resnet_audio.py

# Copyright (c) OpenMMLab. All rights reserved.
import torch.nn as nn
import torch.utils.checkpoint as cp
from mmcv.cnn import ConvModule, constant_init, kaiming_init
from mmcv.runner import load_checkpoint
from torch.nn.modules.batchnorm import _BatchNorm
from torch.nn.modules.utils import _ntuple

from ...utils import get_root_logger
from ..builder import BACKBONES


class Bottleneck2dAudio(nn.Module):
    """Bottleneck2D block for ResNet2D.

    Args:
        inplanes (int): Number of channels for the input in first conv3d layer.
        planes (int): Number of channels produced by some norm/conv3d layers.
        stride (int | tuple[int]): Stride in the conv layer. Default: 1.
        dilation (int): Spacing between kernel elements. Default: 1.
        downsample (nn.Module): Downsample layer. Default: None.
        factorize (bool): Whether to factorize kernel. Default: True.
        norm_cfg (dict):
            Config for norm layers. required keys are `type` and
            `requires_grad`. Default: None.
        with_cp (bool): Use checkpoint or not. Using checkpoint will save some
            memory while slowing down the training speed. Default: False.
    """
    expansion = 4

    def __init__(self,
                 inplanes,
                 planes,
                 stride=2,
                 dilation=1,
                 downsample=None,
                 factorize=True,
                 norm_cfg=None,
                 with_cp=False):
        super().__init__()

        self.inplanes = inplanes
        self.planes = planes
        self.stride = stride
        self.dilation = dilation
        self.factorize = factorize
        self.norm_cfg = norm_cfg
        self.with_cp = with_cp

        self.conv1_stride = 1
        self.conv2_stride = stride

        conv1_kernel_size = (1, 1)
        conv1_padding = 0
        conv2_kernel_size = (3, 3)
        conv2_padding = (dilation, dilation)
        self.conv1 = ConvModule(
            inplanes,
            planes,
            kernel_size=conv1_kernel_size,
            padding=conv1_padding,
            dilation=dilation,
            norm_cfg=self.norm_cfg,
            bias=False)
        self.conv2 = ConvModule(
            planes,
            planes,
            kernel_size=conv2_kernel_size,
            stride=stride,
            padding=conv2_padding,
            dilation=dilation,
            bias=False,
            conv_cfg=dict(type='ConvAudio') if factorize else dict(
                type='Conv'),
            norm_cfg=None,
            act_cfg=None)
        self.conv3 = ConvModule(
            2 * planes if factorize else planes,
            planes * self.expansion,
            kernel_size=1,
            bias=False,
            norm_cfg=self.norm_cfg,
            act_cfg=None)

        self.relu = nn.ReLU(inplace=True)
        self.downsample = downsample

    def forward(self, x):

        def _inner_forward(x):
            identity = x
            out = self.conv1(x)
            out = self.conv2(out)
            out = self.conv3(out)

            if self.downsample is not None:
                identity = self.downsample(x)
            out += identity

            return out

        if self.with_cp and x.requires_grad:
            out = cp.checkpoint(_inner_forward, x)
        else:
            out = _inner_forward(x)

        out = self.relu(out)

        return out


@BACKBONES.register_module()
class ResNetAudio(nn.Module):
    """ResNet 2d audio backbone. Reference:

        <https://arxiv.org/abs/2001.08740>`_.

