检索结果-内蒙古大学图书馆

Permutation Learning with Only N Parameters: From SoftSort to Self-Organizing Gaussians

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Barthel, Kai Uwe Barthel, Florian Tim Eisert, Peter Visual Computing Group HTW Berlin Berlin Germany Vision and Imaging Technologies Fraunhofer HHI Berlin Germany

Sorting and permutation learning are key concepts in optimization and machine learning, especially when organizing high-dimensional data into meaningful spatial layouts. The Gumbel-Sinkhorn method, while effective, requires N2 parameters to determine a full permutation matrix, making it computationally expensive for large datasets. Low-rank matrix factorization approximations reduce memory requirements to 2NM (with M N), but they still struggle with very large problems. SoftSort, by providing a continuous relaxation of the argsort operator, allows differentiable 1D sorting, but it faces challenges with multidimensional data and complex permutations. In this paper, we present a novel method for learning permutations using only N parameters, which dramatically reduces storage costs. Our approach builds on SoftSort, but extends it by iteratively shuffling the N indices of the elements to be sorted through a separable learning process. This modification significantly improves sorting quality, especially for multidimensional data and complex optimization criteria, and outperforms pure SoftSort. Our method offers improved memory efficiency and scalability compared to existing approaches, while maintaining high-quality permutation learning. Its dramatically reduced memory requirements make it particularly well-suited for large-scale optimization tasks, such as"Self-Organizing Gaussians", where efficient and scalable permutation learning is critical. © 2025, CC BY-NC-ND.

关键词： Matrix factorization

CREATING SORTED GRID LAYOUTS WITH GRADIENT-BASED OPTIMIZATION

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Barthel, Kai Uwe Barthel, Florian Tim Eisert, Peter Hezel, Nico Schall, Konstantin Visual Computing Group HTW Berlin Berlin Germany Fraunhofer HHI Berlin Germany HU Berlin Berlin Germany

visually sorted grid layouts provide an efficient method for organizing high-dimensional vectors in two-dimensional space by aligning spatial proximity with similarity relationships. This approach facilitates the effective sorting of diverse elements ranging from data points to images, and enables the simultaneous visualization of a significant number of elements. However, sorting data on two-dimensional grids is a challenge due to its high complexity. Even for a small 8-by-8 grid with 64 elements, the number of possible arrangements exceeds 1.3 · 1089 - more than the number of atoms in the universe - making brute-force solutions impractical. Although various methods have been proposed to address the challenge of determining sorted grid layouts, none have investigated the potential of gradient-based optimization. In this paper, we present a novel method for grid-based sorting that exploits gradient optimization for the first time. We introduce a novel loss function that balances two opposing goals: ensuring the generation of a"valid" permutation matrix, and optimizing the arrangement on the grid to reflect the similarity between vectors, inspired by metrics that assess the quality of sorted grids. While learning-based approaches are inherently computationally complex, our method shows promising results in generating sorted grid layouts with superior sorting quality compared to existing techniques. Copyright © 2025, The Authors. All rights reserved.

关键词： Vector spaces

Scale-wise Bidirectional Alignment Network for Referring Remote Sensing Image Segmentation

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Li, Kun Vosselman, George Yang, Michael Ying The Faculty of ITC University of Twente Netherlands The Visual Computing Group Department of Computer Science University of Bath United Kingdom

The goal of referring remote sensing image segmentation (RRSIS) is to extract specific pixel-level regions within an aerial image via a natural language expression. Recent advancements, particularly Transformer-based fusion designs, have demonstrated remarkable progress in this domain. However, existing methods primarily focus on refining visual features using language-aware guidance during the cross-modal fusion stage, neglecting the complementary vision-to-language flow. This limitation often leads to irrelevant or suboptimal representations. In addition, the diverse spatial scales of ground objects in aerial images pose significant challenges to the visual perception capabilities of existing models when conditioned on textual inputs. In this paper, we propose an innovative framework called Scale-wise Bidirectional Alignment Network (SBANet) to address these challenges for RRSIS. Specifically, we design a Bidirectional Alignment Module (BAM) with learnable query tokens to selectively and effectively represent visual and linguistic features, emphasizing regions associated with key tokens. BAM is further enhanced with a dynamic feature selection block, designed to provide both macro- and micro-level visual features, preserving global context and local details to facilitate more effective cross-modal interaction. Furthermore, SBANet incorporates a text-conditioned channel and spatial aggregator to bridge the gap between the encoder and decoder, enhancing cross-scale information exchange in complex aerial scenarios. Extensive experiments demonstrate that our proposed method achieves superior performance in comparison to previous state-of-the-art methods on the RRSIS-D and RefSegRS datasets, both quantitatively and qualitatively. The code will be released after publication. Copyright © 2025, The Authors. All rights reserved.

