dvlab-research / PointGroup

PointGroup: Dual-Set Point Grouping for 3D Instance Segmentation

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

[help wanted] A question in model implementation

JLUtangchuan opened this issue · comments

See in pointgroup.py#L283, why semantic_preds >1 ? (ignore Class 0 and Class 1?)

object_idxs = torch.nonzero(semantic_preds > 1).view(-1)

object_idxs = torch.nonzero(semantic_preds > 1).view(-1)

Do you know why this is set? I also have this doubt

class0 and class1 represent wall and floor, those should be ignored during offset prediction