基于增强混合特征的尺度不变人体姿态估计的ScaleFormer架构。

IF 3.9 2区 综合性期刊 Q1 MULTIDISCIPLINARY SCIENCES
Congying Ge, Wei Fu Qin
{"title":"基于增强混合特征的尺度不变人体姿态估计的ScaleFormer架构。","authors":"Congying Ge, Wei Fu Qin","doi":"10.1038/s41598-025-12620-4","DOIUrl":null,"url":null,"abstract":"<p><p>Human pose estimation is a fundamental task in computer vision. However, existing methods face performance fluctuation challenges when processing human targets at different scales, especially in outdoor scenes where target distances and viewing angles frequently change. This paper proposes ScaleFormer, a novel scale-invariant pose estimation framework that effectively addresses multi-scale pose estimation problems by innovatively combining the hierarchical feature extraction capabilities of Swin Transformer with the fine-grained feature enhancement mechanisms of ConvNeXt. We design an adaptive feature representation mechanism that enables the model to maintain consistent performance across different scales. Extensive experiments on the MPII human pose dataset demonstrate that ScaleFormer significantly outperforms existing methods on multiple metrics including PCKh, scale consistency score, and keypoint mean average precision. Notably, under extreme scaling conditions (scaling factor 2.0), ScaleFormer's scale consistency score exceeds the baseline model by 48.8 percentage points. Under 30% random occlusion conditions, keypoint detection accuracy improves by 20.5 percentage points. Experiments further verify the complementary contributions of the two core components. These results indicate that ScaleFormer has significant advantages in practical application scenarios and provides new research directions for the field of pose estimation.</p>","PeriodicalId":21811,"journal":{"name":"Scientific Reports","volume":"15 1","pages":"27754"},"PeriodicalIF":3.9000,"publicationDate":"2025-07-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12311106/pdf/","citationCount":"0","resultStr":"{\"title\":\"ScaleFormer architecture for scale invariant human pose estimation with enhanced mixed features.\",\"authors\":\"Congying Ge, Wei Fu Qin\",\"doi\":\"10.1038/s41598-025-12620-4\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<p><p>Human pose estimation is a fundamental task in computer vision. However, existing methods face performance fluctuation challenges when processing human targets at different scales, especially in outdoor scenes where target distances and viewing angles frequently change. This paper proposes ScaleFormer, a novel scale-invariant pose estimation framework that effectively addresses multi-scale pose estimation problems by innovatively combining the hierarchical feature extraction capabilities of Swin Transformer with the fine-grained feature enhancement mechanisms of ConvNeXt. We design an adaptive feature representation mechanism that enables the model to maintain consistent performance across different scales. Extensive experiments on the MPII human pose dataset demonstrate that ScaleFormer significantly outperforms existing methods on multiple metrics including PCKh, scale consistency score, and keypoint mean average precision. Notably, under extreme scaling conditions (scaling factor 2.0), ScaleFormer's scale consistency score exceeds the baseline model by 48.8 percentage points. Under 30% random occlusion conditions, keypoint detection accuracy improves by 20.5 percentage points. Experiments further verify the complementary contributions of the two core components. These results indicate that ScaleFormer has significant advantages in practical application scenarios and provides new research directions for the field of pose estimation.</p>\",\"PeriodicalId\":21811,\"journal\":{\"name\":\"Scientific Reports\",\"volume\":\"15 1\",\"pages\":\"27754\"},\"PeriodicalIF\":3.9000,\"publicationDate\":\"2025-07-30\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12311106/pdf/\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Scientific Reports\",\"FirstCategoryId\":\"103\",\"ListUrlMain\":\"https://doi.org/10.1038/s41598-025-12620-4\",\"RegionNum\":2,\"RegionCategory\":\"综合性期刊\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q1\",\"JCRName\":\"MULTIDISCIPLINARY SCIENCES\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Scientific Reports","FirstCategoryId":"103","ListUrlMain":"https://doi.org/10.1038/s41598-025-12620-4","RegionNum":2,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"MULTIDISCIPLINARY SCIENCES","Score":null,"Total":0}
引用次数: 0