关键词： Image segmentation

PraK Tool V3: Enhancing Video Item Search Using Localized Text and Texture Queries 31st

学校读者我要写书评

暂无评论

PraK Tool V3: Enhancing Video Item Search Using Localized Te...

31st International Conference on Multimedia Modeling, MMM 2025

作者： Stroh, Michael Kloda, Vojtěch Verner, Benjamin Vopálková, Zuzana Buchmüller, Raphael Jäckl, Bastian Hajko, Jakub Lokoč, Jakub Visual Computing Department of Computer Science Faculty of Natural Sciences Konstanz University Konstanz Germany SIRET Research Group Department of Software Engineering Faculty of Mathematics and Physics Charles University Prague Czech Republic Data Analysis and Visualization Department of Computer Science Faculty of Natural Sciences Konstanz University Konstanz Germany

ISBN: (纸本)9789819620739

We present a third version of the PraK system designed around an effective text-image and image-image search model. The system integrates sub-image search options for localized context search for CLIP and image color/texture and interfaces tailored to specific task types. A user study shows the benefits of sub-image search in homogeneous video collections. We further enhance the user experience of the PraK framework by re-designing the frontend and introducing task-specific views. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.

关键词： Video analysis

3D floor plan recovery from overlapping spherical images

学校读者我要写书评

暂无评论

Computational visual Media 2018年第4期4卷 367-383页

作者： Giovanni Pintore Fabio Ganovelli Ruggero Pintus Roberto Scopigno Enrico Gobbetti CRS4 Visual Computing Group CNR-ISTI Visual Computing Group

We present a novel approach to automatically recover, from a small set of partially overlapping spherical images, an indoor structure representation in terms of a 3D floor plan registered with a set of 3D environment maps. We introduce several improvements over previous approaches based on color and spatial reasoning exploiting Manhattan world priors. In particular, we introduce a new method for geometric context extraction based on a 3D facet representation,which combines color distribution analysis of individual images with sparse multi-view clues. We also introduce an efficient method to combine the facets from different viewpoints in a single consistent model, taking into the reliability of the facet information. The resulting capture and reconstruction pipeline automatically generates 3D multi-room environments in cases where most previous approaches fail, e.g., in the presence of hidden corners and large clutter, without the need for additional dense 3D data or tools. We demonstrate the effectiveness and performance of our approach on different real-world indoor scenes. Our test data is available to allow further studies and comparisons.

关键词： indoor reconstruction spherical panoramic cameras 360 degree photography multi-room environments

Rapidvol: Rapid Reconstruction of 3D Ultrasound Volumes from Sensorless 2D Scans

学校读者我要写书评

暂无评论

Rapidvol: Rapid Reconstruction of 3D Ultrasound Volumes from...

IEEE International Symposium on Biomedical Imaging

作者： Mark C. Eid Pak-Hei Yeung Madeleine K. Wyburd João F. Henriques Ana I.L. Namburete Visual Geometry Group University of Oxford U.K. Oxford Machine Learning in Neuroimaging Lab University of Oxford U.K. College of Computing and Data Science Nanyang Technological University Singapore

ISBN: (数字)9798331520526

ISBN: (纸本)9798331520533

Two-dimensional (2D) freehand ultrasonography is a widely used medical imaging modality, particularly in obstetrics and gynaecology. However, it only captures 2D cross-sectional views of inherently 3D anatomies, losing valuable contextual information. As an alternative to costly 3D ultrasound (US) scanners, 3D volumes can be artificially reconstructed from 2D scans, but this is usually prohibitively slow. Hence, we propose RapidVol: a neural representation framework to speed up slice-to-volume US reconstruction. We use tensor-rank decomposition to decompose the typical 3D volume into tri-planes, which are stored alongside a small neural network. With a set of 2D US scans and their estimated 3D orientation, RapidVol can achieve complete 3D reconstruction. To evaluate our method, we form reconstructions from real fetal brain scans, and then request novel cross-sectional views. Compared to prior fully implicit (e.g. neural radiance field) approaches, our method is over 3x quicker, 46% more accurate, and more robust to errors in pose estimation. We also demonstrate that further speed-up is achievable by reconstructing from a structural prior rather than from random initialisation.

