File size: 6,750 Bytes
3094730
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
# 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)