    Args:
        depth (int): Depth of resnet, from {50, 101, 152}.
        pretrained (str | None): Name of pretrained model.
        in_channels (int): Channel num of input features. Default: 1.
        base_channels (int): Channel num of stem output features. Default: 32.
        num_stages (int): Resnet stages. Default: 4.
        strides (Sequence[int]): Strides of residual blocks of each stage.
            Default: (1, 2, 2, 2).
        dilations (Sequence[int]): Dilation of each stage.
            Default: (1, 1, 1, 1).
        conv1_kernel (int): Kernel size of the first conv layer. Default: 9.
        conv1_stride (int | tuple[int]): Stride of the first conv layer.
            Default: 1.
        frozen_stages (int): Stages to be frozen (all param fixed). -1 means
            not freezing any parameters.
        factorize (Sequence[int]): factorize Dims of each block for audio.
            Default: (1, 1, 0, 0).
        norm_eval (bool): Whether to set BN layers to eval mode, namely, freeze
            running stats (mean and var). Default: False.
        with_cp (bool): Use checkpoint or not. Using checkpoint will save some
            memory while slowing down the training speed. Default: False.
        conv_cfg (dict): Config for norm layers. Default: dict(type='Conv').
        norm_cfg (dict):
            Config for norm layers. required keys are `type` and
            `requires_grad`. Default: dict(type='BN2d', requires_grad=True).
        act_cfg (dict): Config for activate layers.
            Default: dict(type='ReLU', inplace=True).
        zero_init_residual (bool):
            Whether to use zero initialization for residual block,
            Default: True.
    """

    arch_settings = {
        # 18: (BasicBlock2dAudio, (2, 2, 2, 2)),
        # 34: (BasicBlock2dAudio, (3, 4, 6, 3)),
        50: (Bottleneck2dAudio, (3, 4, 6, 3)),
        101: (Bottleneck2dAudio, (3, 4, 23, 3)),
        152: (Bottleneck2dAudio, (3, 8, 36, 3))
    }

    def __init__(self,
                 depth,
                 pretrained,
                 in_channels=1,
                 num_stages=4,
                 base_channels=32,
                 strides=(1, 2, 2, 2),
                 dilations=(1, 1, 1, 1),
                 conv1_kernel=9,
                 conv1_stride=1,
                 frozen_stages=-1,
                 factorize=(1, 1, 0, 0),
                 norm_eval=False,
                 with_cp=False,
                 conv_cfg=dict(type='Conv'),
                 norm_cfg=dict(type='BN2d', requires_grad=True),
                 act_cfg=dict(type='ReLU', inplace=True),
                 zero_init_residual=True):
        super().__init__()
        if depth not in self.arch_settings:
            raise KeyError(f'invalid depth {depth} for resnet')
        self.depth = depth
        self.pretrained = pretrained
        self.in_channels = in_channels
        self.base_channels = base_channels
        self.num_stages = num_stages
        assert 1 <= num_stages <= 4
        self.dilations = dilations
        self.conv1_kernel = conv1_kernel
        self.conv1_stride = conv1_stride
        self.frozen_stages = frozen_stages
        self.stage_factorization = _ntuple(num_stages)(factorize)
        self.norm_eval = norm_eval
        self.with_cp = with_cp
        self.conv_cfg = conv_cfg
        self.norm_cfg = norm_cfg
        self.act_cfg = act_cfg
        self.zero_init_residual = zero_init_residual

        self.block, stage_blocks = self.arch_settings[depth]
        self.stage_blocks = stage_blocks[:num_stages]
        self.inplanes = self.base_channels

        self._make_stem_layer()

        self.res_layers = []
        for i, num_blocks in enumerate(self.stage_blocks):
            stride = strides[i]
            dilation = dilations[i]
            planes = self.base_channels * 2**i
            res_layer = self.make_res_layer(
                self.block,
                self.inplanes,
                planes,
                num_blocks,
                stride=stride,
                dilation=dilation,
                factorize=self.stage_factorization[i],
                norm_cfg=self.norm_cfg,
                with_cp=with_cp)
            self.inplanes = planes * self.block.expansion
            layer_name = f'layer{i + 1}'
            self.add_module(layer_name, res_layer)
            self.res_layers.append(layer_name)

        self.feat_dim = self.block.expansion * self.base_channels * 2**(
            len(self.stage_blocks) - 1)

    @staticmethod
    def make_res_layer(block,
                       inplanes,
                       planes,
                       blocks,
                       stride=1,
                       dilation=1,
                       factorize=1,
                       norm_cfg=None,
                       with_cp=False):
        """Build residual layer for ResNetAudio.

