open-mmlab
Semantic/Instance Segmentation
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?
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.
