def __init__(self, args):
super(deeplabv3plus_res18, self).__init__()
self.backbone = None
self.backbone_layers = None
input_channel = 256
args.TRAIN_BN_MOM = 0.1
self.aspp = ASPP(dim_in=input_channel,
dim_out=args.MODEL_ASPP_OUTDIM,
rate=16 // args.MODEL_OUTPUT_STRIDE,
bn_mom=args.TRAIN_BN_MOM
)
self.dropout1 = nn.Dropout(0.5)
self.upsample4 = nn.UpsamplingBilinear2d(scale_factor=4)
self.upsample_sub = nn.UpsamplingBilinear2d(scale_factor=args.MODEL_OUTPUT_STRIDE // 4)
indim = 32
self.shortcut_conv = nn.Sequential(
nn.Conv2d(indim, args.MODEL_SHORTCUT_DIM, args.MODEL_SHORTCUT_KERNEL, 1,
padding=args.MODEL_SHORTCUT_KERNEL // 2, bias=True),
SynchronizedBatchNorm2d(args.MODEL_SHORTCUT_DIM, momentum=args.TRAIN_BN_MOM),
nn.ReLU(inplace=True),
)
self.cat_conv = nn.Sequential(
nn.Conv2d(args.MODEL_ASPP_OUTDIM + args.MODEL_SHORTCUT_DIM, args.MODEL_ASPP_OUTDIM, 3, 1, padding=1,
bias=True),
SynchronizedBatchNorm2d(args.MODEL_ASPP_OUTDIM, momentum=args.TRAIN_BN_MOM),
nn.ReLU(inplace=True),
nn.Dropout(0.5),
nn.Conv2d(args.MODEL_ASPP_OUTDIM, args.MODEL_ASPP_OUTDIM, 3, 1, padding=1, bias=True),
SynchronizedBatchNorm2d(args.MODEL_ASPP_OUTDIM, momentum=args.TRAIN_BN_MOM),
nn.ReLU(inplace=True),
nn.Dropout(0.1),
)
self.cls_conv = nn.Conv2d(args.MODEL_ASPP_OUTDIM, args.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, SynchronizedBatchNorm2d):
nn.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)
self.backbone = build_backbone(args.MODEL_BACKBONE, os=args.MODEL_OUTPUT_STRIDE)
self.backbone_layers = self.backbone.get_layers()
self.upsample2 = nn.UpsamplingBilinear2d(scale_factor=2)
def __init__(self, cfg):
super(deeplabv3plus_fpn, self).__init__()
self.backbone = None
self.backbone_layers = None
input_channel = 2048
cfg.TRAIN_BN_MOM = 0.1
self.aspp = ASPP(dim_in=input_channel,
dim_out=cfg.MODEL_ASPP_OUTDIM,
rate=16 // cfg.MODEL_OUTPUT_STRIDE,
bn_mom=cfg.TRAIN_BN_MOM)
self.dropout1 = nn.Dropout(0.5)
self.upsample4 = nn.UpsamplingBilinear2d(scale_factor=4)
self.upsample_sub = nn.UpsamplingBilinear2d(
scale_factor=cfg.MODEL_OUTPUT_STRIDE // 4)
indim = 256
self.shortcut_conv = nn.Sequential(
nn.Conv2d(indim,
cfg.MODEL_SHORTCUT_DIM,
cfg.MODEL_SHORTCUT_KERNEL,
1,
padding=cfg.MODEL_SHORTCUT_KERNEL // 2,
bias=True),
SynchronizedBatchNorm2d(cfg.MODEL_SHORTCUT_DIM,
momentum=cfg.TRAIN_BN_MOM),
nn.ReLU(inplace=True),
)
self.cat_conv = nn.Sequential(
nn.Conv2d(cfg.MODEL_ASPP_OUTDIM + cfg.MODEL_SHORTCUT_DIM,
cfg.MODEL_ASPP_OUTDIM,
3,
1,
padding=1,
bias=True),
SynchronizedBatchNorm2d(cfg.MODEL_ASPP_OUTDIM,
