mobilenet_v2.py

"""MobileNet v2 models for Keras.

MobileNetV2 is a general architecture and can be used for multiple use cases.
Depending on the use case, it can use different input layer size and
different width factors. This allows different width models to reduce
the number of multiply-adds and thereby
reduce inference cost on mobile devices.

MobileNetV2 is very similar to the original MobileNet,
except that it uses inverted residual blocks with
bottlenecking features. It has a drastically lower
parameter count than the original MobileNet.
MobileNets support any input size greater
than 32 x 32, with larger image sizes
offering better performance.

The number of parameters and number of multiply-adds
can be modified by using the `alpha` parameter,
which increases/decreases the number of filters in each layer.
By altering the image size and `alpha` parameter,
all 22 models from the paper can be built, with ImageNet weights provided.

The paper demonstrates the performance of MobileNets using `alpha` values of
1.0 (also called 100 % MobileNet), 0.35, 0.5, 0.75, 1.0, 1.3, and 1.4

For each of these `alpha` values, weights for 5 different input image sizes
are provided (224, 192, 160, 128, and 96).


The following table describes the performance of
MobileNet on various input sizes:
------------------------------------------------------------------------
MACs stands for Multiply Adds

 Classification Checkpoint| MACs (M) | Parameters (M)| Top 1 Accuracy| Top 5 Accuracy
--------------------------|------------|---------------|---------|----|-------------
| [mobilenet_v2_1.4_224]  | 582 | 6.06 |          75.0 | 92.5 |
| [mobilenet_v2_1.3_224]  | 509 | 5.34 |          74.4 | 92.1 |
| [mobilenet_v2_1.0_224]  | 300 | 3.47 |          71.8 | 91.0 |
| [mobilenet_v2_1.0_192]  | 221 | 3.47 |          70.7 | 90.1 |
| [mobilenet_v2_1.0_160]  | 154 | 3.47 |          68.8 | 89.0 |
| [mobilenet_v2_1.0_128]  | 99  | 3.47 |          65.3 | 86.9 |
| [mobilenet_v2_1.0_96]   | 56  | 3.47 |          60.3 | 83.2 |
| [mobilenet_v2_0.75_224] | 209 | 2.61 |          69.8 | 89.6 |
| [mobilenet_v2_0.75_192] | 153 | 2.61 |          68.7 | 88.9 |
| [mobilenet_v2_0.75_160] | 107 | 2.61 |          66.4 | 87.3 |
| [mobilenet_v2_0.75_128] | 69  | 2.61 |          63.2 | 85.3 |
| [mobilenet_v2_0.75_96]  | 39  | 2.61 |          58.8 | 81.6 |
| [mobilenet_v2_0.5_224]  | 97  | 1.95 |          65.4 | 86.4 |
| [mobilenet_v2_0.5_192]  | 71  | 1.95 |          63.9 | 85.4 |
| [mobilenet_v2_0.5_160]  | 50  | 1.95 |          61.0 | 83.2 |
| [mobilenet_v2_0.5_128]  | 32  | 1.95 |          57.7 | 80.8 |
| [mobilenet_v2_0.5_96]   | 18  | 1.95 |          51.2 | 75.8 |
| [mobilenet_v2_0.35_224] | 59  | 1.66 |          60.3 | 82.9 |
| [mobilenet_v2_0.35_192] | 43  | 1.66 |          58.2 | 81.2 |
| [mobilenet_v2_0.35_160] | 30  | 1.66 |          55.7 | 79.1 |
| [mobilenet_v2_0.35_128] | 20  | 1.66 |          50.8 | 75.0 |
| [mobilenet_v2_0.35_96]  | 11  | 1.66 |          45.5 | 70.4 |

The weights for all 16 models are obtained and
translated from the Tensorflow checkpoints
from TensorFlow checkpoints found [here]
(https://github.com/tensorflow/models/blob/master/research/slim/nets/mobilenet/README.md).

# Reference

This file contains building code for MobileNetV2, based on
[MobileNetV2: Inverted Residuals and Linear Bottlenecks]
(https://arxiv.org/abs/1801.04381) (CVPR 2018)

Tests comparing this model to the existing Tensorflow model can be
found at [mobilenet_v2_keras]
(https://github.com/JonathanCMitchell/mobilenet_v2_keras)
"""
from __future__ import print_function
from __future__ import absolute_import
from __future__ import division

import os
import warnings
import numpy as np

from . import correct_pad
from . import get_submodules_from_kwargs
from . import imagenet_utils
from .imagenet_utils import decode_predictions
from .imagenet_utils import _obtain_input_shape

# TODO Change path to v1.1
BASE_WEIGHT_PATH = ('https://github.com/JonathanCMitchell/mobilenet_v2_keras/'
                    'releases/download/v1.1/')

backend = None
layers = None
models = None
keras_utils = None


def preprocess_input(x, **kwargs):
    """Preprocesses a numpy array encoding a batch of images.

    # Arguments
        x: a 4D numpy array consists of RGB values within [0, 255].

    # Returns
        Preprocessed array.
    """
    return imagenet_utils.preprocess_input(x, mode='tf', **kwargs)


# This function is taken from the original tf repo.
# It ensures that all layers have a channel number that is divisible by 8
# It can be seen here:
# https://github.com/tensorflow/models/blob/master/research/slim/nets/mobilenet/mobilenet.py


def _make_divisible(v, divisor, min_value=None):
    if min_value is None:
        min_value = divisor
    new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
    # Make sure that round down does not go down by more than 10%.
    if new_v < 0.9 * v:
        new_v += divisor
    return new_v


def MobileNetV2(input_shape=None,
                alpha=1.0,
                include_top=True,
                weights='imagenet',
                input_tensor=None,
                pooling=None,
                classes=1000,
                **kwargs):
    """Instantiates the MobileNetV2 architecture.

    # Arguments
        input_shape: optional shape tuple, to be specified if you would
            like to use a model with an input img resolution that is not
            (224, 224, 3).
            It should have exactly 3 inputs channels (224, 224, 3).
            You can also omit this option if you would like
            to infer input_shape from an input_tensor.
            If you choose to include both input_tensor and input_shape then
            input_shape will be used if they match, if the shapes
            do not match then we will throw an error.
            E.g. `(160, 160, 3)` would be one valid value.
        alpha: controls the width of the network. This is known as the
        width multiplier in the MobileNetV2 paper, but the name is kept for
        consistency with MobileNetV1 in Keras.
            - If `alpha` < 1.0, proportionally decreases the number
                of filters in each layer.
            - If `alpha` > 1.0, proportionally increases the number
                of filters in each layer.
            - If `alpha` = 1, default number of filters from the paper
                 are used at each layer.
        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.
        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.

    # Returns
        A Keras model instance.

    # Raises
        ValueError: in case of invalid argument for `weights`,
            or invalid input shape or invalid alpha, rows when
            weights='imagenet'
    """
    global backend, layers, models, keras_utils
    backend, layers, models, keras_utils = get_submodules_from_kwargs(kwargs)

    if not (weights in {'imagenet', None} or os.path.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 and default size.
    # If both input_shape and input_tensor are used, they should match
    if input_shape is not None and input_tensor is not None:
        try:
            is_input_t_tensor = backend.is_keras_tensor(input_tensor)
        except ValueError:
            try:
                is_input_t_tensor = backend.is_keras_tensor(
                    keras_utils.get_source_inputs(input_tensor))
            except ValueError:
                raise ValueError('input_tensor: ', input_tensor,
                                 'is not type input_tensor')
        if is_input_t_tensor:
            if backend.image_data_format == 'channels_first':
                if backend.int_shape(input_tensor)[1] != input_shape[1]:
                    raise ValueError('input_shape: ', input_shape,
                                     'and input_tensor: ', input_tensor,
                                     'do not meet the same shape requirements')
            else:
                if backend.int_shape(input_tensor)[2] != input_shape[1]:
                    raise ValueError('input_shape: ', input_shape,
                                     'and input_tensor: ', input_tensor,
                                     'do not meet the same shape requirements')
        else:
            raise ValueError('input_tensor specified: ', input_tensor,
                             'is not a keras tensor')

    # If input_shape is None, infer shape from input_tensor
    if input_shape is None and input_tensor is not None:

        try:
            backend.is_keras_tensor(input_tensor)
        except ValueError:
            raise ValueError('input_tensor: ', input_tensor,
                             'is type: ', type(input_tensor),
                             'which is not a valid type')

        if input_shape is None and not backend.is_keras_tensor(input_tensor):
            default_size = 224
        elif input_shape is None and backend.is_keras_tensor(input_tensor):
            if backend.image_data_format() == 'channels_first':
                rows = backend.int_shape(input_tensor)[2]
                cols = backend.int_shape(input_tensor)[3]
            else:
                rows = backend.int_shape(input_tensor)[1]
                cols = backend.int_shape(input_tensor)[2]

            if rows == cols and rows in [96, 128, 160, 192, 224]:
                default_size = rows
            else:
                default_size = 224

    # If input_shape is None and no input_tensor
    elif input_shape is None:
        default_size = 224

    # If input_shape is not None, assume default size
    else:
        if backend.image_data_format() == 'channels_first':
            rows = input_shape[1]
            cols = input_shape[2]
        else:
            rows = input_shape[0]
            cols = input_shape[1]

        if rows == cols and rows in [96, 128, 160, 192, 224]:
            default_size = rows
        else:
            default_size = 224

    input_shape = _obtain_input_shape(input_shape,
                                      default_size=default_size,
                                      min_size=32,
                                      data_format=backend.image_data_format(),
                                      require_flatten=include_top,
                                      weights=weights)

    if backend.image_data_format() == 'channels_last':
        row_axis, col_axis = (0, 1)
    else:
        row_axis, col_axis = (1, 2)
    rows = input_shape[row_axis]
    cols = input_shape[col_axis]

    if weights == 'imagenet':
        if alpha not in [0.35, 0.50, 0.75, 1.0, 1.3, 1.4]:
            raise ValueError('If imagenet weights are being loaded, '
                             'alpha can be one of `0.35`, `0.50`, `0.75`, '
                             '`1.0`, `1.3` or `1.4` only.')

        if rows != cols or rows not in [96, 128, 160, 192, 224]:
            if rows is None:
                rows = 224
                warnings.warn('MobileNet shape is undefined.'
                              ' Weights for input shape'
                              '(224, 224) will be loaded.')

    if backend.image_data_format() != 'channels_last':
        warnings.warn('The MobileNet family of models is only available '
                      'for the input data format "channels_last" '
                      '(width, height, channels). '
                      'However your settings specify the default '
                      'data format "channels_first" (channels, width, height).'
                      ' You should set `image_data_format="channels_last"` '
                      'in your Keras config located at ~/.keras/keras.json. '
                      'The model being returned right now will expect inputs '
                      'to follow the "channels_last" data format.')
        backend.set_image_data_format('channels_last')
        old_data_format = 'channels_first'
    else:
        old_data_format = None

    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

    first_block_filters = _make_divisible(32 * alpha, 8)
    x = layers.ZeroPadding2D(padding=correct_pad(backend, img_input, 3),
                             name='Conv1_pad')(img_input)
    x = layers.Conv2D(first_block_filters,
                      kernel_size=3,
                      strides=(2, 2),
                      padding='valid',
                      use_bias=False,
                      name='Conv1')(x)
    x = layers.BatchNormalization(
        epsilon=1e-3, momentum=0.999, name='bn_Conv1')(x)
    x = layers.ReLU(6., name='Conv1_relu')(x)

    x = _inverted_res_block(x, filters=16, alpha=alpha, stride=1,
                            expansion=1, block_id=0)

    x = _inverted_res_block(x, filters=24, alpha=alpha, stride=2,
                            expansion=6, block_id=1)
    x = _inverted_res_block(x, filters=24, alpha=alpha, stride=1,
                            expansion=6, block_id=2)

    x = _inverted_res_block(x, filters=32, alpha=alpha, stride=2,
                            expansion=6, block_id=3)
    x = _inverted_res_block(x, filters=32, alpha=alpha, stride=1,
                            expansion=6, block_id=4)
    x = _inverted_res_block(x, filters=32, alpha=alpha, stride=1,
                            expansion=6, block_id=5)

    x = _inverted_res_block(x, filters=64, alpha=alpha, stride=2,
                            expansion=6, block_id=6)
    x = _inverted_res_block(x, filters=64, alpha=alpha, stride=1,
                            expansion=6, block_id=7)
    x = _inverted_res_block(x, filters=64, alpha=alpha, stride=1,
                            expansion=6, block_id=8)
    x = _inverted_res_block(x, filters=64, alpha=alpha, stride=1,
                            expansion=6, block_id=9)

    x = _inverted_res_block(x, filters=96, alpha=alpha, stride=1,
                            expansion=6, block_id=10)
    x = _inverted_res_block(x, filters=96, alpha=alpha, stride=1,
                            expansion=6, block_id=11)
    x = _inverted_res_block(x, filters=96, alpha=alpha, stride=1,
                            expansion=6, block_id=12)

    x = _inverted_res_block(x, filters=160, alpha=alpha, stride=2,
                            expansion=6, block_id=13)
    x = _inverted_res_block(x, filters=160, alpha=alpha, stride=1,
                            expansion=6, block_id=14)
    x = _inverted_res_block(x, filters=160, alpha=alpha, stride=1,
                            expansion=6, block_id=15)

    x = _inverted_res_block(x, filters=320, alpha=alpha, stride=1,
                            expansion=6, block_id=16)

