Hugging Face's logo

Spaces:
AIFILMS
/
audioldm-text-to-audio-generation
Runtime error

App Files Community
audioldm-text-to-audio-generation
File size: 7,177 Bytes 
23f4617
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
 
"""
Feature Fusion for Varible-Length Data Processing
AFF/iAFF is referred and modified from https://github.com/YimianDai/open-aff/blob/master/aff_pytorch/aff_net/fusion.py
According to the paper: Yimian Dai et al, Attentional Feature Fusion, IEEE Winter Conference on Applications of Computer Vision, WACV 2021
"""

import torch
import torch.nn as nn


class DAF(nn.Module):
    """
    η›΄ζŽ₯η›ΈεŠ  DirectAddFuse
    """

    def __init__(self):
        super(DAF, self).__init__()

    def forward(self, x, residual):
        return x + residual


class iAFF(nn.Module):
    """
    ε€šη‰ΉεΎθžεˆ iAFF
    """

    def __init__(self, channels=64, r=4, type="2D"):
        super(iAFF, self).__init__()
        inter_channels = int(channels // r)

        if type == "1D":
            # ζœ¬εœ°ζ³¨ζ„εŠ›
            self.local_att = nn.Sequential(
                nn.Conv1d(channels, inter_channels, kernel_size=1, stride=1, padding=0),
                nn.BatchNorm1d(inter_channels),
                nn.ReLU(inplace=True),
                nn.Conv1d(inter_channels, channels, kernel_size=1, stride=1, padding=0),
                nn.BatchNorm1d(channels),
            )

            # ε…¨ε±€ζ³¨ζ„εŠ›
            self.global_att = nn.Sequential(
                nn.AdaptiveAvgPool1d(1),
                nn.Conv1d(channels, inter_channels, kernel_size=1, stride=1, padding=0),
                nn.BatchNorm1d(inter_channels),
                nn.ReLU(inplace=True),
                nn.Conv1d(inter_channels, channels, kernel_size=1, stride=1, padding=0),
                nn.BatchNorm1d(channels),
            )

            # η¬¬δΊŒζ¬‘ζœ¬εœ°ζ³¨ζ„εŠ›
            self.local_att2 = nn.Sequential(
                nn.Conv1d(channels, inter_channels, kernel_size=1, stride=1, padding=0),
                nn.BatchNorm1d(inter_channels),
                nn.ReLU(inplace=True),
                nn.Conv1d(inter_channels, channels, kernel_size=1, stride=1, padding=0),
                nn.BatchNorm1d(channels),
            )
            # η¬¬δΊŒζ¬‘ε…¨ε±€ζ³¨ζ„εŠ›
            self.global_att2 = nn.Sequential(
                nn.AdaptiveAvgPool1d(1),
                nn.Conv1d(channels, inter_channels, kernel_size=1, stride=1, padding=0),
                nn.BatchNorm1d(inter_channels),
                nn.ReLU(inplace=True),
                nn.Conv1d(inter_channels, channels, kernel_size=1, stride=1, padding=0),
                nn.BatchNorm1d(channels),
            )
        elif type == "2D":
            # ζœ¬εœ°ζ³¨ζ„εŠ›
            self.local_att = nn.Sequential(
                nn.Conv2d(channels, inter_channels, kernel_size=1, stride=1, padding=0),
                nn.BatchNorm2d(inter_channels),
                nn.ReLU(inplace=True),
                nn.Conv2d(inter_channels, channels, kernel_size=1, stride=1, padding=0),
                nn.BatchNorm2d(channels),
            )

            # ε…¨ε±€ζ³¨ζ„εŠ›
            self.global_att = nn.Sequential(
                nn.AdaptiveAvgPool2d(1),
                nn.Conv2d(channels, inter_channels, kernel_size=1, stride=1, padding=0),
                nn.BatchNorm2d(inter_channels),
                nn.ReLU(inplace=True),
                nn.Conv2d(inter_channels, channels, kernel_size=1, stride=1, padding=0),
                nn.BatchNorm2d(channels),
            )

