weigq/3d_pose_baseline_pytorch

why 3d pose is similar

Closed this issue · 6 comments

huidemie commented

hello, I visualize the 3d skeleton and it looks like a human. But why the results of different datasets looks very similar ?

weigq commented

Can you provide more details?

huidemie commented

For example, when I visualize one of the 3d results in "sitting down" dataset, it looks like the following picture. I don't think it is "sitting down" action. Besides, 3d result in different datasets looks like the following picture all the same.
1618406692(1)

weigq commented

You used the H36m dataset or other dataset to visualize?

huidemie commented

You used the H36m dataset or other dataset to visualize?

I use H36M

weigq commented

I am not sure which samples you used. Let me guess if you only use the first several frames in each video (like, you only use the first frame of each action)? In h36m, the poses in the first several frames are every similar, where the actor is standing like you ploted.

huidemie commented

I am not sure which samples you used. Let me guess if you only use the first several frames in each video (like, you only use the first frame of each action)? In h36m, the poses in the first several frames are every similar, where the actor is standing like you ploted.

You are right ! Thanks a lot !