Questions about human body reconstruction

[ruiqiyan]

The surface I'm modeling is extremely rough. I've enabled mask_depth_normal, milo, and depth_order, but the results are very poor, as shown in the image.

The detail rendering is quite rough, far inferior to PGSR. How can I adjust it?

[ruiqiyan]

This is the result of PGSR.The details of the facial features will be more three-dimensional.

[ruiqiyan]

While the shoelaces are better than those of the PGSR, the shoes are more prone to damage.

[diego1401]

Hi @ruiqiyan,

Thank you for your interest in our work.
Can you please let me know what is the command you are using to process this scene ?
I answered your previous issue about using our method on datasets without background here.

I think for this kind of dataset extracting with PAM could be extremely beneficial!
For instance here is a capture of the quality I get extracting the Lego scene for the NeRF synthetic dataset.

[diego1401]

Hi @ruiqiyan ,
I thought a bit more about the results you shared, this seems to me that the current run is using too few gaussians.
I would recommend (besides using a similar script to above), tuning the densification parameters in order to get around 1-2M gaussians.