MoE_wavelet/wavelet_function.py at main · btma48/MoE_wavelet · GitHub

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
import time
import pywt
import math
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.autograd import Function

class DWT_Function(Function):
    @staticmethod
    def forward(ctx, x, w_ll,w_lh, w_hl, w_hh, padding):
        x = x.contiguous()
        ctx.save_for_backward(w_ll,w_lh, w_hl, w_hh)
        ctx.shape = x.shape
        ctx.padding = padding

        dim = x.shape[1]

        x_ll = torch.nn.functional.conv2d(x, w_ll.expand(dim, -1, -1, -1), stride = 2, padding=padding, groups = dim)
        x_lh = torch.nn.functional.conv2d(x, w_lh.expand(dim, -1, -1, -1), stride = 2, padding=padding, groups = dim)
        x_hl = torch.nn.functional.conv2d(x, w_hl.expand(dim, -1, -1, -1), stride = 2, padding=padding, groups = dim)
        x_hh = torch.nn.functional.conv2d(x, w_hh.expand(dim, -1, -1, -1), stride = 2, padding=padding, groups = dim)
        x = torch.cat([x_ll, x_lh, x_hl, x_hh], dim=1)
        return x

    @staticmethod
    def backward(ctx, dx):
        if ctx.needs_input_grad[0]:
            w_ll,w_lh, w_hl, w_hh = ctx.saved_tensors
            B, C, H, W = ctx.shape
            dx = dx.view(B, 4, -1, H//2, W//2)
            padding = ctx.padding

            dx = dx.transpose(1,2).reshape(B, -1, H//2, W//2)
            filters = torch.cat([w_ll, w_lh, w_hl, w_hh], dim=0).repeat(C, 1, 1, 1)
            dx = torch.nn.functional.conv_transpose2d(dx, filters, stride=2, padding=padding, groups=C)

        return dx, None, None, None, None, None

class IDWT_Function(Function):
    @staticmethod
    def forward(ctx, x, filters, padding):
        ctx.save_for_backward(filters)
        ctx.shape = x.shape
        ctx.padding = padding

        B, _, H, W = x.shape
        x = x.view(B, 4, -1, H, W).transpose(1, 2)
        C = x.shape[1]
        x = x.reshape(B, -1, H, W)
        filters = filters.repeat(C, 1, 1, 1)
        x = torch.nn.functional.conv_transpose2d(x, filters, stride=2, padding=padding, groups=C)
        return x

    @staticmethod
    def backward(ctx, dx):
        if ctx.needs_input_grad[0]:
            filters = ctx.saved_tensors
            filters = filters[0]
            B, C, H, W = ctx.shape
            C = C // 4
            dx = dx.contiguous()
            padding = ctx.padding

            w_ll,w_lh, w_hl, w_hh = torch.unbind(filters, dim=0)

            x_ll = torch.nn.functional.conv2d(dx, w_ll.unsqueeze(1).expand(C, -1, -1, -1), stride = 2, padding=padding, groups = C)
            x_lh = torch.nn.functional.conv2d(dx, w_lh.unsqueeze(1).expand(C, -1, -1, -1), stride = 2, padding=padding, groups = C)
            x_hl = torch.nn.functional.conv2d(dx, w_hl.unsqueeze(1).expand(C, -1, -1, -1), stride = 2, padding=padding, groups = C)
            x_hh = torch.nn.functional.conv2d(dx, w_hh.unsqueeze(1).expand(C, -1, -1, -1), stride = 2, padding=padding, groups = C)
            dx = torch.cat([x_ll,x_lh, x_hl, x_hh], dim=1)
        return dx, None, None

class IDWT_2D(nn.Module):
    def __init__(self, wave):
        super(IDWT_2D, self).__init__()
        w = pywt.Wavelet(wave)
        rec_hi = torch.Tensor(w.rec_hi)
        rec_lo = torch.Tensor(w.rec_lo)

        w_ll = rec_lo.unsqueeze(0)*rec_lo.unsqueeze(1)
        w_lh = rec_lo.unsqueeze(0)*rec_hi.unsqueeze(1)
        w_hl = rec_hi.unsqueeze(0)*rec_lo.unsqueeze(1)
        w_hh = rec_hi.unsqueeze(0)*rec_hi.unsqueeze(1)

        if w_ll.shape[0] == 2:
            padding = 0
        elif w_ll.shape[0] == 6:
            padding = 2
        self.padding = padding

        w_ll = w_ll.unsqueeze(0).unsqueeze(1)
        w_lh = w_lh.unsqueeze(0).unsqueeze(1)
        w_hl = w_hl.unsqueeze(0).unsqueeze(1)
        w_hh = w_hh.unsqueeze(0).unsqueeze(1)
        filters = torch.cat([w_ll, w_lh, w_hl, w_hh], dim=0)
        self.register_buffer('filters', filters)

    def forward(self, x):
        return IDWT_Function.apply(x, self.filters, self.padding)

class DWT_2D(nn.Module):
    def __init__(self, wave):
        super(DWT_2D, self).__init__()
        w = pywt.Wavelet(wave)
        dec_hi = torch.Tensor(w.dec_hi[::-1])
        dec_lo = torch.Tensor(w.dec_lo[::-1])

        w_ll = dec_lo.unsqueeze(0)*dec_lo.unsqueeze(1)
        w_lh = dec_lo.unsqueeze(0)*dec_hi.unsqueeze(1)
        w_hl = dec_hi.unsqueeze(0)*dec_lo.unsqueeze(1)
        w_hh = dec_hi.unsqueeze(0)*dec_hi.unsqueeze(1)

        if w_ll.shape[0] == 2:
            padding = 0
        elif w_ll.shape[0] == 6:
            padding = 2
        self.padding = padding

        self.register_buffer('w_ll', w_ll.unsqueeze(0).unsqueeze(0))
        self.register_buffer('w_lh', w_lh.unsqueeze(0).unsqueeze(0))
        self.register_buffer('w_hl', w_hl.unsqueeze(0).unsqueeze(0))
        self.register_buffer('w_hh', w_hh.unsqueeze(0).unsqueeze(0))

    def forward(self, x):
        return DWT_Function.apply(x, self.w_ll, self.w_lh, self.w_hl, self.w_hh, self.padding)