search

saidwivedi / TokenHMR

[CVPR 2024] TokenHMR: Advancing Human Mesh Recovery with a Tokenized Pose Representation
https://tokenhmr.is.tue.mpg.de
Other
213 stars 18 forks source link

Test results of training with SD. #11

Closed zhanganny closed 1 month ago

zhanganny commented 2 months ago

Hello, thanks for your great job! I'm confused why the test results of TokenHMR on SD is not showed in paper? I wonder whether its performance can surpass the other method listed (such as HMR2.0、CLIFF、HybrIK). Have you ever tried to compute the metrics on SD?

Looking forward to your reply!

image
saidwivedi commented 2 months ago

TokenHMR enables training methods that use large scale in-the-wild pseudo-GT data with the goal of generalising for difficult poses while not compromising on 3D accuracy. Since SD contains very less in-the-wild pseudo-GT data, we never trained on this train split. However, feel free to train a version with SD data and let us know your findings.

saidwivedi commented 1 month ago

Closing the issue now, if you have further questions, feel free to ask.