Introduction¶
This notebook presents ResNet50 architecutre build from scratch and applied to Oxford VGG Flowers 17 dataset.
References
- Deep Residual Learning for Image Recognition (2015) by Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun
Imports¶
In [1]:
import os
import numpy as np
import matplotlib.pyplot as plt
In [2]:
import tensorflow as tf
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
with tf.Session(config=config):
pass # init sessin with allow_growth
VGG Flowers 17 Dataset¶
In [3]:
dataset_location = '../Datasets/vgg-flowers-17/'
In [4]:
datafile = os.path.join(dataset_location, '17flowers.npz')
if not os.path.isfile(datafile):
print('Run StanfordDogs.ipynb notebook first!')
raise
In [5]:
npzfile = np.load(datafile)
train_images_raw = npzfile['train_images']
train_labels_raw = npzfile['train_labels']
valid_images_raw = npzfile['valid_images']
valid_labels_raw = npzfile['valid_labels']
In [6]:
train_images = tf.keras.applications.resnet50.preprocess_input(train_images_raw.astype(np.float32))
valid_images = tf.keras.applications.resnet50.preprocess_input(valid_images_raw.astype(np.float32))
print('train_images.shape:', train_images.shape)
print('valid_images.shape:', valid_images.shape)
print('valid_images:\n', valid_images[0,:,:,0].round()) # first image, red channel
In [7]:
train_labels = tf.keras.utils.to_categorical(train_labels_raw)
valid_labels = tf.keras.utils.to_categorical(valid_labels_raw)
print('train_labels.shape:', train_labels.shape)
print('valid_labels.shape:', valid_labels.shape)
print('valid_labels:\n', valid_labels)
ResNet-50 From Scratch¶
In [8]:
from tensorflow.keras.layers import Input
from tensorflow.keras.layers import Conv2D, BatchNormalization, Activation, MaxPooling2D, Add
from tensorflow.keras.layers import GlobalAveragePooling2D, Flatten, Dense
In [9]:
def residual_block(X_start, filters, name, reduce=False, res_conv2d=False):
"""
Residual building block used by ResNet-50
"""
nb_filters_1, nb_filters_2, nb_filters_3 = filters
strides_1 = [2,2] if reduce else [1,1]
X = Conv2D(filters=nb_filters_1, kernel_size=[1,1], strides=strides_1, padding='same', name=name)(X_start)
X = BatchNormalization()(X) # default axis-1 is ok
X = Activation('relu')(X)
X = Conv2D(filters=nb_filters_2, kernel_size=[3,3], strides=[1,1], padding='same')(X)
X = BatchNormalization()(X)
X = Activation('relu')(X)
X = Conv2D(filters=nb_filters_3, kernel_size=[1,1], strides=[1,1], padding='same')(X)
X = BatchNormalization()(X)
if res_conv2d:
X_res = Conv2D(filters=nb_filters_3, kernel_size=[1,1], strides=strides_1, padding='same')(X_start)
X_res = BatchNormalization()(X_res)
else:
X_res = X_start
X = Add()([X, X_res])
X = Activation('relu')(X)
return X
In [12]:
def resnet50(input_shape, nb_classes):
assert len(input_shape) == 3
X_input = Input(shape=input_shape)
# conv1
X = Conv2D(filters=64, kernel_size=[7,7], strides=[2,2], padding='same', name='conv1')(X_input)
X = BatchNormalization(name='bn_conv1')(X)
X = Activation('relu')(X)
X = MaxPooling2D([3,3], strides=[2,2])(X)
# conv2_x
X = residual_block(X, filters=[64, 64, 256], name='conv2_a', reduce=False, res_conv2d=True)
X = residual_block(X, filters=[64, 64, 256], name='conv2_b')
X = residual_block(X, filters=[64, 64, 256], name='conv2_c')
# conv3_x
X = residual_block(X, filters=[128, 128, 512], name='conv3_a', reduce=True, res_conv2d=True)
X = residual_block(X, filters=[128, 128, 512], name='conv3_b')
X = residual_block(X, filters=[128, 128, 512], name='conv3_c')
X = residual_block(X, filters=[128, 128, 512], name='conv3_d')
# conv4_x
X = residual_block(X, filters=[256, 256, 1024], name='conv4_a', reduce=True, res_conv2d=True)
X = residual_block(X, filters=[256, 256, 1024], name='conv4_b')
X = residual_block(X, filters=[256, 256, 1024], name='conv4_c')
X = residual_block(X, filters=[256, 256, 1024], name='conv4_d')
X = residual_block(X, filters=[256, 256, 1024], name='conv4_e')
X = residual_block(X, filters=[256, 256, 1024], name='conv4_f')
# conv5_x
X = residual_block(X, filters=[512, 512, 2048], name='conv5_a', reduce=True, res_conv2d=True)
X = residual_block(X, filters=[512, 512, 2048], name='conv5_b')
X = residual_block(X, filters=[512, 512, 2048], name='conv5_c')
X = GlobalAveragePooling2D(name='avg_pool')(X)
X = Flatten()(X)
X = Dense(units=nb_classes, activation='softmax')(X)
model = tf.keras.models.Model(inputs=X_input, outputs=X)
model.compile(optimizer=tf.keras.optimizers.Adam(lr=0.0001),
loss='categorical_crossentropy', metrics=['accuracy'])
return model
In [13]:
model = resnet50(input_shape=[224, 224, 3], nb_classes=17)
# model.summary() # this should be identical to tf.keras.applications.ResNet50()
In [14]:
hist = model.fit(x=train_images, y=train_labels, batch_size=34, epochs=30,
validation_data=(valid_images, valid_labels), verbose=2)
In [15]:
fig, (ax1, ax2) = plt.subplots(nrows=1, ncols=2, figsize=[16,6])
ax1.plot(hist.history['loss'], label='train_loss')
ax1.plot(hist.history['val_loss'], label='valid_loss')
ax1.legend()
ax2.plot(hist.history['acc'], label='train_acc')
ax2.plot(hist.history['val_acc'], label='valid_acc')
ax2.legend();
In [ ]: