def __init__(
self,
in_channels=1,
out_channels=32,
input_dim=312,
hidden_dim=32,
output_dim=10,
):
super(cnn1d_ser, self).__init__()
self.classifier = nn.Sequential(
nn.Conv1d(in_channels, out_channels, 5, stride=1, padding=2),
nn.BatchNorm1d(out_channels),
nn.ReLU(),
nn.Dropout(0.5),
nn.Conv1d(out_channels, out_channels, 5, stride=1, padding=2),
nn.BatchNorm1d(out_channels),
nn.ReLU(),
nn.Dropout(0.5),
nn.Flatten(),
nn.Linear(input_dim * out_channels, hidden_dim),
nn.BatchNorm1d(hidden_dim),
nn.ReLU(),
nn.Dropout(0.5),
nn.Linear(hidden_dim, output_dim),
)
def __init__(self, num_speakers=2) -> None:
super(simple_CNN, self).__init__()
self.convs = nn.Sequential(
nn.Conv1d(1, 16, 100, stride=10),
nn.BatchNorm1d(16),
nn.ReLU(),
nn.Conv1d(16, 64, 21, stride=10),
nn.BatchNorm1d(64),
nn.ReLU(),
nn.Conv1d(64, 64, 5, stride=5),
nn.BatchNorm1d(64),
nn.ReLU(),
)
self.linears = nn.Sequential(nn.Linear(1 * 6 * 64, 128),
nn.Linear(128, num_speakers))
def build_norm(norm, dim):
"""
Build normalize layer
LN cost more memory than BN
"""
if norm not in ["cLN", "gLN", "BN"]:
raise RuntimeError("Unsupported normalize layer: {}".format(norm))
if norm == "cLN":
return ChannelWiseLayerNorm(dim, elementwise_affine=True)
elif norm == "BN":
return nn.BatchNorm1d(dim)
else:
return GlobalChannelLayerNorm(dim, elementwise_affine=True)
def get_bn_graph():
model = nn.BatchNorm1d(6)
model.eval()
model.to_global(flow.env.all_device_placement("cpu"), flow.sbp.broadcast)
class Testgraph(flow.nn.Graph):
def __init__(self, model):
super(Testgraph, self).__init__()
self.module = model
def build(self, x):
return self.module(x)
test_graph = Testgraph(model)
return test_graph
def __init__(self, num_classes):
super(ResReid, self).__init__()
self.in_planes = 2048
self.model = ResNet(block=Bottleneck,
layers=[3, 4, 6, 3],
last_stride=1)
self.gap = nn.AdaptiveAvgPool2d(1)
self.bnneck = nn.BatchNorm1d(self.in_planes)
self.bnneck.bias.requires_grad_(False) # no shift
self.num_classes = num_classes
nn.init.normal_(self.bnneck.weight, 1, 0.02)
nn.init.constant_(self.bnneck.bias, 0)
self.classifier = nn.Linear(self.in_planes,
self.num_classes,
bias=False)
nn.init.normal_(self.classifier.weight, 0, 0.01)
def __init__(self,
in_planes,
out_planes,
kernel_size=3,
stride=1,
groups=1):
padding = (kernel_size - 1) // 2
super(ConvBNReLU, self).__init__(
nn.Conv1d(
in_planes,
out_planes,
kernel_size,
stride,
padding,
groups=groups,
bias=False,
),
nn.BatchNorm1d(out_planes),
nn.ReLU6(),
)
def __init__(self, channels, kernel_size, bias=True, dropout=0.0):
super(ConformerConvolutionModule, self).__init__()
assert kernel_size % 2 == 1
self.pointwise_conv1 = nn.Linear(channels, 2 * channels, bias=bias)
self.depthwise_conv = nn.Conv1d(
channels,
channels,
kernel_size,
stride=1,
padding=(kernel_size - 1) // 2,
groups=channels,
bias=bias,
)
self.batch_norm = nn.BatchNorm1d(channels)
self.pointwise_conv2 = nn.Linear(channels, channels, bias=bias)
self.dropout = nn.Dropout(dropout)
def __init__(self, inp, oup, stride, expand_ratio):
super(InvertedResidual, self).__init__()
self.stride = stride
assert stride in [1, 2]
hidden_dim = int(round(inp * expand_ratio))
