search

Nicholasli1995 / EvoSkeleton

Official project website for the CVPR 2020 paper (Oral Presentation) "Cascaded deep monocular 3D human pose estimation wth evolutionary training data"
https://arxiv.org/abs/2006.07778
MIT License
332 stars 44 forks source link

About 2D anchor of the cropped image #80

Open BaymaxCEO opened 2 years ago

BaymaxCEO commented 2 years ago

May I ask a question？ When estimating the 2D pose from the cropped image, how is the anchor of the cropped image determined

Nicholasli1995 commented 2 years ago

May I ask a question？ When estimating the 2D pose from the cropped image, how is the anchor of the cropped image determined

Do you mean the bounding box? You may use a human detector in you application.

BaymaxCEO commented 2 years ago

May I ask a question？ When estimating the 2D pose from the cropped image, how is the anchor of the cropped image determined

Do you mean the bounding box? You may use a human detector in you application.

So could you please provide the bounding box network used in this work Thank you so much！

Nicholasli1995 commented 2 years ago

May I ask a question？ When estimating the 2D pose from the cropped image, how is the anchor of the cropped image determined

Do you mean the bounding box? You may use a human detector in you application.

So could you please provide the bounding box network used in this work Thank you so much！

Hi, this repo does not contain a person detector since the 3D pose estimation evaluation assumes the persons are already detected. You may directly use the provided boxes from the H3.6M dataset for experiments. Or you may use https://github.com/Nicholasli1995/EgoNet/blob/master/libs/common/img_proc.py#L495 which computes a bounding box to enclose the annotated 2D keypoints.