摘要

人体姿态估计是计算机视觉中的一项基本任务。然而,现有方法在处理不同尺度的人体目标时面临性能波动的挑战,特别是在目标距离和视角频繁变化的室外场景中。本文提出了一种新的尺度不变姿态估计框架ScaleFormer,通过创新地将Swin Transformer的分层特征提取能力与ConvNeXt的细粒度特征增强机制相结合,有效地解决了多尺度姿态估计问题。我们设计了一种自适应特征表示机制,使模型在不同尺度上保持一致的性能。在MPII人体姿态数据集上的大量实验表明,ScaleFormer在PCKh、尺度一致性评分和关键点平均精度等多个指标上都明显优于现有方法。值得注意的是,在极端缩放条件下(缩放因子2.0),ScaleFormer的缩放一致性得分比基线模型高出48.8个百分点。在30%随机遮挡条件下,关键点检测准确率提高20.5个百分点。实验进一步验证了两个核心组件的互补性贡献。这些结果表明,ScaleFormer在实际应用场景中具有显著的优势,为姿态估计领域提供了新的研究方向。
本文章由计算机程序翻译,如有差异,请以英文原文为准。

ScaleFormer architecture for scale invariant human pose estimation with enhanced mixed features.

ScaleFormer architecture for scale invariant human pose estimation with enhanced mixed features.

ScaleFormer architecture for scale invariant human pose estimation with enhanced mixed features.

ScaleFormer architecture for scale invariant human pose estimation with enhanced mixed features.

Human pose estimation is a fundamental task in computer vision. However, existing methods face performance fluctuation challenges when processing human targets at different scales, especially in outdoor scenes where target distances and viewing angles frequently change. This paper proposes ScaleFormer, a novel scale-invariant pose estimation framework that effectively addresses multi-scale pose estimation problems by innovatively combining the hierarchical feature extraction capabilities of Swin Transformer with the fine-grained feature enhancement mechanisms of ConvNeXt. We design an adaptive feature representation mechanism that enables the model to maintain consistent performance across different scales. Extensive experiments on the MPII human pose dataset demonstrate that ScaleFormer significantly outperforms existing methods on multiple metrics including PCKh, scale consistency score, and keypoint mean average precision. Notably, under extreme scaling conditions (scaling factor 2.0), ScaleFormer's scale consistency score exceeds the baseline model by 48.8 percentage points. Under 30% random occlusion conditions, keypoint detection accuracy improves by 20.5 percentage points. Experiments further verify the complementary contributions of the two core components. These results indicate that ScaleFormer has significant advantages in practical application scenarios and provides new research directions for the field of pose estimation.

求助全文
通过发布文献求助,成功后即可免费获取论文全文。 去求助
来源期刊
Scientific Reports
Scientific Reports Natural Science Disciplines-
CiteScore
7.50
自引率
4.30%
发文量
19567
审稿时长
3.9 months
期刊介绍: We publish original research from all areas of the natural sciences, psychology, medicine and engineering. You can learn more about what we publish by browsing our specific scientific subject areas below or explore Scientific Reports by browsing all articles and collections. Scientific Reports has a 2-year impact factor: 4.380 (2021), and is the 6th most-cited journal in the world, with more than 540,000 citations in 2020 (Clarivate Analytics, 2021). •Engineering Engineering covers all aspects of engineering, technology, and applied science. It plays a crucial role in the development of technologies to address some of the world''s biggest challenges, helping to save lives and improve the way we live. •Physical sciences Physical sciences are those academic disciplines that aim to uncover the underlying laws of nature — often written in the language of mathematics. It is a collective term for areas of study including astronomy, chemistry, materials science and physics. •Earth and environmental sciences Earth and environmental sciences cover all aspects of Earth and planetary science and broadly encompass solid Earth processes, surface and atmospheric dynamics, Earth system history, climate and climate change, marine and freshwater systems, and ecology. It also considers the interactions between humans and these systems. •Biological sciences Biological sciences encompass all the divisions of natural sciences examining various aspects of vital processes. The concept includes anatomy, physiology, cell biology, biochemistry and biophysics, and covers all organisms from microorganisms, animals to plants. •Health sciences The health sciences study health, disease and healthcare. This field of study aims to develop knowledge, interventions and technology for use in healthcare to improve the treatment of patients.
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:604180095
Book学术官方微信