self.use_res_connect = self.stride == 1 and inp == oup
layers = []
if expand_ratio != 1:
# pw
layers.append(ConvBNReLU(inp, hidden_dim, kernel_size=1))
layers.extend([
# dw
ConvBNReLU(hidden_dim,
hidden_dim,
stride=stride,
groups=hidden_dim),
# pw-linear
nn.Conv1d(hidden_dim, oup, 1, 1, 0, bias=False),
nn.BatchNorm1d(oup),
])
self.conv = nn.Sequential(*layers)
def __init__(self, options):
super(SincNet, self).__init__()
self.cnn_N_filt = options["cnn_N_filt"]
self.cnn_len_filt = options["cnn_len_filt"]
self.cnn_max_pool_len = options["cnn_max_pool_len"]
self.cnn_act = options["cnn_act"]
self.cnn_drop = options["cnn_drop"]
self.cnn_use_laynorm = options["cnn_use_laynorm"]
self.cnn_use_batchnorm = options["cnn_use_batchnorm"]
self.cnn_use_laynorm_inp = options["cnn_use_laynorm_inp"]
self.cnn_use_batchnorm_inp = options["cnn_use_batchnorm_inp"]
self.input_dim = int(options["input_dim"])
self.fs = options["fs"]
self.N_cnn_lay = len(options["cnn_N_filt"])
self.conv = nn.ModuleList([])
self.bn = nn.ModuleList([])
self.ln = nn.ModuleList([])
self.act = nn.ModuleList([])
self.drop = nn.ModuleList([])
if self.cnn_use_laynorm_inp:
self.ln0 = LayerNorm(self.input_dim)
if self.cnn_use_batchnorm_inp:
self.bn0 = nn.BatchNorm1d([self.input_dim], momentum=0.05)
current_input = self.input_dim
for i in range(self.N_cnn_lay):
N_filt = int(self.cnn_N_filt[i])
len_filt = int(self.cnn_len_filt[i])
# dropout
self.drop.append(nn.Dropout(p=self.cnn_drop[i]))
# activation
self.act.append(act_fun(self.cnn_act[i]))
# layer norm initialization
self.ln.append(
LayerNorm((
N_filt,
int((current_input - self.cnn_len_filt[i] + 1) /
self.cnn_max_pool_len[i]),
)))
self.bn.append(
nn.BatchNorm1d(
N_filt,
int((current_input - self.cnn_len_filt[i] + 1) /
self.cnn_max_pool_len[i]),
momentum=0.05,
))
if i == 0:
self.conv.append(
SincConv_fast(self.cnn_N_filt[0], self.cnn_len_filt[0],
self.fs))
else:
self.conv.append(
nn.Conv1d(self.cnn_N_filt[i - 1], self.cnn_N_filt[i],
self.cnn_len_filt[i]))
current_input = int((current_input - self.cnn_len_filt[i] + 1) /
self.cnn_max_pool_len[i])
self.out_dim = current_input * N_filt
def __init__(self, options):
super(MLP, self).__init__()
self.input_dim = int(options["input_dim"])
self.fc_lay = options["fc_lay"]
self.fc_drop = options["fc_drop"]
self.fc_use_batchnorm = options["fc_use_batchnorm"]
self.fc_use_laynorm = options["fc_use_laynorm"]
self.fc_use_laynorm_inp = options["fc_use_laynorm_inp"]
self.fc_use_batchnorm_inp = options["fc_use_batchnorm_inp"]
self.fc_act = options["fc_act"]
self.wx = nn.ModuleList([])
self.bn = nn.ModuleList([])
self.ln = nn.ModuleList([])
self.act = nn.ModuleList([])
self.drop = nn.ModuleList([])
# input layer normalization
if self.fc_use_laynorm_inp:
self.ln0 = LayerNorm(self.input_dim)
# input batch normalization
if self.fc_use_batchnorm_inp:
self.bn0 = nn.BatchNorm1d([self.input_dim], momentum=0.05)
self.N_fc_lay = len(self.fc_lay)
current_input = self.input_dim
# Initialization of hidden layers
for i in range(self.N_fc_lay):
# dropout
self.drop.append(nn.Dropout(p=self.fc_drop[i]))
# activation
self.act.append(act_fun(self.fc_act[i]))
add_bias = True
