def __init__(self,
num_classes,
input_channels=3,
use_xyz=True,
type_r=None):
super(Pointnet2SSG, self).__init__()
self.SA_modules = nn.ModuleList()
self.SA_modules.append(
PointnetSAModule(
npoint=512,
radius=0.2,
nsample=32,
mlp=[input_channels, 64, 64, 128],
use_xyz=True,
))
self.SA_modules.append(
PointnetSAModule(
npoint=128,
radius=0.4,
nsample=64,
mlp=[128, 128, 128, 256],
use_xyz=use_xyz,
))
self.SA_modules.append(
PointnetSAModule(mlp=[256, 256, 512, 1024], use_xyz=use_xyz))
self.FC_layer = (pt_utils.Seq(1024).fc(512, bn=True).dropout(0.5).fc(
256, bn=True).dropout(0.5).fc(num_classes, activation=None))
def __init__(self, input_channels=3, use_xyz=True):
super(Pointnet_Backbone, self).__init__()
self.SA_modules = nn.ModuleList()
self.SA_modules.append(
PointnetSAModule(
radius=0.3,
nsample=32,
mlp=[input_channels, 64, 64, 128],
use_xyz=use_xyz,
))
self.SA_modules.append(
PointnetSAModule(
radius=0.5,
nsample=32,
mlp=[128, 128, 128, 256],
use_xyz=use_xyz,
))
self.SA_modules.append(
PointnetSAModule(
radius=0.7,
nsample=32,
mlp=[256, 256, 256, 256],
use_xyz=use_xyz,
))
self.cov_final = nn.Conv1d(256, 256, kernel_size=1)
def __init__(self,
num_classes,
input_channels=3,
use_xyz=True,
type_r='qpu',
loss='ce'):
super(PointNetPoseEstimate, self).__init__()
self.PN = nn.ModuleList()
if type_r == 'qpu':
self.PN.append(
PointnetSAModuleQPU(npoint=256,
radius=0.4,
nsample=32,
mlp=nn.Sequential(
qpu_layers.QPU(4 * 8, 64 * 4),
qpu_layers.KeepImPart(dim=-1,
type_r='angle'),
),
use_xyz=use_xyz))
else:
self.PN.append(
PointnetSAModule(
npoint=256,
radius=0.4,
nsample=32,
mlp=[input_channels, 64, 64, 64 * 3],
use_xyz=True,
))
self.PN.append(
PointnetSAModule(
npoint=128,
radius=0.4,
nsample=16, # 64
# mlp=[32, 128],
mlp=[64 * 3, 192, 192, 256],
use_xyz=use_xyz))
self.PN.append(
PointnetSAModule(mlp=[256, 256, 512, 1024], use_xyz=use_xyz))
# self.PN.append(PointnetSAModule(mlp=[256, 384, 384, 512], use_xyz=use_xyz))
self.loss = loss
self.step = 5
self.mix = lambda x, y: torch.cat([x.squeeze(-1), y.squeeze(-1)], -1)
featShape = (512 * 2, 384)
# featShape = (512,128)
if loss == 'ce': # include ['ce']
self.pred_theta = self.multi_cls_head(*featShape, 360 // self.step)
self.pred_alpha = self.multi_cls_head(*featShape, 180 // self.step)
self.pred_beta = self.multi_cls_head(*featShape, 180 // self.step)
elif loss == 'weighted_ce':
self.pred_theta = self.multi_cls_head(*featShape, 360 // self.step,
-1)
self.pred_alpha = self.multi_cls_head(*featShape, 180 // self.step,
-1)
self.pred_beta = self.multi_cls_head(*featShape, 180 // self.step,
-1)
else: # include ['mse']
self.pred_theta = self.regression_head(*featShape, 1)
self.pred_alpha = self.regression_head(*featShape, 1)
self.pred_beta = self.regression_head(*featShape, 1)
def __init__(self, num_classes, input_channels=3, use_xyz=True):
super().__init__()
self.SA_modules = nn.ModuleList()
self.SA_modules.append(
PointnetSAModuleMSG(npoint=512,
radii=[0.1, 0.2, 0.4],
nsamples=[15, 32, 128],
mlps=[[input_channels, 32, 32, 64],
[input_channels, 64, 64, 128],
[input_channels, 64, 96, 128]],
use_xyz=use_xyz))
input_channels = 64 + 128 + 128
self.SA_modules.append(
PointnetSAModuleMSG(npoint=128,
radii=[0.2, 0.4, 0.8],
nsamples=[32, 64, 128],
mlps=[[input_channels, 64, 64, 128],
[input_channels, 128, 128, 256],
[input_channels, 128, 128, 256]],
use_xyz=use_xyz))
self.SA_modules.append(
PointnetSAModule(mlp=[128 + 256 + 256, 256, 512, 1024],
use_xyz=use_xyz))
self.FC_layer = nn.Sequential(
pt_utils.FC(1024, 512, bn=True), nn.Dropout(p=0.5),
pt_utils.FC(512, 256, bn=True), nn.Dropout(p=0.5),
pt_utils.FC(256, num_classes, activation=None))
def __init__(self, num_classes, input_channels=3, use_xyz=True):
super(Pointnet2MSG, self).__init__()
self.SA_modules = nn.ModuleList()
self.SA_modules.append(
PointnetSAModuleMSG(
npoint=512,
radii=[0.1, 0.2, 0.4],
nsamples=[16, 32, 128],
mlps=[
[input_channels, 32, 32, 64],
[input_channels, 64, 64, 128],
[input_channels, 64, 96, 128],
],
use_xyz=use_xyz,
))
input_channels = 64 + 128 + 128
self.SA_modules.append(
PointnetSAModuleMSG(
npoint=128,
radii=[0.2, 0.4, 0.8],
nsamples=[32, 64, 128],
mlps=[
[input_channels, 64, 64, 128],
[input_channels, 128, 128, 256],
[input_channels, 128, 128, 256],
],
use_xyz=use_xyz,
))
self.SA_modules.append(
PointnetSAModule(mlp=[128 + 256 + 256, 256, 512, 1024],
use_xyz=use_xyz))
self.FC_layer = (pt_utils.Seq(1024).fc(512, bn=True).dropout(0.5).fc(
256, bn=True).dropout(0.5).fc(num_classes, activation=None))
def __init__(self, num_classes, input_channels=3, use_xyz=True):
super(Pointnet2SSG, self).__init__()
self.SA_modules = nn.ModuleList()
self.SA_modules.append(
PointnetSAModule(
npoint=1024,
radius=0.1,
nsample=32,
mlp=[input_channels, 32, 32, 64],
use_xyz=use_xyz,
))
self.SA_modules.append(
PointnetSAModule(
npoint=256,
radius=0.2,
nsample=32,
mlp=[64, 64, 64, 128],
use_xyz=use_xyz,
))
self.SA_modules.append(
PointnetSAModule(
npoint=64,
radius=0.4,
nsample=32,
mlp=[128, 128, 128, 256],
use_xyz=use_xyz,
))
self.SA_modules.append(
PointnetSAModule(
npoint=16,
radius=0.8,
nsample=32,
mlp=[256, 256, 256, 512],
use_xyz=use_xyz,
))
self.FP_modules = nn.ModuleList()
self.FP_modules.append(
PointnetFPModule(mlp=[128 + input_channels, 128, 128, 128]))
self.FP_modules.append(PointnetFPModule(mlp=[256 + 64, 256, 128]))
self.FP_modules.append(PointnetFPModule(mlp=[256 + 128, 256, 256]))
self.FP_modules.append(PointnetFPModule(mlp=[512 + 256, 256, 256]))
self.FC_layer = (pt_utils.Seq(128).conv1d(
128, bn=True).dropout().conv1d(num_classes, activation=None))
def __init__(self, num_classes, input_channels=3, use_xyz=True):
super(Pointnet2SSGQPU, self).__init__()
self.SA_modules = nn.ModuleList()
self.SA_modules.append(
PointnetSAModuleQPU(
npoint=256,
radius=0.4,
nsample=32,
mlp=nn.Sequential(
qpu_layers.QPU(4*8, 64*4),
qpu_layers.KeepRealPart(dim=-1),
nn.Linear(64, 128)
),
use_xyz=use_xyz
)
)
self.SA_modules.append(
PointnetSAModule(
npoint=128,
radius=0.4,
nsample=16, # 64
# mlp=[32, 128],
mlp=[128, 128, 128, 256],
use_xyz=use_xyz
)
)
self.SA_modules.append(
PointnetSAModule(mlp=[256, 256, 512, 1024], use_xyz=use_xyz)
)
self.FC_layer = (
pt_utils.Seq(1024)
.fc(256, bn=True)
.dropout(0.5)
.fc(128, bn=True)
.dropout(0.5)
.fc(num_classes, activation=None)
)
def __init__(self,
num_classes,
input_channels=3,
use_xyz=True,
type_r='r'):
super(Pointnet2SSGQPU, self).__init__()
if type_r == 'no_r':
kp = qpu_layers.KeepImPart
temp_size = 64 * 3
else:
kp = qpu_layers.KeepRealPart
temp_size = 64 + 63 * ('inner' in type_r)
self.SA_modules = nn.ModuleList()
self.SA_modules.append(
PointnetSAModuleQPU(npoint=256,
radius=0.4,
nsample=32,
mlp=nn.Sequential(
qpu_layers.QPU(4 * 8, 64 * 4),
kp(dim=-1, type_r=type_r),
nn.Linear(temp_size, 128)),
use_xyz=use_xyz))
self.SA_modules.append(
PointnetSAModule(
npoint=128,
