def __init__(self,
ratio=None,
radius=None,
down_conv_nn=None,
kp_points=16,
nb_feature=0,
*args,
**kwargs):
super(KPConv, self).__init__(FPSSampler(ratio),
RadiusNeighbourFinder(radius), *args,
**kwargs)
self.ratio = ratio
self.radius = radius
in_features, out_features = down_conv_nn
# KPCONV arguments
self.in_features = in_features
self.out_features = out_features
self.kp_points = kp_points
self._conv = PointKernel(self.kp_points,
self.in_features,
self.out_features,
radius=self.radius)
def __init__(self,
ratio=None,
radius=None,
down_conv_nn=None,
kp_points=16,
nb_feature=0,
*args,
**kwargs):
super(ResidualBKPConv, self).__init__(FPSSampler(ratio),
RadiusNeighbourFinder(radius),
*args, **kwargs)
self.ratio = ratio
self.radius = radius
self.max_num_neighbors = kwargs.get("max_num_neighbors", 64)
in_features, out_features = down_conv_nn
# KPCONV arguments
self.in_features = in_features
self.out_features = out_features
self.kp_points = kp_points
self.pre_mlp = nn.Linear(self.in_features, self.in_features // 2)
self._conv = PointKernel(
self.kp_points,
self.in_features // 2,
self.in_features // 2,
radius=self.radius,
)
self.post_mlp = nn.Linear(self.in_features // 2, out_features)
self.shortcut_mlp = nn.Linear(self.in_features, self.out_features)
def __init__(
self,
inN=None,
outN=None,
K=None,
D=None,
C1=None,
C2=None,
hidden_channel=None,
*args,
**kwargs,
):
super(PointCNNConvDown, self).__init__(FPSSampler(outN / inN),
DilatedKNNNeighbourFinder(K, D))
self._conv = XConv(C1, C2, 3, K, hidden_channels=hidden_channel)
def __init__(self,
ratio=None,
radius=None,
down_conv_nn=None,
kp_points=16,
nb_feature=0,
is_strided=True,
kp_extent=1,
density_parameter=1,
*args,
**kwargs):
super(BaseKPConvPartialDense, self).__init__(
FPSSampler(ratio),
RadiusNeighbourFinder(radius,
max_num_neighbors=kwargs.get(
"max_num_neighbors", 64),
conv_type=self.CONV_TYPE), *args, **kwargs)
self.ratio = ratio
self.radius = radius
self.is_strided = is_strided
if len(down_conv_nn) == 2:
in_features, out_features = down_conv_nn
intermediate_features = None
elif len(down_conv_nn) == 3:
in_features, intermediate_features, out_features = down_conv_nn
else:
raise NotImplementedError
# KPCONV arguments
self.in_features = in_features
self.out_features = out_features
self.intermediate_features = intermediate_features
self.kp_points = kp_points
# Dataset ~ Model parameters
self.kp_extent = kp_extent
self.density_parameter = density_parameter
# PARAMTERS IMPORTANT FOR SHADOWING
self.shadow_features_fill = 0.0
self.shadow_points_fill_ = float(10e6)
def __init__(self,
ratio=None,
radius=None,
radius_num_point=None,
down_conv_nn=None,
nb_feature=None,
*args,
**kwargs):
super(SAModule, self).__init__(
FPSSampler(ratio=ratio),
MultiscaleRadiusNeighbourFinder(
radius, max_num_neighbors=radius_num_point), *args, **kwargs)
local_nn = MLP(down_conv_nn) if down_conv_nn is not None else None
self._conv = PointConv(local_nn=local_nn, global_nn=None)
self._radius = radius
self._ratio = ratio
self._num_points = radius_num_point