def forward(self, features, num_voxels, coors):
device = features.device
dtype = features.dtype
# Find distance of x, y, and z from cluster center
points_mean = features[:, :, :3].sum(
dim=1, keepdim=True) / num_voxels.type_as(features).view(-1, 1, 1)
f_cluster = features[:, :, :3] - points_mean
# Find distance of x, y, and z from pillar center
f_center = torch.zeros_like(features[:, :, :2])
f_center[:, :, 0] = features[:, :, 0] - (
coors[:, 3].to(dtype).unsqueeze(1) * self.vx + self.x_offset)
f_center[:, :, 1] = features[:, :, 1] - (
coors[:, 2].to(dtype).unsqueeze(1) * self.vy + self.y_offset)
# Combine together feature decorations
features_ls = [features, f_cluster, f_center]
if self._with_distance:
points_dist = torch.norm(features[:, :, :3], 2, 2, keepdim=True)
features_ls.append(points_dist)
features = torch.cat(features_ls, dim=-1)
# The feature decorations were calculated without regard to whether pillar was empty. Need to ensure that
# empty pillars remain set to zeros.
voxel_count = features.shape[1]
mask = get_paddings_indicator(num_voxels, voxel_count, axis=0)
mask = torch.unsqueeze(mask, -1).type_as(features)
features *= mask
# Forward pass through PFNLayers
for pfn in self.pfn_layers:
features = pfn(features)
return features.squeeze()
def forward(self, features, num_voxels, coors):
# print(coors.shape, num_voxels.shape, features.shape)
# print(torch.max(num_voxels), torch.min(num_voxels), torch.mean(num_voxels.float()), num_voxels.shape)
# print(torch.sum(num_voxels>=8).float()/num_voxels.shape[0])
device = features.device
dtype = features.dtype
# print(features.shape, num_voxels.shape, coors.shape)
# print(coors[0])
# print(num_voxels[0])
# Find distance of x, y, and z from cluster center
points_mean = features[:, :, :3].sum(
dim=1, keepdim=True) / num_voxels.type_as(features).view(-1, 1, 1)
f_cluster = features[:, :, :3] - points_mean
# Find distance of x, y, and z from pillar center
f_center = torch.zeros_like(features[:, :, :2])
f_center[:, :, 0] = features[:, :, 0] - (
coors[:, 3].to(dtype).unsqueeze(1) * self.vx + self.x_offset)
f_center[:, :, 1] = features[:, :, 1] - (
coors[:, 2].to(dtype).unsqueeze(1) * self.vy + self.y_offset)
# Combine together feature decorations
features_ls = [features, f_cluster, f_center]
if self._with_distance:
points_dist = torch.norm(features[:, :, :3], 2, 2, keepdim=True)
features_ls.append(points_dist)
features_out = torch.cat(features_ls, dim=-1)