        Args:
            block (nn.Module): Residual module to be built.
            inplanes (int): Number of channels for the input feature
                in each block.
            planes (int): Number of channels for the output feature
                in each block.
            blocks (int): Number of residual blocks.
            stride (Sequence[int]): Strides of residual blocks of each stage.
                Default: (1, 2, 2, 2).
            dilation (int): Spacing between kernel elements. Default: 1.
            factorize (int | Sequence[int]): Determine whether to factorize
                for each block. Default: 1.
            norm_cfg (dict):
                Config for norm layers. required keys are `type` and
                `requires_grad`. Default: None.
            with_cp (bool): Use checkpoint or not. Using checkpoint will save
                some memory while slowing down the training speed.
                Default: False.

        Returns:
            A residual layer for the given config.
        """
        factorize = factorize if not isinstance(
            factorize, int) else (factorize, ) * blocks
        assert len(factorize) == blocks
        downsample = None
        if stride != 1 or inplanes != planes * block.expansion:
            downsample = ConvModule(
                inplanes,
                planes * block.expansion,
                kernel_size=1,
                stride=stride,
                bias=False,
                norm_cfg=norm_cfg,
                act_cfg=None)

        layers = []
        layers.append(
            block(
                inplanes,
                planes,
                stride,
                dilation,
                downsample,
                factorize=(factorize[0] == 1),
                norm_cfg=norm_cfg,
                with_cp=with_cp))
        inplanes = planes * block.expansion
        for i in range(1, blocks):
            layers.append(
                block(
                    inplanes,
                    planes,
                    1,
                    dilation,
                    factorize=(factorize[i] == 1),
                    norm_cfg=norm_cfg,
                    with_cp=with_cp))

        return nn.Sequential(*layers)

    def _make_stem_layer(self):
        """Construct the stem layers consists of a conv+norm+act module and a
        pooling layer."""
        self.conv1 = ConvModule(
            self.in_channels,
            self.base_channels,
            kernel_size=self.conv1_kernel,
            stride=self.conv1_stride,
            bias=False,
            conv_cfg=dict(type='ConvAudio', op='sum'),
            norm_cfg=self.norm_cfg,
            act_cfg=self.act_cfg)

    def _freeze_stages(self):
        """Prevent all the parameters from being optimized before
        ``self.frozen_stages``."""
        if self.frozen_stages >= 0:
            self.conv1.bn.eval()
            for m in [self.conv1.conv, self.conv1.bn]:
                for param in m.parameters():
                    param.requires_grad = False

        for i in range(1, self.frozen_stages + 1):
            m = getattr(self, f'layer{i}')
            m.eval()
            for param in m.parameters():
                param.requires_grad = False

    def init_weights(self):
        """Initiate the parameters either from existing checkpoint or from
        scratch."""
        if isinstance(self.pretrained, str):
            logger = get_root_logger()
            logger.info(f'load model from: {self.pretrained}')

            load_checkpoint(self, self.pretrained, strict=False, logger=logger)

        elif self.pretrained is None:
            for m in self.modules():
                if isinstance(m, nn.Conv2d):
                    kaiming_init(m)
                elif isinstance(m, _BatchNorm):
                    constant_init(m, 1)

            if self.zero_init_residual:
                for m in self.modules():
                    if isinstance(m, Bottleneck2dAudio):
                        constant_init(m.conv3.bn, 0)

        else:
            raise TypeError('pretrained must be a str or None')

    def forward(self, x):
        """Defines the computation performed at every call.

        Args:
            x (torch.Tensor): The input data.

        Returns:
            torch.Tensor: The feature of the input samples extracted
            by the backbone.
        """
        x = self.conv1(x)
        for layer_name in self.res_layers:
            res_layer = getattr(self, layer_name)
            x = res_layer(x)
        return x

    def train(self, mode=True):
        """Set the optimization status when training."""
        super().train(mode)
        self._freeze_stages()
        if mode and self.norm_eval:
            for m in self.modules():
                if isinstance(m, _BatchNorm):
                    m.eval()