Open jaquilio opened 1 year ago
According to issue https://github.com/open-mmlab/mmhuman3d/issues/84 - "point cloud-based methods are under development and will be added in MMHuman3D by 2022 H1. Please check our Roadmap for upcoming features.
I'm don't believe this ever made it in - would also love to know if this is still planned. @caizhongang looks to currently be working on https://github.com/caizhongang/PointHPS which maybe will one day make it into MMHuman3D. Otherwise https://github.com/hanabi7/VoteHMR also aims to reconstruct humans from point cloud, however trained models + human models are unavailable.
It seems to me that, so far, the mmhuman3d project has focused on things like human shape modeling, pose estimation etc., and not on semantic/instance segmentation. Is semantic/instance segmentation of humans in 3D point clouds intended to come under the scope of mmhuman3d project or is the plan to subsume it within the scope of mmdetection3d project?