MEEV Body Mesh Estimation On Egocentric Video Nicolas Monet1and Dongyoon Wee1

2025-05-02 0 0 3.24MB 5 页 10玖币

侵权投诉

MEEV: Body Mesh Estimation On Egocentric

Video

Nicolas Monet1and Dongyoon Wee1

NAVER CLOVA Video

Abstract. This technical report introduces our solution, MEEV, pro-

posed to the EgoBody Challenge at ECCV 2022. Captured from head-

mounted devices, the dataset consists of human body shape and mo-

tion of interacting people. The EgoBody dataset has challenges such as

occluded body or blurry image. In order to overcome the challenges,

MEEV is designed to exploit multiscale features for rich spatial infor-

mation. Besides, to overcome the limited size of dataset, the model is

pre-trained with the dataset aggregated 2D and 3D pose estimation

datasets. Achieving 82.30 for MPJPE and 92.93 for MPVPE, MEEV

has won the EgoBody Challenge at ECCV 2022, which shows the ef-

fectiveness of the proposed method. The code is available at https:

//github.com/clovaai/meev

Keywords: body pose estimation, body mesh estimation, video, EgoB-

ody, egocentric

1 Introduction

The EgoBody Challenge[12] deals with the 3D human pose and shape estima-

tion task from monocular RGB image. Speciﬁcally, they propose a dataset with

images captured from egocentric views, which makes challenging problems such

as high occlusion and severe motion blur compared to the other datasets used

in pose or shape estimation task, e.g., 3DPW [7].

In order to tackle the challenging task, we proposed MEEV, which exploit

multi-scale features from image encoder and pre-training with large datasets,

e.g., AGORA [10]

In the ﬁrst section, we describe our approach and then details the experiments

we made with our results.

2 Our Approach

Extending from the work of Moon et al. [9], MEEV additionally use the feature

from higher resolution to supplement spatial information.

The input image resolution I∈R3×256×192 is cropped from the image us-

ing the bounding box of the person. We use a ConvNext [5] large backbone for

the input image. As shown in Fig. 1, the image features F0∈R1536×8×6and

arXiv:2210.14165v1 [cs.CV] 21 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：

MEEV:BodyMeshEstimationOnEgocentricVideoNicolasMonet1andDongyoonWee1NAVERCLOVAVideoAbstract.Thistechnicalreportintroducesoursolution,MEEV,pro-posedtotheEgoBodyChallengeatECCV2022.Capturedfromhead-mounteddevices,thedatasetconsistsofhumanbodyshapeandmo-tionofinteractingpeople.TheEgoBodydatasethaschall...

展开>> 收起<<

MEEV Body Mesh Estimation On Egocentric Video Nicolas Monet1and Dongyoon Wee1.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！

MEEV Body Mesh Estimation On Egocentric Video Nicolas Monet1and Dongyoon Wee1

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: