Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Add NASNet models #8714

Merged
merged 12 commits into from
Dec 12, 2017
Merged

Add NASNet models #8714

Changes from 1 commit
Commits
Show all changes
12 commits
Select commit Hold shift + click to select a range
9a62e76
Add NASNet models and associated tests
titu1994 Dec 7, 2017
2dc0bda
Remove mention of weight decay and auxiliary branch from comments
titu1994 Dec 7, 2017
50a3e29
Fix PEP8
titu1994 Dec 7, 2017
318bcc7
Correct implementation hyperlink
titu1994 Dec 8, 2017
5462200
Make corrections as requested (other than regularizers)
titu1994 Dec 10, 2017
77a2096
Remove dropout
titu1994 Dec 10, 2017
073c826
Fix PEP8
titu1994 Dec 10, 2017
15fc1d1
Remove regularization and dropout
titu1994 Dec 10, 2017
9c7385c
Update applications.md
titu1994 Dec 10, 2017
6a6d608
Fix requested changes
titu1994 Dec 11, 2017
9c3712c
Fix PEP8
titu1994 Dec 11, 2017
49250b4
Add check before calling K.int_shape
titu1994 Dec 11, 2017
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
Next Next commit
Remove regularization and dropout
  • Loading branch information
@titu1994
titu1994 committed Dec 10, 2017
commit 15fc1d1068df3f03a2b7d529e50efb8965fc0ea9
92 changes: 34 additions & 58 deletions keras/applications/nasnet.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -42,7 +42,6 @@
from ..layers import Input
from ..layers import Activation
from ..layers import Dense
from ..layers import Dropout
from ..layers import BatchNormalization
from ..layers import MaxPooling2D
from ..layers import AveragePooling2D
Expand All @@ -54,18 +53,17 @@
from ..layers import Cropping2D
from ..layers import concatenate
from ..layers import add
from ..regularizers import l2
from ..utils.data_utils import get_file
from ..engine.topology import get_source_inputs
from ..applications.imagenet_utils import _obtain_input_shape
from ..applications.inception_v3 import preprocess_input
from ..applications.imagenet_utils import decode_predictions
from .. import backend as K

NASNET_MOBILE_WEIGHT_PATH = 'https://github.com/titu1994/Keras-NASNet/releases/download/v1.0/NASNet-mobile.h5'
NASNET_MOBILE_WEIGHT_PATH_NO_TOP = 'https://github.com/titu1994/Keras-NASNet/releases/download/v1.0/NASNet-mobile-no-top.h5'
NASNET_LARGE_WEIGHT_PATH = 'https://github.com/titu1994/Keras-NASNet/releases/download/v1.1/NASNet-large.h5'
NASNET_LARGE_WEIGHT_PATH_NO_TOP = 'https://github.com/titu1994/Keras-NASNet/releases/download/v1.1/NASNet-large-no-top.h5'
NASNET_MOBILE_WEIGHT_PATH = 'https://github.com/titu1994/Keras-NASNet/releases/download/v1.2/NASNet-mobile.h5'
NASNET_MOBILE_WEIGHT_PATH_NO_TOP = 'https://github.com/titu1994/Keras-NASNet/releases/download/v1.2/NASNet-mobile-no-top.h5'
NASNET_LARGE_WEIGHT_PATH = 'https://github.com/titu1994/Keras-NASNet/releases/download/v1.2/NASNet-large.h5'
NASNET_LARGE_WEIGHT_PATH_NO_TOP = 'https://github.com/titu1994/Keras-NASNet/releases/download/v1.2/NASNet-large-no-top.h5'


def NASNet(input_shape=None,
Expand All @@ -74,14 +72,14 @@ def NASNet(input_shape=None,
stem_filters=96,
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can we name this to something more explicit? Most users will not understand what this is without further context.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

How about num_stem_block_filters?

skip_reduction=True,
filter_multiplier=2,
weight_decay=5e-5,
include_top=True,
weights=None,
input_tensor=None,
pooling=None,
classes=1000,
default_size=None):
'''Instantiates a NASNet model.
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Introduce a blank line after this


