-
Notifications
You must be signed in to change notification settings - Fork 5
/
model.py
206 lines (173 loc) · 9.18 KB
/
model.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
import torch
from torch import nn
from torch.nn import init
import torch.nn.functional as F
from convlstm import ConvLSTM
from dcn_v2 import DCN
# vgg choice
base = {'vgg': [64, 64, 'M', 128, 128, 'M', 256, 256, 256, 'M', 512, 512, 512, 'M', 512, 512, 512, 'M']}
# vgg16
def vgg16(cfg, i=3, batch_norm=False):
layers = []
in_channels = i
for v in cfg:
if v == 'M':
layers += [nn.MaxPool2d(kernel_size=2, stride=2)]
else:
conv2d = nn.Conv2d(in_channels, v, kernel_size=3, padding=1)
if batch_norm:
layers += [conv2d, nn.BatchNorm2d(v), nn.ReLU(inplace=True)]
else:
layers += [conv2d, nn.ReLU(inplace=True)]
in_channels = v
return layers
# something originate from EDVR(https://github.com/xinntao/EDVR)
class PCD_Align(nn.Module):
''' Alignment module using Pyramid, Cascading and Deformable convolution
with 3 pyramid levels.
'''
def __init__(self, nf=512, groups=8):
super(PCD_Align, self).__init__()
# L3: level 3, 1/4 spatial size
self.L3_offset_conv1 = nn.Conv2d(nf * 2, nf, 3, 1, 1, bias=True) # concat for diff
self.L3_offset_conv2 = nn.Conv2d(nf, nf, 3, 1, 1, bias=True)
self.L3_dcnpack = DCN(nf, nf, kernel_size=(3, 3), stride=1, padding=1, dilation=1, deformable_groups=groups)
# L2: level 2, 1/2 spatial size
self.L2_offset_conv1 = nn.Conv2d(nf * 2, nf, 3, 1, 1, bias=True) # concat for diff
self.L2_offset_conv2 = nn.Conv2d(nf * 2, nf, 3, 1, 1, bias=True) # concat for offset
self.L2_offset_conv3 = nn.Conv2d(nf, nf, 3, 1, 1, bias=True)
self.L2_dcnpack = DCN(nf, nf, kernel_size=(3, 3), stride=1, padding=1, dilation=1, deformable_groups=groups)
self.L2_fea_conv = nn.Conv2d(nf * 2, nf, 3, 1, 1, bias=True) # concat for fea
# L1: level 1, original spatial size
self.L1_offset_conv1 = nn.Conv2d(nf * 2, nf, 3, 1, 1, bias=True) # concat for diff
self.L1_offset_conv2 = nn.Conv2d(nf * 2, nf, 3, 1, 1, bias=True) # concat for offset
self.L1_offset_conv3 = nn.Conv2d(nf, nf, 3, 1, 1, bias=True)
self.L1_dcnpack = DCN(nf, nf, kernel_size=(3, 3), stride=1, padding=1, dilation=1, deformable_groups=groups)
self.L1_fea_conv = nn.Conv2d(nf * 2, nf, 3, 1, 1, bias=True) # concat for fea
# Cascading DCN
self.cas_offset_conv1 = nn.Conv2d(nf * 2, nf, 3, 1, 1, bias=True) # concat for diff
self.cas_offset_conv2 = nn.Conv2d(nf, nf, 3, 1, 1, bias=True)
self.cas_dcnpack = DCN(nf, nf, kernel_size=(3, 3), stride=1, padding=1, dilation=1, deformable_groups=groups)
self.lrelu = nn.LeakyReLU(negative_slope=0.1, inplace=True)
def forward(self, nbr_fea_l, ref_fea_l):
'''align other neighboring frames to the reference frame in the feature level
nbr_fea_l, ref_fea_l: [L1, L2, L3], each with [B,C,H,W] features
'''
# L3
L3_offset = torch.cat((nbr_fea_l[2], ref_fea_l[2]), dim=1)
L3_offset = self.lrelu(self.L3_offset_conv1(L3_offset))
L3_offset = self.lrelu(self.L3_offset_conv2(L3_offset))
L3_fea = self.lrelu(self.L3_dcnpack(L3_offset, nbr_fea_l[2]))
# L2
L2_offset = torch.cat((nbr_fea_l[1], ref_fea_l[1]), dim=1)
L2_offset = self.lrelu(self.L2_offset_conv1(L2_offset))
L3_offset = F.interpolate(L3_offset, scale_factor=2, mode='bilinear', align_corners=False)
L2_offset = self.lrelu(self.L2_offset_conv2(torch.cat((L2_offset, L3_offset * 2), dim=1)))
L2_offset = self.lrelu(self.L2_offset_conv3(L2_offset))
L2_fea = self.L2_dcnpack(L2_offset, nbr_fea_l[1])
L3_fea = F.interpolate(L3_fea, scale_factor=2, mode='bilinear', align_corners=False)
L2_fea = self.lrelu(self.L2_fea_conv(torch.cat((L2_fea, L3_fea), dim=1)))
# L1
L1_offset = torch.cat((nbr_fea_l[0], ref_fea_l[0]), dim=1)
L1_offset = self.lrelu(self.L1_offset_conv1(L1_offset))
L2_offset = F.interpolate(L2_offset, scale_factor=2, mode='bilinear', align_corners=False)
L1_offset = self.lrelu(self.L1_offset_conv2(torch.cat((L1_offset, L2_offset * 2), dim=1)))
L1_offset = self.lrelu(self.L1_offset_conv3(L1_offset))
L1_fea = self.L1_dcnpack(L1_offset, nbr_fea_l[0])
L2_fea = F.interpolate(L2_fea, scale_factor=2, mode='bilinear', align_corners=False)
