def __init__(self, num_classes):
super(UNet, self).__init__()
self.backbone = mobilenet_v2(pretrained=True)
self.up_convs = nn.ModuleList([
ConvNormAct(1280, 256),
ConvNormAct(352, 128),
ConvNormAct(160, 64)
])
self.cls_conv = nn.Conv2d(88, num_classes, 3, padding=1)
initialize_weights(self.up_convs)
initialize_weights(self.cls_conv)
def __init__(self, num_classes):
super(UNet, self).__init__()
self.stages = resnet34(
pretrained=True, replace_stride_with_dilation=[False, False,
True]).stages
self.up_convs = nn.ModuleList([
ConvNormAct(512, 128),
ConvNormAct(256, 64),
ConvNormAct(128, 64)
])
self.cls_conv = nn.Conv2d(128, num_classes, 3, padding=1)
initialize_weights(self.up_convs)
initialize_weights(self.cls_conv)
def __init__(
self,
num_classes,
img_size=(416, 416),
anchors=[
[[116, 90], [156, 198], [373, 326]],
[[30, 61], [62, 45], [59, 119]],
[[10, 13], [16, 30], [33, 23]],
]):
super(YOLOV3, self).__init__()
self.backbone = mobilenet_v2(pretrained=True)
depth = 5
width = [512, 256, 128]
planes_list = [1280, 96, 32]
self.spp = nn.Sequential(ConvNormAct(1280, 320, 1, activate=nn.ReLU(True)), SPP())
self.fpn = FPN(planes_list, width, depth)
self.head = nn.ModuleList([])
self.yolo_layers = nn.ModuleList([])
for i in range(3):
self.head.append(
nn.Sequential(
SeparableConvNormAct(width[i],
width[i],
activate=nn.ReLU(True)),
nn.Conv2d(width[i],
len(anchors[i]) * (6 + num_classes), 1),
))
self.yolo_layers.append(
YOLOLayer(
anchors=np.float32(anchors[i]),
nc=num_classes,
img_size=img_size,
yolo_index=i,
))
initialize_weights(self.fpn)
initialize_weights(self.head)
def __init__(self, num_classes):
super(DeepLabV3Plus, self).__init__()
self.backbone = resnet50(
pretrained=True, replace_stride_with_dilation=[False, False, True])
self.project = ConvNormAct(256, 128, 1)
self.aspp = ASPP(2048, 256, [6, 12, 18])
self.cls_conv = nn.Conv2d(384, num_classes, 3, padding=1)
# init weight and bias
initialize_weights(self.aspp)
initialize_weights(self.project)
initialize_weights(self.cls_conv)