def __init__(self, cfg):
super(DeeplabV3Plus, self).__init__()
self.backbone = resnet50_atrous(pretrained=True, os=cfg.OUTPUT_STRIDE)
input_channel = 2048
self.aspp = ASPP(in_chans=input_channel, out_chans=cfg.ASPP_OUTDIM, rate=16//cfg.OUTPUT_STRIDE)
self.dropout1 = nn.Dropout(0.5)
self.upsample4 = nn.UpsamplingBilinear2d(scale_factor=4)
self.upsample_sub = nn.UpsamplingBilinear2d(scale_factor=cfg.OUTPUT_STRIDE//4)
indim = 256
self.shortcut_conv = nn.Sequential(
nn.Conv2d(indim, cfg.SHORTCUT_DIM, cfg.SHORTCUT_KERNEL, 1, padding=cfg.SHORTCUT_KERNEL//2,bias=False),
nn.BatchNorm2d(cfg.SHORTCUT_DIM),
nn.ReLU(inplace=True),
)
self.cat_conv = nn.Sequential(
nn.Conv2d(cfg.ASPP_OUTDIM+cfg.SHORTCUT_DIM, cfg.ASPP_OUTDIM, 3, 1, padding=1,bias=False),
nn.BatchNorm2d(cfg.ASPP_OUTDIM),
nn.ReLU(inplace=True),
nn.Dropout(0.5),
nn.Conv2d(cfg.ASPP_OUTDIM, cfg.ASPP_OUTDIM, 3, 1, padding=1, bias=False),
nn.BatchNorm2d(cfg.ASPP_OUTDIM),
nn.ReLU(inplace=True),
nn.Dropout(0.1),
)
self.cls_conv = nn.Conv2d(cfg.ASPP_OUTDIM, cfg.NUM_CLASSES, 1, 1, padding=0)
for m in self.modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
elif isinstance(m, nn.BatchNorm2d):
nn.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)
def __init__(self, cfg):
super(DeeplabV3Plus, self).__init__()
# 定义DeeplavV3Plus的backbone使用res50_atrous 返回值是outstride=2, 4, 8, 1 时不同layers下的x list
self.backbone = resnet50_atrous(pretrained=True, os=cfg.OUTPUT_STRIDE)
input_channel = 2048
# 定义aspp, 默认参数为：ASPP(2048, 256, 1)
self.aspp = ASPP(in_chans=input_channel,
out_chans=cfg.ASPP_OUTDIM,
rate=16 // cfg.OUTPUT_STRIDE)
# 定义droput
self.dropout1 = nn.Dropout(0.5)
# 定义双线性插值上采样UpsamplingBilinear2d，默认参数为4倍
self.upsample4 = nn.UpsamplingBilinear2d(scale_factor=4)
self.upsample_sub = nn.UpsamplingBilinear2d(
scale_factor=cfg.OUTPUT_STRIDE // 4)
indim = 256
# 定义shortcut_conv(Sequential操作),nn.Conv2d(256, 48, 1, 1, 0, bias=False)+bn+relu
self.shortcut_conv = nn.Sequential(
nn.Conv2d(indim,
cfg.SHORTCUT_DIM,
cfg.SHORTCUT_KERNEL,
1,
padding=cfg.SHORTCUT_KERNEL // 2,
bias=False),
nn.BatchNorm2d(cfg.SHORTCUT_DIM),
nn.ReLU(inplace=True),
)
# 定义cat_conv(Sequential操作),nn.Conv2d(256+48, 256, 3, 1, 1, bias=False)+bn+relu+dropout+...
# nn.Conv2d(256, 256, 3, 1, 1, bias=False)+bn+relu+dropout
self.cat_conv = nn.Sequential(
nn.Conv2d(cfg.ASPP_OUTDIM + cfg.SHORTCUT_DIM,
cfg.ASPP_OUTDIM,
3,
1,
padding=1,
bias=False),
nn.BatchNorm2d(cfg.ASPP_OUTDIM),
nn.ReLU(inplace=True),
nn.Dropout(0.5),
nn.Conv2d(cfg.ASPP_OUTDIM,
cfg.ASPP_OUTDIM,
3,
1,
padding=1,
bias=False),
nn.BatchNorm2d(cfg.ASPP_OUTDIM),
nn.ReLU(inplace=True),
nn.Dropout(0.1),
)
# 定义cls.conv = nn.Conv2d(256, 8, 1, 1, 0)
self.cls_conv = nn.Conv2d(cfg.ASPP_OUTDIM,
cfg.NUM_CLASSES,
1,
1,
padding=0)
# 每层进行kaiming初始化
for m in self.modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight,
mode='fan_out',
nonlinearity='relu')
elif isinstance(m, nn.BatchNorm2d):
nn.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)