modal Motion Generation
Mingyuan Zhang, Ziwei Liu
GenerationTransformerDiffusion modelVideoTextMultimodalityAudio
🎯 What it does: Developed a unified large-scale multimodal motion generation model, LMM, trained on multi-task and multimodal data, achieving performance comparable to or better than specialized models on 9 public benchmarks.
Large-Scale Multi-Hypotheses Cell Tracking Using Ultrametric Contours Maps
Jordão Bragantini (Chan Zuckerberg Biohub), Loïc A Royer (Chan Zuckerberg Biohub)
Object TrackingSegmentationConvolutional Neural NetworkBiomedical DataBenchmark
🎯 What it does: Hierarchical segmentation is used to generate cell candidate regions, and integer linear programming (ILP) selects non-overlapping segmentations across adjacent time frames to enable large-scale 3D cell tracking;
Large-scale Reinforcement Learning for Diffusion Models
Yinan Zhang (Pinterest ATG), Dmitry Kislyuk (Pinterest ATG)
GenerationOptimizationReinforcement Learning from Human FeedbackReinforcement LearningDiffusion modelImageText
🎯 What it does: Propose a scalable reinforcement learning framework for multi-objective fine-tuning of large-scale diffusion models, enhancing the quality, composition, and fairness of text-to-image generation.
LASS3D: Language-Assisted Semi-Supervised 3D Semantic Segmentation with Progressive Unreliable Data Exploitation
Jianan Li (University of Chinese Academy of Sciences), Qiulei Dong (University of Chinese Academy of Sciences)
SegmentationLarge Language ModelVision Language ModelContrastive LearningTextMultimodalityPoint Cloud
🎯 What it does: Proposes a language-assisted semi-supervised 3D semantic segmentation method called LASS3D, which leverages a large language-visual model to generate multi-layer text descriptions and integrates them into point cloud features, while designing a strategy to progressively utilize unreliable points.
Latent Diffusion Prior Enhanced Deep Unfolding for Snapshot Spectral Compressive Imaging
Zongliang Wu (Zhejiang University), Xin Yuan (Westlake University)
RestorationTransformerDiffusion modelImage
🎯 What it does: This paper proposes a Deep Unfolding Network (DUN) combining the prior of Latent Diffusion Models (LDM) for three-dimensional spectral image reconstruction in single-shot compressive spectral imaging (CASSI).
Latent Guard: a Safety Framework for Text-to-image Generation
Runtao Liu (Hong Kong University of Science and Technology), Fabio Pizzati (University of Oxford)
GenerationSafty and PrivacyLarge Language ModelVision Language ModelContrastive LearningText
🎯 What it does: Proposed the Latent Guard framework, which achieves safe filtering in text-to-image generation by learning latent space detection of blacklisted concepts on the text encoder.
Latent-INR: A Flexible Framework for Implicit Representations of Videos with Discriminative Semantics
Shishira R Maiya (University of Maryland), Abhinav Shrivastava (University of Maryland)
RetrievalCompressionRepresentation LearningVision Language ModelNeural Radiance FieldVideo
🎯 What it does: Proposes a Latent-INR framework that decouples space and time, using frame-level latent vectors and a hypernetwork to generate video INR, enabling multi-task functions such as compression, interpolation, retrieval, and dialogue.
LatentEditor: Text Driven Local Editing of 3D Scenes
Umar Khalid (University of Central Florida), Chen Chen (University of Central Florida)
Image TranslationGenerationLarge Language ModelPrompt EngineeringVision Language ModelDiffusion modelNeural Radiance FieldImageTextMultimodalityBenchmark
🎯 What it does: The study proposes the LatentEditor framework, which enables precise local editing of 3D NeRF scenes using only text prompts.
latentSplat: Autoencoding Variational Gaussians for Fast Generalizable 3D Reconstruction
Christopher Wewer (Max Planck Institute for Informatics), Jan E. Lenssen
GenerationData SynthesisTransformerNeural Radiance FieldAuto EncoderGenerative Adversarial NetworkGaussian SplattingVideo
🎯 What it does: Propose a lightweight variational Gaussian-based autoencoder (latentSplat) that enables fast generation of high-quality 3D reconstructions and novel view rendering from two views.
LATTE3D: Large-scale Amortized Text-To-Enhanced3D Synthesis
Kevin Xie (Nvidia), Xiaohui Zeng (Nvidia)
GenerationData SynthesisDiffusion modelTextPoint CloudMesh
🎯 What it does: Built a large-scale amortized text-to-enhanced 3D generation framework called Latte3D, capable of generating high-quality, textured 3D meshes from a single text prompt within 400ms, and supports fast test-time optimization and stylization.
