def __init__(self,cfg,feature_extracter,dynamic_seghead=DynamicSegHead(),freeze_bn=False):
super(IntVOS,self).__init__()
self.feature_extracter=feature_extracter
self.feature_extracter.cls_conv=nn.Sequential()
self.feature_extracter.upsample4=nn.Sequential()
self.semantic_embedding=None
self.seperate_conv=nn.Conv2d(cfg.MODEL_ASPP_OUTDIM,cfg.MODEL_ASPP_OUTDIM,kernel_size=3,stride=1,padding=1,groups=cfg.MODEL_ASPP_OUTDIM)
self.bn1 = SynchronizedBatchNorm2d(cfg.MODEL_ASPP_OUTDIM, momentum=cfg.TRAIN_BN_MOM)
self.relu1=nn.ReLU(True)
self.embedding_conv = nn.Conv2d(cfg.MODEL_ASPP_OUTDIM,cfg.MODEL_SEMANTIC_EMBEDDING_DIM,1,1)
self.relu2=nn.ReLU(True)
self.bn2 = SynchronizedBatchNorm2d(cfg.MODEL_SEMANTIC_EMBEDDING_DIM, momentum=cfg.TRAIN_BN_MOM)
self.semantic_embedding=nn.Sequential(*[self.seperate_conv,self.bn1,self.relu1,self.embedding_conv,self.bn2,self.relu2])
# self.semantic_embedding=nn.Sequential(*[self.seperate_conv,self.relu1,self.embedding_conv,self.relu2])
for m in self.semantic_embedding:
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
#########################
self.dynamic_seghead = DynamicSegHead()
if cfg.MODEL_GLOBAL_ATTEN:
self.global_atten=Global_Atten()
if cfg.INTER_USE_GLOBAL:
self.inter_seghead = IntSegHead2(in_dim=cfg.MODEL_SEMANTIC_EMBEDDING_DIM+4)
else:
self.inter_seghead = IntSegHead2(in_dim=cfg.MODEL_SEMANTIC_EMBEDDING_DIM+3)
if freeze_bn:
self.freeze_bn()
def __init__(self,in_dim,out_dim):
super(_res_block,self).__init__()
self.conv1=nn.Conv2d(in_dim,out_dim,kernel_size=3,stride=1,padding=1)
self.relu1=nn.ReLU()
self.bn1 = SynchronizedBatchNorm2d(out_dim, momentum=cfg.TRAIN_BN_MOM)
self.conv2=nn.Conv2d(out_dim,out_dim,kernel_size=3,stride=1,padding=1)
self.relu2=nn.ReLU()
self.bn2 = SynchronizedBatchNorm2d(out_dim, momentum=cfg.TRAIN_BN_MOM)
def __init__(self,in_dim,out_dim,kernel_size=7):
super(_split_separable_conv2d,self).__init__()
self.conv1=nn.Conv2d(in_dim,in_dim,kernel_size=kernel_size,stride=1,padding=int((kernel_size-1)/2),groups=in_dim)
self.relu1=nn.ReLU(True)
self.bn1 = SynchronizedBatchNorm2d(in_dim, momentum=cfg.TRAIN_BN_MOM)
self.conv2=nn.Conv2d(in_dim,out_dim,kernel_size=1,stride=1)
self.relu2=nn.ReLU(True)
self.bn2 = SynchronizedBatchNorm2d(out_dim, momentum=cfg.TRAIN_BN_MOM)
nn.init.kaiming_normal_(self.conv1.weight,mode='fan_out',nonlinearity='relu')
nn.init.kaiming_normal_(self.conv2.weight,mode='fan_out',nonlinearity='relu')
def __init__(self,in_dim=(cfg.MODEL_SEMANTIC_EMBEDDING_DIM+3),emb_dim=cfg.MODEL_HEAD_EMBEDDING_DIM):
super(IntSegHead,self).__init__()
self.conv1 = nn.Conv2d(in_dim,emb_dim,kernel_size=7,stride=1,padding=3)
self.bn1 = SynchronizedBatchNorm2d(emb_dim, momentum=cfg.TRAIN_BN_MOM)
self.relu1 = nn.ReLU()
self.res1 = _res_block(emb_dim,emb_dim)
self.res2 = _res_block(emb_dim,emb_dim)
# self.aspp = ASPP(emb_dim,SynchronizedBatchNorm2d)
self.conv2 = nn.Conv2d(256,emb_dim,kernel_size=3,stride=1,padding=1)
self.bn2 = SynchronizedBatchNorm2d(emb_dim, momentum=cfg.TRAIN_BN_MOM)
self.relu2 = nn.ReLU()
self.conv3 = nn.Conv2d(emb_dim,1,1,1)
def __init__(self,in_dim,out_dim=256):
super(InterFuse,self).__init__()
self.conv1 = nn.Conv2d(cfg.MODEL_SEMANTIC_EMBEDDING_DIM,256,1,1)
self.conv2 = nn.Conv2d(256,256,1,1)
self.conv3 = nn.Conv2d(in_dim,256,1,1)
self.conv4 = nn.Conv2d(256,out_dim,1,1)
self.relu = nn.ReLU()
self.sigmoid = nn.Sigmoid()
self.bn1 = SynchronizedBatchNorm2d(256, momentum=cfg.TRAIN_BN_MOM)
