Jove
Visualize
联系我们
JoVE
x logofacebook logolinkedin logoyoutube logo
关于 JoVE
概览领导团队博客JoVE 帮助中心
作者
出版流程编辑委员会范围与政策同行评审常见问题投稿
图书馆员
用户评价订阅访问资源图书馆顾问委员会常见问题
研究
JoVE JournalMethods CollectionsJoVE Encyclopedia of Experiments存档
教育
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab Manual教师资源中心教师网站
使用条款与条件
隐私政策
政策

相关实验视频

Updated: Jun 3, 2025

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
03:31

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications

Published on: December 15, 2023

463

身份模型转换用于提高对象检测网络的性能和效率.

Zhongyuan Lu1, Jin Liu1, Miaozhong Xu1

  • 1The State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, 129 Luoyu Road, Wuhan, 430079, China.

Neural networks : the official journal of the International Neural Network Society
|January 8, 2025
PubMed
概括
此摘要是机器生成的。

相关概念视频

您也可能阅读

相关文章

通过共同作者、期刊和引用图与本文相关的文章。

排序
Same author

Intravenous lipid emulsion improves recovery time and quality from isoflurane anaesthesia: a double-blind clinical trial.

Basic & clinical pharmacology & toxicology·2014
Same author

Improvement of matrix-assisted laser desorption/ionization time-of-flight mass spectrometry for identification of clinically important Candida species.

Clinical laboratory·2014
Same author

Genetic engineering of the green alga Chlorella zofingiensis: a modified norflurazon-resistant phytoene desaturase gene as a dominant selectable marker.

Applied microbiology and biotechnology·2014
Same author

Successful one-stage extraction of an intracardiac and intravenous leiomyoma through the right atrium under transesophageal ultrasound monitoring.

Canadian journal of anaesthesia = Journal canadien d'anesthesie·2014
Same author

A particle swarm optimization variant with an inner variable learning strategy.

TheScientificWorldJournal·2014
Same author

Integrative Analysis of Cancer Diagnosis Studies with Composite Penalization.

Scandinavian journal of statistics, theory and applications·2014

身份模型转换 (IMT) 是一种新的技术,可以在没有性能损失的情况下修改网络结构. 这种方法显著减少了培训时间,并提高了网络性能,使得快速的架构变化和更好的优化潜力.

科学领域:

  • 计算机科学 计算机科学
  • 人工智能的人工智能
  • 机器学习 机器学习

背景情况:

  • 修改神经网络架构对于性能提升至关重要.
  • 层次的修改往往导致预训练的重量不匹配,导致耗时和资源密集的微调.
  • 现有的方法难以实现高效和有效的建筑转型.

研究的目的:

  • 引入一种新的技术,即身份模型转换 (IMT),用于有效修改神经网络.
  • 为了应对预训练重量不匹配和漫长的微调过程的挑战.
  • 为了实现快速的建筑过渡,并推导出改进模型的家族.

主要方法:

  • 身份模型转换 (IMT) 使用严格的代数转换来保持结构变化之前和之后的输出平等.
  • 这保证了原始模型性能的保存.
  • IMT促进了分析的延续,以生成一个相关模型的家族.

主要成果:

  • IMT保留了原始模型的性能,同时允许结构修改.
  • 训练时间显著减少,在DOTA 1.5数据集上微调YOLOv4-Rot节省了94.76%.
  • 在多个数据集中观察到一致的性能改善:AI-TOD (9.89%),DOTA1.5 (6.94%),coco2017 (2.36%) 和MRSAText (4.86%).
关键词:
深度学习是一种深度学习.身份的转变身份的转变模型转移 模型转移对象检测检测对象检测对象检测

更多相关视频

A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
04:23

A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images

Published on: April 21, 2023

1.7K
A Step-by-Step Implementation of DeepBehavior, Deep Learning Toolbox for Automated Behavior Analysis
05:41

A Step-by-Step Implementation of DeepBehavior, Deep Learning Toolbox for Automated Behavior Analysis

Published on: February 6, 2020

9.3K

相关实验视频

Last Updated: Jun 3, 2025

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
03:31

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications

Published on: December 15, 2023

463
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
04:23

A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images

Published on: April 21, 2023

1.7K
A Step-by-Step Implementation of DeepBehavior, Deep Learning Toolbox for Automated Behavior Analysis
05:41

A Step-by-Step Implementation of DeepBehavior, Deep Learning Toolbox for Automated Behavior Analysis

Published on: February 6, 2020

9.3K

结论:

  • 身份模型转换 (IMT) 为神经网络架构修改提供了一种高效和有效的解决方案.
  • 它弥合了快速模型转换的差距,使得优化模型家族的衍生成为可能.
  • 在对象检测任务中,IMT显示了大量的时间节省和性能提升.