def __init__(self,
num_input_layers,
num_outputs,
loss_func,
reduce_error_func_dict,
hist_error_func_dict,
text_error_func_dict,
output_scaling,
resnet_type: str,
learning_rate,
cosine_annealing_steps,
weight_decay,
dtype=torch.float32,
track_ideal_metrics=False):
super().__init__(num_input_layers,
num_outputs,
loss_func,
reduce_error_func_dict,
hist_error_func_dict,
text_error_func_dict,
output_scaling,
resnet_type,
learning_rate,
cosine_annealing_steps,
weight_decay,
dtype=torch.float32,
track_ideal_metrics=False)
self.model = self.resnet_dict[resnet_type](pretrained=False,
num_classes=num_outputs)
# altering resnet to fit more than 3 input layers
if resnet_type.startswith('res'):
self.model.conv1 = nn.Conv2d(num_input_layers,
64,
kernel_size=7,
stride=2,
padding=3,
bias=False)
if resnet_type.startswith('mobile'):
self.model.features[0] = ConvBNReLU(self.num_input_layers,
_make_divisible(32.0, 8),
stride=2,
norm_layer=nn.BatchNorm2d)
self.plotter_val_data = None
self.plotter_train_data = None
self.type(dst_type=dtype)
def __init__(self,
num_classes=1000,
width_mult=1.0,
inverted_residual_setting=None,
round_nearest=8):
"""
MobileNet V2 main class
Args:
num_classes (int): Number of classes
width_mult (float): Width multiplier - adjusts number of channels in each layer by this amount
inverted_residual_setting: Network structure
round_nearest (int): Round the number of channels in each layer to be a multiple of this number
Set to 1 to turn off rounding
"""
super(MobileNetV2, self).__init__()
block = InvertedResidual
input_channel = 32
last_channel = 1280
if inverted_residual_setting is None:
inverted_residual_setting = [
# t, c, n, s
[1, 16, 1, 1],
[6, 24, 2, 2],
[6, 32, 3, 2],
[6, 64, 4, 2],
[6, 96, 3, 1],
[6, 160, 3, 2],
[6, 320, 1, 1],
]
# only check the first element, assuming user knows t,c,n,s are required
if not inverted_residual_setting or len(
inverted_residual_setting[0]) != 4:
raise ValueError("inverted_residual_setting should be non-empty "
"or a 4-element list, got {}".format(
inverted_residual_setting))
# building first layer
input_channel = _make_divisible(input_channel * width_mult,
round_nearest)
self.last_channel = _make_divisible(
last_channel * max(1.0, width_mult), round_nearest)
features = [ConvBNReLU(3, input_channel, stride=2)]
# building inverted residual blocks
for expand_ratio, num_channels, num_layers, block_stride in inverted_residual_setting:
output_channel = _make_divisible(num_channels * width_mult,
round_nearest)
for i in range(num_layers):
stride = block_stride if i == 0 else 1
features.append(
block(input_channel,
output_channel,
stride,
expand_ratio=expand_ratio))
input_channel = output_channel
# building last several layers
features.append(
ConvBNReLU(input_channel, self.last_channel, kernel_size=1))
# make it nn.Sequential
self.features = nn.Sequential(*features)
# building classifier
self.classifier = nn.Sequential(
nn.Dropout(0.2),
nn.Linear(self.last_channel, num_classes),
)
# weight initialization
for module in self.modules():
if isinstance(module, nn.Conv2d):
nn.init.kaiming_normal_(module.weight, mode='fan_out')
if module.bias is not None:
nn.init.zeros_(module.bias)
elif isinstance(module, nn.BatchNorm2d):
nn.init.ones_(module.weight)
nn.init.zeros_(module.bias)
elif isinstance(module, nn.Linear):
nn.init.normal_(module.weight, 0, 0.01)
nn.init.zeros_(module.bias)
def __init__(self, input_channels: int, squeeze_factor: int = 4):
super().__init__()
squeeze_channels = _make_divisible(input_channels // squeeze_factor, 8)
self.fc1 = nn.Conv2d(input_channels, squeeze_channels, 1)
self.fc2 = nn.Conv2d(squeeze_channels, input_channels, 1)
def adjust_channels(channels: int, width_mult: float):
return _make_divisible(channels * width_mult, 8)
def __init__(self,
num_classes=1000,
width_mult=1.0,
version="v1",
round_nearest=8):
"""
MobileNet V2 main class
Args:
num_classes (int): Number of classes
width_mult (float): Width multiplier - adjusts number of channels in each layer by this amount
round_nearest (int): Round the number of channels in each layer to be a multiple of this number
Set to 1 to turn off rounding
"""
super(MobileNet, self).__init__()
input_channel = 32
if version == "v2":
settings = [
# t, c, n, s
[1, 16, 1, 1],
[6, 24, 2, 2],
[6, 32, 3, 2],
[6, 64, 4, 2],
[6, 96, 3, 1],
[6, 160, 3, 2],
[6, 320, 1, 1],
]
last_channel = 1280
layer = mobilenet.InvertedResidual
elif version == "v1":
settings = [
# t, c, n, s
[1, 64, 1, 1],
[1, 128, 2, 2],
[1, 256, 2, 2],
[1, 512, 6, 2],
[1, 1024, 2, 2],
]
last_channel = 1024
layer = SepConvBNReLU
self.settings = settings
self.version = version
# building first layer
input_channel = mobilenet._make_divisible(input_channel * width_mult,
round_nearest)
self.last_channel = mobilenet._make_divisible(
last_channel * max(1.0, width_mult), round_nearest)
self.conv1 = mobilenet.ConvBNReLU(3, input_channel, stride=2)
# building inverted residual blocks
for j, (t, c, n, s) in enumerate(settings):
output_channel = mobilenet._make_divisible(c * width_mult,
round_nearest)
layers = []
for i in range(n):
stride = s if i == 0 else 1
layers.append(
layer(input_channel,
output_channel,
stride=stride,
expand_ratio=t))
input_channel = output_channel
self.add_module("layer{}".format(j + 1), nn.Sequential(*layers))
# building last several layers
if self.version == "v2":
self.head_conv = mobilenet.ConvBNReLU(input_channel,
self.last_channel,
kernel_size=1)
# building classifier
self.classifier = nn.Sequential(
nn.Dropout(0.2),
nn.Linear(self.last_channel, num_classes),
)
# weight initialization
for m in self.modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight, mode="fan_out")
if m.bias is not None:
nn.init.zeros_(m.bias)
elif isinstance(m, nn.BatchNorm2d):
nn.init.ones_(m.weight)
nn.init.zeros_(m.bias)
elif isinstance(m, nn.Linear):
nn.init.normal_(m.weight, 0, 0.01)
nn.init.zeros_(m.bias)
