search

MarilynKeller / OSSO

From a body shape, infer the anatomic skeleton.
Other
210 stars 32 forks source link

Issue with Intersection of Segments in Exported Skeleton Model and Potential Manually Modification #9

Open hwy1992129 opened 1 year ago

hwy1992129 commented 1 year ago

I have encountered an issue with the exported skeleton model where the model segments appear to intersect. The involved program components are register_star, infer_skeleton, and pose_skeleton. image image

I'm considering a potential workaround and need some guidance: Is it possible to manually edit the skeleton model to resolve the intersection problem and then use this modified model as an input for the pose_skeleton function? Would this approach create any unforeseen issues? Looking forward to your insights.

MarilynKeller commented 1 year ago

We don't have a loss to avoid those specific interpenetrations. You could edit the final skeleton mesh to correct for those interpenetrations. If you edit the lying down inferred skeleton before reposing, it should help for the spine, but no guarranty that the scapula will not intersect with the thorax at all after reposing. I think a way to do would be to add a loss to avoid interpenetration between the scapula and the thorax by fitting ellipsoids to them.

hwy1992129 commented 1 year ago

@MarilynKeller It seems that the ribs sometimes are not in the correct position, as shown in the images... image image

MarilynKeller commented 1 year ago

Can you specify what you mean with "Not in the correct position" ? We do have light interpenetrations between the ribs and the skin in some poses as our rib cage is rigid. For T12 and Rib 12, their relative position is the one that was in the template we used from Anatomy Transfer.

hwy1992129 commented 1 year ago

I mean T12 and Rib12.... For some results, the Rib12 may stick to T11, because of the scale and translation.... I check the template. It is correct.