Spaces:
Runtime error
Runtime error
File size: 11,883 Bytes
412c852 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 |
# Copyright (c) OpenMMLab. All rights reserved.
import torch
import torch.nn as nn
from mmcv.cnn import ConvModule
from mmengine.model import BaseModule
from mmseg.registry import MODELS
from ..utils import resize
class SpatialPath(BaseModule):
"""Spatial Path to preserve the spatial size of the original input image
and encode affluent spatial information.
Args:
in_channels(int): The number of channels of input
image. Default: 3.
num_channels (Tuple[int]): The number of channels of
each layers in Spatial Path.
Default: (64, 64, 64, 128).
Returns:
x (torch.Tensor): Feature map for Feature Fusion Module.
"""
def __init__(self,
in_channels=3,
num_channels=(64, 64, 64, 128),
conv_cfg=None,
norm_cfg=dict(type='BN'),
act_cfg=dict(type='ReLU'),
init_cfg=None):
super().__init__(init_cfg=init_cfg)
assert len(num_channels) == 4, 'Length of input channels \
of Spatial Path must be 4!'
self.layers = []
for i in range(len(num_channels)):
layer_name = f'layer{i + 1}'
self.layers.append(layer_name)
if i == 0:
self.add_module(
layer_name,
ConvModule(
in_channels=in_channels,
out_channels=num_channels[i],
kernel_size=7,
stride=2,
padding=3,
conv_cfg=conv_cfg,
norm_cfg=norm_cfg,
act_cfg=act_cfg))
elif i == len(num_channels) - 1:
self.add_module(
layer_name,
ConvModule(
in_channels=num_channels[i - 1],
out_channels=num_channels[i],
kernel_size=1,
stride=1,
padding=0,
conv_cfg=conv_cfg,
norm_cfg=norm_cfg,
act_cfg=act_cfg))
else:
self.add_module(
layer_name,
ConvModule(
in_channels=num_channels[i - 1],
out_channels=num_channels[i],
kernel_size=3,
stride=2,
padding=1,
conv_cfg=conv_cfg,
norm_cfg=norm_cfg,
act_cfg=act_cfg))
def forward(self, x):
for i, layer_name in enumerate(self.layers):
layer_stage = getattr(self, layer_name)
x = layer_stage(x)
return x
class AttentionRefinementModule(BaseModule):
"""Attention Refinement Module (ARM) to refine the features of each stage.
Args:
in_channels (int): The number of input channels.
out_channels (int): The number of output channels.
Returns:
x_out (torch.Tensor): Feature map of Attention Refinement Module.
"""
def __init__(self,
in_channels,
out_channel,
conv_cfg=None,
norm_cfg=dict(type='BN'),
act_cfg=dict(type='ReLU'),
init_cfg=None):
super().__init__(init_cfg=init_cfg)
self.conv_layer = ConvModule(
in_channels=in_channels,
out_channels=out_channel,
kernel_size=3,
stride=1,
padding=1,
conv_cfg=conv_cfg,
norm_cfg=norm_cfg,
act_cfg=act_cfg)
self.atten_conv_layer = nn.Sequential(
nn.AdaptiveAvgPool2d((1, 1)),
ConvModule(
in_channels=out_channel,
out_channels=out_channel,
kernel_size=1,
bias=False,
conv_cfg=conv_cfg,
norm_cfg=norm_cfg,
act_cfg=None), nn.Sigmoid())
def forward(self, x):
x = self.conv_layer(x)
x_atten = self.atten_conv_layer(x)
x_out = x * x_atten
return x_out
class ContextPath(BaseModule):
"""Context Path to provide sufficient receptive field.
Args:
backbone_cfg:(dict): Config of backbone of
Context Path.
context_channels (Tuple[int]): The number of channel numbers
of various modules in Context Path.
Default: (128, 256, 512).
align_corners (bool, optional): The align_corners argument of
resize operation. Default: False.
Returns:
x_16_up, x_32_up (torch.Tensor, torch.Tensor): Two feature maps
undergoing upsampling from 1/16 and 1/32 downsampling
feature maps. These two feature maps are used for Feature
Fusion Module and Auxiliary Head.
"""
def __init__(self,
backbone_cfg,
context_channels=(128, 256, 512),
align_corners=False,
conv_cfg=None,
norm_cfg=dict(type='BN'),
act_cfg=dict(type='ReLU'),
init_cfg=None):
super().__init__(init_cfg=init_cfg)
assert len(context_channels) == 3, 'Length of input channels \
of Context Path must be 3!'