momentum=cfg.TRAIN_BN_MOM),
nn.ReLU(inplace=True),
nn.Dropout(0.5),
nn.Conv2d(cfg.MODEL_ASPP_OUTDIM,
cfg.MODEL_ASPP_OUTDIM,
3,
1,
padding=1,
bias=True),
SynchronizedBatchNorm2d(cfg.MODEL_ASPP_OUTDIM,
momentum=cfg.TRAIN_BN_MOM),
nn.ReLU(inplace=True),
nn.Dropout(0.1),
)
self.cls_conv = nn.Conv2d(cfg.MODEL_ASPP_OUTDIM,
cfg.MODEL_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, SynchronizedBatchNorm2d):
nn.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)
self.backbone = build_backbone(cfg.MODEL_BACKBONE,
os=cfg.MODEL_OUTPUT_STRIDE)
self.backbone_layers = self.backbone.get_layers()
# ----------- fpn 结构-----------------
in_layer0 = 256 # 1/4
in_layer1 = 512 # 1/8
in_layer2 = 1024 # 1/16
in_layer3 = 2048 # 1/16
self.feat3_conv = nn.Conv2d(in_layer3, in_layer3, 1, 1, padding=0)
self.feat3_convS2 = nn.Conv2d(in_layer1, in_layer3, 3, 2, padding=1)
self.feat2_conv = nn.Conv2d(in_layer1, in_layer1, 1, 1, padding=0)
self.feat2_deconv = nn.ConvTranspose2d(in_layer3,
in_layer1,
2,
2,
padding=0)
self.feat1_conv = nn.Conv2d(in_layer0, in_layer0, 1, 1, padding=0)
self.feat1_deconv = nn.ConvTranspose2d(in_layer1,
in_layer0,
2,
2,
padding=0)
def __init__(self, cfg):
super(deeplabv3plus_mutilAspp, self).__init__()
self.backbone = None
self.backbone_layers = None
input_channel = 2048
cfg.TRAIN_BN_MOM = 0.1
self.aspp = ASPP(
dim_in=input_channel,
dim_out=cfg.MODEL_ASPP_OUTDIM, # 256
rate=16 // cfg.MODEL_OUTPUT_STRIDE,
bn_mom=cfg.TRAIN_BN_MOM)
input_channel_layer1 = 512
# 用于1/8特征图做aspp
self.aspp_layer1 = ASPP_size4(
dim_in=input_channel_layer1,
dim_out=cfg.MODEL_ASPP_OUTDIM, # 256
rate=16 // cfg.MODEL_OUTPUT_STRIDE,
bn_mom=cfg.TRAIN_BN_MOM)
input_channel_layer0 = 256
# 用于1/4特征图做aspp
self.aspp_layer0 = ASPP_size3(
dim_in=input_channel_layer0,
dim_out=cfg.MODEL_ASPP_OUTDIM, # 256
rate=16 // cfg.MODEL_OUTPUT_STRIDE,
bn_mom=cfg.TRAIN_BN_MOM)
self.dropout1 = nn.Dropout(0.5)
self.upsample4 = nn.UpsamplingBilinear2d(scale_factor=4)
self.upsample_sub = nn.UpsamplingBilinear2d(
scale_factor=cfg.MODEL_OUTPUT_STRIDE // 4)
self.upsample_sub_layer1 = nn.UpsamplingBilinear2d(
scale_factor=cfg.MODEL_OUTPUT_STRIDE // 8)
indim = 256
self.shortcut_conv = nn.Sequential(
nn.Conv2d(indim,
cfg.MODEL_SHORTCUT_DIM,
cfg.MODEL_SHORTCUT_KERNEL,
1,
padding=cfg.MODEL_SHORTCUT_KERNEL // 2,
bias=True),
SynchronizedBatchNorm2d(cfg.MODEL_SHORTCUT_DIM,
momentum=cfg.TRAIN_BN_MOM),
nn.ReLU(inplace=True),
)
self.cat_conv = nn.Sequential(
nn.Conv2d(cfg.MODEL_ASPP_OUTDIM * 2 + cfg.MODEL_SHORTCUT_DIM,
cfg.MODEL_ASPP_OUTDIM,
3,
1,
padding=1,
bias=True),
SynchronizedBatchNorm2d(cfg.MODEL_ASPP_OUTDIM,
momentum=cfg.TRAIN_BN_MOM),
nn.ReLU(inplace=True),
nn.Dropout(0.5),
nn.Conv2d(cfg.MODEL_ASPP_OUTDIM,
cfg.MODEL_ASPP_OUTDIM,
3,
1,
padding=1,
bias=True),
SynchronizedBatchNorm2d(cfg.MODEL_ASPP_OUTDIM,
momentum=cfg.TRAIN_BN_MOM),
nn.ReLU(inplace=True),
nn.Dropout(0.1),
)
self.cls_conv = nn.Conv2d(cfg.MODEL_ASPP_OUTDIM,
cfg.MODEL_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, SynchronizedBatchNorm2d):
nn.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)
self.backbone = build_backbone(cfg.MODEL_BACKBONE,
os=cfg.MODEL_OUTPUT_STRIDE)
self.backbone_layers = self.backbone.get_layers()
def __init__(self, cfg):
super(deeplabv3plus_baseOc, self).__init__()
self.backbone = None
self.backbone_layers = None
backbone_out = 2048
oc_out = 1024
'''
layers torch.Size([2, 2048, 32, 32])
feature_aspp torch.Size([2, 256, 32, 32])
'''
self.context = nn.Sequential(
nn.Conv2d(backbone_out, oc_out, kernel_size=3, stride=1,
padding=1), nn.BatchNorm2d(oc_out), nn.ReLU(),
BaseOC_Module(in_channels=oc_out,
out_channels=oc_out,
key_channels=oc_out // 2,
value_channels=oc_out - oc_out // 2,
dropout=0.05,
sizes=([1])))
cfg.TRAIN_BN_MOM = 0.1
self.aspp = ASPP(dim_in=oc_out,
dim_out=cfg.MODEL_ASPP_OUTDIM,
rate=16 // cfg.MODEL_OUTPUT_STRIDE,
bn_mom=cfg.TRAIN_BN_MOM)
self.dropout1 = nn.Dropout(0.5)
self.upsample4 = nn.UpsamplingBilinear2d(scale_factor=4)
self.upsample_sub = nn.UpsamplingBilinear2d(
scale_factor=cfg.MODEL_OUTPUT_STRIDE // 4)
indim = 256
self.shortcut_conv = nn.Sequential(
nn.Conv2d(indim,
cfg.MODEL_SHORTCUT_DIM,
cfg.MODEL_SHORTCUT_KERNEL,
1,
padding=cfg.MODEL_SHORTCUT_KERNEL // 2,
bias=True),
SynchronizedBatchNorm2d(cfg.MODEL_SHORTCUT_DIM,
momentum=cfg.TRAIN_BN_MOM),
nn.ReLU(inplace=True),
)
self.cat_conv = nn.Sequential(
nn.Conv2d(cfg.MODEL_ASPP_OUTDIM + cfg.MODEL_SHORTCUT_DIM,
cfg.MODEL_ASPP_OUTDIM,
3,
1,
padding=1,
bias=True),
SynchronizedBatchNorm2d(cfg.MODEL_ASPP_OUTDIM,
momentum=cfg.TRAIN_BN_MOM),
nn.ReLU(inplace=True),
nn.Dropout(0.5),
nn.Conv2d(cfg.MODEL_ASPP_OUTDIM,
cfg.MODEL_ASPP_OUTDIM,
3,
1,
padding=1,
bias=True),
SynchronizedBatchNorm2d(cfg.MODEL_ASPP_OUTDIM,
momentum=cfg.TRAIN_BN_MOM),
nn.ReLU(inplace=True),
nn.Dropout(0.1),
)
self.cls_conv = nn.Conv2d(cfg.MODEL_ASPP_OUTDIM,
cfg.MODEL_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, SynchronizedBatchNorm2d):
nn.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)
self.backbone = build_backbone(cfg.MODEL_BACKBONE,
os=cfg.MODEL_OUTPUT_STRIDE)
self.backbone_layers = self.backbone.get_layers()