Skip to content
Snippets Groups Projects
Select Git revision
  • master default protected
1 result
Blame Permalink
model.py 5.32 KiB
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 






import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.nn.init as init

# TODO: update wth https://github.com/antao97/dgcnn.pytorch/blob/07d534c2702905010ec9991619f552d8cacae45b/model.py#L166
# TODO: There are mode conv layers there


class EdgeConv(nn.Module):
    def __init__(self, in_channels, out_channels):
        super(EdgeConv, self).__init__()
        self.in_channels = in_channels
        self.conv = nn.Sequential(
            nn.Conv2d(2*in_channels, out_channels, kernel_size=1, bias=False),
            nn.BatchNorm2d(out_channels),
            nn.ReLU()
        )

    def forward(self, x, k=20):
        #batch_size, num_points, in_channels

        batch_size, num_points, feature_dim = x.shape
        x = x.view(batch_size, num_points,feature_dim )
        knn_indices = self.knn(x, k)
        knn_gathered = self.gather_neighbors(x, knn_indices)
        edge_features = torch.cat((knn_gathered - x.unsqueeze(2).repeat(1,1,k,1), knn_gathered), dim=1)
        edge_features = edge_features.view(batch_size, 2*feature_dim, num_points, k)
        return self.conv(edge_features).transpose(2, 1)

    @staticmethod
    def knn(x, k):
        """Find the indices of the k nearest neighbors for each point in the input tensor."""
        batch_size, num_points, _ = x.shape
        x_expanded = x.unsqueeze(2).expand(-1, -1, num_points, -1)
        x_transposed = x.unsqueeze(1).expand(-1, num_points, -1, -1)
        distances = torch.norm(x_expanded - x_transposed, dim=-1)
        _, indices = distances.topk(k=k, dim=-1, largest=False)
        return indices
    
    def gather_neighbors(self, x, knn_indices):
        batch_size, num_points, _ = x.shape
        _, _, k = knn_indices.shape
        x_expanded = x.unsqueeze(2).repeat(1, 1, num_points, 1)
        neighbors = torch.gather(x_expanded, 2, knn_indices.view(batch_size, num_points, k, 1).repeat(1, 1, 1, self.in_channels))
        return neighbors
    
    
class Transform_Net(nn.Module):
    def __init__(self, k=3):
        super(Transform_Net, self).__init__()
        self.k = k

        self.conv1 = nn.Conv1d(k, 64, 1)
        self.conv2 = nn.Conv1d(64, 128, 1)
        self.conv3 = nn.Conv1d(128, 1024, 1)

        self.fc1 = nn.Linear(1024, 512)
        self.fc2 = nn.Linear(512, 256)
        self.fc3 = nn.Linear(256, k*k)

        self.bn1 = nn.BatchNorm1d(64)
        self.bn2 = nn.BatchNorm1d(128)
        self.bn3 = nn.BatchNorm1d(1024)
        self.bn4 = nn.BatchNorm1d(512)
        self.bn5 = nn.BatchNorm1d(256)

    def forward(self, x):
        # Input shape: (batch_size, k, num_points)
        x = x.transpose(2, 1)
        x = F.relu(self.bn1(self.conv1(x)))
        x = F.relu(self.bn2(self.conv2(x)))
        x = F.relu(self.bn3(self.conv3(x)))

        x = torch.max(x, 2, keepdim=True)[0]
        x = x.view(-1, 1024)

        x = F.relu(self.bn4(self.fc1(x)))
        x = F.relu(self.bn5(self.fc2(x)))
        x = self.fc3(x)

        identity = torch.eye(self.k, dtype=x.dtype, device=x.device)
        transform = x.view(-1, self.k, self.k) + identity

        transform = transform.transpose(2, 1)

        return transform

class DGCNN(nn.Module):
    def __init__(self, num_classes):
        super(DGCNN, self).__init__()
        self.transform_net = Transform_Net()
        self.edge_conv1 = EdgeConv(3, 64)
        self.edge_conv2 = EdgeConv(64, 128)
        self.edge_conv3 = EdgeConv(128, 256)
        self.edge_conv4 = EdgeConv(256, 512)
        self.bn5 = nn.BatchNorm1d(1024)
        self.conv5 = nn.Sequential(nn.Conv1d(512, 1024, kernel_size=1, bias=False),
                                   self.bn5,
                                   nn.LeakyReLU(negative_slope=0.2))
        self.linear1 = nn.Linear(2048, 512, bias=False)

        self.fc = nn.Sequential(
            nn.Linear(512, 256),
            nn.BatchNorm1d(256),
            nn.ReLU(),
            nn.Dropout(p=0.5),
            nn.Linear(256, 128),
            nn.BatchNorm1d(128),
            nn.ReLU(),
            nn.Dropout(p=0.5),
            nn.Linear(128, num_classes),
        )

    def forward(self, x):
        # Apply Transform_Net on input point cloud
        batch_size = x.size(0)

        trans_matrix = self.transform_net(x)
        x = torch.bmm(x, trans_matrix)
        x1 = self.edge_conv1(x)
        x1 = x1.max(dim=-1, keepdim=False)[0] 
        # print("x1 shape: ", x1.shape)
        x2 = self.edge_conv2(x1)
        x2 = x2.max(dim=-1, keepdim=False)[0]
        # print("x2 shape: ", x2.shape)
        x3 = self.edge_conv3(x2)
        x3 = x3.max(dim=-1, keepdim=False)[0]
        # print("x3 shape: ", x3.shape)
        x4 = self.edge_conv4(x3)
        x4 = x4.max(dim=-1, keepdim=False)[0]
        # print("x4 shape: ", x4.shape)
        # x5 = torch.cat((x1, x2, x3, x4), dim=1)  # (batch_size, 64+64+128+256, num_points)
        # x6 = F.adaptive_max_pool1d(x5, 1).view(batch_size, -1)           # (batch_size, emb_dims, num_points) -> (batch_size, emb_dims)
        # x7 = F.adaptive_avg_pool1d(x5, 1).view(batch_size, -1)           # (batch_size, emb_dims, num_points) -> (batch_size, emb_dims)
        # x8 = torch.cat((x6, x7), 1)              # (batch_size, emb_dims*2)

        # x8 = F.leaky_relu(self.bn6(self.linear1(x8)), negative_slope=0.2) # (batch_size, emb_dims*2) -> (batch_size, 512)

        x9 = torch.max(x4, dim=1, keepdim=True)[0]
        x10 = self.fc(x9.squeeze(1))
        return x10