Closed Sharpiless closed 7 months ago
Hi, maybe you should first know about the SMPL series. And some methods could be used to obtain the human_param, like OSX, SMPLerX. Note that we use the SMPL-H, so I suggest using the method that can infer hand parameters. For mask, you can use Grounded-SAM to predict the human and object mask or just manually annotate through this tool .
If you do not use the curvature, the above are enough. For curvature, use
trimesh.curvature.discrete_mean_curvature_measure
or trimesh.curvature.discrete_gaussian_curvature_measure
for mesh data. CouldCompare can also be used to calculate curvatures for point cloud data.
Hi, awesome work! But how can i process a single custom image into the input format, including human_param, mask, and so on?