# The feature decorations were calculated without regard to whether pillar was empty. Need to ensure that
# empty pillars remain set to zeros.
voxel_count = features_out.shape[1]
mask = get_paddings_indicator(num_voxels, voxel_count, axis=0)
mask = torch.unsqueeze(mask, -1).type_as(features_out)
# print(torch.sum(features[:,:,:3] - (features*mask)[:,:,:3]))
features_out = features_out * mask
# Forward pass through PFNLayers
for pfn in self.pfn_layers:
# print(features.shape)
# features = batch_process(features, pfn, num_batches=20)
features_out = batch_process(features_out, pfn, num_batches=10)
# features_out = pfn(features_out)
# print(features.grad_fn)
# print(mask)
features_out = features_out * mask
# print(mask.shape, features.shape)
features_max = torch.max(features_out, dim=1, keepdim=True)[0]
if pfn.last_vfe:
features_out = features_max
else:
# features_max = torch.max(x, dim=1, keepdim=True)[0]
features_repeat = features_max.repeat(1, features_out.shape[1],
1)
features_out = torch.cat([features_out, features_repeat],
dim=2)
# print(features.shape)
return features_out.squeeze()
def forward(self, features, num_voxels, coors, batch_size): #coors坐标
device = features.device
dtype = features.dtype #float32
# Find distance of x, y, and z from cluster center 求x, y, z到聚类中心的距离
points_mean = features[:, :, :3].sum(
dim=1, keepdim=True) / num_voxels.type_as(features).view(-1, 1, 1)
aa = features[:, :, :3].sum(dim=1, keepdim=True) #每个柱子60个点相加，只加x,y,z
bb = num_voxels.type_as(features).view(-1, 1, 1).shape
f_cluster = features[:, :, :3] - points_mean
# Find distance of x, y, and z from pillar center 求x, y, z到柱子中心的距离
f_center = torch.zeros_like(features[:, :, :2])
f_center[:, :, 0] = features[:, :, 0] - (
coors[:, 3].to(dtype).unsqueeze(1) * self.vx + self.x_offset)
f_center[:, :, 1] = features[:, :, 1] - (
coors[:, 2].to(dtype).unsqueeze(1) * self.vy + self.y_offset)
# Combine together feature decorations 组合特征装饰
if features.shape[2] == 4:
features_ls = [features, f_cluster,
f_center] #加上到中心点距离和到底面中心点距离变成9维度
if features.shape[2] == 6:
features_ls = [features, f_cluster]
if self._with_distance: #false
points_dist = torch.norm(features[:, :, :3], 2, 2, keepdim=True)
features_ls.append(points_dist)
features = torch.cat(features_ls, dim=-1) #feature shape
a = features.shape #torch.Size([50929, 60, 9]) #[3444,60,23]radar
# The feature decorations were calculated without regard to whether pillar was empty. Need to ensure that
# empty pillars remain set to zeros.特征装饰的计算不考虑柱子是否空。需要确保
# 空柱子仍然设置为零
voxel_count = features.shape[1] #一个柱子60的点
mask = get_paddings_indicator(num_voxels, voxel_count,
axis=0) #根据voxel的点数设置mask
mask = torch.unsqueeze(mask, -1).type_as(features)
features *= mask #特征mask掩膜(60,9,41924)
# Forward pass through PFNLayers
for pfn in self.pfn_layers:
features = pfn(features, batch_size)
# print(features.shape) #[38093,1,64]
return features.squeeze()
def forward(self, features, num_voxels, coors):
# Find distance of x, y, and z from cluster center
## ref: https://github.com/traveller59/second.pytorch/issues/144
num_voxels_set_0_to_1 = num_voxels.clone()
num_voxels_set_0_to_1[num_voxels_set_0_to_1 == 0] = 1
points_mean = features[:, :, :3].sum(
dim=1,
keepdim=True) / num_voxels_set_0_to_1.type_as(features).view(
-1, 1, 1)
#points_mean = features[:, :, :3].sum(dim=1, keepdim=True) / num_voxels.type_as(features).view(-1, 1, 1)
f_cluster = features[:, :, :3] - points_mean
# Find distance of x, y, and z from pillar center
f_center = torch.zeros_like(features[:, :, :2])
f_center[:, :, 0] = features[:, :, 0] - (
coors[:, 3].float().unsqueeze(1) * self.vx + self.x_offset)
f_center[:, :, 1] = features[:, :, 1] - (
coors[:, 2].float().unsqueeze(1) * self.vy + self.y_offset)
# Combine together feature decorations
features_ls = [features, f_cluster, f_center]
if self._with_distance:
points_dist = torch.norm(features[:, :, :3], 2, 2, keepdim=True)
features_ls.append(points_dist)
features = torch.cat(features_ls, dim=-1)
# The feature decorations were calculated without regard to whether pillar was empty. Need to ensure that
# empty pillars remain set to zeros.
voxel_count = features.shape[1]
mask = get_paddings_indicator(num_voxels, voxel_count, axis=0)
mask = torch.unsqueeze(mask, -1).type_as(features)
features *= mask ## (N,60,9) ##(N,100,9)
features = self.VoxelFeature_TA(points_mean, features)
# Forward pass through PFNLayers
for pfn in self.pfn_layers:
features = pfn(features)
return features.squeeze()