            # η¬¬δΊŒζ¬‘ζœ¬εœ°ζ³¨ζ„εŠ›
            self.local_att2 = nn.Sequential(
                nn.Conv2d(channels, inter_channels, kernel_size=1, stride=1, padding=0),
                nn.BatchNorm2d(inter_channels),
                nn.ReLU(inplace=True),
                nn.Conv2d(inter_channels, channels, kernel_size=1, stride=1, padding=0),
                nn.BatchNorm2d(channels),
            )
            # η¬¬δΊŒζ¬‘ε…¨ε±€ζ³¨ζ„εŠ›
            self.global_att2 = nn.Sequential(
                nn.AdaptiveAvgPool2d(1),
                nn.Conv2d(channels, inter_channels, kernel_size=1, stride=1, padding=0),
                nn.BatchNorm2d(inter_channels),
                nn.ReLU(inplace=True),
                nn.Conv2d(inter_channels, channels, kernel_size=1, stride=1, padding=0),
                nn.BatchNorm2d(channels),
            )
        else:
            raise f"the type is not supported"

        self.sigmoid = nn.Sigmoid()

    def forward(self, x, residual):
        flag = False
        xa = x + residual
        if xa.size(0) == 1:
            xa = torch.cat([xa, xa], dim=0)
            flag = True
        xl = self.local_att(xa)
        xg = self.global_att(xa)
        xlg = xl + xg
        wei = self.sigmoid(xlg)
        xi = x * wei + residual * (1 - wei)

        xl2 = self.local_att2(xi)
        xg2 = self.global_att(xi)
        xlg2 = xl2 + xg2
        wei2 = self.sigmoid(xlg2)
        xo = x * wei2 + residual * (1 - wei2)
        if flag:
            xo = xo[0].unsqueeze(0)
        return xo


class AFF(nn.Module):
    """
    ε€šη‰ΉεΎθžεˆ AFF
    """

    def __init__(self, channels=64, r=4, type="2D"):
        super(AFF, self).__init__()
        inter_channels = int(channels // r)

        if type == "1D":
            self.local_att = nn.Sequential(
                nn.Conv1d(channels, inter_channels, kernel_size=1, stride=1, padding=0),
                nn.BatchNorm1d(inter_channels),
                nn.ReLU(inplace=True),
                nn.Conv1d(inter_channels, channels, kernel_size=1, stride=1, padding=0),
                nn.BatchNorm1d(channels),
            )
            self.global_att = nn.Sequential(
                nn.AdaptiveAvgPool1d(1),
                nn.Conv1d(channels, inter_channels, kernel_size=1, stride=1, padding=0),
                nn.BatchNorm1d(inter_channels),
                nn.ReLU(inplace=True),
                nn.Conv1d(inter_channels, channels, kernel_size=1, stride=1, padding=0),
                nn.BatchNorm1d(channels),
            )
        elif type == "2D":
            self.local_att = nn.Sequential(
                nn.Conv2d(channels, inter_channels, kernel_size=1, stride=1, padding=0),
                nn.BatchNorm2d(inter_channels),
                nn.ReLU(inplace=True),
                nn.Conv2d(inter_channels, channels, kernel_size=1, stride=1, padding=0),
                nn.BatchNorm2d(channels),
            )
            self.global_att = nn.Sequential(
                nn.AdaptiveAvgPool2d(1),
                nn.Conv2d(channels, inter_channels, kernel_size=1, stride=1, padding=0),
                nn.BatchNorm2d(inter_channels),
                nn.ReLU(inplace=True),
                nn.Conv2d(inter_channels, channels, kernel_size=1, stride=1, padding=0),
                nn.BatchNorm2d(channels),
            )
        else:
            raise f"the type is not supported."

        self.sigmoid = nn.Sigmoid()

    def forward(self, x, residual):
        flag = False
        xa = x + residual
        if xa.size(0) == 1:
            xa = torch.cat([xa, xa], dim=0)
            flag = True
        xl = self.local_att(xa)
        xg = self.global_att(xa)
        xlg = xl + xg
        wei = self.sigmoid(xlg)
        xo = 2 * x * wei + 2 * residual * (1 - wei)
        if flag:
            xo = xo[0].unsqueeze(0)
        return xo