def __init__(self,
backbone='resnet',
output_stride=16,
num_classes=19,
sync_bn=True,
freeze_bn=False,
args=None,
separate=False):
super(DeepLab, self).__init__()
if backbone == 'drn':
output_stride = 8
if sync_bn == True:
BatchNorm = SynchronizedBatchNorm2d
else:
BatchNorm = nn.BatchNorm2d
self.backbone = build_backbone(backbone, output_stride, BatchNorm,
args)
self.aspp = build_aspp(backbone, output_stride, BatchNorm, args,
separate)
self.decoder = build_decoder(num_classes, backbone, BatchNorm, args,
separate)
if freeze_bn:
self.freeze_bn()
def __init__(self, backbone='resnet', output_stride=16, num_classes=21,
sync_bn=False, freeze_bn=False):
super(DeepLab, self).__init__()
if backbone == 'drn':
output_stride = 8
if backbone == 'resnet':
link_in = 1024
link_out = 1024
elif backbone == 'mobilenet':
link_in = 64
link_out = 64
if sync_bn == True:
BatchNorm = SynchronizedBatchNorm2d
else:
BatchNorm = nn.BatchNorm2d
self.backbone = build_backbone(backbone, output_stride, BatchNorm)
self.aspp = build_aspp(backbone, output_stride, BatchNorm)
self.link_conv = nn.Sequential(nn.Conv2d(link_in, link_out, kernel_size=1, stride=1, padding=0, bias=False))
self.last_conv = nn.Sequential(nn.Conv2d(64, 64, kernel_size=3, stride=1, padding=1, bias=False),
BatchNorm(64),
nn.ReLU(),
nn.Dropout(0.1),
nn.Conv2d(64, num_classes, kernel_size=1, stride=1))
self._init_weight()
if freeze_bn:
self.freeze_bn()
def __init__(self,
args,
backbone='resnet',
output_stride=16,
num_classes=21,
sync_bn=True,
freeze_bn=False):
super(DeepLab, self).__init__()
self.args = args
if backbone == 'drn':
output_stride = 8
if sync_bn == True:
BatchNorm = SynchronizedBatchNorm2d
else:
BatchNorm = nn.BatchNorm2d
self.backbone = build_backbone(backbone, output_stride, BatchNorm)
self.aspp = build_aspp(backbone, output_stride, BatchNorm)
if self.args.use_kinematic == False:
self.decoder = build_decoder(num_classes, backbone, BatchNorm)
else:
self.decoder = build_decoder_kinematic(backbone, BatchNorm)
self.kinematic_layer = build_kinematic_graph(BatchNorm)
self.freeze_bn = freeze_bn
def __init__(self,
backbone='resnet101',
output_stride=16,
num_classes=21,
sync_bn=True,
freeze_bn=False,
enable_interpolation=True,
pretrained_path=None,
norm_layer=nn.BatchNorm2d,
enable_aspp=True):
super(DeepLab, self).__init__()
self.enable_aspp = enable_aspp
if backbone == 'drn':
output_stride = 8
BatchNorm = norm_layer
self.backbone = build_backbone(backbone,
output_stride,
BatchNorm,
pretrained_path=pretrained_path)
self.aspp = build_aspp(backbone,
output_stride,
BatchNorm,
enable_aspp=self.enable_aspp)
self.decoder = build_decoder(num_classes, backbone, BatchNorm)
self.enable_interpolation = enable_interpolation
def __init__(self,
backbone='resnet',
output_stride=16,
num_classes=19,
use_ABN=True,
freeze_bn=False,
args=None,
separate=False):
super(DeepLab, self).__init__()
if backbone == 'drn':
output_stride = 8
if use_ABN:
BatchNorm = ABN
else:
BatchNorm = NaiveBN
self.backbone = build_backbone(backbone, output_stride, BatchNorm,
args)
