csp_darknet.py

# Copyright 2022 The KerasCV Authors
#
# 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
#
#     https://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.

"""DarkNet models for KerasCV.
Reference:
    - [YoloV4 Paper](https://arxiv.org/abs/1804.02767)
    - [CSPNet Paper](https://arxiv.org/pdf/1911.11929)
    - [YoloX Paper](https://arxiv.org/abs/2107.08430)
    - [YoloX implementation](https://github.com/ultralytics/yolov3)
"""

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers

from keras_cv.models import utils
from keras_cv.models.__internal__.darknet_utils import CrossStagePartial
from keras_cv.models.__internal__.darknet_utils import DarknetConvBlock
from keras_cv.models.__internal__.darknet_utils import DarknetConvBlockDepthwise
from keras_cv.models.__internal__.darknet_utils import Focus
from keras_cv.models.__internal__.darknet_utils import SpatialPyramidPoolingBottleneck


def CSPDarkNet(
    include_rescaling,
    include_top,
    depth_multiplier=1.0,
    width_multiplier=1.0,
    use_depthwise=False,
    classes=None,
    weights=None,
    input_shape=(None, None, 3),
    input_tensor=None,
    pooling=None,
    classifier_activation="softmax",
    name=None,
    **kwargs,
):
    """Instantiates the CSPDarkNet architecture.

    Although the DarkNet architecture is commonly used for detection tasks, it is
    possible to extract the intermediate dark2 to dark5 layers from the model for
    creating a feature pyramid Network.

    Reference:
        - [YoloV4 Paper](https://arxiv.org/abs/1804.02767)
        - [CSPNet Paper](https://arxiv.org/pdf/1911.11929)
        - [YoloX Paper](https://arxiv.org/abs/2107.08430)
        - [YoloX implementation](https://github.com/ultralytics/yolov3)
    For transfer learning use cases, make sure to read the
    [guide to transfer learning & fine-tuning](https://keras.io/guides/transfer_learning/).

    Args:
        include_rescaling: whether or not to Rescale the inputs.If set to True,
            inputs will be passed through a `Rescaling(1/255.0)` layer.
        include_top: whether to include the fully-connected layer at the top of
            the network.  If provided, `classes` must be provided.
        depth_multiplier: A float value used to calculate the base depth of the model
            this changes based the detection model being used. Defaults to 1.0.
        width_multiplier: A float value used to calculate the base width of the model
            this changes based the detection model being used. Defaults to 1.0.
        use_depthwise: a boolean value used to decide whether a depthwise conv block
            should be used over a regular darknet block. Defaults to False
        classes: optional number of classes to classify images into, only to be
            specified if `include_top` is True.
        weights: one of `None` (random initialization), or a pretrained weight
            file path.
        input_tensor: optional Keras tensor (i.e. output of `layers.Input()`)
            to use as image input for the model.
        input_shape: optional shape tuple, defaults to (None, None, 3).
        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.
        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.

        name: (Optional) name to pass to the model.  Defaults to "DarkNet".
    Returns:
        A `keras.Model` instance.
    """
    if weights and not tf.io.gfile.exists(weights):
        raise ValueError(
            "The `weights` argument should be either `None` or the path to the "
            f"weights file to be loaded. Weights file not found at location: {weights}"
        )

    if include_top and not classes:
        raise ValueError(
            "If `include_top` is True, you should specify `classes`. Received: "
            f"classes={classes}"
        )

    ConvBlock = DarknetConvBlockDepthwise if use_depthwise else DarknetConvBlock

    base_channels = int(width_multiplier * 64)
    base_depth = max(round(depth_multiplier * 3), 1)

    inputs = utils.parse_model_inputs(input_shape, input_tensor)

    x = inputs
    if include_rescaling:
        x = layers.Rescaling(1 / 255.0)(x)

    # stem
    x = Focus(name="stem_focus")(x)
    x = DarknetConvBlock(base_channels, kernel_size=3, strides=1, name="stem_conv")(x)

    # dark2
    x = ConvBlock(base_channels * 2, kernel_size=3, strides=2, name="dark2_conv")(x)
    x = CrossStagePartial(
        base_channels * 2,
        num_bottlenecks=base_depth,
        use_depthwise=use_depthwise,
        name="dark2_csp",
    )(x)

    # dark3
    x = ConvBlock(base_channels * 4, kernel_size=3, strides=2, name="dark3_conv")(x)
    x = CrossStagePartial(
        base_channels * 4,
        num_bottlenecks=base_depth * 3,
        use_depthwise=use_depthwise,
        name="dark3_csp",
    )(x)

    # dark4
    x = ConvBlock(base_channels * 8, kernel_size=3, strides=2, name="dark4_conv")(x)
    x = CrossStagePartial(
        base_channels * 8,
        num_bottlenecks=base_depth * 3,
        use_depthwise=use_depthwise,
        name="dark4_csp",
    )(x)

    # dark5
    x = ConvBlock(base_channels * 16, kernel_size=3, strides=2, name="dark5_conv")(x)
    x = SpatialPyramidPoolingBottleneck(
        base_channels * 16, hidden_filters=base_channels * 8, name="dark5_spp"
    )(x)
    x = CrossStagePartial(
        base_channels * 16,
        num_bottlenecks=base_depth,
        residual=False,
        use_depthwise=use_depthwise,
        name="dark5_csp",
    )(x)

    if include_top:
        x = layers.GlobalAveragePooling2D(name="avg_pool")(x)
        x = layers.Dense(classes, activation=classifier_activation, name="predictions")(
            x
        )
    elif pooling == "avg":
        x = layers.GlobalAveragePooling2D(name="avg_pool")(x)
    elif pooling == "max":
        x = layers.GlobalMaxPooling2D(name="max_pool")(x)

    model = keras.Model(inputs, x, name=name, **kwargs)

    if weights is not None:
        model.load_weights(weights)
    return model