Note that only TensorFlow is supported for now,
therefore it only works with the data format
`image_data_format='channels_last'` in your Keras config
Expand Down Expand Up @@ -113,7 +111,6 @@ def NASNet(input_shape=None,
of filters in each layer.
- If `filter_multiplier` = 1, default number of filters from the
paper are used at each layer.
weight_decay: l2 regularization weight
include_top: whether to include the fully-connected
layer at the top of the network.
weights: `None` (random initialization) or
Expand Down Expand Up @@ -208,51 +205,47 @@ def NASNet(input_shape=None,
if not skip_reduction:
x = Conv2D(stem_filters, (3, 3), strides=(2, 2), padding='valid',
use_bias=False, name='stem_conv1',
kernel_initializer='he_normal',
kernel_regularizer=l2(weight_decay))(img_input)
kernel_initializer='he_normal')(img_input)
else:
x = Conv2D(stem_filters, (3, 3), strides=(1, 1), padding='same',
use_bias=False, name='stem_conv1',
kernel_initializer='he_normal',
kernel_regularizer=l2(weight_decay))(img_input)
kernel_initializer='he_normal')(img_input)

x = BatchNormalization(axis=channel_dim, momentum=0.9997,
epsilon=1e-3, name='stem_bn1')(x)

p = None
if not skip_reduction: # imagenet / mobile mode
x, p = _reduction_A(x, p, filters // (filter_multiplier ** 2),
weight_decay, id='stem_1')
x, p = _reduction_A(x, p, filters // filter_multiplier, weight_decay,
id='stem_2')
id='stem_1')
x, p = _reduction_A(x, p, filters // filter_multiplier, id='stem_2')

for i in range(num_blocks):
x, p = _normal_A(x, p, filters, weight_decay, id='%d' % (i))
x, p = _normal_A(x, p, filters, id='%d' % (i))

x, p0 = _reduction_A(x, p, filters * filter_multiplier, weight_decay,
x, p0 = _reduction_A(x, p, filters * filter_multiplier,
id='reduce_%d' % (num_blocks))

p = p0 if not skip_reduction else p

for i in range(num_blocks):
x, p = _normal_A(x, p, filters * filter_multiplier, weight_decay,
x, p = _normal_A(x, p, filters * filter_multiplier,
id='%d' % (num_blocks + i + 1))

x, p0 = _reduction_A(x, p, filters * filter_multiplier ** 2, weight_decay,
x, p0 = _reduction_A(x, p, filters * filter_multiplier ** 2,
id='reduce_%d' % (2 * num_blocks))

p = p0 if not skip_reduction else p

for i in range(num_blocks):
x, p = _normal_A(x, p, filters * filter_multiplier ** 2, weight_decay,
x, p = _normal_A(x, p, filters * filter_multiplier ** 2,
id='%d' % (2 * num_blocks + i + 1))

x = Activation('relu')(x)

if include_top:
x = GlobalAveragePooling2D()(x)
x = Dense(classes, activation='softmax', name='predictions',
kernel_regularizer=l2(weight_decay))(x)
x = Dense(classes, activation='softmax', name='predictions')(x)
else:
if pooling == 'avg':
x = GlobalAveragePooling2D()(x)
Expand Down Expand Up @@ -280,7 +273,7 @@ def NASNet(input_shape=None,

weights_file = get_file(model_name, weight_path,
cache_subdir='models')
model.load_weights(weights_file, by_name=True)
model.load_weights(weights_file)

elif default_size == 331: # large version
if include_top:
Expand All @@ -292,7 +285,7 @@ def NASNet(input_shape=None,

weights_file = get_file(model_name, weight_path,
cache_subdir='models')
model.load_weights(weights_file, by_name=True)
model.load_weights(weights_file)
else:
raise ValueError(
'ImageNet weights can only be loaded on NASNetLarge'
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

"with" would fit better than "on" I think

Expand All @@ -313,6 +306,7 @@ def NASNetLarge(input_shape=None,
pooling=None,
classes=1000):
'''Instantiates a NASNet model in ImageNet mode.

Note that only TensorFlow is supported for now,
therefore it only works with the data format
`image_data_format='channels_last'` in your Keras config
Expand Down Expand Up @@ -364,7 +358,6 @@ def NASNetLarge(input_shape=None,
stem_filters=96,
skip_reduction=False,
filter_multiplier=2,
weight_decay=5e-5,
include_top=include_top,
weights=weights,
input_tensor=input_tensor,
Expand All @@ -380,6 +373,7 @@ def NASNetMobile(input_shape=None,
pooling=None,
classes=1000):
'''Instantiates a Mobile NASNet model in ImageNet mode.