radius=0.4,
nsample=16, # 64
# mlp=[32, 128],
mlp=[128, 128, 128, 256],
use_xyz=use_xyz))
self.SA_modules.append(
PointnetSAModule(mlp=[256, 256, 512, 1024], use_xyz=use_xyz))
self.FC_layer = (pt_utils.Seq(1024).fc(256, bn=True).dropout(0.5).fc(
128, bn=True).dropout(0.5).fc(num_classes, activation=None))
def __init__(self, num_classes, input_channels=3, use_xyz=True):
super().__init__()
self.SA_modules = nn.ModuleList()
self.SA_modules.append(
PointnetSAModule(npoint=512,
radius=0.2,
nsample=64,
mlp=[input_channels, 64, 64, 128],
use_xyz=use_xyz))
self.SA_modules.append(
PointnetSAModule(npoint=128,
radius=0.4,
nsample=64,
mlp=[128, 128, 128, 256],
use_xyz=use_xyz))
self.SA_modules.append(
PointnetSAModule(mlp=[256, 256, 512, 1024], use_xyz=use_xyz))
self.FC_layer = nn.Sequential(
pt_utils.FC(1024, 512, bn=True), nn.Dropout(p=0.5),
pt_utils.FC(512, 256, bn=True), nn.Dropout(p=0.5),
pt_utils.FC(256, num_classes, activation=None))
def __init__(self, input_channels=3, use_xyz=True):
super(PCEncoder, self).__init__()
self.SA_modules = nn.ModuleList()
self.SA_modules.append(
PointnetSAModule(
npoint=512,
radius=0.2,
nsample=64,
mlp=[input_channels, 64, 64, 128],
use_xyz=use_xyz,
))
self.SA_modules.append(
PointnetSAModule(
npoint=128,
radius=0.4,
nsample=64,
mlp=[128, 128, 128, 256],
use_xyz=use_xyz,
))
self.SA_modules.append(
PointnetSAModule(mlp=[256, 256, 512, 1024], use_xyz=use_xyz))
self.FC_layer = (pt_utils.Seq(1024).fc(256, bn=False, activation=None))
def __init__(self,
num_classes,
input_channels=3,
use_xyz=True,
n_strxmani=None,
align=False):
super(Pointnet2MSG_manimix, self).__init__()
self.SA_modules = nn.ModuleList()
self.SA_modules.append(
PointnetSAModuleMSG(
npoint=512,
radii=[0.1, 0.2, 0.4],
nsamples=[16, 32, 128],
mlps=[
[input_channels, 32, 32, 64],
[input_channels, 64, 64, 128],
[input_channels, 64, 96, 128],
],
use_xyz=use_xyz,
))
input_channels = 64 + 128 + 128
self.SA_modules.append(
PointnetSAModuleMSG(
npoint=128,
radii=[0.2, 0.4, 0.8],
nsamples=[32, 64, 128],
mlps=[
[input_channels, 64, 64, 128],
[input_channels, 128, 128, 256],
[input_channels, 128, 128, 256],
],
use_xyz=use_xyz,
))
self.SA_modules.append(
PointnetSAModule(mlp=[128 + 256 + 256, 256, 512, 1024],
use_xyz=use_xyz))
self.FC_layer = (pt_utils.Seq(1024).fc(512, bn=True).dropout(0.5).fc(
256, bn=True).dropout(0.5).fc(num_classes, activation=None))
if n_strxmani is not None:
self.mixupbias_layer = (pt_utils.Seq(n_strxmani).fc(
64, bn=True).fc(num_classes, activation=None))
self.EMD = emd_module.emdModule()
self.cd = CD()
self.align = align
def __init__(self, input_channels=3, use_xyz=True, objective=False):
super(Pointnet_Tracking, self).__init__()
self.backbone_net = Pointnet_Backbone(input_channels, use_xyz)
self.cosine = nn.CosineSimilarity(dim=1)
self.mlp = pt_utils.SharedMLP([4 + 256, 256, 256, 256], bn=True)
self.FC_layer_cla = (pt_utils.Seq(256).conv1d(256, bn=True).conv1d(
256, bn=True).conv1d(1, activation=None))
self.fea_layer = (pt_utils.Seq(256).conv1d(256, bn=True).conv1d(
256, activation=None))
self.vote_layer = (pt_utils.Seq(3 + 256).conv1d(256, bn=True).conv1d(
256, bn=True).conv1d(3 + 256, activation=None))
self.vote_aggregation = PointnetSAModule(radius=0.3,
nsample=16,
mlp=[1 + 256, 256, 256, 256],
use_xyz=use_xyz)
self.num_proposal = 64
self.FC_proposal = (pt_utils.Seq(256).conv1d(256, bn=True).conv1d(
256, bn=True).conv1d(3 + 1 + 1, activation=None))
