def select_patches(self,
pts,
refer_pts,
vicinity,
num_points_per_patch=1024):
gc.collect()
pts = torch.FloatTensor(pts).cuda().unsqueeze(0)
refer_pts = torch.FloatTensor(refer_pts).cuda().unsqueeze(0)
group_idx = pnt2.ball_query(vicinity, num_points_per_patch, pts,
refer_pts)
pts_trans = pts.transpose(1, 2).contiguous()
new_points = pnt2.grouping_operation(pts_trans, group_idx)
new_points = new_points.permute([0, 2, 3, 1])
mask = group_idx[:, :, 0].unsqueeze(2).repeat(1, 1,
num_points_per_patch)
mask = (group_idx == mask).float()
mask[:, :, 0] = 0
mask[:, :, num_points_per_patch - 1] = 1
mask = mask.unsqueeze(3).repeat([1, 1, 1, 3])
new_pts = refer_pts.unsqueeze(2).repeat(
[1, 1, num_points_per_patch, 1])
local_patches = new_points * (1 -
mask).float() + new_pts * mask.float()
# local_patches = list(local_patches.squeeze(0).detach().cpu().numpy())
local_patches = local_patches.squeeze(0)
del mask
del new_points
del group_idx
del new_pts
del pts
del pts_trans
return local_patches
def forward(self, x, feature, num_pool):
x = x.contiguous()
xyz_flipped = x.transpose(1, 2).contiguous()
x_sub = (pointnet2_utils.gather_operation(
x, pointnet2_utils.furthest_point_sample(xyz_flipped,
num_pool)).transpose(
1, 2).contiguous())
# sub_index = self.knn(x_sub, xyz_flipped).int()
sub_index = pointnet2_utils.ball_query(0.2 * self.nlayer, self.k,
xyz_flipped, x_sub)
x = pointnet2_utils.grouping_operation(x, sub_index)
x = torch.max(x, dim=-1)[0]
# x = soft_pool2d(x, [1, x.shape[-1]]).squeeze(-1)
feature = pointnet2_utils.grouping_operation(feature, sub_index)
feature = self.mlp(feature)
feature = torch.max(feature, dim=-1)[0]
# feature = soft_pool2d(feature, [1, feature.shape[-1]]).squeeze(-1)
return x, feature
def pool(xyz, points, k, npoint):
xyz_flipped = xyz.transpose(1, 2).contiguous()
new_xyz = pointnet2_utils.gather_operation(
xyz_flipped, pointnet2_utils.furthest_point_sample(
xyz_flipped, npoint)).transpose(1, 2).contiguous()
_, idx = knn_point(k, xyz, new_xyz)
new_points = torch.max(pointnet2_utils.grouping_operation(
points.permute(0, 2, 1).contiguous(),
idx.int().permute(0, 2, 1).contiguous()).permute(0, 3, 2, 1),
dim=2).values
return new_xyz, new_points
def sample_and_group(xyz, points, npoint, nsample, radius, use_xyz=True):
"""
Args:
xyz: Tensor, (B, 3, N)
points: Tensor, (B, f, N)
npoint: int
nsample: int
radius: float
use_xyz: boolean
Returns:
new_xyz: Tensor, (B, 3, npoint)
new_points: Tensor, (B, 3 | f+3 | f, npoint, nsample)
idx_local: Tensor, (B, npoint, nsample)
grouped_xyz: Tensor, (B, 3, npoint, nsample)
"""
xyz_flipped = xyz.permute(0, 2, 1).contiguous() # (B, N, 3)
new_xyz = gather_operation(xyz,
furthest_point_sample(xyz_flipped,
npoint)) # (B, 3, npoint)
idx = ball_query(radius, nsample, xyz_flipped,
new_xyz.permute(0, 2,
1).contiguous()) # (B, npoint, nsample)
grouped_xyz = grouping_operation(xyz, idx) # (B, 3, npoint, nsample)
grouped_xyz -= new_xyz.unsqueeze(3).repeat(1, 1, 1, nsample)
if points is not None:
grouped_points = grouping_operation(points,
idx) # (B, f, npoint, nsample)
if use_xyz:
new_points = torch.cat([grouped_xyz, grouped_points], 1)
else:
new_points = grouped_points
else:
new_points = grouped_xyz
return new_xyz, new_points, idx, grouped_xyz
def group(xyz, points, k, dilation=1, use_xyz=False):
_, idx = knn_point(k * dilation + 1, xyz, xyz)
idx = idx[:, :, 1::dilation].int().contiguous()
# print("xyz",xyz.shape,idx.shape)
xyz_trans = xyz.transpose(1, 2).contiguous()
grouped_xyz = pointnet2_utils.grouping_operation(
xyz_trans,
idx.int().permute(0, 2, 1).contiguous()) # (batch_size, npoint, k, 3)
# print("GRO",grouped_xyz.shape,xyz.shape)
grouped_xyz -= torch.unsqueeze(xyz_trans, 2) # translation normalization
if points is not None:
# print(points.shape)
grouped_points = pointnet2_utils.grouping_operation(
points.permute(0, 2, 1).contiguous(),
idx.int().permute(
0, 2, 1).contiguous()) # (batch_size, npoint, k, channel)
if use_xyz:
grouped_points = torch.cat(
[grouped_xyz, grouped_points],
dim=-1) # (batch_size, npoint, k, 3+channel)
else:
grouped_points = grouped_xyz
return grouped_xyz, grouped_points, idx
def sphere_query_new(pts, new_pts, radius, nsample):
"""
:param pts: all points, [B. N. 3]
:param new_pts: query points, [B, S. 3]
:param radius: local sperical radius
:param nsample: max sample number in local sphere
:return:
"""
device = pts.device
B, N, C = pts.shape
_, S, _ = new_pts.shape
pts = pts.contiguous()
new_pts = new_pts.contiguous()
group_idx = pnt2.ball_query(radius, nsample, pts, new_pts)
mask = group_idx[:, :, 0].unsqueeze(2).repeat(1, 1, nsample)
mask = (group_idx == mask).float()
mask[:, :, 0] = 0
mask1 = (group_idx[:, :, 0] == 0).unsqueeze(2).float()
mask1 = torch.cat([mask1, torch.zeros_like(mask)[:, :, :-1]], dim=2)
mask = mask + mask1
# C implementation
pts_trans = pts.transpose(1, 2).contiguous()
new_points = pnt2.grouping_operation(pts_trans,
group_idx) # (B, 3, npoint, nsample)
new_points = new_points.permute([0, 2, 3, 1])
# replace the wrong points using new_pts
mask = mask.unsqueeze(3).repeat([1, 1, 1, 3])
n_points = new_points * (1 - mask).float()
del mask
del new_points
del group_idx
del new_pts
del pts
del pts_trans
return n_points
