前言
上一篇筆者使用如何閱讀深度學習論文的方法閱讀了 ResNet
。爲了加深理解,本文帶大家使用 TensorFlow 2 Keras 實現 ResNet 網絡。
網絡結構
從論文了解到,可以根據網絡的深度開發不同類型的 ResNet
,如:ResNet-34
、ResNet-50
、ResNet-152
,甚至可以自定義網絡的深度,來滿足需求。因此,ResNet
更像是一種範式,可能叫 ResNets
會更合適些。
ResNet
可以看做是 VGG
的升級版,區別在於 ResNet
中使用的快捷連接(shortcuts)。在下圖中,我們可以看到 VGG
的架構以及 34
層 ResNet
。
圖 3.
ImageNet
的網絡架構例子。左:作爲參考的VGG-19
模型[41]。中:具有34
個參數層的簡單網絡(36
億FLOPs
)。右:具有34
個參數層的殘差網絡(36
億FLOPs
)。帶點的快捷連接增加了維度。表 1 顯示了更多細節和其它變種。
而對於不同的網絡類型,其中的構建塊(building block)也不一樣,如下圖所示:
圖 5.
ImageNet
的深度殘差函數 F。左:ResNet-34
的構建塊(在56×56
的特徵圖上),如圖 3。右:ResNet-50/101/152
的 “bottleneck”構建塊。
論文也提供了不同深度的 ResNet
的架構圖,如下所示:
表 1.
ImageNet
架構。構建塊顯示在括號中(也可看圖 5),以及構建塊的堆疊數量。下采樣通過步長爲2
的conv3_1
,conv4_1
和conv5_1
執行。
實現
下面以 ResNet-50
爲例,其他的類型都是類似的。
注:源碼已經上傳 https://github.com/CatchZeng/YiAI-examples/blob/master/papers/ResNet/ResNet.py,需要的同學可以參考。
首先,先實現下堆疊的殘差結構(上圖紅色框部分)。
def ResNet50(include_top=True,
input_shape=None,
pooling=None,
classes=1000):
"""Instantiates the ResNet50 architecture."""
# 堆疊的殘差結構
def stack_fn(x):
x = stack(x, 64, 3, stride1=1, name='conv2')
x = stack(x, 128, 4, name='conv3')
x = stack(x, 256, 6, name='conv4')
return stack(x, 512, 3, name='conv5')
return ResNet(stack_fn, 'resnet50', include_top, input_shape, pooling, classes)
def stack(x, filters, blocks, stride1=2, name=None):
"""A set of stacked residual blocks.
Args:
x: input tensor.
filters: integer, filters of the bottleneck layer in a block.
blocks: integer, blocks in the stacked blocks.
stride1: default 2, stride of the first layer in the first block.
name: string, stack label.
Returns:
Output tensor for the stacked blocks.
"""
x = block(x, filters, stride=stride1, name=name + '_block1')
# 沒有增加維度的時候可以做恆等快捷連接,不需要 conv_shortcut,可以參考圖 3。
for i in range(2, blocks + 1):
x = block(x, filters, conv_shortcut=False,
name=name + '_block' + str(i))
return x
def block(x, filters, kernel_size=3, stride=1, conv_shortcut=True, name=None):
"""A residual block.
Args:
x: input tensor.
filters: integer, filters of the bottleneck layer.
kernel_size: default 3, kernel size of the bottleneck layer.
stride: default 1, stride of the first layer.
conv_shortcut: default True, use convolution shortcut if True,
otherwise identity shortcut.
name: string, block label.
Returns:
Output tensor for the residual block.
