def inception_block(
x: TensorType,
specs: Tuple,
channel_axis: int,
name: str,
weight_suffix: Optional[str] = "weights",
conv_suffix: Optional[str] = "",
norm_suffix: Optional[str] = "/BatchNorm") -> TensorType:
"""Inception block.
Args:
x: input_tensor
specs: Number of filters per branch.
channel_axis: channel dimension
name: Prefix for ths block.
weight_suffix: Name of learn-able parameters in conv.
conv_suffix: Suffix for conv layer.
norm_suffix: Suffix for batch norm.
Returns: Concatenated output of inception block.
"""
(br0, br1, br2, br3) = specs # ((64,), (96,128), (16,32), (32,))
branch_0 = conv_norm_relu(x,
br0[0],
1,
1,
name=name + "/Branch_0/Conv2d_0a_1x1",
weight_suffix=weight_suffix,
conv_suffix=conv_suffix,
norm_suffix=norm_suffix)
branch_1 = conv_norm_relu(x,
br1[0],
1,
1,
name=name + "/Branch_1/Conv2d_0a_1x1",
weight_suffix=weight_suffix,
conv_suffix=conv_suffix,
norm_suffix=norm_suffix)
branch_1 = conv_norm_relu(branch_1,
br1[1],
3,
3,
name=name + "/Branch_1/Conv2d_0b_3x3",
weight_suffix=weight_suffix,
conv_suffix=conv_suffix,
norm_suffix=norm_suffix)
branch_2 = conv_norm_relu(x,
br2[0],
1,
1,
name=name + "/Branch_2/Conv2d_0a_1x1",
weight_suffix=weight_suffix,
conv_suffix=conv_suffix,
norm_suffix=norm_suffix)
if name == "InceptionV1/Mixed_5b":
branch_2b_name = name + "/Branch_2/Conv2d_0a_3x3"
else:
branch_2b_name = name + "/Branch_2/Conv2d_0b_3x3"
branch_2 = conv_norm_relu(branch_2,
br2[1],
3,
3,
name=branch_2b_name,
weight_suffix=weight_suffix,
conv_suffix=conv_suffix,
norm_suffix=norm_suffix)
branch_3 = max_pool(x,
3,
strides=1,
padding='same',
name=name + "/Branch_3/Conv2d_0a_max")
branch_3 = conv_norm_relu(branch_3,
br3[0],
1,
1,
name=name + "/Branch_3/Conv2d_0b_1x1",
weight_suffix=weight_suffix,
conv_suffix=conv_suffix,
norm_suffix=norm_suffix)
x = concat([branch_0, branch_1, branch_2, branch_3],
axis=channel_axis,
name=name + "_Concatenated")
return x
def build_model(self, img_input: TensorType) -> TensorType:
"""Build graph using img_input as input.
Args:
img_input: 4D Image input tensor of shape (batch, height, width, channels)
Returns:
`Tensor` holding output probabilities per class, shape (batch, num_classes)
"""
x = conv_norm_relu(img_input, 32, 3, strides=2, padding='VALID')
x = conv_norm_relu(x, 32, 3, padding='VALID')
x = conv_norm_relu(
x,
64,
3,
)
x = max_pool(x, 3, strides=2)
x = conv_norm_relu(x, 80, 1, padding='VALID')
x = conv_norm_relu(x, 192, 3, padding='VALID')
x = max_pool(x, 3, strides=2)
# mixed 0: 35 x 35 x 256
branch1x1 = conv_norm_relu(x, 64, 1)
branch5x5 = conv_norm_relu(x, 48, 1)
branch5x5 = conv_norm_relu(branch5x5, 64, 5)
branch3x3dbl = conv_norm_relu(x, 64, 1)
branch3x3dbl = conv_norm_relu(branch3x3dbl, 96, 3)
branch3x3dbl = conv_norm_relu(branch3x3dbl, 96, 3)
branch_pool = avg_pool(x, 3, strides=1, padding='SAME')
branch_pool = conv_norm_relu(branch_pool, 32, 1)
x = concat([branch1x1, branch5x5, branch3x3dbl, branch_pool],
axis=-1,
name='mixed0')
# mixed 1: 35 x 35 x 288
branch1x1 = conv_norm_relu(x, 64, 1)
branch5x5 = conv_norm_relu(x, 48, 1)
branch5x5 = conv_norm_relu(branch5x5, 64, 5)
branch3x3dbl = conv_norm_relu(x, 64, 1)
branch3x3dbl = conv_norm_relu(branch3x3dbl, 96, 3)
branch3x3dbl = conv_norm_relu(branch3x3dbl, 96, 3)
