MCPcopy
hub / github.com/Pointcept/PointTransformerV3 / __init__

Method __init__

model.py:754–779  ·  view source on GitHub ↗
(
        self,
        in_channels,
        embed_channels,
        norm_layer=None,
        act_layer=None,
    )

Source from the content-addressed store, hash-verified

752
753class Embedding(PointModule):
754 def __init__(
755 self,
756 in_channels,
757 embed_channels,
758 norm_layer=None,
759 act_layer=None,
760 ):
761 super().__init__()
762 self.in_channels = in_channels
763 self.embed_channels = embed_channels
764
765 # TODO: check remove spconv
766 self.stem = PointSequential(
767 conv=spconv.SubMConv3d(
768 in_channels,
769 embed_channels,
770 kernel_size=5,
771 padding=1,
772 bias=False,
773 indice_key="stem",
774 )
775 )
776 if norm_layer is not None:
777 self.stem.add(norm_layer(embed_channels), name="norm")
778 if act_layer is not None:
779 self.stem.add(act_layer(), name="act")
780
781 def forward(self, point: Point):
782 point = self.stem(point)

Callers

nothing calls this directly

Calls 3

PointSequentialClass · 0.85
addMethod · 0.80
__init__Method · 0.45

Tested by

no test coverage detected