0
# Pre-trained Models
1

2
Extensive collection of pre-trained computer vision models with ImageNet weights for transfer learning, feature extraction, and fine-tuning across various architectures including ResNet, EfficientNet, VGG, MobileNet, and Inception families.
3

4
## Capabilities
5

6
### ResNet Models
7

8
Residual networks with skip connections for deep architectures.
9

10
```python { .api }
11
def ResNet50(include_top=True, weights='imagenet', input_tensor=None, input_shape=None, 
12
             pooling=None, classes=1000, classifier_activation='softmax', **kwargs):
13
    """
14
    ResNet50 architecture.
15
    
16
    Args:
17
        include_top (bool): Whether to include classification head
18
        weights (str): Pre-trained weights ('imagenet' or None)
19
        input_tensor: Optional input tensor  
20
        input_shape (tuple): Input shape for images
21
        pooling (str): Pooling mode ('avg', 'max', or None)
22
        classes (int): Number of classes for classification
23
        classifier_activation (str): Activation for final layer
24
        
25
    Returns:
26
        Model: ResNet50 model
27
    """
28

29
def ResNet101(include_top=True, weights='imagenet', **kwargs): ...
30
def ResNet152(include_top=True, weights='imagenet', **kwargs): ...
31
def ResNet50V2(include_top=True, weights='imagenet', **kwargs): ...
32
def ResNet101V2(include_top=True, weights='imagenet', **kwargs): ...
33
def ResNet152V2(include_top=True, weights='imagenet', **kwargs): ...
34
```
35

36
### EfficientNet Models
37

38
Compound scaled convolutional networks optimized for efficiency and accuracy.
39

40
```python { .api }
41
def EfficientNetB0(include_top=True, weights='imagenet', input_tensor=None, **kwargs): ...
42
def EfficientNetB1(include_top=True, weights='imagenet', **kwargs): ...
43
def EfficientNetB2(include_top=True, weights='imagenet', **kwargs): ...
44
def EfficientNetB3(include_top=True, weights='imagenet', **kwargs): ...
45
def EfficientNetB4(include_top=True, weights='imagenet', **kwargs): ...
46
def EfficientNetB5(include_top=True, weights='imagenet', **kwargs): ...
47
def EfficientNetB6(include_top=True, weights='imagenet', **kwargs): ...
48
def EfficientNetB7(include_top=True, weights='imagenet', **kwargs): ...
49

50
def EfficientNetV2B0(include_top=True, weights='imagenet', **kwargs): ...
51
def EfficientNetV2B1(include_top=True, weights='imagenet', **kwargs): ...
52
def EfficientNetV2B2(include_top=True, weights='imagenet', **kwargs): ...
53
def EfficientNetV2B3(include_top=True, weights='imagenet', **kwargs): ...
54
def EfficientNetV2S(include_top=True, weights='imagenet', **kwargs): ...
55
def EfficientNetV2M(include_top=True, weights='imagenet', **kwargs): ...
56
def EfficientNetV2L(include_top=True, weights='imagenet', **kwargs): ...
57
```
58

59
### VGG Models
60

61
Visual Geometry Group architectures with small convolution filters.
62

63
```python { .api }
64
def VGG16(include_top=True, weights='imagenet', input_tensor=None, **kwargs): ...
65
def VGG19(include_top=True, weights='imagenet', input_tensor=None, **kwargs): ...
66
```
67

68
### MobileNet Models
69

70
Lightweight models optimized for mobile and embedded devices.
71

72
```python { .api }
73
def MobileNet(include_top=True, weights='imagenet', input_tensor=None, **kwargs): ...
74
def MobileNetV2(include_top=True, weights='imagenet', input_tensor=None, **kwargs): ...
75
def MobileNetV3Small(include_top=True, weights='imagenet', **kwargs): ...
76
def MobileNetV3Large(include_top=True, weights='imagenet', **kwargs): ...
77
```
78

79
### Other Architectures
80

81
```python { .api }
82
def InceptionV3(include_top=True, weights='imagenet', **kwargs): ...
83
def InceptionResNetV2(include_top=True, weights='imagenet', **kwargs): ...
84
def Xception(include_top=True, weights='imagenet', **kwargs): ...
85
def DenseNet121(include_top=True, weights='imagenet', **kwargs): ...
86
def DenseNet169(include_top=True, weights='imagenet', **kwargs): ...
87
def DenseNet201(include_top=True, weights='imagenet', **kwargs): ...
88
def NASNetMobile(include_top=True, weights='imagenet', **kwargs): ...
89
def NASNetLarge(include_top=True, weights='imagenet', **kwargs): ...
90
```
91

92
## Usage Examples
93

94
### Transfer Learning
95

96
```python
97
import keras
98
from keras.applications import ResNet50
99
from keras import layers
100

101
# Load pre-trained model without top classification layer
102
base_model = ResNet50(weights='imagenet', include_top=False, input_shape=(224, 224, 3))
103

104
# Freeze base model weights
105
base_model.trainable = False
106

107
# Add custom classification head
108
model = keras.Sequential([
109
    base_model,
110
    layers.GlobalAveragePooling2D(),
111
    layers.Dense(128, activation='relu'),
112
    layers.Dropout(0.2),
113
    layers.Dense(num_classes, activation='softmax')
114
])
115

116
model.compile(
117
    optimizer='adam',
118
    loss='sparse_categorical_crossentropy', 
119
    metrics=['accuracy']
120
)
121
```
122

123
### Feature Extraction
124

125
```python
126
from keras.applications import EfficientNetB0
127
from keras.applications.efficientnet import preprocess_input
128

129
# Load model for feature extraction
130
model = EfficientNetB0(weights='imagenet', include_top=False, pooling='avg')
131

132
# Preprocess images
133
x = preprocess_input(images)
134

135
# Extract features
136
features = model.predict(x)
137
```