self.aspp = build_aspp(backbone, output_stride, BatchNorm, args,
separate)
self.decoder = build_decoder(num_classes, backbone, BatchNorm, args,
separate)
if freeze_bn:
self.freeze_bn()
def __init__(self, backbone='resnet', output_stride=16, num_classes=21,
sync_bn=True, freeze_bn=False):
super(DeepLab, self).__init__()
if backbone == 'drn':
output_stride = 8
if sync_bn == True:
BatchNorm = SynchronizedBatchNorm2d
else:
BatchNorm = nn.BatchNorm2d
self.backbone = build_backbone(backbone, output_stride, BatchNorm)
self.aspp = build_aspp(backbone, output_stride, BatchNorm)
self.decoder = build_decoder(num_classes, backbone, BatchNorm)
# self.last_conv = nn.Sequential(nn.Conv2d(256, 256, kernel_size=3, stride=1, padding=1, bias=False),
# BatchNorm(256),
# nn.ReLU(),
# nn.Dropout(0.5),
# nn.Conv2d(256, 256, kernel_size=3, stride=1, padding=1, bias=False),
# BatchNorm(256),
# nn.ReLU(),
# nn.Dropout(0.1),
# nn.Conv2d(256, 2, kernel_size=1, stride=1))
self._init_weight()
if freeze_bn:
self.freeze_bn()
def __init__(self,
backbone='seresnext101',
output_stride=16,
num_classes=5,
sync_bn=True,
freeze_bn=False):
super(DeepLab, self).__init__()
if backbone == 'drn':
output_stride = 8
if sync_bn:
BatchNorm = SynchronizedBatchNorm2d
else:
BatchNorm = nn.BatchNorm2d
self.backbone = build_backbone(backbone, output_stride, BatchNorm)
#print('bacbone')
self.aspp = build_aspp(backbone, output_stride, BatchNorm)
#print('aspp')
self.decoder = build_decoder(num_classes, backbone, BatchNorm)
#print('decoder')
#self.up = DeconvBlock(304,256,BatchNorm=BatchNorm,n_iter=2)
#self.last_conv = nn.Conv2d(256, num_classes, kernel_size=1, stride=1)
if freeze_bn:
self.freeze_bn()
def __init__(self,
backbone='resnet',
output_stride=16,
num_classes=21,
sync_bn=True,
freeze_bn=False,
cp_path=None):
super(DeepLab, self).__init__()
if backbone == 'drn':
output_stride = 8
if sync_bn == True:
BatchNorm = SynchronizedBatchNorm2d
else:
BatchNorm = nn.BatchNorm2d
self.backbone = build_backbone(backbone, output_stride, BatchNorm)
self.aspp = build_aspp(backbone, output_stride, BatchNorm)
self.decoder = build_decoder(21, backbone, BatchNorm)
if (cp_path is not None):
cp = torch.load(cp_path, map_location='cpu')
self.load_state_dict(cp['state_dict'])
if freeze_bn:
self.freeze_bn()
# predict two classes (background / foreground)
self.decoder.last_conv[-1] = nn.Conv2d(256, 2, kernel_size=1, stride=1)
def __init__(self,
backbone='resnet',
output_stride=16,
num_classes=21,
sync_bn=True,
freeze_bn=False):
super(DeepLab, self).__init__()
if backbone == 'drn':
output_stride = 8
if sync_bn == True:
BatchNorm = SynchronizedBatchNorm2d
else:
BatchNorm = nn.BatchNorm2d
self.backbone = build_backbone(backbone, output_stride, BatchNorm)
self.aspp = build_aspp(backbone, output_stride, BatchNorm)
self.decoder = build_decoder(num_classes, backbone, BatchNorm)
self.sr_decoder = build_sr_decoder(num_classes, backbone, BatchNorm)
self.pointwise = torch.nn.Sequential(
torch.nn.Conv2d(num_classes, 3, 1),
torch.nn.BatchNorm2d(3), #添加了BN层
torch.nn.ReLU(inplace=True))
self.up_sr_1 = nn.ConvTranspose2d(64, 64, 2, stride=2)
self.up_edsr_1 = EDSRConv(64, 64)
self.up_sr_2 = nn.ConvTranspose2d(64, 32, 2, stride=2)
self.up_edsr_2 = EDSRConv(32, 32)
self.up_sr_3 = nn.ConvTranspose2d(32, 16, 2, stride=2)
self.up_edsr_3 = EDSRConv(16, 16)
self.up_conv_last = nn.Conv2d(16, 3, 1)
self.freeze_bn = freeze_bn
def __init__(
self,
Norm,
backbone="resnet",
output_stride=16,
num_classes=3,
freeze_bn=False,
abn=False,
):
super(DeepLabv3, self).__init__()
self.abn = abn
if backbone == "drn":
output_stride = 8
if Norm == "gn":
norm = gn
elif Norm == "bn":
norm = bn
elif Norm == "syncbn":
norm = syncbn
self.backbone = build_backbone(backbone,
output_stride,
Norm,
dec=False,
abn=abn)
self.aspp = build_aspp(backbone, output_stride, norm, dec=False)
if freeze_bn:
self.freeze_bn()
def __init__(self, backbone='resnet', output_stride=16, num_classes=21,
sync_bn=True, freeze_bn=False, pretrain=True):
super(DeepLabX, self).__init__()
self.num_classes = num_classes
if backbone == 'drn':
output_stride = 8
if sync_bn == True:
BatchNorm = SynchronizedBatchNorm2d
else:
BatchNorm = nn.BatchNorm2d
self.backbone = build_backbone(backbone, output_stride, BatchNorm)
self.aspp = build_aspp(backbone, output_stride, BatchNorm)
# self.aspp = build_psp()
# self.aspp = build_naiveGCE()
self.decoder = build_decoder(num_classes, backbone, BatchNorm)
if freeze_bn:
self.freeze_bn()
if pretrain:
self._load_pretrain()
# change the last inference layer for binary segmentation mask
last_conv = list(self.decoder.last_conv.children())
self.decoder.last_conv = nn.Sequential(*last_conv[:-1])
self.decoder.last_conv.add_module('8', nn.Conv2d(256, 2, 1, 1))
def __init__(self,
backbone='resnet_multiscale',
output_stride=16,
num_classes=21,
sync_bn=True,
freeze_bn=False):
super(DeepLabCA, self).__init__()
if backbone == 'drn':
output_stride = 8
if sync_bn == True:
BatchNorm = SynchronizedBatchNorm2d
else:
BatchNorm = nn.BatchNorm2d
self.backbone = build_backbone(backbone, output_stride, BatchNorm)
self.aspp = build_aspp(backbone, output_stride, BatchNorm)
self.decoder = build_decoder(num_classes, backbone, BatchNorm)
self.avg_pool = nn.AdaptiveAvgPool2d(32)
# self.se=RCAB(2048+1024+512+256+256,1,16)
self.ca = CAM_Module()
in_channels = 2048 + 1024 + 512 + 256 + 256
inter_channels = in_channels // 4
# self.conv5c = nn.Sequential(nn.Conv2d(in_channels, inter_channels, 3, padding=1, bias=False),
# BatchNorm(inter_channels),
# nn.ReLU())
# self.conv5 = nn.Sequential(nn.Conv2d(inter_channels, inter_channels, 3, padding=1, bias=False),
# BatchNorm(inter_channels),
# nn.ReLU())
# self.conv6 = nn.Sequential(nn.Dropout2d(0.1, False), nn.Conv2d(inter_channels, num_classes, 1))
if freeze_bn:
self.freeze_bn()
def __init__(self, args, num_classes=21):
super(DeepLab, self).__init__()
self.args = args
output_stride = args.out_stride
if args.backbone == 'drn':
output_stride = 8
if args.backbone.split('-')[0] == 'efficientnet':
output_stride = 32
if args.norm == 'gn': norm = gn
elif args.norm == 'bn': norm = bn
elif args.norm == 'syncbn': norm = syncbn
else:
print(args.norm, "normalization is not implemented")
raise NotImplementedError
self.backbone = build_backbone(args)
self.aspp = build_aspp(args.backbone, args.out_stride, norm)
self.decoder = build_decoder(num_classes, args.backbone, norm)
self.classifier = nn.Linear(300, num_classes)
if self.args.freeze_bn:
self.freeze_bn()
def __init__(self,
args,
num_classes=21,
freeze_bn=False,
abn=False,
deep_dec=True):
super(DeepLab, self).__init__()
self.args = args
self.abn = abn
self.deep_dec = deep_dec # if True, it deeplabv3+, otherwise, deeplabv3
output_stride = args.out_stride
if args.backbone == "drn":
output_stride = 8
if args.backbone.split("-")[0] == "efficientnet":
output_stride = 32
if args.norm == "gn":
norm = gn
elif args.norm == "bn":
norm = bn
elif args.norm == "syncbn":
norm = syncbn
else:
print(args.norm, "normalization is not implemented")
raise NotImplementedError
self.backbone = build_backbone(args)
self.aspp = build_aspp(args.backbone, args.out_stride, norm)
if self.deep_dec:
self.decoder = build_decoder(num_classes, args.backbone, norm)
if freeze_bn:
self.freeze_bn()
def __init__(self,
backbone='resnet',
output_stride=16,
num_classes=21,
sync_bn=True,
freeze_bn=False,
use_iou=True):
super(DeepLab, self).__init__()
self.use_iou = use_iou
if backbone == 'drn':
output_stride = 8
if sync_bn == True:
BatchNorm = SynchronizedBatchNorm2d
else:
BatchNorm = nn.BatchNorm2d
self.backbone = build_backbone(backbone, output_stride, BatchNorm)
self.aspp = build_aspp(backbone, output_stride, BatchNorm)
self.decoder = build_decoder(num_classes, backbone, BatchNorm)
if self.use_iou:
self.maskiou = build_maskiou(num_classes, BatchNorm)
if freeze_bn:
self.freeze_bn()
def __init__(self, low_level_inplanes):
super(Deeplab, self).__init__()
#aspp
self.aspp = build_aspp('', 16, BatchNorm)
#decoder
self.conv1 = nn.Conv2d(low_level_inplanes,
48,
1,
bias=False,
groups=cfg.RESNETS.NUM_GROUPS)
self.bn1 = BatchNorm(48)
self.relu = nn.ReLU()
self.last_conv = nn.Sequential(
nn.Conv2d(304,
256,
kernel_size=3,
stride=1,
padding=1,
bias=False,
groups=cfg.RESNETS.NUM_GROUPS), BatchNorm(256),
nn.ReLU(), nn.Dropout(0.5),
nn.Conv2d(256,
256,
kernel_size=3,
stride=1,
padding=1,
bias=False,
groups=cfg.RESNETS.NUM_GROUPS), BatchNorm(256),
nn.ReLU(), nn.Dropout(0.1),
nn.Conv2d(256, cfg.MODEL.NUM_CLASSES, kernel_size=1, stride=1))
self._init_weight()
def __init__(self, args, backbone='resnet', output_stride=16, num_classes=4,
sync_bn=False, freeze_bn=False, depth=50):
super(DeepLab, self).__init__()
if backbone == 'drn':
output_stride = 8
if sync_bn == True:
BatchNorm = SynchronizedBatchNorm2d
else:
BatchNorm = nn.BatchNorm2d
self.args = args
if self.args.oly_s1 and not self.args.oly_s2:
in_channel = 2
pretrn=False
elif not self.args.oly_s1 and self.args.oly_s2 and not self.args.rgb:
in_channel = 10
pretrn = False
elif not self.args.oly_s1 and self.args.oly_s2 and self.args.rgb:
in_channel = 3
pretrn = True
elif not self.args.oly_s1 and not self.args.oly_s2 and not self.args.rgb:
in_channel = 12
pretrn = False