self.bn2 = SynchronizedBatchNorm2d(256, momentum=cfg.TRAIN_BN_MOM)
nn.init.kaiming_normal_(self.conv1.weight,mode='fan_out',nonlinearity='relu')
nn.init.kaiming_normal_(self.conv2.weight,mode='fan_out',nonlinearity='relu')
nn.init.kaiming_normal_(self.conv3.weight,mode='fan_out',nonlinearity='relu')
nn.init.kaiming_normal_(self.conv4.weight,mode='fan_out',nonlinearity='relu')
def __init__(self,in_dim=(cfg.MODEL_SEMANTIC_EMBEDDING_DIM+1),out_dim=cfg.MODEL_HEAD_EMBEDDING_DIM,kernel_size=7):
super(Global_Atten,self).__init__()
self.conv1 = nn.Conv2d(in_dim,out_dim,kernel_size=kernel_size,stride=1,padding=int((kernel_size-1)/2))
self.relu1=nn.ReLU(True)
self.bn1 = SynchronizedBatchNorm2d(out_dim, momentum=cfg.TRAIN_BN_MOM)
self.conv2 = nn.Conv2d(out_dim,1,kernel_size=1,stride=1)
nn.init.kaiming_normal_(self.conv1.weight,mode='fan_out',nonlinearity='relu')
nn.init.kaiming_normal_(self.conv2.weight,mode='fan_out',nonlinearity='relu')
def __init__(self, cfg):
super(deeplabv3plus, self).__init__()
self.cfg=cfg
self.backbone = None
self.backbone_layers = None
input_channel = 2048
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)
self.upsample4 = nn.functional.interpolate
self.upsample_sub=nn.functional.interpolate
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()
def __init__(self, in_fea=[256, 512, 1024], mid_fea=256, out_fea=1):
super(Edge_Module, self).__init__()
self.conv1 = nn.Sequential(
nn.Conv2d(in_fea[0],
mid_fea,
kernel_size=1,
padding=0,
dilation=1,
bias=False),
SynchronizedBatchNorm2d(mid_fea, momentum=cfg.TRAIN_BN_MOM))
self.conv2 = nn.Sequential(
nn.Conv2d(in_fea[1],
mid_fea,
kernel_size=1,
padding=0,
dilation=1,
bias=False),
SynchronizedBatchNorm2d(mid_fea, momentum=cfg.TRAIN_BN_MOM))
self.conv3 = nn.Sequential(
nn.Conv2d(in_fea[2],
mid_fea,
kernel_size=1,
padding=0,
dilation=1,
bias=False),
SynchronizedBatchNorm2d(mid_fea, momentum=cfg.TRAIN_BN_MOM))
self.conv4 = nn.Conv2d(mid_fea,
out_fea,
kernel_size=3,
padding=1,
dilation=1,
bias=True)
self.conv5 = nn.Conv2d(out_fea * 3,
out_fea,
kernel_size=1,
padding=0,
dilation=1,
bias=True)
def __init__(self, cfg, feature_extracter):
super(IntVOS, self).__init__()
self.feature_extracter = feature_extracter ##embedding extractor
self.feature_extracter.cls_conv = nn.Sequential()
self.feature_extracter.upsample4 = nn.Sequential()
self.semantic_embedding = None
self.seperate_conv = nn.Conv2d(cfg.MODEL_ASPP_OUTDIM,
cfg.MODEL_ASPP_OUTDIM,
kernel_size=3,
stride=1,
padding=1,
groups=cfg.MODEL_ASPP_OUTDIM)
self.bn1 = SynchronizedBatchNorm2d(cfg.MODEL_ASPP_OUTDIM,
momentum=cfg.TRAIN_BN_MOM)
self.relu1 = nn.ReLU(True)
self.embedding_conv = nn.Conv2d(cfg.MODEL_ASPP_OUTDIM,
cfg.MODEL_SEMANTIC_EMBEDDING_DIM, 1, 1)
self.relu2 = nn.ReLU(True)
self.bn2 = SynchronizedBatchNorm2d(cfg.MODEL_SEMANTIC_EMBEDDING_DIM,
momentum=cfg.TRAIN_BN_MOM)
self.semantic_embedding = nn.Sequential(*[
self.seperate_conv, self.bn1, self.relu1, self.embedding_conv,
self.bn2, self.relu2
])
for m in self.semantic_embedding:
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight,
mode='fan_out',
nonlinearity='relu')
self.dynamic_seghead = DynamicSegHead() #propagation segm head
if cfg.MODEL_USEIntSeg:
self.inter_seghead = IntSegHead(
in_dim=cfg.MODEL_SEMANTIC_EMBEDDING_DIM + 3)
else:
self.inter_seghead = DynamicSegHead(
in_dim=cfg.MODEL_SEMANTIC_EMBEDDING_DIM +
2) #interaction segm head