LaWa: Using Latent Space for In-Generation Image Watermarking
Ahmad Rezaei (University of British Columbia), Yong Zhang (Huawei Technologies Canada Co. Ltd.)
GenerationSafty and PrivacyDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: Implemented LaWa, an in-generation image watermarking method within Latent Diffusion Models (LDM), which can directly embed watermarks during the generation process.
Layer-Wise Relevance Propagation with Conservation Property for ResNet
Seitaro Otsuki (Keio University), Komei Sugiura (Keio University)
ClassificationExplainability and InterpretabilityConvolutional Neural NetworkImage
🎯 What it does: Extend Layer-wise Relevance Propagation (LRP) for the ResNet architecture by introducing Relevance Splitting at residual blocks and skip connections, and incorporating Heat Quantization to balance the explanation heatmaps.
LayerDiff: Exploring Text-guided Multi-layered Composable Image Synthesis via Layer-Collaborative Diffusion Model
Runhui Huang (Sun Yat-sen University), Hang Xu (Huawei Noah's Ark Lab)
GenerationData SynthesisDiffusion modelImageText
🎯 What it does: Propose LayerDiff, a text-driven multi-layer composable image synthesis method based on layer-cooperative diffusion models.
Layered Rendering Diffusion Model for Controllable Zero-Shot Image Synthesis
Zipeng Qi (Beihang University), Fei Ye (Beihang University)
GenerationDiffusion modelImageText
🎯 What it does: Proposed a two-stage hierarchical rendering diffusion model, LRDiff, which can achieve high-precision spatially controlled image synthesis with text + bounding boxes/instance masks without model fine-tuning.
LayeredFlow: A Real-World Benchmark for Non-Lambertian Multi-Layer Optical Flow
Hongyu Wen (Princeton University), Jia Deng (Princeton University)
Data SynthesisConvolutional Neural NetworkRecurrent Neural NetworkOptical FlowImageBenchmark
🎯 What it does: Propose the LayeredFlow benchmark and synthetic dataset, providing multi-layer non-Laplacian optical flow annotations, and design a multi-layer optical flow prediction network based on RAFT.
Layout-Corrector: Alleviating Layout Sticking Phenomenon in Discrete Diffusion Model
Shoma Iwai (Tohoku University), Shinichiro Omachi (Tohoku University)
GenerationTransformerDiffusion modelImage
🎯 What it does: Proposes an external evaluation module called Layout-Corrector, which can detect and re-mask discordant elements during layout generation by discrete diffusion models (DDM), prompting the model to regenerate more harmonious layouts.
LayoutDETR: Detection Transformer Is a Good Multimodal Layout Designer
Ning Yu (Salesforce Research), Ran Xu (Salesforce Research)
GenerationTransformerAuto EncoderGenerative Adversarial NetworkImageTextMultimodality
🎯 What it does: Proposed a multi-modal graphic layout generation framework called LayoutDETR based on DETR, which can generate layouts that meet design requirements given background images and foreground images/text.
LayoutFlow: Flow Matching for Layout Generation
Julian Jorge Andrade Guerreiro (University of Tokyo), Hideki Nakayama (University of Tokyo)
GenerationTransformerFlow-based ModelImageOrdinary Differential Equation
🎯 What it does: This paper proposes LayoutFlow, a layout generation model based on flow matching, which uses ODE-based flows to migrate sampling points from a simple prior distribution to the target layout distribution, significantly reducing the number of sampling steps and accelerating inference;
Lazy Diffusion Transformer for Interactive Image Editing
Yotam Nitzan (Adobe Research), Michaël Gharbi (Adobe Research)
GenerationTransformerDiffusion modelScore-based ModelAuto EncoderImage
🎯 What it does: This paper proposes an interactive image editing framework called LazyDiffusion, which achieves efficient local image updates by one-time encoding the entire image context and then generating only the pixels in the masked regions during multi-step diffusion decoding.
LCM-Lookahead for Encoder-based Text-to-Image Personalization
Rinon Gal (NVIDIA), Danny Cohen-Or
GenerationDiffusion modelAuto EncoderImageTextMultimodality
🎯 What it does: This paper proposes a method that utilizes an LCM (Latent Consistency Model) to generate high-quality preview images, incorporating an image space identity loss into encoder training to enhance the personalization and identity preservation capabilities of text-to-image models.
