439 lines
16 KiB
Python
439 lines
16 KiB
Python
# Copyright 2017 The TensorFlow Authors. All Rights Reserved.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
# ==============================================================================
|
|
|
|
"""Inception-ResNet V2 model for Keras.
|
|
|
|
Reference:
|
|
- [Inception-v4, Inception-ResNet and the Impact of
|
|
Residual Connections on Learning](https://arxiv.org/abs/1602.07261)
|
|
(AAAI 2017)
|
|
"""
|
|
|
|
import tensorflow.compat.v2 as tf
|
|
|
|
import keras
|
|
from keras import backend
|
|
from keras import layers as keras_layers
|
|
from keras.applications import imagenet_utils
|
|
from keras.engine import training
|
|
from keras.layers import VersionAwareLayers
|
|
from keras.utils import data_utils
|
|
from keras.utils import layer_utils
|
|
|
|
# isort: off
|
|
from tensorflow.python.util.tf_export import keras_export
|
|
|
|
BASE_WEIGHT_URL = (
|
|
"https://storage.googleapis.com/tensorflow/"
|
|
"keras-applications/inception_resnet_v2/"
|
|
)
|
|
layers = None
|
|
|
|
|
|
@keras_export(
|
|
"keras.applications.inception_resnet_v2.InceptionResNetV2",
|
|
"keras.applications.InceptionResNetV2",
|
|
)
|
|
def InceptionResNetV2(
|
|
include_top=True,
|
|
weights="imagenet",
|
|
input_tensor=None,
|
|
input_shape=None,
|
|
pooling=None,
|
|
classes=1000,
|
|
classifier_activation="softmax",
|
|
**kwargs,
|
|
):
|
|
"""Instantiates the Inception-ResNet v2 architecture.
|
|
|
|
Reference:
|
|
- [Inception-v4, Inception-ResNet and the Impact of
|
|
Residual Connections on Learning](https://arxiv.org/abs/1602.07261)
|
|
(AAAI 2017)
|
|
|
|
This function returns a Keras image classification model,
|
|
optionally loaded with weights pre-trained on ImageNet.
|
|
|
|
For image classification use cases, see
|
|
[this page for detailed examples](
|
|
https://keras.io/api/applications/#usage-examples-for-image-classification-models).
|
|
|
|
For transfer learning use cases, make sure to read the
|
|
[guide to transfer learning & fine-tuning](
|
|
https://keras.io/guides/transfer_learning/).
|
|
|
|
Note: each Keras Application expects a specific kind of input preprocessing.
|
|
For InceptionResNetV2, call
|
|
`tf.keras.applications.inception_resnet_v2.preprocess_input`
|
|
on your inputs before passing them to the model.
|
|
`inception_resnet_v2.preprocess_input`
|
|
will scale input pixels between -1 and 1.
|
|
|
|
Args:
|
|
include_top: whether to include the fully-connected
|
|
layer at the top of the network.
|
|
weights: one of `None` (random initialization),
|
|
'imagenet' (pre-training on ImageNet),
|
|
or the path to the weights file to be loaded.
|
|
input_tensor: optional Keras tensor (i.e. output of `layers.Input()`)
|
|
to use as image input for the model.
|
|
input_shape: optional shape tuple, only to be specified
|
|
if `include_top` is `False` (otherwise the input shape
|
|
has to be `(299, 299, 3)` (with `'channels_last'` data format)
|
|
or `(3, 299, 299)` (with `'channels_first'` data format).
|
|
It should have exactly 3 inputs channels,
|
|
and width and height should be no smaller than 75.
|
|
E.g. `(150, 150, 3)` would be one valid value.
|
|
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 block.
|
|
- `'avg'` means that global average pooling
|
|
will be applied to the output of the
|
|
last convolutional block, 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"`.
|
|
**kwargs: For backwards compatibility only.
|
|
|
|
Returns:
|
|
A `keras.Model` instance.
|
|
"""
|
|
global layers
|
|
if "layers" in kwargs:
|
|
layers = kwargs.pop("layers")
|
|
else:
|
|
layers = VersionAwareLayers()
|
|
if kwargs:
|
|
raise ValueError(f"Unknown argument(s): {kwargs}")
|
|
if not (weights in {"imagenet", None} or tf.io.gfile.exists(weights)):
|
|
raise ValueError(
|
|
"The `weights` argument should be either "
|
|
"`None` (random initialization), `imagenet` "
|
|
"(pre-training on ImageNet), "
|
|
"or the path to the weights file to be loaded."
|
|
)
|
|
|
|
if weights == "imagenet" and include_top and classes != 1000:
|
|
raise ValueError(
|
|
'If using `weights` as `"imagenet"` with `include_top`'
|
|
" as true, `classes` should be 1000"
|
|
)
|
|
|
|
# Determine proper input shape
|
|
input_shape = imagenet_utils.obtain_input_shape(
|
|
input_shape,
|
|
default_size=299,
|
|
min_size=75,
|
|
data_format=backend.image_data_format(),
|
|
require_flatten=include_top,
|
|
weights=weights,
|
|
)
|
|
|
|
if input_tensor is None:
|
|
img_input = layers.Input(shape=input_shape)
|
|
else:
|
|
if not backend.is_keras_tensor(input_tensor):
|
|
img_input = layers.Input(tensor=input_tensor, shape=input_shape)
|
|
else:
|
|
img_input = input_tensor
|
|
|
|
# Stem block: 35 x 35 x 192
|
|
x = conv2d_bn(img_input, 32, 3, strides=2, padding="valid")
|
|
x = conv2d_bn(x, 32, 3, padding="valid")
|
|
x = conv2d_bn(x, 64, 3)
|
|
x = layers.MaxPooling2D(3, strides=2)(x)
|
|
x = conv2d_bn(x, 80, 1, padding="valid")
|
|
x = conv2d_bn(x, 192, 3, padding="valid")
|
|
x = layers.MaxPooling2D(3, strides=2)(x)
|
|
|
|
# Mixed 5b (Inception-A block): 35 x 35 x 320
|
|
branch_0 = conv2d_bn(x, 96, 1)
|
|
branch_1 = conv2d_bn(x, 48, 1)
|
|
branch_1 = conv2d_bn(branch_1, 64, 5)
|
|
branch_2 = conv2d_bn(x, 64, 1)
|
|
branch_2 = conv2d_bn(branch_2, 96, 3)
|
|
branch_2 = conv2d_bn(branch_2, 96, 3)
|
|
branch_pool = layers.AveragePooling2D(3, strides=1, padding="same")(x)
|
|
branch_pool = conv2d_bn(branch_pool, 64, 1)
|
|
branches = [branch_0, branch_1, branch_2, branch_pool]
|
|
channel_axis = 1 if backend.image_data_format() == "channels_first" else 3
|
|
x = layers.Concatenate(axis=channel_axis, name="mixed_5b")(branches)
|
|
|
|
# 10x block35 (Inception-ResNet-A block): 35 x 35 x 320
|
|
for block_idx in range(1, 11):
|
|
x = inception_resnet_block(
|
|
x, scale=0.17, block_type="block35", block_idx=block_idx
|
|
)
|
|
|
|
# Mixed 6a (Reduction-A block): 17 x 17 x 1088
|
|
branch_0 = conv2d_bn(x, 384, 3, strides=2, padding="valid")
|
|
branch_1 = conv2d_bn(x, 256, 1)
|
|
branch_1 = conv2d_bn(branch_1, 256, 3)
|
|
branch_1 = conv2d_bn(branch_1, 384, 3, strides=2, padding="valid")
|
|
branch_pool = layers.MaxPooling2D(3, strides=2, padding="valid")(x)
|
|
branches = [branch_0, branch_1, branch_pool]
|
|
x = layers.Concatenate(axis=channel_axis, name="mixed_6a")(branches)
|
|
|
|
# 20x block17 (Inception-ResNet-B block): 17 x 17 x 1088
|
|
for block_idx in range(1, 21):
|
|
x = inception_resnet_block(
|
|
x, scale=0.1, block_type="block17", block_idx=block_idx
|
|
)
|
|
|
|
# Mixed 7a (Reduction-B block): 8 x 8 x 2080
|
|
branch_0 = conv2d_bn(x, 256, 1)
|
|
branch_0 = conv2d_bn(branch_0, 384, 3, strides=2, padding="valid")
|
|
branch_1 = conv2d_bn(x, 256, 1)
|
|
branch_1 = conv2d_bn(branch_1, 288, 3, strides=2, padding="valid")
|
|
branch_2 = conv2d_bn(x, 256, 1)
|
|
branch_2 = conv2d_bn(branch_2, 288, 3)
|
|
branch_2 = conv2d_bn(branch_2, 320, 3, strides=2, padding="valid")
|
|
branch_pool = layers.MaxPooling2D(3, strides=2, padding="valid")(x)
|
|
branches = [branch_0, branch_1, branch_2, branch_pool]
|
|
x = layers.Concatenate(axis=channel_axis, name="mixed_7a")(branches)
|
|
|
|
# 10x block8 (Inception-ResNet-C block): 8 x 8 x 2080
|
|
for block_idx in range(1, 10):
|
|
x = inception_resnet_block(
|
|
x, scale=0.2, block_type="block8", block_idx=block_idx
|
|
)
|
|
x = inception_resnet_block(
|
|
x, scale=1.0, activation=None, block_type="block8", block_idx=10
|
|
)
|
|
|
|
# Final convolution block: 8 x 8 x 1536
|
|
x = conv2d_bn(x, 1536, 1, name="conv_7b")
|
|
|
|
if include_top:
|
|
# Classification block
|
|
x = layers.GlobalAveragePooling2D(name="avg_pool")(x)
|
|
imagenet_utils.validate_activation(classifier_activation, weights)
|
|
x = layers.Dense(
|
|
classes, activation=classifier_activation, name="predictions"
|
|
)(x)
|
|
else:
|
|
if pooling == "avg":
|
|
x = layers.GlobalAveragePooling2D()(x)
|
|
elif pooling == "max":
|
|
x = layers.GlobalMaxPooling2D()(x)
|
|
|
|
# Ensure that the model takes into account
|
|
# any potential predecessors of `input_tensor`.
|
|
if input_tensor is not None:
|
|
inputs = layer_utils.get_source_inputs(input_tensor)
|
|
else:
|
|
inputs = img_input
|
|
|
|
# Create model.
|
|
model = training.Model(inputs, x, name="inception_resnet_v2")
|
|
|
|
# Load weights.
|
|
if weights == "imagenet":
|
|
if include_top:
|
|
fname = "inception_resnet_v2_weights_tf_dim_ordering_tf_kernels.h5"
|
|
weights_path = data_utils.get_file(
|
|
fname,
|
|
BASE_WEIGHT_URL + fname,
|
|
cache_subdir="models",
|
|
file_hash="e693bd0210a403b3192acc6073ad2e96",
|
|
)
|
|
else:
|
|
fname = (
|
|
"inception_resnet_v2_weights_"
|
|
"tf_dim_ordering_tf_kernels_notop.h5"
|
|
)
|
|
weights_path = data_utils.get_file(
|
|
fname,
|
|
BASE_WEIGHT_URL + fname,
|
|
cache_subdir="models",
|
|
file_hash="d19885ff4a710c122648d3b5c3b684e4",
|
|
)
|
|
model.load_weights(weights_path)
|
|
elif weights is not None:
|
|
model.load_weights(weights)
|
|
|
|
return model
|
|
|
|
|
|
def conv2d_bn(
|
|
x,
|
|
filters,
|
|
kernel_size,
|
|
strides=1,
|
|
padding="same",
|
|
activation="relu",
|
|
use_bias=False,
|
|
name=None,
|
|
):
|
|
"""Utility function to apply conv + BN.
|
|
|
|
Args:
|
|
x: input tensor.
|
|
filters: filters in `Conv2D`.
|
|
kernel_size: kernel size as in `Conv2D`.
|
|
strides: strides in `Conv2D`.
|
|
padding: padding mode in `Conv2D`.
|
|
activation: activation in `Conv2D`.
|
|
use_bias: whether to use a bias in `Conv2D`.
|
|
name: name of the ops; will become `name + '_ac'` for the activation
|
|
and `name + '_bn'` for the batch norm layer.
|
|
|
|
Returns:
|
|
Output tensor after applying `Conv2D` and `BatchNormalization`.
|
|
"""
|
|
x = layers.Conv2D(
|
|
filters,
|
|
kernel_size,
|
|
strides=strides,
|
|
padding=padding,
|
|
use_bias=use_bias,
|
|
name=name,
|
|
)(x)
|
|
if not use_bias:
|
|
bn_axis = 1 if backend.image_data_format() == "channels_first" else 3
|
|
bn_name = None if name is None else name + "_bn"
|
|
x = layers.BatchNormalization(axis=bn_axis, scale=False, name=bn_name)(
|
|
x
|
|
)
|
|
if activation is not None:
|
|
ac_name = None if name is None else name + "_ac"
|
|
x = layers.Activation(activation, name=ac_name)(x)
|
|
return x
|
|
|
|
|
|
@keras.utils.register_keras_serializable()
|
|
class CustomScaleLayer(keras_layers.Layer):
|
|
def __init__(self, scale, **kwargs):
|
|
super().__init__(**kwargs)
|
|
self.scale = scale
|
|
|
|
def get_config(self):
|
|
config = super().get_config()
|
|
config.update({"scale": self.scale})
|
|
return config
|
|
|
|
def call(self, inputs):
|
|
return inputs[0] + inputs[1] * self.scale
|
|
|
|
|
|
def inception_resnet_block(x, scale, block_type, block_idx, activation="relu"):
|
|
"""Adds an Inception-ResNet block.