Note that only TensorFlow is supported for now,
therefore it only works with the data format
`image_data_format='channels_last'` in your Keras config
Expand Down Expand Up @@ -429,7 +423,6 @@ def NASNetMobile(input_shape=None,
stem_filters=32,
skip_reduction=False,
filter_multiplier=2,
weight_decay=4e-5,
include_top=include_top,
weights=weights,
input_tensor=input_tensor,
Expand All @@ -439,7 +432,7 @@ def NASNetMobile(input_shape=None,


def _separable_conv_block(ip, filters, kernel_size=(3, 3), strides=(1, 1),
weight_decay=5e-5, id=None):
id=None):
'''Adds 2 blocks of [relu-separable conv-batchnorm]

# Arguments:
Expand All @@ -460,24 +453,22 @@ def _separable_conv_block(ip, filters, kernel_size=(3, 3), strides=(1, 1),
x = SeparableConv2D(filters, kernel_size, strides=strides,
name='separable_conv_1_%s' % id,
padding='same', use_bias=False,
kernel_initializer='he_normal',
kernel_regularizer=l2(weight_decay))(x)
kernel_initializer='he_normal')(x)
x = BatchNormalization(axis=channel_dim, momentum=0.9997,
epsilon=1e-3,
name='separable_conv_1_bn_%s' % (id))(x)
x = Activation('relu')(x)
x = SeparableConv2D(filters, kernel_size,
name='separable_conv_2_%s' % id,
padding='same', use_bias=False,
kernel_initializer='he_normal',
kernel_regularizer=l2(weight_decay))(x)
kernel_initializer='he_normal')(x)
x = BatchNormalization(axis=channel_dim, momentum=0.9997,
epsilon=1e-3,
name='separable_conv_2_bn_%s' % (id))(x)
return x


def _adjust_block(p, ip, filters, weight_decay=5e-5, id=None):
def _adjust_block(p, ip, filters, id=None):
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Don't use id, it's a reserved keyword. Maybe block_id or something like that.

'''
Adjusts the input `previou path` to match the shape of the `input`
or situations where the output number of filters needs to be changed
Expand Down Expand Up @@ -506,17 +497,15 @@ def _adjust_block(p, ip, filters, weight_decay=5e-5, id=None):
p1 = AveragePooling2D((1, 1), strides=(2, 2), padding='valid',
name='adjust_avg_pool_1_%s' % id)(p)
p1 = Conv2D(filters // 2, (1, 1), padding='same',
use_bias=False, kernel_regularizer=l2(weight_decay),
name='adjust_conv_1_%s' % id,
use_bias=False, name='adjust_conv_1_%s' % id,
kernel_initializer='he_normal')(p1)

p2 = ZeroPadding2D(padding=((0, 1), (0, 1)))(p)
p2 = Cropping2D(cropping=((1, 0), (1, 0)))(p2)
p2 = AveragePooling2D((1, 1), strides=(2, 2), padding='valid',
name='adjust_avg_pool_2_%s' % id)(p2)
p2 = Conv2D(filters // 2, (1, 1), padding='same',
use_bias=False, kernel_regularizer=l2(weight_decay),
name='adjust_conv_2_%s' % id,
use_bias=False, name='adjust_conv_2_%s' % id,
kernel_initializer='he_normal')(p2)

p = concatenate([p1, p2], axis=channel_dim)
Expand All @@ -529,15 +518,14 @@ def _adjust_block(p, ip, filters, weight_decay=5e-5, id=None):
p = Activation('relu')(p)
p = Conv2D(filters, (1, 1), strides=(1, 1), padding='same',
name='adjust_conv_projection_%s' % id,
use_bias=False, kernel_regularizer=l2(weight_decay),
kernel_initializer='he_normal')(p)
use_bias=False, kernel_initializer='he_normal')(p)
p = BatchNormalization(axis=channel_dim, momentum=0.9997,
epsilon=1e-3,
name='adjust_bn_%s' % id)(p)
return p


def _normal_A(ip, p, filters, weight_decay=5e-5, id=None):
def _normal_A(ip, p, filters, id=None):
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Don't use id keyword

function names should be snake case (no caps)

'''Adds a Normal cell for NASNet-A (Fig. 4 in the paper)

# Arguments:
Expand All @@ -553,31 +541,26 @@ def _normal_A(ip, p, filters, weight_decay=5e-5, id=None):
channel_dim = 1 if K.image_data_format() == 'channels_first' else -1

with K.name_scope('normal_A_block_%s' % id):
p = _adjust_block(p, ip, filters, weight_decay, id)
p = _adjust_block(p, ip, filters, id)

h = Activation('relu')(ip)
h = Conv2D(filters, (1, 1), strides=(1, 1), padding='same',
name='normal_conv_1_%s' % id,
use_bias=False, kernel_initializer='he_normal',
kernel_regularizer=l2(weight_decay))(h)
use_bias=False, kernel_initializer='he_normal')(h)
h = BatchNormalization(axis=channel_dim, momentum=0.9997,
epsilon=1e-3,
name='normal_bn_1_%s' % id)(h)