self.backbone = MODELS.build(backbone_cfg)
self.align_corners = align_corners
self.arm16 = AttentionRefinementModule(context_channels[1],
context_channels[0])
self.arm32 = AttentionRefinementModule(context_channels[2],
context_channels[0])
self.conv_head32 = ConvModule(
in_channels=context_channels[0],
out_channels=context_channels[0],
kernel_size=3,
stride=1,
padding=1,
conv_cfg=conv_cfg,
norm_cfg=norm_cfg,
act_cfg=act_cfg)
self.conv_head16 = ConvModule(
in_channels=context_channels[0],
out_channels=context_channels[0],
kernel_size=3,
stride=1,
padding=1,
conv_cfg=conv_cfg,
norm_cfg=norm_cfg,
act_cfg=act_cfg)
self.gap_conv = nn.Sequential(
nn.AdaptiveAvgPool2d((1, 1)),
ConvModule(
in_channels=context_channels[2],
out_channels=context_channels[0],
kernel_size=1,
stride=1,
padding=0,
conv_cfg=conv_cfg,
norm_cfg=norm_cfg,
act_cfg=act_cfg))
def forward(self, x):
x_4, x_8, x_16, x_32 = self.backbone(x)
x_gap = self.gap_conv(x_32)
x_32_arm = self.arm32(x_32)
x_32_sum = x_32_arm + x_gap
x_32_up = resize(input=x_32_sum, size=x_16.shape[2:], mode='nearest')
x_32_up = self.conv_head32(x_32_up)
x_16_arm = self.arm16(x_16)
x_16_sum = x_16_arm + x_32_up
x_16_up = resize(input=x_16_sum, size=x_8.shape[2:], mode='nearest')
x_16_up = self.conv_head16(x_16_up)
return x_16_up, x_32_up
class FeatureFusionModule(BaseModule):
"""Feature Fusion Module to fuse low level output feature of Spatial Path
and high level output feature of Context Path.
Args:
in_channels (int): The number of input channels.
out_channels (int): The number of output channels.
Returns:
x_out (torch.Tensor): Feature map of Feature Fusion Module.
"""
def __init__(self,
in_channels,
out_channels,
conv_cfg=None,
norm_cfg=dict(type='BN'),
act_cfg=dict(type='ReLU'),
init_cfg=None):
super().__init__(init_cfg=init_cfg)
self.conv1 = ConvModule(
in_channels=in_channels,
out_channels=out_channels,
kernel_size=1,
stride=1,
padding=0,
conv_cfg=conv_cfg,
norm_cfg=norm_cfg,
act_cfg=act_cfg)
self.gap = nn.AdaptiveAvgPool2d((1, 1))
self.conv_atten = nn.Sequential(
ConvModule(
in_channels=out_channels,
out_channels=out_channels,
kernel_size=1,
stride=1,
padding=0,
bias=False,
conv_cfg=conv_cfg,
norm_cfg=norm_cfg,
act_cfg=act_cfg), nn.Sigmoid())
def forward(self, x_sp, x_cp):
x_concat = torch.cat([x_sp, x_cp], dim=1)
x_fuse = self.conv1(x_concat)
x_atten = self.gap(x_fuse)
# Note: No BN and more 1x1 conv in paper.
x_atten = self.conv_atten(x_atten)
x_atten = x_fuse * x_atten
x_out = x_atten + x_fuse
return x_out
@MODELS.register_module()
class BiSeNetV1(BaseModule):
"""BiSeNetV1 backbone.
This backbone is the implementation of `BiSeNet: Bilateral
Segmentation Network for Real-time Semantic
Segmentation <https://arxiv.org/abs/1808.00897>`_.
Args:
backbone_cfg:(dict): Config of backbone of
Context Path.
in_channels (int): The number of channels of input
image. Default: 3.
spatial_channels (Tuple[int]): Size of channel numbers of
various layers in Spatial Path.
Default: (64, 64, 64, 128).
context_channels (Tuple[int]): Size of channel numbers of
various modules in Context Path.
Default: (128, 256, 512).
out_indices (Tuple[int] | int, optional): Output from which stages.
Default: (0, 1, 2).
align_corners (bool, optional): The align_corners argument of
resize operation in Bilateral Guided Aggregation Layer.
Default: False.
out_channels(int): The number of channels of output.
It must be the same with `in_channels` of decode_head.
Default: 256.
"""
def __init__(self,
backbone_cfg,
in_channels=3,
spatial_channels=(64, 64, 64, 128),
context_channels=(128, 256, 512),
out_indices=(0, 1, 2),
align_corners=False,
out_channels=256,
conv_cfg=None,
norm_cfg=dict(type='BN', requires_grad=True),
act_cfg=dict(type='ReLU'),
init_cfg=None):
super().__init__(init_cfg=init_cfg)
assert len(spatial_channels) == 4, 'Length of input channels \
of Spatial Path must be 4!'
assert len(context_channels) == 3, 'Length of input channels \
of Context Path must be 3!'
self.out_indices = out_indices
self.align_corners = align_corners
self.context_path = ContextPath(backbone_cfg, context_channels,
self.align_corners)
self.spatial_path = SpatialPath(in_channels, spatial_channels)
self.ffm = FeatureFusionModule(context_channels[1], out_channels)
self.conv_cfg = conv_cfg
self.norm_cfg = norm_cfg
self.act_cfg = act_cfg
def forward(self, x):
# stole refactoring code from Coin Cheung, thanks
x_context8, x_context16 = self.context_path(x)
x_spatial = self.spatial_path(x)
x_fuse = self.ffm(x_spatial, x_context8)
outs = [x_fuse, x_context8, x_context16]
outs = [outs[i] for i in self.out_indices]
return tuple(outs)
|