TensorFlow ResNet18

一、模型

可调整的预处理层卷积核大小 

1.主模型 

from keras import layers, Sequential, Model
from keras.layers import Conv2D, BatchNormalization, ReLU,MaxPool2D,Layer,GlobalAvgPool2D,Dense,Input


class ResNet18(Model):
    def __init__(self, class_num,in_shape, pre_filter_size=7):
        """
        :param in_shape: the shape of each example
        :param class_num: the number of classes
        :param pre_filter_size: the size of filters in the preprocessing layer
        """
        super().__init__()
        # preprocessing layer
        self.pl = Sequential([
            Conv2D(64,(pre_filter_size,pre_filter_size),strides=2,padding='same'),
            BatchNormalization(),
            ReLU(),
            MaxPool2D(pool_size=(3,3),strides=2,padding='same')
        ])

        self.input_layer = Input(in_shape)  # used to reveal output shape

        # Residual Blocks
        self.block1 =