L1_fea = self.L1_fea_conv(torch.cat((L1_fea, L2_fea), dim=1))
# Cascading
offset = torch.cat((L1_fea, ref_fea_l[0]), dim=1)
offset = self.lrelu(self.cas_offset_conv1(offset))
offset = self.lrelu(self.cas_offset_conv2(offset))
L1_fea = self.lrelu(self.cas_dcnpack(offset, L1_fea))
return L1_fea
class Model(nn.Module):
def __init__(self, base):
super(Model, self).__init__()
self.center = 5 // 2
self.base = nn.ModuleList(base)
self.extract = [16, 23, 30]
self.nf = 512
self.groups = 8
self.pcd_align = PCD_Align(nf=self.nf, groups=self.groups)
self.final_conv = nn.Sequential(nn.Conv2d(512, 1, 3, 1, 1))
self.final_sigmoid = nn.Sequential(nn.Sigmoid())
self.att_sigmoid = nn.Sequential(nn.Sigmoid())
self.L1_conv = nn.Sequential(nn.Conv2d(256, 512, 1, 1))
self.lstm_forw = ConvLSTM(input_size=(32, 40), input_dim=512, hidden_dim=[512], kernel_size=(3, 3),
padding=1, dilation=1, num_layers=1, batch_first=True, bias=True, return_all_layers=False)
self.lstm_back = ConvLSTM(input_size=(32, 40), input_dim=512, hidden_dim=[512], kernel_size=(3, 3),
padding=1, dilation=1, num_layers=1, batch_first=True, bias=True, return_all_layers=False)
self.sc_fusion = nn.Conv2d(5 * self.nf, self.nf, 1, 1, bias=True)
def forward(self, x, last_forw_state=None, last_back_state=None):
B, N, C, H, W = x.size()
# (B,N,C,H,W) --> (BN,C,H,W)
x = torch.cat(torch.split(x, split_size_or_sections=1, dim=0), dim=1).squeeze()
## extract vgg p3, p4, p5 features
p = list()
for k in range(len(self.base)):
x = self.base[k](x)
if k in self.extract:
p.append(x)
# (BN,C,H,W) --> (B,N,C,H,W)
L1_fea = torch.stack(torch.split(self.L1_conv(p[0]), split_size_or_sections=N, dim=0), dim=0)
L2_fea = torch.stack(torch.split(p[1], split_size_or_sections=N, dim=0), dim=0)
L3_fea = torch.stack(torch.split(p[2], split_size_or_sections=N, dim=0), dim=0)
# (B,N,C,H,W) --> (B,5,C,H,W)*group
L1_fea_split = torch.split(L1_fea, split_size_or_sections=5, dim=1)
L2_fea_split = torch.split(L2_fea, split_size_or_sections=5, dim=1)
L3_fea_split = torch.split(L3_fea, split_size_or_sections=5, dim=1)
## pcd group align
# (B,5,C,H,W)*group --> (B,N/5,C,H,W)
aligned_fea_group = []
for gr in range(len(L1_fea_split)):
ref_fea_l = [
L1_fea_split[gr][:, self.center, :, :, :].clone(), L2_fea_split[gr][:, self.center, :, :, :].clone(),
L3_fea_split[gr][:, self.center, :, :, :].clone()
]
aligned_fea = []
for sp in range(5):
nbr_fea_l = [
L1_fea_split[gr][:, sp, :, :, :].clone(), L2_fea_split[gr][:, sp, :, :, :].clone(),
L3_fea_split[gr][:, sp, :, :, :].clone()
]
aligned_fea.append(self.pcd_align(nbr_fea_l, ref_fea_l))
aligned_fea_result = torch.cat(aligned_fea, dim=1)
fused_fea = self.sc_fusion(aligned_fea_result)
aligned_fea_group.append(fused_fea)
final_aligned_fea_group = torch.stack(aligned_fea_group, dim=1)
reversed_feature = torch.stack(aligned_fea_group[::-1], dim=1)
## Bi convlstm
# (B,N/5,C,H,W) --> (B,N/5,C,H,W)
# forward convlstm
lstm_forw_output, lstm_forw_output_last_state = self.lstm_forw(final_aligned_fea_group, last_forw_state)
lstm_forw_output = lstm_forw_output[0]
# backforw convlstm
lstm_back_output, lstm_back_output_last_state = self.lstm_forw(reversed_feature, last_back_state)
lstm_back_output = lstm_back_output[0]
lstm_output = lstm_forw_output + lstm_back_output
# (B,N/5,C,H,W) --> (BN/5,C,H,W)
lstm_output = torch.cat(torch.split(lstm_output, split_size_or_sections=1, dim=0), dim=1).squeeze(dim=0)
## decoder
x = self.final_conv(lstm_output)
x = F.interpolate(x, scale_factor=4, mode='bilinear', align_corners=False)
out = self.final_sigmoid(x)
# different outs for train and test
if B==1 and N==5:
outs = torch.squeeze(out)
else:
outs = torch.stack(torch.split(out, split_size_or_sections=N//5, dim=0), dim=0)
return outs, lstm_forw_output_last_state, lstm_back_output_last_state
# build the whole network
def build_model():
return Model(vgg16(base['vgg'], 3))
# weight init
def xavier(param):
init.xavier_uniform_(param)
def weights_init(m):
if isinstance(m, nn.Conv2d):
xavier(m.weight.data)
elif isinstance(m, nn.BatchNorm2d):
init.constant_(m.weight, 1)
init.constant_(m.bias, 0)