def __init__(self):
super().__init__()
self.resnet50 = ResNet50(config.backbone_freeze_at, False)
self.FPN = FPN(self.resnet50, 3, 7)
self.R_Head = RetinaNet_Head()
self.R_Anchor = RetinaNet_Anchor()
self.R_Criteria = RetinaNet_Criteria()
def __init__(self):
super().__init__()
# self.resnet152 = ResNet152()
self.resnet50 = ResNet50(config.backbone_freeze_at, False)
self.FPN = FPN(self.resnet50, 2, 6)
self.RPN = RPN(config.rpn_channel)
self.RCNN = RCNN()
assert config.num_classes == 2, 'Only support two class(1fg/1bg).'
def __init__(self):
super().__init__()
self.resnet50 = ResNet50(config.backbone_freeze_at, False)
self.FPN = FPN(self.resnet50, 2, 6)
self.RPN = RPN(config.rpn_channel)
self.RCNN = RCNN()
self.Cascade_0 = Cascade('cascade_0')
self.Cascade_1 = Cascade('cascade_1')
def __init__(self):
super().__init__()
# ----------------------- build the backbone ------------------------ #
self.resnet50 = ResNet50()
# ------------ freeze the weights of resnet stage1 and stage 2 ------ #
if config.backbone_freeze_at >= 1:
for p in self.resnet50.conv1.parameters():
p.requires_grad = False
if config.backbone_freeze_at >= 2:
for p in self.resnet50.layer1.parameters():
p.requires_grad = False
# -------------------------- build the FPN -------------------------- #
self.backbone = FPN(self.resnet50)
# -------------------------- build the RPN -------------------------- #
self.RPN = RPN(config.rpn_channel)
# ----------------------- build the RCNN head ----------------------- #
self.Cascade_0 = Cascade('cascade_0')
#self.Cascade_1 = Cascade('cascade_1')
self.RCNN = RCNN()
# -------------------------- input Tensor --------------------------- #
self.inputs = {
"image":
mge.tensor(
np.random.random([2, 3, 224, 224]).astype(np.float32),
dtype="float32",
),
"im_info":
mge.tensor(
np.random.random([2, 5]).astype(np.float32),
dtype="float32",
),
"gt_boxes":
mge.tensor(
np.random.random([2, 100, 5]).astype(np.float32),
dtype="float32",
),
}