dim 96 when unnorm 3D keypoints

una-dinosauria / 3d-pose-baseline

A simple baseline for 3d human pose estimation in tensorflow. Presented at ICCV 17.

MIT License

1.41k stars 355 forks source link

dim 96 when unnorm 3D keypoints #199

Closed andresherrera97 closed 3 years ago

andresherrera97 commented 3 years ago

Hello, I've been seeing your code and I can't understand why when unnormalizing the 3D keypoint you use an input of dimension 48 and an output of dimension 96. What is the use for the extra dimensions? Thanks in advance.

una-dinosauria commented 3 years ago

96 (96/3=32) is the original data. It contains some extra markers that don't have degrees of freedom; ie, they are located at a constant offset from other markers. 48 (48/3=16) is the number of 3d points we work with.