What if there's only half a person in the video

fabro66 / GAST-Net-3DPoseEstimation

A Graph Attention Spatio-temporal Convolutional Networks for 3D Human Pose Estimation in Video (GAST-Net)

MIT License

314 stars 69 forks source link

What if there's only half a person in the video #37

Closed GuaiYiHu closed 1 year ago

GuaiYiHu commented 3 years ago

This part convert knees and ankles to hips if knees and ankles are invisible in videos, where I don't think it's very reasonable. https://github.com/fabro66/GAST-Net-3DPoseEstimation/blob/783310d0b26f9ff0b47e5036b7c6b864dfb120ef/tools/preprocess.py#L96

fabro66 commented 3 years ago

Hi~ sorry for not reply promptly.

If the joints are occluded, the coordinates of the keypoints obtained by the 2D detector will have a large deviation, which will seriously affect the 2D-to-3D reconstruction, so we adopt this preprocessing.

Looking forward to hearing if you have any other better solutions.

GuaiYiHu commented 3 years ago

Having seen the expose(https://github.com/vchoutas/expose) has solved this problem.

fabro66 commented 3 years ago

Get it. Thanks for the suggestions.