Learn from the Learnt: Source-Free Active Domain Adaptation via Contrastive Sampling and Visual Persistence
Mengyao Lyu (Tsinghua University), Guiguang Ding (Tsinghua University)
Domain AdaptationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes a source-free active domain adaptation (SFADA) framework called Learn from the Learnt (LFTL), which addresses scenarios where source data is unavailable and only a minimal amount of target annotations are available.
Learn to Memorize and to Forget: A Continual Learning Perspective of Dynamic SLAM
Baicheng Li, Hongbin Zha (Peking University)
Neural Radiance FieldSimultaneous Localization and MappingImage
🎯 What it does: This paper proposes a dense neural SLAM framework for continuous learning in dynamic environments, utilizing experience replay and a learnable dynamic object classifier to achieve 'memory' and 'forgetting' of the map.
Learn to Optimize Denoising Scores: A Unified and Improved Diffusion Prior for 3D Generation
Xiaofeng Yang (Nanyang Technological University), Guosheng Lin (Institute for Infocomm Research, A*STAR)
GenerationOptimizationDiffusion modelNeural Radiance FieldImageTextPoint CloudMeshBenchmark
🎯 What it does: In 3D generation tasks, a unified framework called LODS is proposed, which iteratively optimizes 3D models and diffusion priors to improve generation quality.
Learn to Preserve and Diversify: Parameter-Efficient Group with Orthogonal Regularization for Domain Generalization
Jiajun Hu (Nanjing University), Yang Gao (Nanjing University)
Domain AdaptationRepresentation LearningTransformerImageBenchmark
🎯 What it does: Proposed a Parameter-Efficient Group with Orthogonal Regularization (PEGO) framework, which injects multiple LoRA modules into pre-trained Vision Transformers and applies orthogonal regularization to maintain pre-trained knowledge while enabling diverse learning, thereby enhancing domain generalization performance.
Learned HDR Image Compression for Perceptually Optimal Storage and Display
Peibei Cao (City University of Hong Kong), Kede Ma (Xellar Biosystems)
CompressionConvolutional Neural NetworkImage
🎯 What it does: Designed and implemented an end-to-end learning HDR image compression framework called EPIC-HDR, achieving dual optimization for HDR image storage and display, generating LDR images compatible with LDR displays and side information for HDR reconstruction.
Learned Image Enhancement via Color Naming
David Serrano-Lozano (Computer Vision Center), Javier Vazquez-Corral (Computer Vision Center)
RestorationConvolutional Neural NetworkImage
🎯 What it does: Decompose images through color naming, learn Bezier curves corresponding to each color for global color adjustment, and obtain an image enhancement model by fusing the results using an attention mechanism.
Learned Neural Physics Simulation for Articulated 3D Human Pose Reconstruction
Misha Andriluka, Cristian Sminchisescu (Google DeepMind)
Pose EstimationComputational EfficiencyRecurrent Neural NetworkVideoPoint CloudTime SeriesPhysics Related
🎯 What it does: Propose a differentiable, fast neural network physics simulation framework called LARP for learning and simulating joint-driven human motion with contact.
Learned Rate Control for Frame-Level Adaptive Neural Video Compression via Dynamic Neural Network
Chenhao Zhang (Peking University), Wei Gao (Peking University)
CompressionAuto EncoderVideo
🎯 What it does: Propose a neural video compression framework combining a dynamic routing autoencoder and a rate control agent (DRA+RCA), which achieves variable bitrate and precise rate control per frame while optimizing rate-distortion-complexity.
Learning 3D Geometry and Feature Consistent Gaussian Splatting for Object Removal
Yuxin Wang (Hong Kong University of Science and Technology), Dan Xu (Hong Kong University of Science and Technology)
RestorationGenerationDepth EstimationNeural Radiance FieldGaussian SplattingImage
🎯 What it does: Designed a 3D Gaussian Splatting-based object removal method called GScream, achieving the removal of specified objects in already captured scenes while maintaining geometric and texture consistency.
Learning 3D-aware GANs from Unposed Images with Template Feature Field
Xinya Chen (Zhejiang University), Yiyi Liao (Zhejiang University)
GenerationPose EstimationNeural Radiance FieldGenerative Adversarial NetworkContrastive LearningImage
🎯 What it does: This paper proposes a method called Template Feature Field (TeFF), which can train 3D-aware GANs from unannotated images without camera pose annotations, achieving real-time camera pose estimation during training and learning complete 3D shapes.
