una-dinosauria / 3d-pose-baseline

A simple baseline for 3d human pose estimation in tensorflow. Presented at ICCV 17.
MIT License
1.42k stars 354 forks source link

[Question] Are the 2D inputs not normalized with respect to the hip joint? #137

Closed Nicholasli1995 closed 5 years ago

Nicholasli1995 commented 5 years ago

Thanks for the good work! I wonder why the 2D inputs are no normalized to the hip joint while the 3D outputs are.

If I understand correctly, there is a postprocess_3d function but not a 2D counterpart. Without normalizing/post-processing the 2D inputs, the network is not translational invariant for 3D pose estimation.

una-dinosauria commented 5 years ago

See #73

una-dinosauria commented 5 years ago

No activity. Please reopen if you still have questions.