def __init__(self, load_path=None):
"""Construct MobileNetV3Tiny class.
:param load_path: path for saved model
"""
super(MobileNetV3Tiny, self).__init__()
input_channel = 9
features = [
conv_bn_relu6(inchannel=3,
outchannel=input_channel,
kernel=3,
stride=2)
]
for _, lst in enumerate(self.inverted_residual_setting):
output_channel = lst[1]
features.append(
InvertedResidual(inp=input_channel,
oup=output_channel,
stride=lst[2],
expand_ratio=lst[0]))
input_channel = output_channel
self.block = OutlistSequential(*features, out_list=[3, 6, 13, 17])
if load_path is not None and is_torch_backend():
import torch
self.load_state_dict(torch.load(load_path), strict=False)
def __init__(self, load_path=None, width_mult=1.0, round_nearest=8):
"""Construct MobileNetV3Tiny class.
:param load_path: path for saved model
"""
super(MobileNetV2Tiny, self).__init__()
input_channel = 32
input_channel = _make_divisible(input_channel * width_mult,
round_nearest)
features = [conv_bn_relu6(3, input_channel, 3, 2)]
for t, c, n, s in self.inverted_residual_setting:
output_channel = _make_divisible(c * width_mult, round_nearest)
for i in range(n):
stride = s if i == 0 else 1
features.append(
InvertedResidual(inp=input_channel,
oup=output_channel,
stride=stride,
expand_ratio=t))
input_channel = output_channel
self.block = OutlistSequential(*features[:18], out_list=[3, 6, 13, 17])