def __init__(self, input_channels=3, use_xyz=True, objective=False):
super(Pointnet_Tracking, self).__init__()
self.backbone_net = Pointnet_Backbone(input_channels, use_xyz)
self.cosine = nn.CosineSimilarity(dim=1)
self.mlp = pt_utils.SharedMLP([4 + 256, 256, 256, 256], bn=True)
# SharedMLP(
# (layer0): Conv2d(
# (conv): Conv2d(260, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
# (normlayer): BatchNorm2d(
# (bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
# )
# (activation): ReLU(inplace=True)
# )
# (layer1): Conv2d(
# (conv): Conv2d(256, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
# (normlayer): BatchNorm2d(
# (bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
# )
# (activation): ReLU(inplace=True)
# )
# (layer2): Conv2d(
# (conv): Conv2d(256, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
# (normlayer): BatchNorm2d(
# (bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
# )
# (activation): ReLU(inplace=True)
# )
# )
self.FC_layer_cla = (pt_utils.Seq(256).conv1d(256, bn=True).conv1d(
256, bn=True).conv1d(1, activation=None))
# Seq(
# (0): Conv1d(
# (conv): Conv1d(256, 256, kernel_size=(1,), stride=(1,), bias=False)
# (normlayer): BatchNorm1d(
# (bn): BatchNorm1d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
# )
# (activation): ReLU(inplace=True)
# )
# (1): Conv1d(
# (conv): Conv1d(256, 256, kernel_size=(1,), stride=(1,), bias=False)
# (normlayer): BatchNorm1d(
# (bn): BatchNorm1d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
# )
# (activation): ReLU(inplace=True)
# )
# (2): Conv1d(
# (conv): Conv1d(256, 1, kernel_size=(1,), stride=(1,))
# )
# )
self.fea_layer = (pt_utils.Seq(256).conv1d(256, bn=True).conv1d(
256, activation=None))
# Seq(
# (0): Conv1d(
# (conv): Conv1d(256, 256, kernel_size=(1,), stride=(1,), bias=False)
# (normlayer): BatchNorm1d(
# (bn): BatchNorm1d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
# )
# (activation): ReLU(inplace=True)
# )
# (1): Conv1d(
# (conv): Conv1d(256, 256, kernel_size=(1,), stride=(1,))
# )
# )
self.vote_layer = (pt_utils.Seq(3 + 256).conv1d(256, bn=True).conv1d(
256, bn=True).conv1d(3 + 256, activation=None))
# Seq(
# (0): Conv1d(
# (conv): Conv1d(259, 256, kernel_size=(1,), stride=(1,), bias=False)
# (normlayer): BatchNorm1d(
# (bn): BatchNorm1d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
# )
# (activation): ReLU(inplace=True)
# )
# (1): Conv1d(
# (conv): Conv1d(256, 256, kernel_size=(1,), stride=(1,), bias=False)
# (normlayer): BatchNorm1d(
# (bn): BatchNorm1d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
# )
# (activation): ReLU(inplace=True)
# )
# (2): Conv1d(
# (conv): Conv1d(256, 259, kernel_size=(1,), stride=(1,))
# )
# )
self.vote_aggregation = PointnetSAModule(
radius=0.3,
nsample=16,
mlp=[1 + 256, 256, 256, 256],
use_xyz=use_xyz) # 根据投票聚合投影候选中心点
# PointnetSAModule(
# (groupers): ModuleList(
# (0): QueryAndGroup()
# )
# (mlps): ModuleList(
# (0): SharedMLP(
# (layer0): Conv2d(
# (conv): Conv2d(260, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
# (normlayer): BatchNorm2d(
# (bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
# )
# (activation): ReLU(inplace=True)
# )
# (layer1): Conv2d(
# (conv): Conv2d(256, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
# (normlayer): BatchNorm2d(
# (bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
# )
# (activation): ReLU(inplace=True)
# )
# (layer2): Conv2d(
# (conv): Conv2d(256, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
# (normlayer): BatchNorm2d(
# (bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
# )
# (activation): ReLU(inplace=True)
# )
# )
# )
# )
self.num_proposal = 64 # 实际上是_PointnetSAModuleBase(nn.Module)的forward中的npoint输入参数
