-
Notifications
You must be signed in to change notification settings - Fork 136
/
pytorch_pretrained.py
180 lines (157 loc) · 6.38 KB
/
pytorch_pretrained.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
import sys
from typing import Union
import torch.nn as nn
from torchvision import models
from dffml.util.entrypoint import entrypoint
from dffml.model.model import Model
from dffml.base import config, field
from .pytorch_base import PyTorchModelConfig, PyTorchModelContext
from .utils import create_layer
class LayersNotFound(Exception):
"""
Raised when add_layers is set to True but no layers are provided.
"""
@config
class PyTorchPreTrainedModelConfig(PyTorchModelConfig):
pretrained: bool = field(
"Load Pre-trained model weights", default=True,
)
trainable: bool = field(
"Tweak pretrained model by training again", default=False
)
add_layers: bool = field(
"Replace the last layer of the pretrained model", default=False,
)
layers: Union[
dict, nn.ModuleDict, nn.Sequential, nn.ModuleList, nn.Module
] = field(
"Extra layers to replace the last layer of the pretrained model",
default=None,
)
class PyTorchPretrainedContext(PyTorchModelContext):
def __init__(self, parent):
super().__init__(parent)
if self.parent.config.add_layers and self.parent.config.layers is None:
raise LayersNotFound(
"add_layers is set to True but no layers are provided."
)
def createModel(self):
"""
Generates a model
"""
if self._model is not None:
return self._model
self.logger.debug(
"Loading model with classifications(%d): %r",
len(self.classifications),
self.classifications,
)
model = getattr(models, self.parent.PYTORCH_MODEL)(
pretrained=self.parent.config.pretrained
)
for param in model.parameters():
param.require_grad = self.parent.config.trainable
if self.parent.config.add_layers:
if self.parent.config.layers.__class__.__base__.__name__ in [
"ModuleDict",
"Sequential",
"ModuleList",
"Module",
]:
layers = nn.Sequential()
for name, module in self.parent.config.layers.named_children():
layers.add_module(name, module)
else:
layers = [
create_layer(value)
for key, value in self.parent.config.layers.items()
]
if self.parent.LAST_LAYER_TYPE == "classifier_sequential":
if len(layers) > 1:
layers = [nn.Sequential(*layers)]
model.classifier = nn.Sequential(
*list(model.classifier.children())[:-1] + layers
)
elif self.parent.LAST_LAYER_TYPE == "classifier_linear":
if len(layers) == 1:
model.classifier = layers[0]
elif len(layers) > 1:
model.classifier = nn.Sequential(*layers)
else:
if len(layers) == 1:
model.fc = layers[0]
elif len(layers) > 1:
model.fc = nn.Sequential(*layers)
self._model = model.to(self.device)
return self._model
def set_model_parameters(self):
if self.parent.LAST_LAYER_TYPE == "classifier_sequential":
self.model_parameters = (
self._model.parameters()
if self.parent.config.trainable
else self._model.classifier[-1].parameters()
)
elif self.parent.LAST_LAYER_TYPE == "classifier_linear":
self.model_parameters = (
self._model.parameters()
if self.parent.config.trainable
else self._model.classifier.parameters()
)
else:
self.model_parameters = (
self._model.parameters()
if self.parent.config.trainable
else self._model.fc.parameters()
)
class PyTorchPreTrainedModel(Model):
def __init__(self, config) -> None:
super().__init__(config)
for model_name, name, last_layer_type in [
("alexnet", "AlexNet", "classifier_sequential"),
("densenet121", "DenseNet121", "classifier_linear"),
("densenet161", "DenseNet161", "classifier_linear"),
("densenet169", "DenseNet169", "classifier_linear"),
("densenet201", "DenseNet201", "classifier_linear"),
("mnasnet0_5", "MnasNet0_5", "classifier_sequential"),
("mnasnet1_0", "MnasNet1_0", "classifier_sequential"),
("mobilenet_v2", "MobileNetV2", "classifier_sequential"),
("vgg11", "VGG11", "classifier_sequential"),
("vgg11_bn", "VGG11BN", "classifier_sequential"),
("vgg13", "VGG13", "classifier_sequential"),
("vgg13_bn", "VGG13BN", "classifier_sequential"),
("vgg16", "VGG16", "classifier_sequential"),
("vgg16_bn", "VGG16BN", "classifier_sequential"),
("vgg19", "VGG19", "classifier_sequential"),
("vgg19_bn", "VGG19BN", "classifier_sequential"),
("googlenet", "GoogleNet", "fully_connected"),
("inception_v3", "InceptionV3", "fully_connected"),
("resnet101", "ResNet101", "fully_connected"),
("resnet152", "ResNet152", "fully_connected"),
("resnet18", "ResNet18", "fully_connected"),
("resnet34", "ResNet34", "fully_connected"),
("resnet50", "ResNet50", "fully_connected"),
("resnext101_32x8d", "ResNext101_32x8D", "fully_connected"),
("resnext50_32x4d", "ResNext50_32x4D", "fully_connected"),
("shufflenet_v2_x0_5", "ShuffleNetV2x0_5", "fully_connected"),
("shufflenet_v2_x1_0", "ShuffleNetV2x1_0", "fully_connected"),
("wide_resnet101_2", "WideResNet101_2", "fully_connected"),
("wide_resnet50_2", "WideResNet50_2", "fully_connected"),
]:
cls_config = type(
name + "ModelConfig", (PyTorchPreTrainedModelConfig,), {},
)
cls_context = type(name + "ModelContext", (PyTorchPretrainedContext,), {},)
dffml_cls = type(
name + "Model",
(PyTorchPreTrainedModel,),
{
"CONFIG": cls_config,
"CONTEXT": cls_context,
"PYTORCH_MODEL": model_name,
"LAST_LAYER_TYPE": last_layer_type,
},
)
dffml_cls = entrypoint(model_name)(dffml_cls)
setattr(sys.modules[__name__], cls_config.__qualname__, cls_config)
setattr(sys.modules[__name__], cls_context.__qualname__, cls_context)
setattr(sys.modules[__name__], dffml_cls.__qualname__, dffml_cls)