branch_pool = avg_pool(x, 3, strides=1, padding='SAME')
branch_pool = conv_norm_relu(branch_pool, 64, 1)
x = concat([branch1x1, branch5x5, branch3x3dbl, branch_pool],
axis=-1,
name='mixed1')
# mixed 2: 35 x 35 x 288
branch1x1 = conv_norm_relu(x, 64, 1)
branch5x5 = conv_norm_relu(x, 48, 1)
branch5x5 = conv_norm_relu(branch5x5, 64, 5)
branch3x3dbl = conv_norm_relu(x, 64, 1)
branch3x3dbl = conv_norm_relu(branch3x3dbl, 96, 3)
branch3x3dbl = conv_norm_relu(branch3x3dbl, 96, 3)
branch_pool = avg_pool(x, 3, strides=1, padding='SAME')
branch_pool = conv_norm_relu(branch_pool, 64, 1)
x = concat([branch1x1, branch5x5, branch3x3dbl, branch_pool],
axis=-1,
name='mixed2')
# mixed 3: 17 x 17 x 768
branch3x3 = conv_norm_relu(x, 384, 3, strides=2, padding='VALID')
branch3x3dbl = conv_norm_relu(x, 64, 1)
branch3x3dbl = conv_norm_relu(branch3x3dbl, 96, 3)
branch3x3dbl = conv_norm_relu(branch3x3dbl,
96,
3,
strides=2,
padding='VALID')
branch_pool = max_pool(x, 3, 2)
x = concat([branch3x3, branch3x3dbl, branch_pool],
axis=-1,
name='mixed3')
# mixed 4: 17 x 17 x 768
branch1x1 = conv_norm_relu(x, 192, 1)
branch7x7 = conv_norm_relu(x, 128, 1)
branch7x7 = conv_norm_relu(branch7x7, 128, 1, 7)
branch7x7 = conv_norm_relu(branch7x7, 192, 7, 1)
branch7x7dbl = conv_norm_relu(x, 128, 1, 1)
branch7x7dbl = conv_norm_relu(branch7x7dbl, 128, 7, 1)
branch7x7dbl = conv_norm_relu(branch7x7dbl, 128, 1, 7)
branch7x7dbl = conv_norm_relu(branch7x7dbl, 128, 7, 1)
branch7x7dbl = conv_norm_relu(branch7x7dbl, 192, 1, 7)
branch_pool = avg_pool(x, 3, strides=1, padding='SAME')
branch_pool = conv_norm_relu(branch_pool, 192, 1, 1)
x = concat([branch1x1, branch7x7, branch7x7dbl, branch_pool],
axis=-1,
name='mixed4')
# mixed 5, 6: 17 x 17 x 768
for i in range(2):
branch1x1 = conv_norm_relu(x, 192, 1, 1)
branch7x7 = conv_norm_relu(x, 160, 1, 1)
branch7x7 = conv_norm_relu(branch7x7, 160, 1, 7)
branch7x7 = conv_norm_relu(branch7x7, 192, 7, 1)
branch7x7dbl = conv_norm_relu(x, 160, 1, 1)
branch7x7dbl = conv_norm_relu(branch7x7dbl, 160, 7, 1)
branch7x7dbl = conv_norm_relu(branch7x7dbl, 160, 1, 7)
branch7x7dbl = conv_norm_relu(branch7x7dbl, 160, 7, 1)
branch7x7dbl = conv_norm_relu(branch7x7dbl, 192, 1, 7)
branch_pool = avg_pool(x, 3, strides=1, padding='SAME')
branch_pool = conv_norm_relu(branch_pool, 192, 1, 1)
x = concat([branch1x1, branch7x7, branch7x7dbl, branch_pool],
axis=-1,
name='mixed' + str(5 + i))
# mixed 7: 17 x 17 x 768
branch1x1 = conv_norm_relu(x, 192, 1, 1)
branch7x7 = conv_norm_relu(x, 192, 1, 1)
branch7x7 = conv_norm_relu(branch7x7, 192, 1, 7)
branch7x7 = conv_norm_relu(branch7x7, 192, 7, 1)
branch7x7dbl = conv_norm_relu(x, 192, 1, 1)
branch7x7dbl = conv_norm_relu(branch7x7dbl, 192, 7, 1)
branch7x7dbl = conv_norm_relu(branch7x7dbl, 192, 1, 7)
branch7x7dbl = conv_norm_relu(branch7x7dbl, 192, 7, 1)
branch7x7dbl = conv_norm_relu(branch7x7dbl, 192, 1, 7)
branch_pool = avg_pool(x, 3, strides=1, padding='SAME')
branch_pool = conv_norm_relu(branch_pool, 192, 1, 1)
x = concat([branch1x1, branch7x7, branch7x7dbl, branch_pool],
axis=-1,
name='mixed7')
# mixed 8: 8 x 8 x 1280
branch3x3 = conv_norm_relu(x, 192, 1)
branch3x3 = conv_norm_relu(branch3x3,
320,
3,
strides=2,
padding='VALID')
