-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmodel.py
147 lines (118 loc) · 5.29 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
<<<<<<< HEAD
import pytorch_lightning as pl
import torch
import torch.nn as nn
import torch.optim as optim
import config
import pandas as pd
import torch.nn.functional as F
import math
import torchvision.models as models
# x.conv1
# x.bn1
# x.relu
# x.maxpool
# x.layer1
# x.layer2
# x.layer3
# x.layer4
# x.avgpool
# x.fc
class Discriminator(nn.Module):
def __init__(self, num_classes,num_features=config.FEATURE_DIM):
super(Discriminator, self).__init__()
self.num_classes = num_classes
resnet101 = models.resnet101(pretrained=True)
# self.resnet50.fc = nn.Identity()
# Remove the last fully connected layer and the average pooling layer
modules = list(resnet101.children())[:-2]
self.resnet101_features = nn.Sequential(*modules)
self.embed_layer = nn.Linear(2048, num_features) # 2048 is the number of features from the ResNet-50 output
self.fc = nn.Linear(num_features, num_classes)
self.prelu = nn.PReLU()
# self.attn_layer = SelfAttentionLayer(in_dim=num_features,hidden_dim=num_features//2)
def make_grad(self,model, requires_grad=True):
for param in model.parameters():
param.requires_grad = requires_grad
def forward(self, x):
with torch.no_grad():
x = self.resnet101_features(x)
x = F.adaptive_avg_pool2d(x, (1, 1))
x = torch.flatten(x, 1)
_features = self.prelu(self.embed_layer(x))
y = self.fc(F.dropout(_features,p=0.4))
return _features , y
# class Discriminator(nn.Module):
# def __init__(self, num_classes,num_features=config.FEATURE_DIM):
# super(Discriminator, self).__init__()
# self.num_classes = num_classes
# self.resnet = models.resnet50(pretrained=True)
# self.make_grad(self.resnet.conv1,requires_grad=False)
# self.make_grad(self.resnet.bn1,requires_grad=False)
# self.make_grad(self.resnet.relu,requires_grad=False)
# self.make_grad(self.resnet.maxpool,requires_grad=False)
# self.make_grad(self.resnet.layer1,requires_grad=False)
# self.make_grad(self.resnet.layer2,requires_grad=False)
# self.make_grad(self.resnet.layer3,requires_grad=False)
# self.make_grad(self.resnet.layer4,requires_grad=False)
# self.embedding = AttentionEmbedding(in_features=2048,out_features=num_features)
# self.fc = nn.Linear(in_features=num_features,out_features=num_classes)
# self.dropout = nn.Dropout(p=0.5)
# self.prelu = nn.PReLU()
# def make_grad(self,model, requires_grad=True):
# for param in model.parameters():
# param.requires_grad = requires_grad
# def forward(self, x):
# with torch.no_grad():
# x = self.resnet.conv1(x)
# x = self.resnet.bn1(x)
# x = self.resnet.relu(x)
# x = self.resnet.maxpool(x)
# x = self.resnet.layer1(x)
# x = self.resnet.layer2(x)
# x = self.resnet.layer3(x)
# x = self.resnet.layer4(x)
# x = F.adaptive_avg_pool2d(x, (1, 1))
# x = torch.flatten(x, 1)
# # x = self.resnet.avgpool(x)
# x = self.prelu(self.embedding(x))
# y = self.fc(self.dropout(x))
# return x,y
class EfficientNetV2(nn.Module):
def __init__(self,num_classes,num_features=config.FEATURE_DIM):
super().__init__()
self.num_classes = num_classes
self.num_features = num_features
self.net = models.efficientnet_v2_l(pretrained=True)
self.net.classifier = nn.Identity()
self.embed_vec = nn.Linear(config.EFFICIENT_NETV2_IN_FEATURES,num_features)
self.fc = nn.Linear(num_features,num_classes)
self.dropout = nn.Dropout(p=0.3)
def forward(self,x):
with torch.no_grad():
x = self.net(x)
_features = self.embed_vec(x)
y = self.dropout(self.fc(F.relu(_features)))
return _features , y
class HybridCNNTransformer(nn.Module):
def __init__(self,num_classes=82):
super().__init__()
self.swin_feature_dim = 1024
self.resnet_feature_dim = 8192
self.swin_transformer = models.swin_b(pretrained=True)
self.resnet = models.resnet101(pretrained=True)
self.swin_feature_extractor = nn.Sequential(*list(self.swin_transformer.children())[:-1])
self.resnet_feature_extractor = nn.Sequential(*list(self.resnet.children())[:-2])
self.embedd_layer = nn.Linear(self.swin_feature_dim + self.resnet_feature_dim , config.FEATURE_DIM)
self.fc = nn.Linear(config.FEATURE_DIM,num_classes)
self.prelu = nn.PReLU(num_parameters=1,init=0.2)
def forward(self,x):
with torch.no_grad():
swin_out = self.swin_feature_extractor(x)
resnet_out = self.resnet_feature_extractor(x)
resnet_out = F.adaptive_avg_pool2d(resnet_out, (2, 2))
resnet_out = torch.flatten(resnet_out, 1)
freeze_features = torch.cat([swin_out,resnet_out],dim=-1)
embed_features = self.prelu(self.embedd_layer(freeze_features))
classify = self.fc(embed_features)
return embed_features , classify