InterHuman data has self-penetrations?

[nicolasugrinovic]

Hi, thanks for the great code.
I am processing the InterHuman dataset and as I visualize data from the motions/ folder I see that several sequences contain self-penetrations (as shown in the figure bellow). Is this expected or am I missing something? I use SMPL-X to generate the meshes with neutral gender and 10 shape parameters.

[tr3e]

oh in this version, Interhuman only supports SMPL-H shape PCs. Please kindly use SMPL-H instead.

[nicolasugrinovic]

thanks for the reply, with SMPL-H self-penetrations are reduced. However, now I see most people with almost the same body shape and large bodies, like the following. Is this right?

[nancy-ux]

hi, I am trying to visualize data too. I have used the SMPL-h. But penetrations between the two humans still exist. (from the 2124.txt frame166-171) Do you also meet the problem? Or is there something wrong with my code? Could you share the code with me? Thank you very much.

[nicolasugrinovic]

I would say that penetrations are allowed to happen, perhaps due to pose mismatch given that the capture system is only camera-based (i.e., no additional sensors). In some cases, poses may be better captured.
Now, this can also be a problem of not having the correct body shape information. Maybe the authors can shed some light into these issues @tr3e

[ggg66]

Hello, could you please share the visualization tools you are using? Thank you!