MC-hands-1M A glove-wearing hand dataset for pose estimation Prodromos Boutis

2025-05-02 0 0 1.39MB 5 页 10玖币

侵权投诉

MC-hands-1M: A glove-wearing hand dataset for

pose estimation

Prodromos Boutis , Zisis Batzos , Konstantinos Konstantoudakis ,

Anastasios Dimou , Petros Daras

Centre for Research & Technology Hellas

Abstract. Nowadays, the need for large amounts of carefully and com-

plexly annotated data for the training of computer vision modules con-

tinues to grow. Furthermore, although the research community presents

state of the art solutions to many problems, there exist special cases,

such as the pose estimation and tracking of a glove-wearing hand, where

the general approaches tend to be unable to provide an accurate solution

or fail completely. In this work, we are presenting a synthetic dataset1for

3D pose estimation of glove-wearing hands, in order to depict the value

of data synthesis in computer vision. The dataset is used to ﬁne-tune a

public hand joint detection model, achieving signiﬁcant performance in

both synthetic and real images of glove-wearing hands.

Keywords: synthetic dataset, 3D hand pose estimation, gloved hands

1 Introduction

Computer vision models, targeting more complex problems, are evolving at an

incredible pace, resulting in an insatiable appetite for more datasets, whose size

and annotations’ detail are becoming a limiting factor. Therefore, in literature,

the utilization of synthetic visual data, from domain adaptation techniques [12]

to the deployment of GANs [5], and from the Cut-Paste approach [6] to video

games’ scenes [8], regularly combined with corresponding real data, has become

an established technique over the last decade.

Speciﬁcally, hand pose estimation is a well-studied problem with a variety

of depth- [1] and color-based [4] solutions, deploying diﬀerent machine-learning

methods [11,10]. However, many applications, in the context of hazardous work

environments and sports, necessitate the use of gloves. Existing hand detection

and tracking AI algorithms, trained on real [3,13] and synthetic [2,7] bare-hand

datasets, exhibit signiﬁcantly reduced performance or fail altogether in gloved

hand scenarios, as they depend deeply on the canvas of the human skin’s colors.

Hence, there is a clear need for a gloved-hand dataset with ground truth for the

joints’ positions, allowing the training or re-training of AI algorithms capable of

estimating poses and/or tracking hands wearing gloves of diverse size and color.

1The dataset is public and can be found at https://www.zenodo.org/record/

7194271/

arXiv:2210.10428v1 [cs.CV] 19 Oct 2022

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

MC-hands-1M:Aglove-wearinghanddatasetforposeestimationProdromosBoutis,ZisisBatzos,KonstantinosKonstantoudakis,AnastasiosDimou,PetrosDarasCentreforResearch&TechnologyHellasAbstract.Nowadays,theneedforlargeamountsofcarefullyandcom-plexlyannotateddataforthetrainingofcomputervisionmodulescon-tinuestogro...

展开>> 收起<<

MC-hands-1M A glove-wearing hand dataset for pose estimation Prodromos Boutis.pdf

共5页,预览1页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

MC-hands-1M A glove-wearing hand dataset for pose estimation Prodromos Boutis

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: