检索结果-内蒙古大学图书馆

Few Shots Are All You Need: A Progressive Few Shot Learning Approach for Low Resource Handwriting Recognition

学校读者我要写书评

暂无评论

arXiv 2021年

作者： Souibgui, Mohamed Ali Fornés, Alicia Kessentini, Yousri Megyesi, Beáta Computer Vision Center Computer Science Department Universitat Autònoma de Barcelona Spain Digital Research Center of Sfax B.P. 275 Sakiet Ezzit Sfax3021 Tunisia SMRTS : Laboratory of Signals SysteMs ARtificial Intelligence and Networks Tunisia Department of Linguistics and Philology Uppsala University Sweden

Handwritten text recognition in low resource scenarios, such as manuscripts with rare alphabets, is a challenging problem. The main difficulty comes from the very few annotated data and the limited linguistic information (e.g. dictionaries and language models). Thus, we propose a few-shot learning-based handwriting recognition approach that significantly reduces the human labor annotation process, requiring only few images of each alphabet symbol. The method consists in detecting all the symbols of a given alphabet in a textline image and decoding the obtained similarity scores to the final sequence of transcribed symbols. Our model is first pretrained on synthetic line images generated from any alphabet, even though different from the target domain. A second training step is then applied to diminish the gap between the source and target data. Since this retraining would require annotation of thousands of handwritten symbols together with their bounding boxes, we propose to avoid such human effort through an unsupervised progressive learning approach that automatically assigns pseudo-labels to the non-annotated data. The evaluation on different manuscript datasets show that our model can lead to competitive results with a significant reduction in human effort. The code will be publicly available in this repository: https://***/dali92002/HTRbyMatching. © 2021, CC BY.

关键词： Character recognition

Physical Adversarial Attack Meets computer vision: A Decade Survey

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Wei, Hui Tang, Hao Jia, Xuemei Wang, Zhixiang Yu, Hanxun Li, Zhubo Satoh, Shin'ichi Van Gool, Luc Wang, Zheng School of Computer Science National Engineering Research Center for Multimedia Software Wuhan University Wuhan China National Key Laboratory for Multimedia Information Processing School of Computer Science Peking University Beijing100871 China Colleage of Software and Technology Zhejiang University Hangzhou China School of Cyber Science and Engineering Wuhan University Wuhan China Digital Content and Media Sciences Research Division National Institute of Informatics Japan Department of Information and Communication Engineering Graduate School of Information Science and Technology The University of Tokyo Japan Computer Vision Lab of ETH Zurich Zürich8092 Switzerland KU Leuven Leuven3000 Belgium INSAIT Sofia Bulgaria

Despite the impressive achievements of Deep Neural Networks (DNNs) in computer vision, their vulnerability to adversarial attacks remains a critical concern. Extensive research has demonstrated that incorporating sophisticated perturbations into input images can lead to a catastrophic degradation in DNNs’ performance. This perplexing phenomenon not only exists in the digital space but also in the physical world. Consequently, it becomes imperative to evaluate the security of DNNs-based systems to ensure their safe deployment in real-world scenarios, particularly in security-sensitive applications. To facilitate a profound understanding of this topic, this paper presents a comprehensive overview of physical adversarial attacks. Firstly, we distill four general steps for launching physical adversarial attacks. Building upon this foundation, we uncover the pervasive role of artifacts carrying adversarial perturbations in the physical world. These artifacts influence each step. To denote them, we introduce a new term: adversarial medium. Then, we take the first step to systematically evaluate the performance of physical adversarial attacks, taking the adversarial medium as a first attempt. Our proposed evaluation metric, hiPAA, comprises six perspectives: Effectiveness, Stealthiness, Robustness, Practicability, Aesthetics, and Economics. We also provide comparative results across task categories, together with insightful observations and suggestions for future research directions. Copyright © 2022, The Authors. All rights reserved.

关键词： Deep neural networks

Open-Pose 3D Zero-Shot Learning: Benchmark and Challenges

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Zhao, Weiguang Yang, Guanyu Zhang, Rui Jiang, Chenru Yang, Chaolong Yan, Yuyao Hussain, Amir Huang, Kaizhu Department of Computer Science University of Liverpool LiverpoolL69 7ZX United Kingdom Department of Foundational Mathematics Xi’an Jiaotong-Liverpool University Suzhou215123 China Department of Mechatronics and Robotics Xi’an Jiaotong-Liverpool University Suzhou215123 China School of Robotic Xi’an Jiaotong-Liverpool University Suzhou215123 China School of Computing Edinburgh Napier University EdinburghEH11 4BN United Kingdom Data Science Research Center Duke Kunshan University Kunshan215316 China

