mmyolo.models.necks.cspnext_pafpn 源代码

# Copyright (c) OpenMMLab. All rights reserved.
import math
from typing import Sequence

import torch.nn as nn
from mmcv.cnn import ConvModule, DepthwiseSeparableConvModule
from mmdet.models.backbones.csp_darknet import CSPLayer
from mmdet.utils import ConfigType, OptMultiConfig

from mmyolo.registry import MODELS
from .base_yolo_neck import BaseYOLONeck


@MODELS.register_module()
class CSPNeXtPAFPN(BaseYOLONeck):
    """Path Aggregation Network with CSPNeXt blocks.

    Args:
        in_channels (Sequence[int]): Number of input channels per scale.
        out_channels (int): Number of output channels (used at each scale)
        deepen_factor (float): Depth multiplier, multiply number of
            blocks in CSP layer by this amount. Defaults to 1.0.
        widen_factor (float): Width multiplier, multiply number of
            channels in each layer by this amount. Defaults to 1.0.
        num_csp_blocks (int): Number of bottlenecks in CSPLayer.
            Defaults to 3.
        use_depthwise (bool): Whether to use depthwise separable convolution in
            blocks. Defaults to False.
        expand_ratio (float): Ratio to adjust the number of channels of the
            hidden layer. Defaults to 0.5.
        upsample_cfg (dict): Config dict for interpolate layer.
            Default: `dict(scale_factor=2, mode='nearest')`
        conv_cfg (dict, optional): Config dict for convolution layer.
            Default: None, which means using conv2d.
        norm_cfg (dict): Config dict for normalization layer.
            Default: dict(type='BN')
        act_cfg (dict): Config dict for activation layer.
            Default: dict(type='SiLU', inplace=True)
        init_cfg (dict or list[dict], optional): Initialization config dict.
            Default: None.
    """

    def __init__(
        self,
        in_channels: Sequence[int],
        out_channels: int,
        deepen_factor: float = 1.0,
        widen_factor: float = 1.0,
        num_csp_blocks: int = 3,
        freeze_all: bool = False,
        use_depthwise: bool = False,
        expand_ratio: float = 0.5,
        upsample_cfg: ConfigType = dict(scale_factor=2, mode='nearest'),
        conv_cfg: bool = None,
        norm_cfg: ConfigType = dict(type='BN'),
        act_cfg: ConfigType = dict(type='SiLU', inplace=True),
        init_cfg: OptMultiConfig = dict(
            type='Kaiming',
            layer='Conv2d',
            a=math.sqrt(5),
            distribution='uniform',
            mode='fan_in',
            nonlinearity='leaky_relu')
    ) -> None:
        self.num_csp_blocks = round(num_csp_blocks * deepen_factor)
        self.conv = DepthwiseSeparableConvModule \
            if use_depthwise else ConvModule
        self.upsample_cfg = upsample_cfg
        self.expand_ratio = expand_ratio
        self.conv_cfg = conv_cfg

        super().__init__(
            in_channels=[
                int(channel * widen_factor) for channel in in_channels
            ],
            out_channels=int(out_channels * widen_factor),
            deepen_factor=deepen_factor,
            widen_factor=widen_factor,
            freeze_all=freeze_all,
            norm_cfg=norm_cfg,
            act_cfg=act_cfg,
            init_cfg=init_cfg)

    def build_reduce_layer(self, idx: int) -> nn.Module:
        """build reduce layer.

        Args:
            idx (int): layer idx.

        Returns:
            nn.Module: The reduce layer.
        """
        if idx == len(self.in_channels) - 1:
            layer = self.conv(
                self.in_channels[idx],
                self.in_channels[idx - 1],
                1,
                norm_cfg=self.norm_cfg,
                act_cfg=self.act_cfg)
        else:
            layer = nn.Identity()

        return layer

    def build_upsample_layer(self, *args, **kwargs) -> nn.Module:
        """build upsample layer."""
        return nn.Upsample(**self.upsample_cfg)

    def build_top_down_layer(self, idx: int) -> nn.Module:
        """build top down layer.

        Args:
            idx (int): layer idx.

        Returns:
            nn.Module: The top down layer.
        """
        if idx == 1:
            return CSPLayer(
                self.in_channels[idx - 1] * 2,
                self.in_channels[idx - 1],
                num_blocks=self.num_csp_blocks,
                add_identity=False,
                use_cspnext_block=True,
                expand_ratio=self.expand_ratio,
                conv_cfg=self.conv_cfg,
                norm_cfg=self.norm_cfg,
                act_cfg=self.act_cfg)
        else:
            return nn.Sequential(
                CSPLayer(
                    self.in_channels[idx - 1] * 2,
                    self.in_channels[idx - 1],
                    num_blocks=self.num_csp_blocks,
                    add_identity=False,
                    use_cspnext_block=True,
                    expand_ratio=self.expand_ratio,
                    conv_cfg=self.conv_cfg,
                    norm_cfg=self.norm_cfg,
                    act_cfg=self.act_cfg),
                self.conv(
                    self.in_channels[idx - 1],
                    self.in_channels[idx - 2],
                    kernel_size=1,
                    norm_cfg=self.norm_cfg,
                    act_cfg=self.act_cfg))

    def build_downsample_layer(self, idx: int) -> nn.Module:
        """build downsample layer.

        Args:
            idx (int): layer idx.

        Returns:
            nn.Module: The downsample layer.
        """
        return self.conv(
            self.in_channels[idx],
            self.in_channels[idx],
            kernel_size=3,
            stride=2,
            padding=1,
            norm_cfg=self.norm_cfg,
            act_cfg=self.act_cfg)

    def build_bottom_up_layer(self, idx: int) -> nn.Module:
        """build bottom up layer.

        Args:
            idx (int): layer idx.

        Returns:
            nn.Module: The bottom up layer.
        """
        return CSPLayer(
            self.in_channels[idx] * 2,
            self.in_channels[idx + 1],
            num_blocks=self.num_csp_blocks,
            add_identity=False,
            use_cspnext_block=True,
            expand_ratio=self.expand_ratio,
            conv_cfg=self.conv_cfg,
            norm_cfg=self.norm_cfg,
            act_cfg=self.act_cfg)

    def build_out_layer(self, idx: int) -> nn.Module:
        """build out layer.

        Args:
            idx (int): layer idx.

        Returns:
            nn.Module: The out layer.
        """
        return self.conv(
            self.in_channels[idx],
            self.out_channels,
            3,
            padding=1,
            norm_cfg=self.norm_cfg,
            act_cfg=self.act_cfg)