-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathNaive_cross_line_net.py
226 lines (184 loc) · 8.27 KB
/
Naive_cross_line_net.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
import torch
import torch.nn as nn
import torch.nn.functional as F
import numpy as np
import matplotlib.pyplot as plt
class _1024_pointnet(nn.Module):
def __init__(self, cat_num=4, dim=2):
super(_1024_pointnet, self).__init__()
self.cat_num = cat_num
self._dim = dim
self.conv1 = torch.nn.Conv1d(self._dim, 64, 1)
self.conv2 = torch.nn.Conv1d(64, 128, 1)
self.conv3 = torch.nn.Conv1d(128, 128, 1)
self.conv4 = torch.nn.Conv1d(128, 512, 1)
self.conv5 = torch.nn.Conv1d(512, 1024, 1)
self.bn1 = nn.BatchNorm1d(64)
self.bn2 = nn.BatchNorm1d(128)
self.bn3 = nn.BatchNorm1d(128)
self.bn4 = nn.BatchNorm1d(512)
self.bn5 = nn.BatchNorm1d(1024)
self.convs0 = torch.nn.Conv1d(2880, 1024, 1)
self.convs1 = torch.nn.Conv1d(1024, 512, 1)
self.convs2 = torch.nn.Conv1d(512, 256, 1)
self.convs3 = torch.nn.Conv1d(256, 128, 1)
self.convs4 = torch.nn.Conv1d(128, cat_num, 1)
self.bns0 = nn.BatchNorm1d(1024)
self.bns1 = nn.BatchNorm1d(512)
self.bns2 = nn.BatchNorm1d(256)
self.bns3 = nn.BatchNorm1d(128)
def forward(self, points):
B, D, N = points.size()
out1 = F.relu(self.bn1(self.conv1(points)))#[B, 64, N]
out2 = F.relu(self.bn2(self.conv2(out1)))#[B, 128, N]
out3 = F.relu(self.bn3(self.conv3(out2)))#[B, 128, N]
out4 = F.relu(self.bn4(self.conv4(out3)))#[B, 512, N]
out5 = F.relu(self.bn5(self.conv5(out4))) # [B, 1024, N]
out_max = torch.max(out5, 2, keepdim=True)[0]
out_max = out_max.view(-1, 1024)#[B, 1024]
expand = out_max.view(-1, 1024, 1).repeat(1, 1, N)#[B, 1024, N]
concat_r = torch.cat([expand, out1, out2, out3, out4, out5], 1)##[B, 64+128+128+512+1024+1024=2880, N]
net = F.relu(self.bns0(self.convs0(concat_r)))
net = F.relu(self.bns1(self.convs1(net)))
net = F.relu(self.bns2(self.convs2(net)))
net = F.relu(self.bns3(self.convs3(net)))
net = self.convs4(net)
return net
class _512_pointnet(nn.Module):
def __init__(self, cat_num=4, dim=2):
super(_512_pointnet, self).__init__()
self.cat_num = cat_num
self._dim = dim
self.conv1 = torch.nn.Conv1d(self._dim, 64, 1)
self.conv2 = torch.nn.Conv1d(64, 128, 1)
self.conv3 = torch.nn.Conv1d(128, 128, 1)
self.conv4 = torch.nn.Conv1d(128, 512, 1)
self.bn1 = nn.BatchNorm1d(64)
self.bn2 = nn.BatchNorm1d(128)
self.bn3 = nn.BatchNorm1d(128)
self.bn4 = nn.BatchNorm1d(512)
self.convs1 = torch.nn.Conv1d(1344, 256, 1)
self.convs2 = torch.nn.Conv1d(256, 256, 1)
self.convs3 = torch.nn.Conv1d(256, 128, 1)
self.convs4 = torch.nn.Conv1d(128, cat_num, 1)
self.bns1 = nn.BatchNorm1d(256)
self.bns2 = nn.BatchNorm1d(256)
self.bns3 = nn.BatchNorm1d(128)
def forward(self, points):
B, D, N = points.size()
out1 = F.relu(self.bn1(self.conv1(points)))#[B, 64, N]
out2 = F.relu(self.bn2(self.conv2(out1)))#[B, 128, N]
out3 = F.relu(self.bn3(self.conv3(out2)))#[B, 128, N]
out4 = F.relu(self.bn4(self.conv4(out3)))#[B, 512, N]
out_max = torch.max(out4, 2, keepdim=True)[0]
out_max = out_max.view(-1, 512)#[B, 512]
expand = out_max.view(-1, 512, 1).repeat(1, 1, N)#[B, 512, 840]
concat_r = torch.cat([expand, out1, out2, out3, out4], 1)##[B, 64+128+128+512+512=1344, N]
net = F.relu(self.bns1(self.convs1(concat_r)))
net = F.relu(self.bns2(self.convs2(net)))
net = F.relu(self.bns3(self.convs3(net)))
net = self.convs4(net)
return net
class _256_pointnet(nn.Module):
def __init__(self, cat_num=4, dim=2):
super(_256_pointnet, self).__init__()
self.cat_num = cat_num
self._dim = dim
self.conv1 = torch.nn.Conv1d(self._dim, 64, 1)
self.conv2 = torch.nn.Conv1d(64, 128, 1)
self.conv3 = torch.nn.Conv1d(128, 128, 1)
self.conv4 = torch.nn.Conv1d(128, 256, 1)