With the explosive 3D data growth, the urgency of utilizing zero-shot learning to facilitate data labeling becomes evident. Recently, methods transferring language or language-image pre-training models like Contrastive Language-Image Pre-training (CLIP) to 3D vision have made significant progress in the 3D zero-shot classification task. These methods primarily focus on 3D object classification with an aligned pose;such a setting is, however, rather restrictive, which overlooks the recognition of 3D objects with open poses typically encountered in real-world scenarios, such as an overturned chair or a lying teddy bear. To this end, we propose a more realistic and challenging scenario named open-pose 3D zero-shot classification, focusing on the recognition of 3D objects regardless of their orientation. First, we revisit the current research on 3D zero-shot classification, and propose two benchmark datasets specifically designed for the open-pose setting. We empirically validate many of the most popular methods in the proposed open-pose benchmark. Our investigations reveal that most current 3D zero-shot classification models suffer from poor performance, indicating a substantial exploration room towards the new direction. Furthermore, we study a concise pipeline with an iterative angle refinement mechanism that automatically optimizes one ideal angle to classify these open-pose 3D objects. In particular, to make validation more compelling and not just limited to existing CLIP-based methods, we also pioneer the exploration of knowledge transfer based on Diffusion models. While the proposed solutions can serve as a new benchmark for open-pose 3D zero-shot classification, we discuss the complexities and challenges of this scenario that remain for further research development. The code is available publicly at https://***/weiguangzhao/Diff-OP3D. © 2023, CC BY-NC-ND.

关键词： Classification (of information)

Towards end-to-end deep learning for autonomous racing: On data collection and a unified architecture for steering and throttle prediction

学校读者我要写书评

暂无评论

arXiv 2021年

作者： Wadekar, Shakti N. Schwartz, Benjamin J. Kannan, Shyam S. Mar, Manuel Manna, Rohan Kumar Chellapandi, Vishnu Gonzalez, Daniel J. Gamal, Aly El School of Electrical and Computer Engineering Purdue University West LafayetteIN47907 United States Computer Information Technology Polytechnic Institute Purdue University West LafayetteIN47907 United States Robotics Research Center Department of Electrical Engineering and Computer Science United States Military Academy West PointNY10996 United States

Deep Neural Networks (DNNs) which are trained end-to-end have been successfully applied to solve complex problems that we have not been able to solve in past decades. Autonomous driving is one of the most complex problems which is yet to be completely solved and autonomous racing adds more complexity and exciting challenges to this problem. Towards the challenge of applying end-to-end learning to autonomous racing, this paper shows results on two aspects: (1) Analyzing the relationship between the driving data used for training and the maximum speed at which the DNN can be successfully applied for predicting steering angle, (2) Neural network architecture and training methodology for learning steering and throttle without any feedback or recurrent connections. Copyright © 2021, The Authors. All rights reserved.

关键词： Deep neural networks

AbdomenAtlas: A Large-Scale, Detailed-Annotated, & Multi-center Dataset for Efficient Transfer Learning and Open Algorithmic Benchmarking

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Li, Wenxuan Qu, Chongyu Chen, Xiaoxi Bassi, Pedro R.A.S. Shi, Yijia Lai, Yuxiang Yu, Qian Xue, Huimin Chen, Yixiong Lin, Xiaorui Tang, Yutong Cao, Yining Han, Haoqi Zhang, Zheyuan Liu, Jiawei Zhang, Tiezheng Ma, Yujiu Wang, Jincheng Zhang, Guang Yuille, Alan Zhou, Zongwei Department of Computer Science Johns Hopkins University United States Department of Bioengineering University of Illinois Urbana-Champaign United States Alma Mater Studiorum University of Bologna Italy Center for Biomolecular Nanotechnologies Istituto Italiano di Tecnologia Italy LKS Faculty of Medicine The University of Hong Kong Hong Kong Department of Computer Science Southeast University China Department of Radiology Southeast University Zhongda Hospital China Department of Medical Oncology The First Hospital of China Medical University China The Second Clinical College China Medical University China Department of Mechanical Engineering The Laboratory of Computational Sensing and Robotics Johns Hopkins University United States Center of Reproductive Medicine Department of Obstetrics and Gynecology Shengjing Hospital of China Medical University China Radiology Department The First Affiliated Hospital School of Medicine Zhejiang University China Department of Health Management The First Affiliated Hospital of Shandong First Medical University Shandong Provincial Qianfoshan Hospital China Shandong Engineering Research Center of Health Management China Shandong Institute of Health Management China

