-
Notifications
You must be signed in to change notification settings - Fork 1
/
lifting.py
277 lines (239 loc) · 9.61 KB
/
lifting.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
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
import math
import torch
import torch.nn as nn
import torch.nn.functional as F
# This file contains the lifting scheme implementation
# There is no complete network definition inside this file.
# Note that it also contains other wavelet transformation
# used in WCNN and DAWN networks.
# To change if we do horizontal first inside the LS
HORIZONTAL_FIRST = True
class Splitting(nn.Module):
def __init__(self, horizontal):
super(Splitting, self).__init__()
# Deciding the stride base on the direction
self.horizontal = horizontal
if(horizontal):
self.conv_even = lambda x: x[:, :, :, ::2]
self.conv_odd = lambda x: x[:, :, :, 1::2]
else:
self.conv_even = lambda x: x[:, :, ::2, :]
self.conv_odd = lambda x: x[:, :, 1::2, :]
def forward(self, x):
'''Returns the odd and even part'''
return (self.conv_even(x), self.conv_odd(x))
class WaveletHaar(nn.Module):
def __init__(self, horizontal):
super(WaveletHaar, self).__init__()
self.split = Splitting(horizontal)
self.norm = math.sqrt(2.0)
def forward(self, x):
'''Returns the approximation and detail part'''
(x_even, x_odd) = self.split(x)
# Haar wavelet definition
d = (x_odd - x_even) / self.norm
c = (x_odd + x_even) / self.norm
return (c, d)
class WaveletHaar2D(nn.Module):
def __init__(self):
super(WaveletHaar2D, self).__init__()
self.horizontal_haar = WaveletHaar(horizontal=True)
self.vertical_haar = WaveletHaar(horizontal=False)
def forward(self, x):
'''Returns (LL, LH, HL, HH)'''
(c, d) = self.horizontal_haar(x)
(LL, LH) = self.vertical_haar(c)
(HL, HH) = self.vertical_haar(d)
#return (LL, LH, HL, HH)
return (LL, HL, LH, HH)
class Wavelet(nn.Module):
"""This module extract wavelet coefficient defined in pywt
and create 2D convolution kernels to be able to use GPU"""
def _coef_h(self, in_planes, coef):
"""Construct the weight matrix for horizontal 2D convolution.
The weights are repeated on the diagonal"""
v = []
for i in range(in_planes):
l = []
for j in range(in_planes):
if i == j:
l.append([[c for c in coef]])
else:
l.append([[0.0 for c in coef]])
v.append(l)
return v
def _coef_v(self, in_planes, coef):
"""Construct the weight matrix for vertical 2D convolution.
The weights are repeated on the diagonal"""
v = []
for i in range(in_planes):
l = []
for j in range(in_planes):
if i == j:
l.append([[c] for c in coef])
else:
l.append([[0.0] for c in coef])
v.append(l)
return v
def __init__(self, in_planes, horizontal, name="db2"):
super(Wavelet, self).__init__()
# Import wavelet coefficients
import pywt
wavelet = pywt.Wavelet(name)
coef_low = wavelet.dec_lo
coef_high = wavelet.dec_hi
# Determine the kernel 2D shape
nb_coeff = len(coef_low)
if horizontal:
kernel_size = (1, nb_coeff)
stride = (1, 2)
pad = (nb_coeff // 2, nb_coeff - 1 - nb_coeff // 2, 0, 0)
weights_low = self._coef_h(in_planes, coef_low)
weights_high = self._coef_h(in_planes, coef_high)
else:
kernel_size = (nb_coeff, 1)
stride = (2, 1)
pad = (0, 0, nb_coeff // 2, nb_coeff - 1 - nb_coeff // 2)
weights_low = self._coef_v(in_planes, coef_low)
weights_high = self._coef_v(in_planes, coef_high)
# TODO: Debug prints
# print("")
# print("Informations: ")
# print("- kernel_size: ", kernel_size)
# print("- stride : ", stride)
# print("- pad : ", pad)
# print("- low : ", weights_low)
# print("- high : ", weights_high)
# Create the conv2D
self.conv_high = nn.Conv2d(
in_planes, in_planes, kernel_size=kernel_size, stride=stride, bias=False)
self.conv_low = nn.Conv2d(
in_planes, in_planes, kernel_size=kernel_size, stride=stride, bias=False)
self.padding = nn.ReflectionPad2d(padding=pad)