self.bn1 = nn.BatchNorm1d(64)
self.bn2 = nn.BatchNorm1d(128)
self.bn3 = nn.BatchNorm1d(128)
self.bn4 = nn.BatchNorm1d(256)
self.convs1 = torch.nn.Conv1d(832, 256, 1)
self.convs2 = torch.nn.Conv1d(256, 256, 1)
self.convs3 = torch.nn.Conv1d(256, 128, 1)
self.convs4 = torch.nn.Conv1d(128, cat_num, 1)
self.bns1 = nn.BatchNorm1d(256)
self.bns2 = nn.BatchNorm1d(256)
self.bns3 = nn.BatchNorm1d(128)
def forward(self, points):
B, D, N = points.size()
out1 = F.relu(self.bn1(self.conv1(points)))#[B, 64, N]
out2 = F.relu(self.bn2(self.conv2(out1)))#[B, 128, N]
out3 = F.relu(self.bn3(self.conv3(out2)))#[B, 128, N]
out4 = F.relu(self.bn4(self.conv4(out3)))#[B, 256, N]
out_max = torch.max(out4, 2, keepdim=True)[0]
out_max = out_max.view(-1, 256)#[B, 512]
expand = out_max.view(-1, 256, 1).repeat(1, 1, N)#[B, 512, N]
concat_r = torch.cat([expand, out1, out2, out3, out4], 1)##[B, 64+128+128+256+256=832, N]
net = F.relu(self.bns1(self.convs1(concat_r)))
net = F.relu(self.bns2(self.convs2(net)))
net = F.relu(self.bns3(self.convs3(net)))
net = self.convs4(net)
return net
class _128_pointnet(nn.Module):
def __init__(self, cat_num=4, dim=2):
super(_128_pointnet, self).__init__()
self.cat_num = cat_num
self._dim = dim
self.conv1 = torch.nn.Conv1d(self._dim, 64, 1)
self.conv2 = torch.nn.Conv1d(64, 128, 1)
self.conv3 = torch.nn.Conv1d(128, 128, 1)
self.bn1 = nn.BatchNorm1d(64)
self.bn2 = nn.BatchNorm1d(128)
self.bn3 = nn.BatchNorm1d(128)
self.convs1 = torch.nn.Conv1d(448, 256, 1)
self.convs3 = torch.nn.Conv1d(256, 128, 1)
self.convs4 = torch.nn.Conv1d(128, cat_num, 1)
self.bns1 = nn.BatchNorm1d(256)
self.bns3 = nn.BatchNorm1d(128)
def forward(self, points):
B, D, N = points.size()
out1 = F.relu(self.bn1(self.conv1(points)))#[B, 64, 840]
out2 = F.relu(self.bn2(self.conv2(out1)))#[B, 128, 840]
out3 = F.relu(self.bn3(self.conv3(out2)))#[B, 128, 840]
out_max = torch.max(out3, 2, keepdim=True)[0]
out_max = out_max.view(-1, 128)#[B, 512]
expand = out_max.view(-1, 128, 1).repeat(1, 1, N)#[B, 512, 840]
concat_r = torch.cat([expand, out1, out2, out3], 1)##[B, 64+128+128+128=448, N]
net = F.relu(self.bns1(self.convs1(concat_r)))
net = F.relu(self.bns3(self.convs3(net)))
net = self.convs4(net)
return net
class _64_pointnet(nn.Module):
def __init__(self, cat_num=4, dim=2):
super(_64_pointnet, self).__init__()
self.cat_num = cat_num
self._dim = dim
self.conv1 = torch.nn.Conv1d(self._dim, 32, 1)
self.conv2 = torch.nn.Conv1d(32, 64, 1)
self.conv3 = torch.nn.Conv1d(64, 64, 1)
self.bn1 = nn.BatchNorm1d(32)
self.bn2 = nn.BatchNorm1d(64)
self.bn3 = nn.BatchNorm1d(64)
self.convs1 = torch.nn.Conv1d(224, 128, 1)
self.convs3 = torch.nn.Conv1d(128, 128, 1)
self.convs4 = torch.nn.Conv1d(128, cat_num, 1)
self.bns1 = nn.BatchNorm1d(128)
self.bns3 = nn.BatchNorm1d(128)
def forward(self, points):
B, D, N = points.size()
out1 = F.relu(self.bn1(self.conv1(points)))#[B, 32, 840]
out2 = F.relu(self.bn2(self.conv2(out1)))#[B, 64, 840]
out3 = F.relu(self.bn3(self.conv3(out2)))#[B, 64, 840]
out_max = torch.max(out3, 2, keepdim=True)[0]
out_max = out_max.view(-1, 64)#[B, 512]
expand = out_max.view(-1, 64, 1).repeat(1, 1, N)#[B, 512, 840]
concat_r = torch.cat([expand, out1, out2, out3], 1)##[B, 32+64+64+64=224, N]
net = F.relu(self.bns1(self.convs1(concat_r)))
net = F.relu(self.bns3(self.convs3(net)))
net = self.convs4(net)
return net
def main():
# model = _1024_pointnet(2, 2)
model = _64_pointnet(2, 2)
xyz = torch.rand(5, 2, 1280)
test_out = (model(xyz))
total_trainable_params = sum(
p.numel() for p in model.parameters() if p.requires_grad)
print(f'{total_trainable_params:,} training parameters.')
if __name__ == '__main__':
main()