We introduce the largest abdominal CT dataset (termed AbdomenAtlas) of 20,460 three-dimensional CT volumes sourced from 112 hospitals across diverse populations, geographies, and facilities. AbdomenAtlas provides 673K high-quality masks of anatomical structures in the abdominal region annotated by a team of 10 radiologists with the help of AI algorithms. We start by having expert radiologists manually annotate 22 anatomical structures in 5,246 CT volumes. Following this, a semi-automatic annotation procedure is performed for the remaining CT volumes, where radiologists revise the annotations predicted by AI, and in turn, AI improves its predictions by learning from revised annotations. Such a large-scale, detailed-annotated, and multi-center dataset is needed for two reasons. Firstly, AbdomenAtlas provides important resources for AI development at scale, branded as large pre-trained models, which can alleviate the annotation workload of expert radiologists to transfer to broader clinical applications. Secondly, AbdomenAtlas establishes a large-scale benchmark for evaluating AI algorithms—the more data we use to test the algorithms, the better we can guarantee reliable performance in complex clinical scenarios. An ISBI & MICCAI challenge named BodyMaps: Towards 3D Atlas of Human Body was launched using a subset of our AbdomenAtlas, aiming to stimulate AI innovation and to benchmark segmentation accuracy, inference efficiency, and domain generalizability. We hope our AbdomenAtlas can set the stage for larger-scale clinical trials and offer exceptional opportunities to practitioners in the medical imaging community. Codes, models, and datasets are available at https://***/dataset. Copyright © 2024, The Authors. All rights reserved.

关键词： Benchmarking

A Lightweight Flexible Semi-Cylindrical Valve for Seamless Integration in Soft Robots Based on the Giant Electrorheological Fluid

学校读者我要写书评

暂无评论

SSRN

SSRN 2022年

作者： Huang, Tiantian Xu, Denan Zhang, Hang Bai, Ou Aravelli, Aparna Zhou, Xiaoxiao Han, Bin School of Mechanical Science and Engineering Huazhong University of Science and Technology Wuhan430070 China Human Cyber-Physical Systems Laboratory The Department of Electrical and Computer Engineering Florida International University Florida33174 United States Applied Research Center Florida International University Florida33174 United States Guangdong Intelligent Robotics Institute Dongguan523808 China

Complete compliance of soft robots is a trending research topic. One significant aspect is the control element. In the current work, a flexible electro-rheological (ER) valve development is chosen due to the advantage of its dynamic power density, convenient electronic control mode and no mechanical moving parts. These properties enable the ER valve to be miniaturized and easily integrated into a fluid-driven soft robot. However, the integration effect of the existing flexible ER valve configuration in soft robots is not ideal. In this study, a novel concentric semi-cylindrical flexible giant ER (GER) valve with a dual flowing direction is proposed for a soft robot to enable a larger deformation bending motion with multiple motion modes. The resulting valve has an overall volume of 50 mm 3 , a mass of 0.88 g, and a maximum controlling pressure of 84 kPa with a pressure change rate of 13.9, which is suitable for most soft robot applications. The pressure difference calculation method for the concentric semi-cylindrical GER valve is discussed. The effects of size, voltages, and flow rates on the valve’s static and dynamic performances, as well as the characteristics of the two pressure regulation modes, are investigated through simulations and experiments. The seamless integration and flexibility of the proposed valve are demonstrated by controlling a typical network actuator with six motion modes. Finally, a soft robot composed of two actuators is fabricated and realized the grasping of multiple everyday objects using different motion modes. © 2022, The Authors. All rights reserved.

关键词： Nanofluidics

Enhancing Knee Contact Force Prediction Using Transformer Networks: A Comparative Study with Lstm and Mlp

学校读者我要写书评

暂无评论

SSRN

SSRN 2025年

作者： Xu, Fashu Wang, Junqing Tong, Bingkui Fang, Xuyi Zhang, Hui Deng, Tao Wei, Qu Nie, Yong Li, Kang Department of Orthopedic Surgery West China Biomedical Big Data Center West China Hospital Sichuan University Sichuan Chengdu China Department of Orthopedic Surgery Orthopedic Research Institute West China Hospital Sichuan University Sichuan Chengdu China State Key Laboratory of Robotics Shenyang Institute of Automation Chinese Academy of Sciences Liaoning Shenyang China College of Computer Science Sichuan University Sichuan Chengdu China College of Electrical Engineering Sichuan University Sichuan Chengdu China School of Mechanical Engineering Sichuan University Sichuan Chengdu China