elif not self.args.oly_s1 and not self.args.oly_s2 and self.args.rgb:
in_channel = 5
pretrn = False
else:
raise NotImplementedError
self.backbone = build_backbone(backbone, in_channel, output_stride, BatchNorm, depth, pretrn)
self.aspp = build_aspp(backbone, output_stride, BatchNorm)
self.decoder = build_decoder(num_classes, backbone, BatchNorm)
if freeze_bn:
self.freeze_bn()
def __init__(self,
backbone='resnet',
output_stride=16,
num_classes=21,
freeze_bn=False):
super(DeepLab, self).__init__()
BatchNorm = nn.BatchNorm2d
self.backbone = build_backbone(backbone, output_stride, BatchNorm)
self.aspp = build_aspp(backbone, output_stride, BatchNorm)
self.decoder = build_decoder(num_classes, backbone, BatchNorm)
self.freeze_bn = freeze_bn
def __init__(self, backbone='resnet', output_stride=16, num_classes=21,
freeze_bn=False):
super(DeepLab, self).__init__()
if backbone == 'drn':
output_stride = 8
BatchNorm = nn.BatchNorm2d
self.backbone = ResNet101(output_stride, BatchNorm)
self.aspp = build_aspp(backbone, output_stride, BatchNorm)
self.decoder = build_decoder(num_classes, backbone, BatchNorm)
if freeze_bn:
self.freeze_bn()
def __init__(self, backbone='resnet', output_stride=16, num_classes=21,
sync_bn=True, freeze_bn=False):
super(Deeplabv3, self).__init__()
if backbone == 'drn':
output_stride = 8
if sync_bn == True:
BatchNorm = SynchronizedBatchNorm2d
else:
BatchNorm = nn.BatchNorm2d
self.backbone = build_backbone(backbone, output_stride, BatchNorm)
self.aspp = build_aspp(backbone, output_stride, BatchNorm)
self.decoder = build_decoder(num_classes, backbone, BatchNorm)
self.freeze_bn = freeze_bn
def __init__(self, backbone='resnet', output_stride=16, num_classes=21,
sync_bn=True, freeze_bn=False):
super(DeepLab, self).__init__()
if backbone == 'drn':
output_stride = 8
if sync_bn == True:
BatchNorm = SynchronizedBatchNorm2d
else:
BatchNorm = nn.BatchNorm2d
self.backbone = build_backbone(backbone, output_stride, BatchNorm)
self.aspp = build_aspp(backbone, output_stride, BatchNorm)
self.decoder = build_decoder(num_classes, backbone, BatchNorm)
if freeze_bn:
self.freeze_bn()
def __init__(self, backbone='resnet', output_stride=16, num_classes=21,
sync_bn=True, freeze_bn=False):
super(DeepLab, self).__init__()
if backbone == 'drn':
output_stride = 8
if sync_bn == True:
BatchNorm = SynchronizedBatchNorm2d
else:
BatchNorm = nn.BatchNorm2d
#self.deconv1 = nn.ConvTranspose2d(21, 21, 1, 4, 0, 0, bias=True)
self.backbone = build_backbone(backbone, output_stride, BatchNorm)
self.aspp = build_aspp(backbone, output_stride, BatchNorm)
self.decoder = build_decoder(num_classes, backbone, BatchNorm)
if freeze_bn:
self.freeze_bn()
def __init__(self,
backbone='resnet',
output_stride=16,
sync_bn=True,
freeze_bn=False):
super().__init__()
if backbone == 'drn':
output_stride = 8
if sync_bn == True:
BatchNorm = SynchronizedBatchNorm2d
else:
BatchNorm = nn.BatchNorm2d
self.backbone = build_backbone(backbone, output_stride, BatchNorm)
self.aspp = build_aspp(backbone, output_stride, BatchNorm)
self.decoder = build_decoder(5, backbone, BatchNorm)
if freeze_bn:
self.freeze_bn()
def __init__(self,
backbone='resnet',
output_stride=16,
num_classes=2,
sync_bn=True,
freeze_bn=False):
super(DeepLab, self).__init__()
if backbone == 'drn':
output_stride = 8
if sync_bn == True:
BatchNorm = SynchronizedBatchNorm2d
else:
BatchNorm = nn.BatchNorm2d
self.cls1 = nn.Sequential(
nn.Conv2d(512, 512, kernel_size=3, padding=1), # fc6
)
self.cls2 = nn.Sequential(
nn.Conv2d(1024, 512, kernel_size=3, padding=1), # fc6
)
self.backbone = build_backbone(backbone, output_stride, BatchNorm)
self.aspp = build_aspp(backbone, output_stride, BatchNorm)
self.decoder = build_decoder(num_classes, backbone, BatchNorm)
self.freeze_bn = freeze_bn
self.multi_graph_conv = build_graph_conv(backbone, BatchNorm)
# self.classifier_6 = nn.Sequential(
# nn.Conv2d(128, 128, kernel_size=3, dilation=1, padding=1), # fc6
# nn.ReLU(inplace=True)
# )
# self.exit_layer = nn.Conv2d(128, 2, kernel_size=1, padding=1)
# self.cos_similarity_func = nn.CosineSimilarity()
############################################################
bins = [6, 13, 26, 52]
self.features = []
for bin in bins: # [1,2,3,6]
self.features.append(
nn.Sequential(
nn.AdaptiveAvgPool2d(bin), # 括号里面的大小是多少，输出来的大小就是多少
))
def __init__(self, concat_channels=64, gcn_dim=514):
super(DeepLabResnet, self).__init__()
self.cnn_feature_grids = [112, 56, 28, 28]
self.normalizer = transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
self.concat_channels = concat_channels
self.feat_size = 28
self.gcn_dim = gcn_dim
self.image_feature_dim = 256
# self.resnet = build_backbone("resnet", 8, SynchronizedBatchNorm2d)
self.resnet = build_backbone("resnet", 8, BatchNorm2d)
self.conv1_concat = _conv_up(64, concat_channels, upflag=False)
self.res1_concat, self.res1_concat_up = _conv_up(
256, concat_channels, 2)
self.res2_concat, self.res2_concat_up = _conv_up(
512, concat_channels, 4)
self.res4_concat, self.res4_concat_up = _conv_up(
2048, concat_channels, 4)
# self.res5_concat = _conv_up(512, concat_channels, upflag=False)
self.edge_annotation_concat = _make_edge_annotation_concat(
self.gcn_dim)
self.edge_annotation_channels = 64 * 5
self.conv_final = _make_conv_final(concat_channels)
# self.final_PSP = build_aspp("resnet", 8, SynchronizedBatchNorm2d)
self.final_PSP = build_aspp("resnet", 8, BatchNorm2d)
self.conv_final_cat = _make_conv_final_cat()
self.prob_conv = nn.Sequential(
nn.Conv2d(512, 256, kernel_size=3, padding=1, bias=False),
nn.BatchNorm2d(256), nn.LeakyReLU(),
nn.Conv2d(256, 128, kernel_size=3, padding=1, bias=False),
nn.BatchNorm2d(128), nn.LeakyReLU(),
nn.Conv2d(128, 64, kernel_size=3, padding=1, bias=False),
nn.BatchNorm2d(64), nn.LeakyReLU(),
nn.Conv2d(64, 2, kernel_size=3, padding=1, bias=False))
def __init__(self,
backbone='resnet',
output_stride=16,
num_classes=15,
sync_bn=True,
freeze_bn=False):
super(DeepLab, self).__init__()
if backbone == 'drn':
output_stride = 8
if sync_bn == True:
BatchNorm = SynchronizedBatchNorm2d
else:
BatchNorm = nn.BatchNorm2d
self.backbone = xception.AlignedXception(output_stride, BatchNorm)