|
|
|
|
This function builds 3 types of Inception-ResNet blocks mentioned
|
|
in the paper, controlled by the `block_type` argument (which is the
|
|
block name used in the official TF-slim implementation):
|
|
- Inception-ResNet-A: `block_type='block35'`
|
|
- Inception-ResNet-B: `block_type='block17'`
|
|
- Inception-ResNet-C: `block_type='block8'`
|
|
|
|
Args:
|
|
x: input tensor.
|
|
scale: scaling factor to scale the residuals (i.e., the output of passing
|
|
`x` through an inception module) before adding them to the shortcut
|
|
branch. Let `r` be the output from the residual branch, the output of
|
|
this block will be `x + scale * r`.
|
|
block_type: `'block35'`, `'block17'` or `'block8'`, determines the network
|
|
structure in the residual branch.
|
|
block_idx: an `int` used for generating layer names. The Inception-ResNet
|
|
blocks are repeated many times in this network. We use `block_idx` to
|
|
identify each of the repetitions. For example, the first
|
|
Inception-ResNet-A block will have `block_type='block35', block_idx=0`,
|
|
and the layer names will have a common prefix `'block35_0'`.
|
|
activation: activation function to use at the end of the block (see
|
|
[activations](../activations.md)). When `activation=None`, no activation
|
|
is applied
|
|
(i.e., "linear" activation: `a(x) = x`).
|
|
|
|
Returns:
|
|
Output tensor for the block.
|
|
|
|
Raises:
|
|
ValueError: if `block_type` is not one of `'block35'`,
|
|
`'block17'` or `'block8'`.
|
|
"""
|
|
if block_type == "block35":
|
|
branch_0 = conv2d_bn(x, 32, 1)
|
|
branch_1 = conv2d_bn(x, 32, 1)
|
|
branch_1 = conv2d_bn(branch_1, 32, 3)
|
|
branch_2 = conv2d_bn(x, 32, 1)
|
|
branch_2 = conv2d_bn(branch_2, 48, 3)
|
|
branch_2 = conv2d_bn(branch_2, 64, 3)
|
|
branches = [branch_0, branch_1, branch_2]
|
|
elif block_type == "block17":
|
|
branch_0 = conv2d_bn(x, 192, 1)
|
|
branch_1 = conv2d_bn(x, 128, 1)
|
|
branch_1 = conv2d_bn(branch_1, 160, [1, 7])
|
|
branch_1 = conv2d_bn(branch_1, 192, [7, 1])
|
|
branches = [branch_0, branch_1]
|
|
elif block_type == "block8":
|
|
branch_0 = conv2d_bn(x, 192, 1)
|
|
branch_1 = conv2d_bn(x, 192, 1)
|
|
branch_1 = conv2d_bn(branch_1, 224, [1, 3])
|
|
branch_1 = conv2d_bn(branch_1, 256, [3, 1])
|
|
branches = [branch_0, branch_1]
|
|
else:
|
|
raise ValueError(
|
|
"Unknown Inception-ResNet block type. "
|
|
'Expects "block35", "block17" or "block8", '
|
|
"but got: " + str(block_type)
|
|
)
|
|
|
|
block_name = block_type + "_" + str(block_idx)
|
|
channel_axis = 1 if backend.image_data_format() == "channels_first" else 3
|
|
mixed = layers.Concatenate(axis=channel_axis, name=block_name + "_mixed")(
|
|
branches
|
|
)
|
|
up = conv2d_bn(
|
|
mixed,
|
|
backend.int_shape(x)[channel_axis],
|
|
1,
|
|
activation=None,
|
|
use_bias=True,
|
|
name=block_name + "_conv",
|
|
)
|
|
|
|
x = CustomScaleLayer(scale)([x, up])
|
|
if activation is not None:
|
|
x = layers.Activation(activation, name=block_name + "_ac")(x)
|
|
return x
|
|
|
|
|
|
@keras_export("keras.applications.inception_resnet_v2.preprocess_input")
|
|
def preprocess_input(x, data_format=None):
|
|
return imagenet_utils.preprocess_input(
|
|
x, data_format=data_format, mode="tf"
|
|
)
|
|
|
|
|
|
@keras_export("keras.applications.inception_resnet_v2.decode_predictions")
|
|
def decode_predictions(preds, top=5):
|
|
return imagenet_utils.decode_predictions(preds, top=top)
|
|
|
|
|
|
preprocess_input.__doc__ = imagenet_utils.PREPROCESS_INPUT_DOC.format(
|
|
mode="",
|
|
ret=imagenet_utils.PREPROCESS_INPUT_RET_DOC_TF,
|
|
error=imagenet_utils.PREPROCESS_INPUT_ERROR_DOC,
|
|
)
|
|
decode_predictions.__doc__ = imagenet_utils.decode_predictions.__doc__
|