# layer norm initialization
self.ln.append(LayerNorm(self.fc_lay[i]))
self.bn.append(nn.BatchNorm1d(self.fc_lay[i], momentum=0.05))
if self.fc_use_laynorm[i] or self.fc_use_batchnorm[i]:
add_bias = False
# Linear operations
self.wx.append(nn.Linear(current_input, self.fc_lay[i], bias=add_bias))
# weight initialization
self.wx[i].weight = nn.Parameter(
flow.Tensor(self.fc_lay[i], current_input).uniform_(
-np.sqrt(0.01 / (current_input + self.fc_lay[i])),
np.sqrt(0.01 / (current_input + self.fc_lay[i])),
)
)
self.wx[i].bias = nn.Parameter(flow.zeros(self.fc_lay[i]))
current_input = self.fc_lay[i]
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 (len(inverted_residual_setting) == 0
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(1, input_channel, stride=2)]
# building inverted residual blocks
for t, c, n, s in 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(
block(input_channel,
output_channel,
stride,
expand_ratio=t))
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[0]),
nn.LogSoftmax(dim=1),
)
self.normalize = nn.BatchNorm1d(6420)
self.maxpool1d = nn.MaxPool1d(3, stride=2)
# weight initialization
for m in self.modules():
if isinstance(m, nn.Conv1d):
nn.init.kaiming_normal_(m.weight, mode="fan_out")
if m.bias is not None:
nn.init.zeros_(m.bias)
elif isinstance(m, nn.BatchNorm1d):
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)
def __init__(
self,
block,
layers,
dropout=0,
num_features=512,
zero_init_residual=False,
groups=1,
width_per_group=64,
replace_stride_with_dilation=None,
fp16=False,
):
super(IResNet, self).__init__()
self.fp16 = fp16
self.inplanes = 64
self.dilation = 1
if replace_stride_with_dilation is None:
replace_stride_with_dilation = [False, False, False]
if len(replace_stride_with_dilation) != 3:
raise ValueError("replace_stride_with_dilation should be None "
"or a 3-element tuple, got {}".format(
replace_stride_with_dilation))
self.groups = groups
self.base_width = width_per_group
self.conv1 = nn.Conv2d(3,
self.inplanes,
kernel_size=3,
stride=1,
padding=1,
bias=False)
self.bn1 = nn.BatchNorm2d(self.inplanes, eps=1e-05)
self.prelu = nn.ReLU(self.inplanes)
self.layer1 = self._make_layer(block, 64, layers[0], stride=2)
self.layer2 = self._make_layer(block,
128,
layers[1],
stride=2,
dilate=replace_stride_with_dilation[0])
self.layer3 = self._make_layer(block,
256,
layers[2],
stride=2,
dilate=replace_stride_with_dilation[1])
self.layer4 = self._make_layer(block,
512,
layers[3],
stride=2,
dilate=replace_stride_with_dilation[2])
self.bn2 = nn.BatchNorm2d(
512 * block.expansion,
eps=1e-05,
)
self.dropout = nn.Dropout(p=dropout, inplace=True)
self.fc = nn.Linear(512 * block.expansion * self.fc_scale,
num_features)
self.features = nn.BatchNorm1d(num_features, eps=1e-05)
nn.init.constant_(self.features.weight, 1.0)
self.features.weight.requires_grad = False
for m in self.modules():
if isinstance(m, nn.Conv2d):
nn.init.normal_(m.weight, 0, 0.1)
elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
nn.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)
if zero_init_residual:
for m in self.modules():
if isinstance(m, IBasicBlock):
nn.init.constant_(m.bn2.weight, 0)