Learning a Dynamic Privacy-preserving Camera Robust to Inversion Attacks
Jiacheng Cheng (University of California San Diego), Nuno Vasconcelos (University of California San Diego)
RecognitionObject DetectionPose EstimationSafty and PrivacyImage
🎯 What it does: Designed a dynamic privacy-preserving camera (DyPP) that randomly samples variable point spread functions (PSF) per image to prevent PSF inversion attacks while maintaining performance in computer vision tasks.
Learning Anomalies with Normality Prior for Unsupervised Video Anomaly Detection
Haoyue Shi (Xi'an Jiaotong University), Wei Tang (University of Illinois Chicago)
Anomaly DetectionConvolutional Neural NetworkVideo
🎯 What it does: Propose an unsupervised video anomaly detection method that utilizes the first and last frames of a video as prior knowledge of normality, and combines normal propagation with loss-weighted self-training to achieve pseudo-label learning.
Learning by Aligning 2D Skeleton Sequences and Multi-Modality Fusion
Quoc-Huy Tran (Retrocausal, Inc.), Zeeshan Zia
RecognitionTransformerContrastive LearningVideoMultimodality
🎯 What it does: Propose a self-supervised video alignment framework based on 2D skeleton heatmaps for fine-grained human action understanding.
Learning Camouflaged Object Detection from Noisy Pseudo Label
Jin Zhang (Beijing Institute of Technology), Fahad Shahbaz Khan (Mohamed bin Zayed University of Artificial Intelligence)
Object DetectionSegmentationConvolutional Neural NetworkTransformerImageBenchmark
🎯 What it does: Propose a weak semi-supervised hidden object detection framework (WSSCOD), which generates high-quality pseudo labels using box prompts and employs a small amount of pixel-level annotations for supervision. A noise correction loss L_NC is introduced to train the main network against pseudo label noise.
Learning Chain of Counterfactual Thought for Bias-Robust Vision-Language Reasoning
Yifeng Zhang (University of Minnesota), Qi Zhao (University of Minnesota)
Prompt EngineeringVision Language ModelDiffusion modelImageTextMultimodalityRetrieval-Augmented GenerationChain-of-Thought
🎯 What it does: This paper proposes the CoBRa dataset and the CoCT method, aiming to enhance the bias-robust reasoning ability of vision-language models through counterfactual thinking.
Learning Cross-hand Policies of High-DOF Reaching and Grasping
Qijin She (National University of Defense Technology), Kai Xu (National University of Defense Technology)
Robotic IntelligenceTransformerReinforcement LearningPoint Cloud
🎯 What it does: This paper proposes a two-stage framework that first uses a unified keypoint displacement-based strategy model to predict cross-hand grasping actions, and then employs a hand-specific adapter to convert these displacements into specific joint control, achieving transferable high-degree-of-freedom grasping across different fingertip hands.
Learning Differentially Private Diffusion Models via Stochastic Adversarial Distillation
Bochao Liu (Chinese Academy of Sciences), Shiming Ge (Chinese Academy of Sciences)
GenerationSafty and PrivacyKnowledge DistillationDiffusion modelImage
🎯 What it does: Based on a teacher diffusion model without differential privacy, a diffusion generative model with differential privacy guarantees (DP-SAD) is trained through random adversarial distillation (with a discriminator) and gradient clipping plus noise addition.
Learning Diffusion Models for Multi-View Anomaly Detection
Chieh Liu (National Tsing Hua University), Tyng-Luh Liu (Academia Sinica)
Anomaly DetectionDiffusion modelMultimodality
🎯 What it does: Studies multi-view anomaly detection by utilizing RGB images under seven different lighting conditions and corresponding 3D normal maps for anomaly localization.
Learning Dual-Level Deformable Implicit Representation for Real-World Scale Arbitrary Super-Resolution
Zhiheng Li (Tsinghua University), Jie Zhou (Tsinghua University)
Super ResolutionConvolutional Neural NetworkImage
🎯 What it does: To address super-resolution tasks with arbitrary scaling ratios in real-world scenarios, the authors constructed the RealArbiSR dataset, which includes both integer and non-integer scaling ratios, and proposed the Dual-Layer Deformable Implicit Representation (DDIR) model. This model simultaneously learns image-level and pixel-level degradation deformations, enabling arbitrary-scale super-resolution under real-world noise degradation conditions.