Background and objectiveKnee contact force (KCF) is pivotal in both the study and degenerative pathologies like Knee Osteoarthritis. Traditionally, they are primarily estimated using musculoskeletal modeling and machine learning predictive algorithms. However, musculoskeletal modeling is limited by intricate experimental protocols, prohibitive equipment cost, and scalability challenges. While machine learning methods, though computationally efficient, exhibit suboptimal predictive accuracy. This study innovatively developed a Transformer model for predicting KCF, benchmarked against long short-term memory (LSTM) and multilayer perceptron (MLP) *** comprehensive database was constructed with gait data from 513 participants, incorporating three degrees of freedom of hip, knee flexion/extension, and ankle dorsiflexion/plantarflexion, constituting four input combinations for the models. Contact forces derived from OpenSim served as the ground truth. Root mean square error (RMSE) and coefficient of determination (R2) were used to evaluate the performance of *** results indicated that the Transformer model demonstrated superior performance, achieving a significantly lower mean root mean square error of 0.02 body weight (BW) compared to MLP (0.54 BW) and LSTM (0.45 BW). Furthermore, the mean R2 of the Transformer network (0.99) was markedly superior to that of MLP (0.45) and LSTM (0.58).ConclusionsThe application of Transformer networks significantly improved the accuracy of KCF prediction, addressing the limitations of traditional musculoskeletal modeling and existing machine learning approaches, which was of paramount importance in both clinical practice and biomechanical research. © 2025, The Authors. All rights reserved.

关键词： Mean square error

Mathematical Modeling of 18F-Fluoromisonidazole (18F-FMISO) Radiopharmaceutical Transport in Vascularized Solid Tumors

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Abazari, Mohammad Amin Soltani, M. Eydi, Faezeh Rahmim, Arman Kashkooli, Farshad Moradi Department of Mechanical Engineering K. N. Toosi University of Technology Tehran Iran Department of Electrical and Computer Engineering Faculty of Engineering School of Optometry and Vision Science Faculty of Science University of Waterloo WaterlooN2L 3G Canada Department of Integrative Oncology BC Cancer Research Institute VancouverBCV5Z 1L3 Canada University of Waterloo WaterlooONN2L 3G Canada Cancer Biology Research Center Cancer Institute of Iran Tehran University of Medical Sciences Tehran Iran Departments of Radiology and Physics University of British Columbia VancouverBC Canada

18F-Fluoromisonidazole (18F-FMISO) is a highly promising positron emission tomography radiopharmaceutical for identifying hypoxic regions in solid tumors. This research employs spatiotemporal multi-scale mathematical modeling to explore how different levels of angiogenesis influence the transport of radiopharmaceuticals within tumors. In this study, two tumor geometries with heterogeneous and uniform distributions of capillary networks were employed to incorporate varying degrees of microvascular density. The synthetic image of the heterogeneous and vascularized tumor was generated by simulating the angiogenesis process. The proposed multi-scale spatiotemporal model accounts for intricate physiological and biochemical factors within the tumor microenvironment, such as the transvascular transport of the radiopharmaceutical agent, its movement into the interstitial space by diffusion and convection mechanisms, and ultimately its uptake by tumor cells. Results showed that both quantitative and semi-quantitative metrics of 18F-FMISO uptake differ spatially and temporally at different stages during tumor growth. The presence of a high microvascular density in uniformly vascularized tumor increases cellular uptake, as it allows for more efficient release and rapid distribution of radiopharmaceutical molecules. This results in enhanced uptake compared to the heterogeneous vascularized tumor. In both heterogeneous and uniform distribution of microvessels in tumors, the diffusion transport mechanism has a more pronounced than convection. The findings of this study shed light on the transport phenomena behind 18F-FMISO radiopharmaceutical distribution and its delivery in the tumor microenvironment, aiding oncologists in their routine decision-making processes. © 2024, CC BY.

关键词： Tumors

Inaugural Issue of the International Journal of Artificial Intelligence and robotics research (IJAIRR): The Emergence of an Interdisciplinary Nexus

学校读者我要写书评

暂无评论

International Journal of Artificial Intelligence and Robotic...

International Journal of Artificial Intelligence and robotics research 2024年第1期1卷

作者： Yu Sun Dong Xu Xiaorui Zhu Computer Science and Engineering Department University of South Florida Tampa FL USA Department of Electrical Engineering and Computer Science University of Missouri-Columbia Columbia MO USA Christopher S. Bond Life Sciences Center University of Missouri–Columbia Columbia MO USA Galaxy Artificial Intelligence and Robotics Research Institute Singapore

For the inaugural issue of the International Journal of Artificial Intelligence and robotics research (IJAIRR), I am honored to present an editorial that encapsulates the essence and ambition of this cutting-edge publication. IJAIRR emerges at a time when Artificial Intelligence and robotics (AIR) are not merely technological novelties but fundamental drivers of progress across various scientific and practical domains. This journal aims to be at the forefront of documenting, analyzing, and guiding the interdisciplinary integration of AI, robotics, and fundamental sciences.

关键词：