with K.name_scope('block_1'):
x1_1 = _separable_conv_block(h, filters, kernel_size=(5, 5),
weight_decay=weight_decay,
id='normal_left1_%s' % id)
x1_2 = _separable_conv_block(p, filters, weight_decay=weight_decay,
id='normal_right1_%s' % id)
x1_2 = _separable_conv_block(p, filters, id='normal_right1_%s' % id)
x1 = add([x1_1, x1_2], name='normal_add_1_%s' % id)

with K.name_scope('block_2'):
x2_1 = _separable_conv_block(p, filters, (5, 5),
weight_decay=weight_decay,
id='normal_left2_%s' % id)
x2_2 = _separable_conv_block(p, filters, (3, 3),
weight_decay=weight_decay,
id='normal_right2_%s' % id)
x2 = add([x2_1, x2_2], name='normal_add_2_%s' % id)

Expand All @@ -594,16 +577,15 @@ def _normal_A(ip, p, filters, weight_decay=5e-5, id=None):
x4 = add([x4_1, x4_2], name='normal_add_4_%s' % id)

with K.name_scope('block_5'):
x5 = _separable_conv_block(h, filters, weight_decay=weight_decay,
id='normal_left5_%s' % id)
x5 = _separable_conv_block(h, filters, id='normal_left5_%s' % id)
x5 = add([x5, h], name='normal_add_5_%s' % id)

x = concatenate([p, x1, x2, x3, x4, x5], axis=channel_dim,
name='normal_concat_%s' % id)
return x, ip


def _reduction_A(ip, p, filters, weight_decay=5e-5, id=None):
def _reduction_A(ip, p, filters, id=None):
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Same remarks as above

'''Adds a Reduction cell for NASNet-A (Fig. 4 in the paper)

# Arguments:
Expand All @@ -619,39 +601,34 @@ def _reduction_A(ip, p, filters, weight_decay=5e-5, id=None):
channel_dim = 1 if K.image_data_format() == 'channels_first' else -1

with K.name_scope('reduction_A_block_%s' % id):
p = _adjust_block(p, ip, filters, weight_decay, id)
p = _adjust_block(p, ip, filters, id)

h = Activation('relu')(ip)
h = Conv2D(filters, (1, 1), strides=(1, 1), padding='same',
name='reduction_conv_1_%s' % id,
use_bias=False, kernel_initializer='he_normal',
kernel_regularizer=l2(weight_decay))(h)
use_bias=False, kernel_initializer='he_normal')(h)
h = BatchNormalization(axis=channel_dim, momentum=0.9997,
epsilon=1e-3,
name='reduction_bn_1_%s' % id)(h)

with K.name_scope('block_1'):
x1_1 = _separable_conv_block(h, filters, (5, 5), strides=(2, 2),
weight_decay=weight_decay,
id='reduction_left1_%s' % id)
x1_2 = _separable_conv_block(p, filters, (7, 7), strides=(2, 2),
weight_decay=weight_decay,
id='reduction_1_%s' % id)
x1 = add([x1_1, x1_2], name='reduction_add_1_%s' % id)

with K.name_scope('block_2'):
x2_1 = MaxPooling2D((3, 3), strides=(2, 2), padding='same',
name='reduction_left2_%s' % id)(h)
x2_2 = _separable_conv_block(p, filters, (7, 7), strides=(2, 2),
weight_decay=weight_decay,
id='reduction_right2_%s' % id)
x2 = add([x2_1, x2_2], name='reduction_add_2_%s' % id)

with K.name_scope('block_3'):
x3_1 = AveragePooling2D((3, 3), strides=(2, 2), padding='same',
name='reduction_left3_%s' % id)(h)
x3_2 = _separable_conv_block(p, filters, (5, 5), strides=(2, 2),
weight_decay=weight_decay,
id='reduction_right3_%s' % id)
x3 = add([x3_1, x3_2], name='reduction_add3_%s' % id)

Expand All @@ -662,7 +639,6 @@ def _reduction_A(ip, p, filters, weight_decay=5e-5, id=None):

with K.name_scope('block_5'):
x5_1 = _separable_conv_block(x1, filters, (3, 3),
weight_decay=weight_decay,
id='reduction_left4_%s' % id)
x5_2 = MaxPooling2D((3, 3), strides=(2, 2), padding='same',
name='reduction_right5_%s' % id)(h)
Expand Down