    # no alpha applied to last conv as stated in the paper:
    # if the width multiplier is greater than 1 we
    # increase the number of output channels
    if alpha > 1.0:
        last_block_filters = _make_divisible(1280 * alpha, 8)
    else:
        last_block_filters = 1280

    x = layers.Conv2D(last_block_filters,
                      kernel_size=1,
                      use_bias=False,
                      name='Conv_1')(x)
    x = layers.BatchNormalization(epsilon=1e-3,
                                  momentum=0.999,
                                  name='Conv_1_bn')(x)
    x = layers.ReLU(6., name='out_relu')(x)

    if include_top:
        x = layers.GlobalAveragePooling2D()(x)
        x = layers.Dense(classes, activation='softmax',
                         use_bias=True, name='Logits')(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 = keras_utils.get_source_inputs(input_tensor)
    else:
        inputs = img_input

    # Create model.
    model = models.Model(inputs, x,
                         name='mobilenetv2_%0.2f_%s' % (alpha, rows))

    # Load weights.
    if weights == 'imagenet':
        if backend.image_data_format() == 'channels_first':
            raise ValueError('Weights for "channels_first" format '
                             'are not available.')

        if include_top:
            model_name = ('mobilenet_v2_weights_tf_dim_ordering_tf_kernels_' +
                          str(alpha) + '_' + str(rows) + '.h5')
            weigh_path = BASE_WEIGHT_PATH + model_name
            weights_path = keras_utils.get_file(
                model_name, weigh_path, cache_subdir='models')
        else:
            model_name = ('mobilenet_v2_weights_tf_dim_ordering_tf_kernels_' +
                          str(alpha) + '_' + str(rows) + '_no_top' + '.h5')
            weigh_path = BASE_WEIGHT_PATH + model_name
            weights_path = keras_utils.get_file(
                model_name, weigh_path, cache_subdir='models')
        model.load_weights(weights_path)
    elif weights is not None:
        model.load_weights(weights)

    if old_data_format:
        backend.set_image_data_format(old_data_format)
    return model


def _inverted_res_block(inputs, expansion, stride, alpha, filters, block_id):
    in_channels = backend.int_shape(inputs)[-1]
    pointwise_conv_filters = int(filters * alpha)
    pointwise_filters = _make_divisible(pointwise_conv_filters, 8)
    x = inputs
    prefix = 'block_{}_'.format(block_id)

    if block_id:
        # Expand
        x = layers.Conv2D(expansion * in_channels,
                          kernel_size=1,
                          padding='same',
                          use_bias=False,
                          activation=None,
                          name=prefix + 'expand')(x)
        x = layers.BatchNormalization(epsilon=1e-3,
                                      momentum=0.999,
                                      name=prefix + 'expand_BN')(x)
        x = layers.ReLU(6., name=prefix + 'expand_relu')(x)
    else:
        prefix = 'expanded_conv_'

    # Depthwise
    if stride == 2:
        x = layers.ZeroPadding2D(padding=correct_pad(backend, x, 3),
                                 name=prefix + 'pad')(x)
    x = layers.DepthwiseConv2D(kernel_size=3,
                               strides=stride,
                               activation=None,
                               use_bias=False,
                               padding='same' if stride == 1 else 'valid',
                               name=prefix + 'depthwise')(x)
    x = layers.BatchNormalization(epsilon=1e-3,
                                  momentum=0.999,
                                  name=prefix + 'depthwise_BN')(x)

    x = layers.ReLU(6., name=prefix + 'depthwise_relu')(x)

    # Project
    x = layers.Conv2D(pointwise_filters,
                      kernel_size=1,
                      padding='same',
                      use_bias=False,
                      activation=None,
                      name=prefix + 'project')(x)
    x = layers.BatchNormalization(
        epsilon=1e-3, momentum=0.999, name=prefix + 'project_BN')(x)

    if in_channels == pointwise_filters and stride == 1:
        return layers.Add(name=prefix + 'add')([inputs, x])
    return x