search

facebookresearch / VideoPose3D

Efficient 3D human pose estimation in video using 2D keypoint trajectories
Other
3.74k stars 756 forks source link

Variations in the Wild results (same environment) #177

Open aditya-neuro opened 3 years ago

aditya-neuro commented 3 years ago

Hi, I've been working on VideoPose3D for quite some time now, I've noticed that sometimes predictions on the wild videos are very accurate and sometimes the opposite.

Can anyone, please let me know why is it happening? I've attached gifs for the above-mentioned cases.

In the following videos, I've plotted ankle points of both Keypoint RCNN and VideoPose3D's ankle points -

Problematic Case

videopose_prob

Ideal Case

videopose_ideal

Both videos are shot in the same environment, then is there something else that I need to keep in mind to get the best possible results?

Thank you, any help would be highly appreciated!

Garfield-kh commented 3 years ago

Hi, any code or data to reproduce this result?

slyfox63 commented 2 years ago

maybe it was a result of the aberration between the person and the background, there're some interference, maybe you could try with some clothes of different color.