search

fabro66 / GAST-Net-3DPoseEstimation

A Graph Attention Spatio-temporal Convolutional Networks for 3D Human Pose Estimation in Video (GAST-Net)
MIT License
313 stars 70 forks source link

How to get the video of demo? #20

Closed flyyyyer closed 3 years ago

flyyyyer commented 3 years ago

How can I get the video as your provided as follows? To get the world coordinate and have the trajectory.

[2020/08/14] We achieve real-time 3D pose estimation. [video] [2020/10/15] We achieve real-time online 3D skeleton-based action recognition with a single RGB camera. [video]

fabro66 commented 3 years ago

Hi~ The trajectory in our video is the virtual world coordinates converted from the image coordinates. For the specific conversion method, please refer to #9.

flyyyyer commented 3 years ago

Hi, Thank you~ So how can I use the "revise_skes_real_time”, what's the value of width should I set? And could you provide the trajectory model such as VideoPose to get the world coordinates?

fabro66 commented 3 years ago

Hi~ 1） The value of width is of little reference, just for better rendering animation. 2） Our rendering animation does not use any trajectory model. For better displaying the rendered animation, only the (x, y) coordinates of the image are converted into values in the world coordinate system based on prior knowledge. Code is here https://github.com/fabro66/GAST-Net-3DPoseEstimation/blob/212c8814eeec57e950184641b498108875385388/tools/preprocess.py#L123