self.aspp = build_aspp(backbone, output_stride, BatchNorm)
self.decoder = build_decoder(num_classes, backbone, BatchNorm)
if freeze_bn:
self.freeze_bn()
def __init__(self,
backbone='resnet',
output_stride=16,
num_classes=21,
sync_bn=True,
freeze_bn=False,
model_path=os.getcwd()):
super(DeepLab, self).__init__()
self.model_path = model_path
if sync_bn == True:
BatchNorm = SynchronizedBatchNorm2d
else:
BatchNorm = nn.BatchNorm2d
self.backbone = build_backbone(backbone, output_stride, BatchNorm,
model_path)
self.aspp = build_aspp(backbone, output_stride, BatchNorm)
self.decoder = build_decoder(num_classes, backbone, BatchNorm)
self.freeze_bn = freeze_bn
def __init__(self,
backbone='resnet',
output_stride=16,
num_classes=21,
sync_bn=True,
freeze_bn=False):
super(DeepLab, self).__init__()
if backbone == 'drn':
output_stride = 8
if sync_bn == True:
BatchNorm = SyncBatchNorm
else:
BatchNorm = nn.BatchNorm2d
self.backbone = build_backbone(backbone, output_stride, BatchNorm)
self.aspp = build_aspp(backbone, output_stride, BatchNorm)
self.encoding = build_encoding(backbone, num_classes, BatchNorm)
self.decoder = build_decoder(num_classes, backbone, BatchNorm)
if freeze_bn:
self.freeze_bn()
def __init__(self,
backbone='resnet',
output_stride=16,
num_classes=21,
sync_bn=True,
freeze_bn=False):
super(DeepLab, self).__init__()
if backbone == 'drn':
output_stride = 8
if sync_bn == True:
BatchNorm = SynchronizedBatchNorm2d
else:
BatchNorm = nn.BatchNorm2d
self.CSG = nn.Linear(CODE_SIZE, np.prod(SLICE_SHAPE), bias=False)
self.backbone = build_backbone(backbone, output_stride, BatchNorm)
self.aspp = build_aspp(backbone, output_stride, BatchNorm, self.CSG)
self.decoder = build_decoder(num_classes, backbone, BatchNorm)
if freeze_bn:
self.freeze_bn()
def __init__(self, backbone='resnet', output_stride=16, num_classes=2,
sync_bn=True, freeze_bn=False):
super(DeepLab, self).__init__()
if backbone == 'drn':
output_stride = 8
if sync_bn == True:
BatchNorm = SynchronizedBatchNorm2d
else:
BatchNorm = nn.BatchNorm2d
self.backbone = build_backbone(backbone, output_stride, BatchNorm)
self.aspp = build_aspp(backbone, output_stride, BatchNorm)
# for k in self.backbone.children():
# for param in k.parameters():
# param.requires_grad = False
# for k in self.aspp.children():
# for param in k.parameters():
# param.requires_grad = False
self.decoder = build_decoder(num_classes, backbone, BatchNorm)
self._load_pretrained_model(backbone)
if freeze_bn:
self.freeze_bn()
def __init__(self,
num_classes,
backbone='resnet',
output_stride=16,
sync_bn=True,
freeze_bn=False):
super(DeepLab, self).__init__()
self.num_classes = num_classes
if backbone == 'drn':
output_stride = 8
if sync_bn == True:
BatchNorm = SynchronizedBatchNorm2d
else:
BatchNorm = nn.BatchNorm2d
self.backbone = build_backbone(backbone, output_stride, BatchNorm)
self.aspp = build_aspp(backbone, output_stride, BatchNorm)
self.decoder_seg = build_decoder(num_classes + 1, backbone, BatchNorm)
self.decoder_box = build_decoder(6, backbone, BatchNorm)
self.pos = None
if freeze_bn:
self.freeze_bn()