"""
bn_axis = 3 if backend.image_data_format() == 'channels_last' else 1
# 當維度增加(圖 3 中的虛線快捷連接)時,我們考慮兩個選項:(A)快捷連接仍然執行恆等映射,額外填充零輸入以增加維度。此選項不會引入額外的參數;(B)方程(2)中的投影快捷連接用於匹配維度(由 1×1 卷積完成)。對於這兩個選項,當快捷連接跨越兩種尺寸的特徵圖時,它們執行時步長爲 2。
if conv_shortcut:
shortcut = layers.Conv2D(
4 * filters, 1, strides=stride, name=name + '_0_conv')(x)
shortcut = layers.BatchNormalization(
axis=bn_axis, epsilon=1.001e-5, name=name + '_0_bn')(shortcut)
# 當輸入和輸出具有相同的維度時(圖 3 中的實線快捷連接)時,可以直接使用恆等快捷連接
else:
shortcut = x
# 1x1xfilters
x = layers.Conv2D(filters, 1, strides=stride, name=name + '_1_conv')(x)
x = layers.BatchNormalization(
axis=bn_axis, epsilon=1.001e-5, name=name + '_1_bn')(x)
x = layers.Activation('relu', name=name + '_1_relu')(x)
# 3x3xfilters
x = layers.Conv2D(
filters, kernel_size, padding='SAME', name=name + '_2_conv')(x)
x = layers.BatchNormalization(
axis=bn_axis, epsilon=1.001e-5, name=name + '_2_bn')(x)
x = layers.Activation('relu', name=name + '_2_relu')(x)
# 1x1x(4倍filters)
x = layers.Conv2D(4 * filters, 1, name=name + '_3_conv')(x)
x = layers.BatchNormalization(
axis=bn_axis, epsilon=1.001e-5, name=name + '_3_bn')(x)
x = layers.Add(name=name + '_add')([shortcut, x])
x = layers.Activation('relu', name=name + '_out')(x)
return x
接着再實現輸入和輸出部分(上圖綠色和藍色框部分),代碼如下:
def ResNet(stack_fn,
model_name='resnet',
include_top=True,
input_shape=None,
pooling=None,
classes=1000,
classifier_activation='softmax'):
"""Instantiates the ResNet, ResNetV2, and ResNeXt architecture.
Args:
stack_fn: a function that returns output tensor for the
stacked residual blocks.
model_name: string, model name.
include_top: whether to include the fully-connected
layer at the top of the network.
input_shape: optional shape tuple, `(224, 224, 3)` (with `channels_last` data format)
or `(3, 224, 224)` (with `channels_first` data format).
It should have exactly 3 inputs channels.
pooling: optional pooling mode for feature extraction
when `include_top` is `False`.
- `None` means that the output of the model will be
the 4D tensor output of the
last convolutional layer.
- `avg` means that global average pooling
will be applied to the output of the
last convolutional layer, and thus
the output of the model will be a 2D tensor.
- `max` means that global max pooling will
be applied.
classes: optional number of classes to classify images
into, only to be specified if `include_top` is True, and
if no `weights` argument is specified.
classifier_activation: A `str` or callable. The activation function to use
on the "top" layer. Ignored unless `include_top=True`. Set
`classifier_activation=None` to return the logits of the "top" layer.
When loading pretrained weights, `classifier_activation` can only
be `None` or `"softmax"`.
Returns:
A `keras.Model` instance.
"""
img_input = layers.Input(shape=input_shape)
bn_axis = 3 if backend.image_data_format() == 'channels_last' else 1
x = layers.ZeroPadding2D(
padding=((3, 3), (3, 3)), name='conv1_pad')(img_input)
# conv 1 7x7, 64, stride 2
x = layers.Conv2D(64, 7, strides=2, use_bias=True,
name='conv1_conv')(x)
x = layers.BatchNormalization(
axis=bn_axis, epsilon=1.001e-5, name='conv1_bn')(x)
x = layers.Activation('relu', name='conv1_relu')(x)
x = layers.ZeroPadding2D(padding=((1, 1), (1, 1)), name='pool1_pad')(x)
# 3x3 max pool, stride 2
x = layers.MaxPooling2D(3, strides=2, name='pool1_pool')(x)
x = stack_fn(x)
if include_top:
# average pool, 1000-d fc, softmax
x = layers.GlobalAveragePooling2D(name='avg_pool')(x)
x = layers.Dense(classes, activation=classifier_activation,
name='predictions')(x)
else:
if pooling == 'avg':
x = layers.GlobalAveragePooling2D(name='avg_pool')(x)
elif pooling == 'max':
x = layers.GlobalMaxPooling2D(name='max_pool')(x)
inputs = img_input
model = Model(inputs, x, name=model_name)
return model
測試
通過與 tensorflow
官方的代碼對比,可以驗證模型編寫的準確性。
if __name__ == '__main__':
model = ResNet50(include_top=True, input_shape=(224, 224, 3), classes=10)
model.summary()
print("----------------------------------------")
from tensorflow.keras.applications import resnet
model2 = resnet.ResNet50(
include_top=True, weights=None, input_shape=(224, 224, 3), classes=10)
model2.summary()
小結
實踐出真知,從閱讀到實踐,是一個提升的過程。在實踐中,不但可以瞭解到實現的細節,而且還能熟悉 TensorFlow 的生態。強烈推薦大家,多看論文,並實踐。