branch7x7x3 = conv_norm_relu(x, 192, 1, 1)
branch7x7x3 = conv_norm_relu(branch7x7x3, 192, 1, 7)
branch7x7x3 = conv_norm_relu(branch7x7x3, 192, 7, 1)
branch7x7x3 = conv_norm_relu(branch7x7x3,
192,
3,
3,
strides=2,
padding='VALID')
branch_pool = max_pool(x, 3, strides=2)
x = concat([branch3x3, branch7x7x3, branch_pool],
axis=-1,
name='mixed8')
# mixed 9: 8 x 8 x 2048
for i in range(2):
branch1x1 = conv_norm_relu(x, 320, 1, 1)
branch3x3 = conv_norm_relu(x, 384, 1, 1)
branch3x3_1 = conv_norm_relu(branch3x3, 384, 1, 3)
branch3x3_2 = conv_norm_relu(branch3x3, 384, 3, 1)
branch3x3 = concat([branch3x3_1, branch3x3_2],
axis=-1,
name='mixed9_' + str(i))
branch3x3dbl = conv_norm_relu(x, 448, 1, 1)
branch3x3dbl = conv_norm_relu(branch3x3dbl, 384, 3, 3)
branch3x3dbl_1 = conv_norm_relu(branch3x3dbl, 384, 1, 3)
branch3x3dbl_2 = conv_norm_relu(branch3x3dbl, 384, 3, 1)
branch3x3dbl = concat([branch3x3dbl_1, branch3x3dbl_2], axis=-1)
branch_pool = avg_pool(x, 3, strides=1, padding='SAME')
branch_pool = conv_norm_relu(branch_pool, 192, 1, 1)
x = concat([branch1x1, branch3x3, branch3x3dbl, branch_pool],
axis=-1,
name='mixed' + str(9 + i))
# Classification block
x = avg_pool(x, kernel_size=8, strides=1, name='avg_pool')
x = squeeze(x, axis=[1, 2], name='squeeze')
x = fully_connected(x, self.num_classes, name='predictions')
x = softmax(x, name='output-prob')
return x
def normal_a_cell(ip, p, filters, block_id=None):
"""Adds a Normal cell for NASNet-A (Fig. 4 in the paper).
Args:
ip: Input tensor `x`
p: Input tensor `p`
filters: Number of output filters
block_id: String block_id
Returns:
A tensorflow tensor
"""
channel_dim = -1
with tf.name_scope('normal_A_block_%s' % block_id):
p = NASNetMobile.adjust_block(p, ip, filters, block_id)
h = layers.relu(ip)
h = layers.conv(h,
filters_out=filters,
kernel_size=(1, 1),
stride=1,
padding='same',
name='normal_conv_1_%s' % block_id,
add_bias=False)
h = layers.norm(h,
axis=channel_dim,
momentum=0.9997,
epsilon=1e-3,
name='normal_bn_1_%s' % block_id)
with tf.name_scope('block_1'):
x1_1 = NASNetMobile.separable_conv_block(
h,
filters,
kernel_size=5,
block_id='normal_left1_%s' % block_id)
x1_2 = NASNetMobile.separable_conv_block(
p, filters, block_id='normal_right1_%s' % block_id)
x1 = x1_1 + x1_2
with tf.name_scope('block_2'):
x2_1 = NASNetMobile.separable_conv_block(
p, filters, 5, block_id='normal_left2_%s' % block_id)
x2_2 = NASNetMobile.separable_conv_block(
p, filters, 3, block_id='normal_right2_%s' % block_id)
x2 = x2_1 + x2_2
with tf.name_scope('block_3'):
x3 = layers.avg_pool(h,
3,
strides=1,
padding='same',
name='normal_left3_%s' % block_id)
x3 = x3 + p
with tf.name_scope('block_4'):
x4_1 = layers.avg_pool(p,
3,
strides=1,
padding='same',
name='normal_left4_%s' % block_id)
x4_2 = layers.avg_pool(p,
3,
strides=1,
padding='same',
name='normal_right4_%s' % block_id)
x4 = x4_1 + x4_2
with tf.name_scope('block_5'):
x5 = NASNetMobile.separable_conv_block(
h, filters, block_id='normal_left5_%s' % block_id)
x5 = x5 + h
x = layers.concat([p, x1, x2, x3, x4, x5],
axis=channel_dim,
name='normal_concat_%s' % block_id)
return x, ip
def reduction_a_cell(ip, p, filters, block_id=None):
"""Adds a Reduction cell for NASNet-A (Fig. 4 in the paper).