Learning Equilibrium Transformation for Gamut Expansion and Color Restoration
Jun Xiao (Hong Kong Polytechnic University), Kin-Man Lam (Hong Kong Polytechnic University)
RestorationConvolutional Neural NetworkImage
🎯 What it does: Proposed a lightweight equipotential transformation network for color gamut expansion and color restoration, without relying on any external metadata, implemented through a root search framework and implicit loop mechanism;
Learning Exhaustive Correlation for Spectral Super-Resolution: Where Spatial-Spectral Attention Meets Linear Dependence
Hongyuan Wang, Youliang Yan (Huawei Noah's Ark Lab)
Super ResolutionTransformerImage
🎯 What it does: Proposes the Exhaustive Correlation Transformer (ECT) to achieve spectral super-resolution restoration from RGB images to hyperspectral images.
Learning from the Web: Language Drives Weakly-Supervised Incremental Learning for Semantic Segmentation
Chang Liu (Xidian University), Yi Niu (Xidian University)
SegmentationKnowledge DistillationRepresentation LearningConvolutional Neural NetworkTransformerVision Language ModelGenerative Adversarial NetworkImageText
🎯 What it does: Achieved weakly supervised incremental semantic segmentation learning using only web images and preserved old class knowledge.
Learning High-resolution Vector Representation from Multi-Camera Images for 3D Object Detection
Zhili Chen (HKUST), Qifeng Chen (HKUST)
Object DetectionAutonomous DrivingComputational EfficiencyRepresentation LearningTransformerImage
🎯 What it does: Propose VectorFormer, a 3D object detection framework based on multi-camera images, which utilizes high-resolution vector representation (Vector Query) combined with low-resolution BEV to achieve fine-grained modeling of 3D space while maintaining efficient inference.
Learning Local Pattern Modularization for Point Cloud Reconstruction from Unseen Classes
Chao Chen (Tsinghua University), Zhizhong Han (Wayne State University)
GenerationAuto EncoderPoint Cloud
🎯 What it does: This paper proposes to achieve point cloud reconstruction for unseen categories by learning local pattern modularization in the object-centered coordinate system, resulting in high-fidelity reconstruction outcomes.
Learning Modality-agnostic Representation for Semantic Segmentation from Any Modalities
Xu Zheng (Hong Kong University of Science and Technology), Lin Wang (Hong Kong University of Science and Technology)
SegmentationKnowledge DistillationRepresentation LearningConvolutional Neural NetworkVision Language ModelMultimodality
🎯 What it does: Learn modality-agnostic representations through knowledge distillation from multi-modal vision-language models, enabling semantic segmentation under any modality combination and visual conditions.
Learning Multimodal Latent Generative Models with Energy-Based Prior
Shiyu Yuan (Stevens Institute of Technology), Tian Han (Stevens Institute of Technology)
GenerationMixture of ExpertsAuto EncoderMultimodalityStochastic Differential Equation
🎯 What it does: Propose a multi-modal latent generative model based on energy-based models (EBM), leveraging EBM prior to enhance latent space expressiveness, achieving semantically consistent generation and cross-modal generation.
Learning Natural Consistency Representation for Face Forgery Video Detection
Daichi Zhang (Chinese Academy of Sciences), Shiming Ge (Chinese Academy of Sciences)
Anomaly DetectionConvolutional Neural NetworkTransformerContrastive LearningVideo
🎯 What it does: This study proposes the NACO method, which extracts frame-level spatial features using a CNN encoder and learns long-term spatial consistency representations via a Transformer, for self-supervised detection of facial forged videos.
Learning Neural Deformation Representation for 4D Dynamic Shape Generation
Gyojin Han (Korea Advanced Institute of Science and Technology), Junmo Kim (Korea Advanced Institute of Science and Technology)
GenerationData SynthesisPose EstimationRecurrent Neural NetworkDiffusion modelVideoPoint CloudMesh
🎯 What it does: Propose a 4D dynamic shape generation framework that combines neural SDF with neural deformation representation, training a diffusion model to generate shapes and motion through latent vectors, achieving high-quality, temporally consistent, and efficient 4D shape generation, conditional generation, and motion retargeting.