关键词： Three-dimensional displays Accuracy Two-dimensional displays Pose estimation Ultrasonography Neural radiance field Probes Obstetrics Image reconstruction Biomedical imaging

Beyond Time and Accuracy: Strategies in visual Problem-Solving 25

学校读者我要写书评

暂无评论

Beyond Time and Accuracy: Strategies in Visual Problem-Solvi...

Proceedings of the 2025 CHI Conference on Human Factors in computing Systems

作者： Eric Mörth Zona Kostic Nils Gehlenborg Hanspeter Pfister Johanna Beyer Carolina Nobre Department of Biomedical Informatics Harvard Medical School Boston Massachusetts USA SEAS Harvard University Boston Massachusetts USA Visual Computing Group Harvard University Cambridge Massachusetts USA University of Toronto Toronto Ontario Canada

Named Entity Recognition for Serbian Legal Documents: Design, Methodology and Dataset Development

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Kalušev, Vladimir Brkljač, Branko Visual Computing & Perception Group The Institute for Artificial Intelligence Research and Development of Serbia Novi Sad Serbia Dept. of Power Electronic and Telecommunication Engineering Faculty of Technical Sciences University of Novi Sad Novi Sad Serbia

Recent advancements in the field of natural language processing (NLP) and especially large language models (LLMs) and their numerous applications have brought research attention to design of different document processing tools and enhancements in the process of document archiving, search and retrieval. Domain of official, legal documents is especially interesting due to vast amount of data generated on the daily basis, as well as the significant community of interested practitioners (lawyers, law offices, administrative workers, state institutions and citizens). Providing efficient ways for automation of everyday work involving legal documents is therefore expected to have significant impact in different fields. In this work we present one LLM based solution for Named Entity Recognition (NER) in the case of legal documents written in Serbian language. It leverages on the pre-trained bidirectional encoder representations from transformers (BERT), which had been carefully adapted to the specific task of identifying and classifying specific data points from textual content. Besides novel dataset development for Serbian language (involving public court rulings), presented system design and applied methodology, the paper also discusses achieved performance metrics and their implications for objective assessment of the proposed solution. Performed cross-validation tests on the created manually labeled dataset with mean F1 score of 0.96 and additional results on the examples of intentionally modified text inputs confirm applicability of the proposed system design and robustness of the developed NER *** Codes 68T10, 68T30, 68T35, 68T50, 91F20 © 2025, CC BY.

关键词： Natural language processing systems

Optimality vs. Generality: Performance Assessment of Meta-heuristics in Educational Timetabling

学校读者我要写书评

暂无评论

IEEE Access 2025年 13卷 75679-75696页

作者： Abdipoor, Sina Yaakob, Razali Goh, Say Leng Abdullah, Salwani Hamdan, Hazlina Kasmiran, Khairul Azhar Universiti Putra Malaysia Faculty of Computer Science and Information Technology Department of Computer Science UPM Serdang Selangor 43400 Malaysia Universiti Malaysia Sabah Labuan International Campus Optimisation and Visual Analytics Research Group Faculty of Computing and Informatics Labuan87000 Malaysia Universiti Kebangsaan Malaysia Faculty of Information Science and Technology UKM Bangi Selangor 43600 Malaysia

Educational timetabling, a principal branch of operations research, presents challenging combinatorial optimization problems widely encountered in educational institutions. Meta-heuristics have commonly been applied to these problems and managed to attain promising performance in terms of optimality. However, their general applicability has been overlooked, hindering their effectiveness as versatile solvers. The limited generalizability of current approaches is the primary hurdle between the literature and real-world applications. This paper addresses this gap by introducing a generality taxonomy and conducting comprehensive theoretical and empirical analyses. This study highlights the adverse impact of extreme parameter tuning on generality, emphasizing the need for more generalized approaches. Furthermore, it introduces a performance assessment framework, penalizing problem-tailored solutions. It also examines the optimality vs. generality performance of the state-of-the-art approaches of the latest university course timetabling benchmark to further reinforce our claim and validate the efficacy of our framework. Our findings indicate that the current literature prioritizes optimality over generality. We believe adopting the proposed assessment framework is crucial for bridging the gap between research and practical applications, enabling fairer comparisons, and encouraging more adaptable approaches. © 2013 IEEE.

关键词： Curricula