self.FC_proposal = (pt_utils.Seq(256).conv1d(256, bn=True).conv1d(
256, bn=True).conv1d(3 + 1 + 1, activation=None))
def __init__(self, input_channels=3, use_xyz=True):
super(Pointnet_Backbone, self).__init__()
self.SA_modules = nn.ModuleList()
self.SA_modules.append(
PointnetSAModule(
radius=0.3,
nsample=32,
mlp=[input_channels, 64, 64, 128],
use_xyz=use_xyz,
))
self.SA_modules.append(
PointnetSAModule(
radius=0.5,
nsample=32,
mlp=[128, 128, 128, 256],
use_xyz=use_xyz,
))
self.SA_modules.append(
PointnetSAModule(
radius=0.7,
nsample=32,
mlp=[256, 256, 256, 256],
use_xyz=use_xyz,
))
# ModuleList(
# (0): PointnetSAModule(
# (groupers): ModuleList(
# (0): QueryAndGroup(radius=0.3, nsample=32)
# )
# (mlps): ModuleList(
# (0): SharedMLP(
# (layer0): Conv2d(
# (conv): Conv2d(3, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
# (normlayer): BatchNorm2d(
# (bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
# )
# (activation): ReLU(inplace=True)
# )
# (layer1): Conv2d(
# (conv): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
# (normlayer): BatchNorm2d(
# (bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
# )
# (activation): ReLU(inplace=True)
# )
# (layer2): Conv2d(
# (conv): Conv2d(64, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
# (normlayer): BatchNorm2d(
# (bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
# )
# (activation): ReLU(inplace=True)
# )
# )
# )
# )
# (1): PointnetSAModule(
# (groupers): ModuleList(
# (0): QueryAndGroup(radius=0.5, nsample=32)
# )
# (mlps): ModuleList(
# (0): SharedMLP(
# (layer0): Conv2d(
# (conv): Conv2d(131, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
# (normlayer): BatchNorm2d(
# (bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
# )
# (activation): ReLU(inplace=True)
# )
# (layer1): Conv2d(
# (conv): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
# (normlayer): BatchNorm2d(
# (bn): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
# )
# (activation): ReLU(inplace=True)
# )
# (layer2): Conv2d(
# (conv): Conv2d(128, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
# (normlayer): BatchNorm2d(
# (bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
# )
# (activation): ReLU(inplace=True)
# )
# )
# )
# )
# (2): PointnetSAModule(
# (groupers): ModuleList(
# (0): QueryAndGroup(radius=0.7, nsample=32)
# )
# (mlps): ModuleList(
# (0): SharedMLP(
# (layer0): Conv2d(
# (conv): Conv2d(259, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
# (normlayer): BatchNorm2d(
# (bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
# )
# (activation): ReLU(inplace=True)
# )
# (layer1): Conv2d(
# (conv): Conv2d(256, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
# (normlayer): BatchNorm2d(
# (bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
# )
# (activation): ReLU(inplace=True)
# )
# (layer2): Conv2d(
# (conv): Conv2d(256, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
# (normlayer): BatchNorm2d(
# (bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
# )
# (activation): ReLU(inplace=True)
# )
# )
# )
# )
# )
self.cov_final = nn.Conv1d(256, 256, kernel_size=1)