Learning Neural Volumetric Pose Features for Camera Localization
Jingyu Lin (University of Science and Technology of China), Jieping Ye (University of Science and Technology of China)
Pose EstimationNeural Radiance FieldContrastive LearningImage
🎯 What it does: Proposes a neural volume pose feature PoseMap based on incremental NeRF to enhance the camera localization accuracy of absolute pose regression (APR) methods;
Learning Non-Linear Invariants for Unsupervised Out-of-Distribution Detection
Lars Doorenbos (University of Bern), Pablo Márquez Neila
Anomaly DetectionRepresentation LearningConvolutional Neural NetworkFlow-based ModelImageTabular
🎯 What it does: Proposed an unsupervised OOD detection framework based on a Volume Preserving Network (VPN), leveraging nonlinear invariant learning to identify out-of-distribution samples.
Learning Pseudo 3D Guidance for View-consistent Texturing with 2D Diffusion
Kehan Li (Peking University), Jie Chen (Peking University)
GenerationData SynthesisDiffusion modelScore-based ModelNeural Radiance FieldTextMesh
🎯 What it does: Utilize a pre-trained deep guidance diffusion model to generate high-quality 3D textures by learning rough but globally consistent pseudo-3D textures, guiding multi-view image generation to achieve high-quality 3D textures that are consistent with text descriptions and viewpoints.
Learning Quantized Adaptive Conditions for Diffusion Models
Yuchen Liang (Peking University), Hanting Chen (Peking University)
GenerationDiffusion modelRectified FlowImageOrdinary Differential Equation
🎯 What it does: Propose a quantized adaptive conditional encoder that reduces the curvature of ODE trajectories through lightweight quantization encoding, thereby improving the sampling quality of diffusion models with minimal function evaluations.
Learning Representation for Multitask Learning through Self-Supervised Auxiliary Learning
Seokwon Shin (Dongguk University), Youngdoo Son (Dongguk University)
ClassificationSegmentationRepresentation LearningImage
🎯 What it does: Propose a regularization method called Dummy Gradient Norm Regularization (DGR), aiming to enhance the universality of representations generated by the shared encoder in multi-task learning.
Learning Representations from Foundation Models for Domain Generalized Stereo Matching
Yongjian Zhang (Sun Yat-Sen University), Yulan Guo (Sun Yat-Sen University)
Depth EstimationDomain AdaptationRepresentation LearningTransformerContrastive LearningImage
🎯 What it does: Propose the FormerStereo framework, integrating ViT-based foundation models (e.g., DINOv2) into the stereo matching pipeline to achieve cross-domain zero-shot stereo matching;
Learning Representations of Satellite Images From Metadata Supervision
Jules Bourcier (Preligens), Jocelyn Chanussot (Univ. Grenoble Alpes)
ClassificationRepresentation LearningContrastive LearningImageMultimodality
🎯 What it does: Pre-train images using metadata such as time and location from satellite images to obtain more semantically expressive visual features.
Learning Scalable Model Soup on a Single GPU: An Efficient Subspace Training Strategy
Tao Li (Shanghai Jiao Tong University), James Kwok (Hong Kong University of Science and Technology)
ClassificationOptimizationComputational EfficiencyConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes an efficient method to construct the model 'soup'—Memory-Efficient Hyperplane Learned Soup (MEHL-Soup)—on a single GPU, addressing the memory and computational bottlenecks of traditional Learned-Soup approaches.
Learning Semantic Latent Directions for Accurate and Controllable Human Motion Prediction
Guowei Xu (University of Electronic Science and Technology of China), Lixin Duan (University of Electronic Science and Technology of China)
GenerationGraph Neural NetworkTime SeriesSequentialBenchmark
🎯 What it does: This paper proposes a method called Semantic Latent Directions (SLD) to construct a semantic latent space in stochastic human motion prediction (SHMP), enabling more accurate, controllable, and diverse future motion prediction.
Learning the Unlearned: Mitigating Feature Suppression in Contrastive Learning
Jihai Zhang (Chinese University of Hong Kong), Bryan Hooi (National University of Singapore)
Representation LearningContrastive LearningImageMultimodality
🎯 What it does: Proposes a multi-stage contrastive learning framework (MCL), which gradually uncovers suppressed features through feature-aware negative sampling, and retains all learned features by cross-stage concatenation in the final stage, addressing feature suppression issues in both single-modal and multi-modal contrastive learning.
Learning to Adapt SAM for Segmenting Cross-domain Point Clouds
Xidong Peng (ShanghaiTech University), Yuexin Ma (ShanghaiTech University)
SegmentationDomain AdaptationKnowledge DistillationConvolutional Neural NetworkTransformerContrastive LearningPoint Cloud
🎯 What it does: Align the point cloud features of the source domain and target domain using SAM's general feature space to achieve unsupervised cross-domain segmentation of 3D point clouds.