# TODO: Debug prints
# print("- low : ", self.conv_low.weight)
# print("- high : ", self.conv_high.weight)
# Replace their weights
self.conv_high.weight = torch.nn.Parameter(
data=torch.Tensor(weights_high), requires_grad=False)
self.conv_low.weight = torch.nn.Parameter(
data=torch.Tensor(weights_low), requires_grad=False)
def forward(self, x):
'''Returns the approximation and detail part'''
x = self.padding(x)
return (self.conv_low(x), self.conv_high(x))
class Wavelet2D(nn.Module):
def __init__(self, in_planes, name="db1"):
super(Wavelet2D, self).__init__()
self.horizontal_wavelet = Wavelet(
in_planes, horizontal=True, name=name)
self.vertical_wavelet = Wavelet(in_planes, horizontal=False, name=name)
def forward(self, x):
'''Returns (LL, LH, HL, HH)'''
(c, d) = self.horizontal_wavelet(x)
(LL, LH) = self.vertical_wavelet(c)
(HL, HH) = self.vertical_wavelet(d)
return (LL, LH, HL, HH)
class LiftingScheme(nn.Module):
def __init__(self, horizontal, in_planes, modified=True, size=[], splitting=True, k_size=4, simple_lifting=False):
super(LiftingScheme, self).__init__()
self.modified = modified
if horizontal:
kernel_size = (1, k_size)
pad = (k_size // 2, k_size - 1 - k_size // 2, 0, 0)
else:
kernel_size = (k_size, 1)
pad = (0, 0, k_size // 2, k_size - 1 - k_size // 2)
self.splitting = splitting
self.split = Splitting(horizontal)
# Dynamic build sequential network
modules_P = []
modules_U = []
prev_size = 1
# HARD CODED Architecture
if simple_lifting:
modules_P += [
nn.ReflectionPad2d(pad),
nn.Conv2d(in_planes, in_planes,
kernel_size=kernel_size, stride=1),
nn.Tanh()
]
modules_U += [
nn.ReflectionPad2d(pad),
nn.Conv2d(in_planes, in_planes,
kernel_size=kernel_size, stride=1),
nn.Tanh()
]
else:
size_hidden = 2
modules_P += [
nn.ReflectionPad2d(pad),
nn.Conv2d(in_planes*prev_size, in_planes*size_hidden,
kernel_size=kernel_size, stride=1),
nn.ReLU()
]
modules_U += [
nn.ReflectionPad2d(pad),
nn.Conv2d(in_planes*prev_size, in_planes*size_hidden,
kernel_size=kernel_size, stride=1),
nn.ReLU()
]
prev_size = size_hidden
# Final dense
modules_P += [
nn.Conv2d(in_planes*prev_size, in_planes,
kernel_size=(1, 1), stride=1),
nn.Tanh()
]
modules_U += [
nn.Conv2d(in_planes*prev_size, in_planes,
kernel_size=(1, 1), stride=1),
nn.Tanh()
]
self.P = nn.Sequential(*modules_P)
self.U = nn.Sequential(*modules_U)
def forward(self, x):
if self.splitting:
(x_even, x_odd) = self.split(x)
else:
(x_even, x_odd) = x
if self.modified:
c = x_even + self.U(x_odd)
d = x_odd - self.P(c)
return (c, d)
else:
d = x_odd - self.P(x_even)
c = x_even + self.U(d)
return (c, d)
class LiftingScheme2D(nn.Module):
def __init__(self, in_planes, share_weights, modified=True, size=[2, 1], kernel_size=4, simple_lifting=False):
super(LiftingScheme2D, self).__init__()
self.level1_lf = LiftingScheme(
horizontal=HORIZONTAL_FIRST, in_planes=in_planes, modified=modified,
size=size, k_size=kernel_size, simple_lifting=simple_lifting)
self.share_weights = share_weights
if share_weights:
self.level2_1_lf = LiftingScheme(
horizontal=not HORIZONTAL_FIRST, in_planes=in_planes, modified=modified,
size=size, k_size=kernel_size, simple_lifting=simple_lifting)
self.level2_2_lf = self.level2_1_lf # Double check this
else:
self.level2_1_lf = LiftingScheme(
horizontal=not HORIZONTAL_FIRST, in_planes=in_planes, modified=modified,
size=size, k_size=kernel_size, simple_lifting=simple_lifting)
self.level2_2_lf = LiftingScheme(
horizontal=not HORIZONTAL_FIRST, in_planes=in_planes, modified=modified,
size=size, k_size=kernel_size, simple_lifting=simple_lifting)
def forward(self, x):
'''Returns (LL, LH, HL, HH)'''
(c, d) = self.level1_lf(x)
(LL, LH) = self.level2_1_lf(c)
(HL, HH) = self.level2_2_lf(d)
return (c, d, LL, LH, HL, HH)
# if __name__ == "__main__":
# input = torch.randn(1, 1, 10, 10)
# #m_harr = WaveletLiftingHaar2D()
# m_wavelet = Wavelet2D(1, name="db2")
# print(input)
# print(m_wavelet(input))
# TODO: Do more experiments with the code