Args:
ip: Input tensor `x`
p: Input tensor `p`
filters: Number of output filters
block_id: String block_id
Returns:
A tf tensor
"""
channel_dim = -1
with tf.name_scope('reduction_A_block_%s' % block_id):
p = NASNetMobile.adjust_block(p, ip, filters, block_id)
h = layers.relu(ip)
h = layers.conv(h,
filters_out=filters,
kernel_size=(1, 1),
stride=1,
padding='same',
name='reduction_conv_1_%s' % block_id,
add_bias=False)
h = layers.norm(h,
axis=channel_dim,
momentum=0.9997,
epsilon=1e-3,
name='reduction_bn_1_%s' % block_id)
h3 = layers.zero_padding(h,
padding=NASNetMobile.correct_pad(
h, (3, 3)),
name='reduction_pad_1_%s' % block_id)
with tf.name_scope('block_1'):
x1_1 = NASNetMobile.separable_conv_block(
h,
filters=filters,
kernel_size=5,
strides=2,
block_id='reduction_left1_%s' % block_id)
x1_2 = NASNetMobile.separable_conv_block(
p,
filters=filters,
kernel_size=7,
strides=2,
block_id='reduction_right1_%s' % block_id)
x1 = x1_1 + x1_2
with tf.name_scope('block_2'):
x2_1 = layers.max_pool(h3,
3,
strides=2,
padding='valid',
name='reduction_left2_%s' % block_id)
x2_2 = NASNetMobile.separable_conv_block(
p,
filters=filters,
kernel_size=7,
strides=2,
block_id='reduction_right2_%s' % block_id)
x2 = x2_1 + x2_2
with tf.name_scope('block_3'):
x3_1 = layers.avg_pool(h3,
3,
strides=2,
padding='valid',
name='reduction_left3_%s' % block_id)
x3_2 = NASNetMobile.separable_conv_block(
p,
filters,
5,
strides=2,
block_id='reduction_right3_%s' % block_id)
x3 = x3_1 + x3_2
with tf.name_scope('block_4'):
x4 = layers.avg_pool(x1,
3,
strides=1,
padding='same',
name='reduction_left4_%s' % block_id)
x4 += x2
with tf.name_scope('block_5'):
x5_1 = NASNetMobile.separable_conv_block(
x1, filters, 3, block_id='reduction_left4_%s' % block_id)
x5_2 = layers.max_pool(h3,
3,
strides=2,
padding='valid',
name='reduction_right5_%s' % block_id)
x5 = x5_1 + x5_2
x = layers.concat([x2, x3, x4, x5],
axis=channel_dim,
name='reduction_concat_%s' % block_id)
return x, ip
def adjust_block(p, ip, filters, block_id=None):
"""Adjusts the input `previous path` to match the shape of the `input`.
Used in situations where the output number of filters needs to be changed.
Args:
p: Input tensor which needs to be modified
ip: Input tensor whose shape needs to be matched
filters: Number of output filters to be matched
block_id: String block_id
Returns:
Adjusted tf tensor.
"""
channel_dim = -1
img_dim = -2
ip_shape = ip.get_shape().as_list()
if p is not None:
p_shape = p.get_shape().as_list()
else:
p_shape = ip_shape
with tf.name_scope('adjust_block'):
if p is None:
p = ip
elif p_shape[img_dim] != ip_shape[img_dim]:
with tf.name_scope('adjust_reduction_block_%s' % block_id):
p = layers.relu(p, name='adjust_relu_1_%s' % block_id)
p1 = layers.avg_pool(p,
1,
strides=2,
padding='valid',
name='adjust_avg_pool_1_%s' %
block_id)
p1 = layers.conv(p1,
filters_out=filters // 2,
kernel_size=(1, 1),
padding='same',
add_bias=False,
name='adjust_conv_1_%s' % block_id)
p2 = layers.zero_padding(p, padding=((0, 1), (0, 1)))
p2 = layers.crop(p2, cropping=((1, 0), (1, 0)))
p2 = layers.avg_pool(p2,
1,
strides=2,
padding='valid',
name='adjust_avg_pool_2_%s' %
block_id)
p2 = layers.conv(p2,
filters_out=filters // 2,
kernel_size=(1, 1),
padding='same',
add_bias=False,
name='adjust_conv_2_%s' % block_id)
p = layers.concat([p1, p2], axis=channel_dim)
p = layers.norm(p,
axis=channel_dim,
momentum=0.9997,
epsilon=1e-3,
name='adjust_bn_%s' % block_id)
elif p_shape[channel_dim] != filters:
with tf.name_scope('adjust_projection_block_%s' % block_id):
p = layers.relu(p)
p = layers.conv(p,
filters_out=filters,
kernel_size=(1, 1),
stride=1,
padding='same',
name='adjust_conv_projection_%s' %
block_id,
add_bias=False)
p = layers.norm(p,
axis=channel_dim,
momentum=0.9997,
epsilon=1e-3,
name='adjust_bn_%s' % block_id)
return p