-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathutils.py
157 lines (142 loc) · 6.23 KB
/
utils.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
import numpy as np
import pandas as pd
import os
import warnings
import torch
import timm
import torch.nn as nn
import config
warnings.filterwarnings(action = "ignore")
def onehotencoder(value):
one_hot_vector = np.zeros(config.NUM_CLASS)
one_hot_vector[value] = 1
one_hot_vector = one_hot_vector.astype(int)
return one_hot_vector
def make_df(all_path):
c = 0
alphabets = config.ALPHABETS
df = pd.DataFrame(columns = ['img_path', 'label'])
for data_path in all_path:
base = os.path.split(data_path)[0]
label = os.path.basename(base)
if label in alphabets:
df.loc[c] = [data_path, label]
c+=1
return df
def generate_backbone_name(input_string):
backbone_name = input_string.lower()
if "resnet" in backbone_name:
return "resnet18"
elif "xception" in backbone_name:
return "xception41"
elif "rexnet" in backbone_name:
return "rexnet_100"
elif "efficientnet" in backbone_name:
return "efficientnetv2_rw_m"
elif "mobilenet" in backbone_name:
return "mobilenetv2_050"
else:
return "Unidentified backbone name, pick from the following [resnet, mobilenet, efficientnet, xception, rexnet]"
def initialize_optimizer(opt_method, model, lr):
if opt_method == "adam":
return torch.optim.Adam(model.parameters(), lr=lr, weight_decay=config.WEIGHT_DECAY)
if opt_method == "adamw":
return torch.optim.AdamW(model.parameters(), lr=lr, weight_decay=config.WEIGHT_DECAY)
if opt_method == "sgd":
return torch.optim.SGD(
model.parameters(), lr=lr, weight_decay=config.WEIGHT_DECAY, nesterov=True, momentum = 0.9
)
def initialize_model(backbone, device, transfer):
if backbone == 'resnet18':
if transfer == True:
original_model = timm.create_model(backbone, pretrained=True)
else:
original_model = timm.create_model(backbone, pretrained=False)
num_features = original_model.fc.in_features
original_model.fc = nn.Linear(num_features, config.NUM_CLASS)
model = original_model.to(device)
elif backbone == 'xception41':
if transfer == True:
original_model = timm.create_model(backbone, pretrained=True)
else:
original_model = timm.create_model(backbone, pretrained=False)
num_features = original_model.head.fc.in_features
original_model.head.fc = nn.Linear(num_features, config.NUM_CLASS)
model = original_model.to(device)
elif backbone == 'rexnet_100':
if transfer == True:
original_model = timm.create_model(backbone, pretrained=True)
else:
original_model = timm.create_model(backbone, pretrained=False)
original_model.global_pool = nn.AdaptiveAvgPool2d(1)
num_features = original_model.head.fc.in_features
original_model.head.fc = nn.Linear(num_features, config.NUM_CLASS)
model = original_model.to(device)
else:
if transfer == True:
original_model = timm.create_model(backbone, pretrained=True)
else:
original_model = timm.create_model(backbone, pretrained=False)
num_features = original_model.classifier.in_features
original_model.classifier = nn.Linear(num_features, config.NUM_CLASS)
model = original_model.to(device)
return model
def load_model(backbone, weight, device):
if backbone == 'resnet18':
original_model = timm.create_model(backbone, pretrained=True)
num_features = original_model.fc.in_features
original_model.fc = nn.Linear(num_features, config.NUM_CLASS)
original_model.load_state_dict(torch.load(weight, map_location=device))
model = original_model.to(device)
elif backbone == 'xception41':
original_model = timm.create_model(backbone, pretrained=True)
num_features = original_model.head.fc.in_features
original_model.head.fc = nn.Linear(num_features, config.NUM_CLASS)
original_model.load_state_dict(torch.load(weight, map_location=device))
model = original_model.to(device)
elif backbone == 'rexnet_100':
original_model = timm.create_model(backbone, pretrained=True)
original_model.global_pool = nn.AdaptiveAvgPool2d(1)
num_features = original_model.head.fc.in_features
original_model.head.fc = nn.Linear(num_features, config.NUM_CLASS)
original_model.load_state_dict(torch.load(weight, map_location=device))
model = original_model.to(device)
else:
original_model = timm.create_model(backbone, pretrained=True)
num_features = original_model.classifier.in_features
original_model.classifier = nn.Linear(num_features, config.NUM_CLASS)
original_model.load_state_dict(torch.load(weight, map_location = device))
model = original_model.to(device)
return model
# def generate_gradcam_layer(model, backbone):
# if backbone == 'rexnet_100':
# layer = [list(model.children())[1][-2].conv_exp.conv,
# list(model.children())[1][-2].conv_dw.conv,
# list(model.children())[1][-2].conv_pwl.conv,
# list(model.children())[1][-1].conv]
# elif backbone == 'resnet18':
# layer = [model.layer4[0].conv1,
# model.layer4[0].conv2,
# model.layer4[-1].conv1,
# model.layer4[-1].conv2]
# elif backbone == 'xception41':
# layer = [model.blocks[-1].stack.conv2.conv_dw,
# model.blocks[-1].stack.conv2.conv_pw,
# model.blocks[-1].stack.conv3.conv_dw,
# model.blocks[-1].stack.conv3.conv_pw]
# else:
# layer = [model.blocks[-1][-1].conv_pw,
# model.blocks[-1][-1].conv_dw,
# model.blocks[-1][-1].conv_pwl,
# model.conv_head]
# return layer
def generate_gradcam_layer(model, backbone):
if backbone == 'rexnet_100':
layer = [list(model.children())[1][-1].conv]
elif backbone == 'resnet18':
layer = [model.layer4[-1].conv2]
elif backbone == 'xception41':
layer = [model.blocks[-1].stack.conv3.conv_pw]
else:
layer = [model.conv_head]
return layer