|
|
import torch |
|
|
import torch.nn as nn |
|
|
from models.resnet import CausalResnet1D |
|
|
|
|
|
|
|
|
class CausalConv1d(nn.Module): |
|
|
def __init__(self, in_channels, out_channels, kernel_size, stride=1, dilation=1): |
|
|
super(CausalConv1d, self).__init__() |
|
|
self.pad = (kernel_size - 1) * dilation + (1 - stride) |
|
|
self.conv = nn.Conv1d( |
|
|
in_channels, |
|
|
out_channels, |
|
|
kernel_size, |
|
|
stride=stride, |
|
|
padding=0, |
|
|
dilation=dilation |
|
|
) |
|
|
|
|
|
def forward(self, x): |
|
|
x = nn.functional.pad(x, (self.pad, 0)) |
|
|
return self.conv(x) |
|
|
|
|
|
|
|
|
class CausalEncoder(nn.Module): |
|
|
def __init__(self, |
|
|
input_emb_width = 272, |
|
|
hidden_size = 1024, |
|
|
down_t = 2, |
|
|
stride_t = 2, |
|
|
width = 1024, |
|
|
depth = 3, |
|
|
dilation_growth_rate = 3, |
|
|
activation='relu', |
|
|
norm=None, |
|
|
latent_dim=16, |
|
|
clip_range = [] |
|
|
): |
|
|
super().__init__() |
|
|
self.clip_range = clip_range |
|
|
self.proj = nn.Linear(width, latent_dim*2) |
|
|
|
|
|
blocks = [] |
|
|
filter_t, pad_t = stride_t * 2, stride_t // 2 |
|
|
|
|
|
|
|
|
blocks.append(CausalConv1d(input_emb_width, width, 3, 1, 1)) |
|
|
blocks.append(nn.ReLU()) |
|
|
|
|
|
for i in range(down_t): |
|
|
input_dim = width |
|
|
block = nn.Sequential( |
|
|
CausalConv1d(input_dim, width, filter_t, stride_t, 1), |
|
|
CausalResnet1D(width, depth, dilation_growth_rate, activation=activation, norm=norm), |
|
|
) |
|
|
blocks.append(block) |
|
|
blocks.append(CausalConv1d(width, hidden_size, 3, 1, 1)) |
|
|
self.model = nn.Sequential(*blocks) |
|
|
|
|
|
def reparameterize(self, mu, logvar): |
|
|
std = torch.exp(0.5 * logvar) |
|
|
eps = torch.randn_like(std) |
|
|
return mu + eps * std |
|
|
|
|
|
def forward(self, x): |
|
|
x = self.model(x) |
|
|
x = x.transpose(1, 2) |
|
|
x = self.proj(x) |
|
|
mu, logvar = x.chunk(2, dim=2) |
|
|
logvar = torch.clamp(logvar, self.clip_range[0], self.clip_range[1]) |
|
|
z = self.reparameterize(mu, logvar) |
|
|
|
|
|
return z, mu, logvar |
|
|
|
|
|
class CausalDecoder(nn.Module): |
|
|
def __init__(self, |
|
|
input_emb_width = 272, |
|
|
hidden_size = 1024, |
|
|
down_t = 2, |
|
|
stride_t = 2, |
|
|
width = 1024, |
|
|
depth = 3, |
|
|
dilation_growth_rate = 3, |
|
|
activation='relu', |
|
|
norm=None |
|
|
): |
|
|
super().__init__() |
|
|
blocks = [] |
|
|
|
|
|
filter_t, pad_t = stride_t * 2, stride_t // 2 |
|
|
blocks.append(CausalConv1d(hidden_size, width, 3, 1, 1)) |
|
|
blocks.append(nn.ReLU()) |
|
|
for i in range(down_t): |
|
|
out_dim = width |
|
|
block = nn.Sequential( |
|
|
CausalResnet1D(width, depth, dilation_growth_rate, reverse_dilation=True, activation=activation, norm=norm), |
|
|
nn.Upsample(scale_factor=2, mode='nearest'), |
|
|
CausalConv1d(width, out_dim, 3, 1, 1) |
|
|
) |
|
|
blocks.append(block) |
|
|
blocks.append(CausalConv1d(width, width, 3, 1, 1)) |
|
|
blocks.append(nn.ReLU()) |
|
|
blocks.append(CausalConv1d(width, input_emb_width, 3, 1, 1)) |
|
|
|
|
|
self.model = nn.Sequential(*blocks) |
|
|
|
|
|
def forward(self, z): |
|
|
z = z.transpose(1, 2) |
|
|
return self.model(z) |
|
|
|