archai/tasks/facial_landmark_detection/model.py

# Copyright (c) Microsoft Corporation.
# Licensed under the MIT license.

import torch
from typing import Callable, List, Optional

from torch import nn, Tensor
from torchvision.models._utils import _make_divisible
from torchvision.ops.misc import Conv2dNormActivation


# Adapted from https://github.com/pytorch/vision/blob/main/torchvision/models/mobilenetv2.py
class CustomInvertedResidual(nn.Module):
    def __init__(
        self,
        inp: int,
        oup: int,
        stride: int,
        expand_ratio: int,
        kernel: int,
        norm_layer: Optional[Callable[..., nn.Module]] = None,
    ) -> None:
        super().__init__()
        self.stride = stride
        if stride not in [1, 2]:
            raise ValueError(f"stride should be 1 or 2 instead of {stride}")

        if norm_layer is None:
            norm_layer = nn.BatchNorm2d

        hidden_dim = int(round(inp * expand_ratio))
        self.use_res_connect = self.stride == 1 and inp == oup

        layers: List[nn.Module] = []
        if expand_ratio != 1:
            # pw
            layers.append(
                Conv2dNormActivation(inp, hidden_dim, kernel_size=1, norm_layer=norm_layer, activation_layer=nn.ReLU6)
            )
        layers.extend(
            [
                # dw
                Conv2dNormActivation(
                    hidden_dim,
                    hidden_dim,
                    kernel_size=kernel,
                    stride=stride,
                    groups=hidden_dim,
                    norm_layer=norm_layer,
                    activation_layer=nn.ReLU6,
                ),
                # pw-linear
                nn.Conv2d(hidden_dim, oup, 1, 1, 0, bias=False),
                norm_layer(oup),
            ]
        )
        self.conv = nn.Sequential(*layers)
        self.out_channels = oup
        self._is_cn = stride > 1

    def forward(self, x: Tensor) -> Tensor:
        if self.use_res_connect:
            return x + self.conv(x)
        else:
            return self.conv(x)


#
# Adapted from torchvision MobileNetV2 to allow for different kernel sizes
# to be passed in instead of default value of 3
class CustomMobileNetV2(nn.Module):
    def __init__(
        self,
        num_classes: int,
        width_mult: float = 1.0,
        inverted_residual_setting: Optional[List[List[int]]] = None,
        round_nearest: int = 8,
        block: Optional[Callable[..., nn.Module]] = None,
        norm_layer: Optional[Callable[..., nn.Module]] = None,
        dropout: float = 0.2,
    ) -> None:
        """
        MobileNet V2 main class

        Args:
            num_classes (int): Number of classes
            width_mult (float): Width multiplier - adjusts number of channels in each layer by this amount
            inverted_residual_setting: Network structure
            round_nearest (int): Round the number of channels in each layer to be a multiple of this number
            Set to 1 to turn off rounding
            block: Module specifying inverted residual building block for mobilenet
            norm_layer: Module specifying the normalization layer to use
            dropout (float): The droupout probability

        """
        super().__init__()

        if block is None:
            block = CustomInvertedResidual

        if norm_layer is None:
            norm_layer = nn.BatchNorm2d

        input_channel = 32
        last_channel = 1280

        if inverted_residual_setting is None:
            inverted_residual_setting = [
                # t, c, n, s, k
                [1, 16, 1, 1, 3],
                [6, 24, 2, 2, 3],
                [6, 32, 3, 2, 3],
                [6, 64, 4, 2, 3],
                [6, 96, 3, 1, 3],
                [6, 160, 3, 2, 3],
                [6, 320, 1, 1, 3],
            ]

        # check inverted_residual_setting for validity - t,c,n,s,k are required
        if len(inverted_residual_setting) == 0 or any(len(ir) != 5 for ir in inverted_residual_setting):
            raise ValueError(
                f"inverted_residual_setting should be non-empty or a 5-element list, got {inverted_residual_setting}"
            )

        # building first layer
        input_channel = _make_divisible(input_channel * width_mult, round_nearest)
        self.last_channel = _make_divisible(last_channel * max(1.0, width_mult), round_nearest)
        features: List[nn.Module] = [
            Conv2dNormActivation(3, input_channel, stride=2, norm_layer=norm_layer, activation_layer=nn.ReLU6)
        ]
        # building inverted residual blocks
        for t, c, n, s, k in inverted_residual_setting:
            output_channel = _make_divisible(c * width_mult, round_nearest)
            for i in range(n):
                stride = s if i == 0 else 1
                features.append(
                    block(input_channel, output_channel, stride, expand_ratio=t, kernel=k, norm_layer=norm_layer)
                )
                input_channel = output_channel
        # building last several layers
        features.append(
            Conv2dNormActivation(
                input_channel, self.last_channel, kernel_size=1, norm_layer=norm_layer, activation_layer=nn.ReLU6
            )
        )
        # make it nn.Sequential
        self.features = nn.Sequential(*features)

        # building classifier
        self.classifier = nn.Sequential(
            nn.Dropout(p=dropout),
            nn.Linear(self.last_channel, num_classes),
        )

        # weight initialization
        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                nn.init.kaiming_normal_(m.weight, mode="fan_out")
                if m.bias is not None:
                    nn.init.zeros_(m.bias)
            elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
                nn.init.ones_(m.weight)
                nn.init.zeros_(m.bias)
            elif isinstance(m, nn.Linear):
                nn.init.normal_(m.weight, 0, 0.01)
                nn.init.zeros_(m.bias)

    def _forward_impl(self, x: Tensor) -> Tensor:
        # This exists since TorchScript doesn't support inheritance, so the superclass method
        # (this one) needs to have a name other than `forward` that can be accessed in a subclass
        x = self.features(x)
        # Cannot use "squeeze" as batch-size can be 1
        x = nn.functional.adaptive_avg_pool2d(x, (1, 1))
        x = torch.flatten(x, 1)
        x = self.classifier(x)
        return x

    def forward(self, x: Tensor) -> Tensor:
        return self._forward_impl(x)