File size: 5,038 Bytes
ffa9b64
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import torch.nn as nn

class MusicCNN(nn.Module):
    def __init__(self, num_classes, dropout_rate=0.3, device="cuda"):
        super(MusicCNN, self).__init__()
        self.device = device

        # Convolutional blocks
        self.conv_block1 = nn.Sequential(
            nn.Conv2d(1, 32, kernel_size=3, padding=1),
            nn.BatchNorm2d(32),
            nn.ReLU(),
            nn.Conv2d(32, 32, kernel_size=3, padding=1),
            nn.BatchNorm2d(32),
            nn.ReLU(),
            nn.MaxPool2d(2, 2),
            nn.Dropout2d(dropout_rate)
        ).to(device)

        self.conv_block2 = nn.Sequential(
            nn.Conv2d(32, 64, kernel_size=3, padding=1),
            nn.BatchNorm2d(64),
            nn.ReLU(),
            nn.Conv2d(64, 64, kernel_size=3, padding=1),
            nn.BatchNorm2d(64),
            nn.ReLU(),
            nn.MaxPool2d(2, 2),
            nn.Dropout2d(dropout_rate)
        ).to(device)

        self.conv_block3 = nn.Sequential(
            nn.Conv2d(64, 128, kernel_size=3, padding=1),
            nn.BatchNorm2d(128),
            nn.ReLU(),
            nn.Conv2d(128, 128, kernel_size=3, padding=1),
            nn.BatchNorm2d(128),
            nn.ReLU(),
            nn.MaxPool2d(2, 2),
            nn.Dropout2d(dropout_rate)
        ).to(device)

        self.fc_layers = None  # Fully connected layers will be initialized later
        self.num_classes = num_classes
        self.dropout_rate = dropout_rate

    def forward(self, x):
        x = self.conv_block1(x)
        x = self.conv_block2(x)
        x = self.conv_block3(x)

        # Flatten dynamically
        x = x.view(x.size(0), -1)

        # Initialize FC layers dynamically
        if self.fc_layers is None:
            fc_input_size = x.size(1)
            self.fc_layers = nn.Sequential(
                nn.Linear(fc_input_size, 512),
                nn.BatchNorm1d(512),
                nn.ReLU(),
                nn.Dropout(self.dropout_rate),
                nn.Linear(512, 256),
                nn.BatchNorm1d(256),
                nn.ReLU(),
                nn.Dropout(self.dropout_rate),
                nn.Linear(256, self.num_classes)
            ).to(self.device)

        x = self.fc_layers(x)
        return x


class MusicCRNN2D(nn.Module):
    def __init__(self, num_classes, dropout_rate=0.1, gru_hidden_size=32, device="cuda"):
        super(MusicCRNN2D, self).__init__()
        self.device = device

        # Input batch normalization
        self.input_bn = nn.BatchNorm2d(1).to(device)

        # Convolutional blocks
        self.conv_block1 = nn.Sequential(
            nn.Conv2d(1, 64, kernel_size=3, padding=1),
            nn.BatchNorm2d(64),
            nn.ELU(),
            nn.MaxPool2d((2, 2)),
            nn.Dropout2d(dropout_rate)
        ).to(device)

        self.conv_block2 = nn.Sequential(
            nn.Conv2d(64, 128, kernel_size=3, padding=1),
            nn.BatchNorm2d(128),
            nn.ELU(),
            nn.MaxPool2d((4, 2)),
            nn.Dropout2d(dropout_rate)
        ).to(device)

        self.conv_block3 = nn.Sequential(
            nn.Conv2d(128, 128, kernel_size=3, padding=1),
            nn.BatchNorm2d(128),
            nn.ELU(),
            nn.MaxPool2d((4, 2)),
            nn.Dropout2d(dropout_rate)
        ).to(device)

        self.conv_block4 = nn.Sequential(
            nn.Conv2d(128, 128, kernel_size=3, padding=1),
            nn.BatchNorm2d(128),
            nn.ELU(),
            nn.MaxPool2d((4, 2)),
            nn.Dropout2d(dropout_rate)
        ).to(device)

        self.gru_stack = None  # GRU layers will be initialized later
        self.classifier = None
        self.num_classes = num_classes
        self.dropout_rate = dropout_rate
        self.gru_hidden_size = gru_hidden_size

    def forward(self, x):
        x = self.input_bn(x)
        x = self.conv_block1(x)
        x = self.conv_block2(x)
        x = self.conv_block3(x)
        x = self.conv_block4(x)

        # Reshape for GRU
        batch_size, _, freq, time = x.shape
        x = x.permute(0, 3, 1, 2)  # (batch, time, channels, freq)
        x = x.reshape(batch_size, time, -1)

        # Initialize GRU dynamically
        if self.gru_stack is None:
            gru_input_size = x.size(2)
            self.gru_stack = nn.GRU(
                input_size=gru_input_size,
                hidden_size=self.gru_hidden_size,
                batch_first=True,
                bidirectional=True,
            ).to(self.device)
            self.classifier = nn.Sequential(
                nn.Dropout(self.dropout_rate * 3),
                nn.Linear(self.gru_hidden_size * 2, self.num_classes)  # * 2 for bidirectional
            ).to(self.device)

        x, _ = self.gru_stack(x)

        # Take the last time step
        x = x[:, -1, :]

        # Classification
        x = self.classifier(x)
        return x