search

open-mmlab / mmpose

OpenMMLab Pose Estimation Toolbox and Benchmark.
https://mmpose.readthedocs.io/en/latest/
Apache License 2.0
5.91k stars 1.26k forks source link

Question about SmoothNet Implementation #1330

Closed vra closed 2 years ago

vra commented 2 years ago

Hi all,

Thanks for sharing the implementation of SmoothNet in this repo. In the paper, the author proposed a motion-aware SmoothNet, which explicitly modeled the velocity and acceleration by another two branches like the main branch. I found the implementation of SmoothNet at current only has one branch. Could you give any hints that is extra two branches necessary or just one path is fine even for different tasks(2D/3D pose estimation and human mesh recovery)?

ly015 commented 2 years ago

@ailingzengzzz Could you please help with this question?

ailingzengzzz commented 2 years ago

Hi @vra,

In this implementation, I only provided the Basic SmoothNet with one branch. Motion-aware SmoothNet may be provided further. Visually, their smoothed results are similar. One path is also fine for 2d,3d, and 6d (arbitrary continuous modality) smoothness.

vra commented 2 years ago

Hi @ailingzengzzz Thanks for your clear explanation. My concern has been resolved and I will close this thread now.